builtin-lock.c 23 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029
  1. // SPDX-License-Identifier: GPL-2.0
  2. #include <errno.h>
  3. #include <inttypes.h>
  4. #include "builtin.h"
  5. #include "perf.h"
  6. #include "util/evlist.h" // for struct evsel_str_handler
  7. #include "util/evsel.h"
  8. #include "util/symbol.h"
  9. #include "util/thread.h"
  10. #include "util/header.h"
  11. #include <subcmd/pager.h>
  12. #include <subcmd/parse-options.h>
  13. #include "util/trace-event.h"
  14. #include "util/debug.h"
  15. #include "util/session.h"
  16. #include "util/tool.h"
  17. #include "util/data.h"
  18. #include <sys/types.h>
  19. #include <sys/prctl.h>
  20. #include <semaphore.h>
  21. #include <pthread.h>
  22. #include <math.h>
  23. #include <limits.h>
  24. #include <linux/list.h>
  25. #include <linux/hash.h>
  26. #include <linux/kernel.h>
  27. #include <linux/zalloc.h>
  28. #include <linux/err.h>
  29. static struct perf_session *session;
  30. /* based on kernel/lockdep.c */
  31. #define LOCKHASH_BITS 12
  32. #define LOCKHASH_SIZE (1UL << LOCKHASH_BITS)
  33. static struct list_head lockhash_table[LOCKHASH_SIZE];
  34. #define __lockhashfn(key) hash_long((unsigned long)key, LOCKHASH_BITS)
  35. #define lockhashentry(key) (lockhash_table + __lockhashfn((key)))
  36. struct lock_stat {
  37. struct list_head hash_entry;
  38. struct rb_node rb; /* used for sorting */
  39. /*
  40. * FIXME: evsel__intval() returns u64,
  41. * so address of lockdep_map should be dealed as 64bit.
  42. * Is there more better solution?
  43. */
  44. void *addr; /* address of lockdep_map, used as ID */
  45. char *name; /* for strcpy(), we cannot use const */
  46. unsigned int nr_acquire;
  47. unsigned int nr_acquired;
  48. unsigned int nr_contended;
  49. unsigned int nr_release;
  50. unsigned int nr_readlock;
  51. unsigned int nr_trylock;
  52. /* these times are in nano sec. */
  53. u64 avg_wait_time;
  54. u64 wait_time_total;
  55. u64 wait_time_min;
  56. u64 wait_time_max;
  57. int discard; /* flag of blacklist */
  58. };
  59. /*
  60. * States of lock_seq_stat
  61. *
  62. * UNINITIALIZED is required for detecting first event of acquire.
  63. * As the nature of lock events, there is no guarantee
  64. * that the first event for the locks are acquire,
  65. * it can be acquired, contended or release.
  66. */
  67. #define SEQ_STATE_UNINITIALIZED 0 /* initial state */
  68. #define SEQ_STATE_RELEASED 1
  69. #define SEQ_STATE_ACQUIRING 2
  70. #define SEQ_STATE_ACQUIRED 3
  71. #define SEQ_STATE_READ_ACQUIRED 4
  72. #define SEQ_STATE_CONTENDED 5
  73. /*
  74. * MAX_LOCK_DEPTH
  75. * Imported from include/linux/sched.h.
  76. * Should this be synchronized?
  77. */
  78. #define MAX_LOCK_DEPTH 48
  79. /*
  80. * struct lock_seq_stat:
  81. * Place to put on state of one lock sequence
  82. * 1) acquire -> acquired -> release
  83. * 2) acquire -> contended -> acquired -> release
  84. * 3) acquire (with read or try) -> release
  85. * 4) Are there other patterns?
  86. */
  87. struct lock_seq_stat {
  88. struct list_head list;
  89. int state;
  90. u64 prev_event_time;
  91. void *addr;
  92. int read_count;
  93. };
  94. struct thread_stat {
  95. struct rb_node rb;
  96. u32 tid;
  97. struct list_head seq_list;
  98. };
  99. static struct rb_root thread_stats;
  100. static struct thread_stat *thread_stat_find(u32 tid)
  101. {
  102. struct rb_node *node;
  103. struct thread_stat *st;
  104. node = thread_stats.rb_node;
  105. while (node) {
  106. st = container_of(node, struct thread_stat, rb);
  107. if (st->tid == tid)
  108. return st;
  109. else if (tid < st->tid)
  110. node = node->rb_left;
  111. else
  112. node = node->rb_right;
  113. }
  114. return NULL;
  115. }
  116. static void thread_stat_insert(struct thread_stat *new)
  117. {
  118. struct rb_node **rb = &thread_stats.rb_node;
  119. struct rb_node *parent = NULL;
  120. struct thread_stat *p;
  121. while (*rb) {
  122. p = container_of(*rb, struct thread_stat, rb);
  123. parent = *rb;
  124. if (new->tid < p->tid)
  125. rb = &(*rb)->rb_left;
  126. else if (new->tid > p->tid)
  127. rb = &(*rb)->rb_right;
  128. else
  129. BUG_ON("inserting invalid thread_stat\n");
  130. }
  131. rb_link_node(&new->rb, parent, rb);
  132. rb_insert_color(&new->rb, &thread_stats);
  133. }
  134. static struct thread_stat *thread_stat_findnew_after_first(u32 tid)
  135. {
  136. struct thread_stat *st;
  137. st = thread_stat_find(tid);
  138. if (st)
  139. return st;
  140. st = zalloc(sizeof(struct thread_stat));
  141. if (!st) {
  142. pr_err("memory allocation failed\n");
  143. return NULL;
  144. }
  145. st->tid = tid;
  146. INIT_LIST_HEAD(&st->seq_list);
  147. thread_stat_insert(st);
  148. return st;
  149. }
  150. static struct thread_stat *thread_stat_findnew_first(u32 tid);
  151. static struct thread_stat *(*thread_stat_findnew)(u32 tid) =
  152. thread_stat_findnew_first;
  153. static struct thread_stat *thread_stat_findnew_first(u32 tid)
  154. {
  155. struct thread_stat *st;
  156. st = zalloc(sizeof(struct thread_stat));
  157. if (!st) {
  158. pr_err("memory allocation failed\n");
  159. return NULL;
  160. }
  161. st->tid = tid;
  162. INIT_LIST_HEAD(&st->seq_list);
  163. rb_link_node(&st->rb, NULL, &thread_stats.rb_node);
  164. rb_insert_color(&st->rb, &thread_stats);
  165. thread_stat_findnew = thread_stat_findnew_after_first;
  166. return st;
  167. }
  168. /* build simple key function one is bigger than two */
  169. #define SINGLE_KEY(member) \
  170. static int lock_stat_key_ ## member(struct lock_stat *one, \
  171. struct lock_stat *two) \
  172. { \
  173. return one->member > two->member; \
  174. }
  175. SINGLE_KEY(nr_acquired)
  176. SINGLE_KEY(nr_contended)
  177. SINGLE_KEY(avg_wait_time)
  178. SINGLE_KEY(wait_time_total)
  179. SINGLE_KEY(wait_time_max)
  180. static int lock_stat_key_wait_time_min(struct lock_stat *one,
  181. struct lock_stat *two)
  182. {
  183. u64 s1 = one->wait_time_min;
  184. u64 s2 = two->wait_time_min;
  185. if (s1 == ULLONG_MAX)
  186. s1 = 0;
  187. if (s2 == ULLONG_MAX)
  188. s2 = 0;
  189. return s1 > s2;
  190. }
  191. struct lock_key {
  192. /*
  193. * name: the value for specify by user
  194. * this should be simpler than raw name of member
  195. * e.g. nr_acquired -> acquired, wait_time_total -> wait_total
  196. */
  197. const char *name;
  198. int (*key)(struct lock_stat*, struct lock_stat*);
  199. };
  200. static const char *sort_key = "acquired";
  201. static int (*compare)(struct lock_stat *, struct lock_stat *);
  202. static struct rb_root result; /* place to store sorted data */
  203. #define DEF_KEY_LOCK(name, fn_suffix) \
  204. { #name, lock_stat_key_ ## fn_suffix }
  205. struct lock_key keys[] = {
  206. DEF_KEY_LOCK(acquired, nr_acquired),
  207. DEF_KEY_LOCK(contended, nr_contended),
  208. DEF_KEY_LOCK(avg_wait, avg_wait_time),
  209. DEF_KEY_LOCK(wait_total, wait_time_total),
  210. DEF_KEY_LOCK(wait_min, wait_time_min),
  211. DEF_KEY_LOCK(wait_max, wait_time_max),
  212. /* extra comparisons much complicated should be here */
  213. { NULL, NULL }
  214. };
  215. static int select_key(void)
  216. {
  217. int i;
  218. for (i = 0; keys[i].name; i++) {
  219. if (!strcmp(keys[i].name, sort_key)) {
  220. compare = keys[i].key;
  221. return 0;
  222. }
  223. }
  224. pr_err("Unknown compare key: %s\n", sort_key);
  225. return -1;
  226. }
  227. static void insert_to_result(struct lock_stat *st,
  228. int (*bigger)(struct lock_stat *, struct lock_stat *))
  229. {
  230. struct rb_node **rb = &result.rb_node;
  231. struct rb_node *parent = NULL;
  232. struct lock_stat *p;
  233. while (*rb) {
  234. p = container_of(*rb, struct lock_stat, rb);
  235. parent = *rb;
  236. if (bigger(st, p))
  237. rb = &(*rb)->rb_left;
  238. else
  239. rb = &(*rb)->rb_right;
  240. }
  241. rb_link_node(&st->rb, parent, rb);
  242. rb_insert_color(&st->rb, &result);
  243. }
  244. /* returns left most element of result, and erase it */
  245. static struct lock_stat *pop_from_result(void)
  246. {
  247. struct rb_node *node = result.rb_node;
  248. if (!node)
  249. return NULL;
  250. while (node->rb_left)
  251. node = node->rb_left;
  252. rb_erase(node, &result);
  253. return container_of(node, struct lock_stat, rb);
  254. }
  255. static struct lock_stat *lock_stat_findnew(void *addr, const char *name)
  256. {
  257. struct list_head *entry = lockhashentry(addr);
  258. struct lock_stat *ret, *new;
  259. list_for_each_entry(ret, entry, hash_entry) {
  260. if (ret->addr == addr)
  261. return ret;
  262. }
  263. new = zalloc(sizeof(struct lock_stat));
  264. if (!new)
  265. goto alloc_failed;
  266. new->addr = addr;
  267. new->name = zalloc(sizeof(char) * strlen(name) + 1);
  268. if (!new->name) {
  269. free(new);
  270. goto alloc_failed;
  271. }
  272. strcpy(new->name, name);
  273. new->wait_time_min = ULLONG_MAX;
  274. list_add(&new->hash_entry, entry);
  275. return new;
  276. alloc_failed:
  277. pr_err("memory allocation failed\n");
  278. return NULL;
  279. }
  280. struct trace_lock_handler {
  281. int (*acquire_event)(struct evsel *evsel,
  282. struct perf_sample *sample);
  283. int (*acquired_event)(struct evsel *evsel,
  284. struct perf_sample *sample);
  285. int (*contended_event)(struct evsel *evsel,
  286. struct perf_sample *sample);
  287. int (*release_event)(struct evsel *evsel,
  288. struct perf_sample *sample);
  289. };
  290. static struct lock_seq_stat *get_seq(struct thread_stat *ts, void *addr)
  291. {
  292. struct lock_seq_stat *seq;
  293. list_for_each_entry(seq, &ts->seq_list, list) {
  294. if (seq->addr == addr)
  295. return seq;
  296. }
  297. seq = zalloc(sizeof(struct lock_seq_stat));
  298. if (!seq) {
  299. pr_err("memory allocation failed\n");
  300. return NULL;
  301. }
  302. seq->state = SEQ_STATE_UNINITIALIZED;
  303. seq->addr = addr;
  304. list_add(&seq->list, &ts->seq_list);
  305. return seq;
  306. }
  307. enum broken_state {
  308. BROKEN_ACQUIRE,
  309. BROKEN_ACQUIRED,
  310. BROKEN_CONTENDED,
  311. BROKEN_RELEASE,
  312. BROKEN_MAX,
  313. };
  314. static int bad_hist[BROKEN_MAX];
  315. enum acquire_flags {
  316. TRY_LOCK = 1,
  317. READ_LOCK = 2,
  318. };
  319. static int report_lock_acquire_event(struct evsel *evsel,
  320. struct perf_sample *sample)
  321. {
  322. void *addr;
  323. struct lock_stat *ls;
  324. struct thread_stat *ts;
  325. struct lock_seq_stat *seq;
  326. const char *name = evsel__strval(evsel, sample, "name");
  327. u64 tmp = evsel__intval(evsel, sample, "lockdep_addr");
  328. int flag = evsel__intval(evsel, sample, "flags");
  329. memcpy(&addr, &tmp, sizeof(void *));
  330. ls = lock_stat_findnew(addr, name);
  331. if (!ls)
  332. return -ENOMEM;
  333. if (ls->discard)
  334. return 0;
  335. ts = thread_stat_findnew(sample->tid);
  336. if (!ts)
  337. return -ENOMEM;
  338. seq = get_seq(ts, addr);
  339. if (!seq)
  340. return -ENOMEM;
  341. switch (seq->state) {
  342. case SEQ_STATE_UNINITIALIZED:
  343. case SEQ_STATE_RELEASED:
  344. if (!flag) {
  345. seq->state = SEQ_STATE_ACQUIRING;
  346. } else {
  347. if (flag & TRY_LOCK)
  348. ls->nr_trylock++;
  349. if (flag & READ_LOCK)
  350. ls->nr_readlock++;
  351. seq->state = SEQ_STATE_READ_ACQUIRED;
  352. seq->read_count = 1;
  353. ls->nr_acquired++;
  354. }
  355. break;
  356. case SEQ_STATE_READ_ACQUIRED:
  357. if (flag & READ_LOCK) {
  358. seq->read_count++;
  359. ls->nr_acquired++;
  360. goto end;
  361. } else {
  362. goto broken;
  363. }
  364. break;
  365. case SEQ_STATE_ACQUIRED:
  366. case SEQ_STATE_ACQUIRING:
  367. case SEQ_STATE_CONTENDED:
  368. broken:
  369. /* broken lock sequence, discard it */
  370. ls->discard = 1;
  371. bad_hist[BROKEN_ACQUIRE]++;
  372. list_del_init(&seq->list);
  373. free(seq);
  374. goto end;
  375. default:
  376. BUG_ON("Unknown state of lock sequence found!\n");
  377. break;
  378. }
  379. ls->nr_acquire++;
  380. seq->prev_event_time = sample->time;
  381. end:
  382. return 0;
  383. }
  384. static int report_lock_acquired_event(struct evsel *evsel,
  385. struct perf_sample *sample)
  386. {
  387. void *addr;
  388. struct lock_stat *ls;
  389. struct thread_stat *ts;
  390. struct lock_seq_stat *seq;
  391. u64 contended_term;
  392. const char *name = evsel__strval(evsel, sample, "name");
  393. u64 tmp = evsel__intval(evsel, sample, "lockdep_addr");
  394. memcpy(&addr, &tmp, sizeof(void *));
  395. ls = lock_stat_findnew(addr, name);
  396. if (!ls)
  397. return -ENOMEM;
  398. if (ls->discard)
  399. return 0;
  400. ts = thread_stat_findnew(sample->tid);
  401. if (!ts)
  402. return -ENOMEM;
  403. seq = get_seq(ts, addr);
  404. if (!seq)
  405. return -ENOMEM;
  406. switch (seq->state) {
  407. case SEQ_STATE_UNINITIALIZED:
  408. /* orphan event, do nothing */
  409. return 0;
  410. case SEQ_STATE_ACQUIRING:
  411. break;
  412. case SEQ_STATE_CONTENDED:
  413. contended_term = sample->time - seq->prev_event_time;
  414. ls->wait_time_total += contended_term;
  415. if (contended_term < ls->wait_time_min)
  416. ls->wait_time_min = contended_term;
  417. if (ls->wait_time_max < contended_term)
  418. ls->wait_time_max = contended_term;
  419. break;
  420. case SEQ_STATE_RELEASED:
  421. case SEQ_STATE_ACQUIRED:
  422. case SEQ_STATE_READ_ACQUIRED:
  423. /* broken lock sequence, discard it */
  424. ls->discard = 1;
  425. bad_hist[BROKEN_ACQUIRED]++;
  426. list_del_init(&seq->list);
  427. free(seq);
  428. goto end;
  429. default:
  430. BUG_ON("Unknown state of lock sequence found!\n");
  431. break;
  432. }
  433. seq->state = SEQ_STATE_ACQUIRED;
  434. ls->nr_acquired++;
  435. ls->avg_wait_time = ls->nr_contended ? ls->wait_time_total/ls->nr_contended : 0;
  436. seq->prev_event_time = sample->time;
  437. end:
  438. return 0;
  439. }
  440. static int report_lock_contended_event(struct evsel *evsel,
  441. struct perf_sample *sample)
  442. {
  443. void *addr;
  444. struct lock_stat *ls;
  445. struct thread_stat *ts;
  446. struct lock_seq_stat *seq;
  447. const char *name = evsel__strval(evsel, sample, "name");
  448. u64 tmp = evsel__intval(evsel, sample, "lockdep_addr");
  449. memcpy(&addr, &tmp, sizeof(void *));
  450. ls = lock_stat_findnew(addr, name);
  451. if (!ls)
  452. return -ENOMEM;
  453. if (ls->discard)
  454. return 0;
  455. ts = thread_stat_findnew(sample->tid);
  456. if (!ts)
  457. return -ENOMEM;
  458. seq = get_seq(ts, addr);
  459. if (!seq)
  460. return -ENOMEM;
  461. switch (seq->state) {
  462. case SEQ_STATE_UNINITIALIZED:
  463. /* orphan event, do nothing */
  464. return 0;
  465. case SEQ_STATE_ACQUIRING:
  466. break;
  467. case SEQ_STATE_RELEASED:
  468. case SEQ_STATE_ACQUIRED:
  469. case SEQ_STATE_READ_ACQUIRED:
  470. case SEQ_STATE_CONTENDED:
  471. /* broken lock sequence, discard it */
  472. ls->discard = 1;
  473. bad_hist[BROKEN_CONTENDED]++;
  474. list_del_init(&seq->list);
  475. free(seq);
  476. goto end;
  477. default:
  478. BUG_ON("Unknown state of lock sequence found!\n");
  479. break;
  480. }
  481. seq->state = SEQ_STATE_CONTENDED;
  482. ls->nr_contended++;
  483. ls->avg_wait_time = ls->wait_time_total/ls->nr_contended;
  484. seq->prev_event_time = sample->time;
  485. end:
  486. return 0;
  487. }
  488. static int report_lock_release_event(struct evsel *evsel,
  489. struct perf_sample *sample)
  490. {
  491. void *addr;
  492. struct lock_stat *ls;
  493. struct thread_stat *ts;
  494. struct lock_seq_stat *seq;
  495. const char *name = evsel__strval(evsel, sample, "name");
  496. u64 tmp = evsel__intval(evsel, sample, "lockdep_addr");
  497. memcpy(&addr, &tmp, sizeof(void *));
  498. ls = lock_stat_findnew(addr, name);
  499. if (!ls)
  500. return -ENOMEM;
  501. if (ls->discard)
  502. return 0;
  503. ts = thread_stat_findnew(sample->tid);
  504. if (!ts)
  505. return -ENOMEM;
  506. seq = get_seq(ts, addr);
  507. if (!seq)
  508. return -ENOMEM;
  509. switch (seq->state) {
  510. case SEQ_STATE_UNINITIALIZED:
  511. goto end;
  512. case SEQ_STATE_ACQUIRED:
  513. break;
  514. case SEQ_STATE_READ_ACQUIRED:
  515. seq->read_count--;
  516. BUG_ON(seq->read_count < 0);
  517. if (seq->read_count) {
  518. ls->nr_release++;
  519. goto end;
  520. }
  521. break;
  522. case SEQ_STATE_ACQUIRING:
  523. case SEQ_STATE_CONTENDED:
  524. case SEQ_STATE_RELEASED:
  525. /* broken lock sequence, discard it */
  526. ls->discard = 1;
  527. bad_hist[BROKEN_RELEASE]++;
  528. goto free_seq;
  529. default:
  530. BUG_ON("Unknown state of lock sequence found!\n");
  531. break;
  532. }
  533. ls->nr_release++;
  534. free_seq:
  535. list_del_init(&seq->list);
  536. free(seq);
  537. end:
  538. return 0;
  539. }
  540. /* lock oriented handlers */
  541. /* TODO: handlers for CPU oriented, thread oriented */
  542. static struct trace_lock_handler report_lock_ops = {
  543. .acquire_event = report_lock_acquire_event,
  544. .acquired_event = report_lock_acquired_event,
  545. .contended_event = report_lock_contended_event,
  546. .release_event = report_lock_release_event,
  547. };
  548. static struct trace_lock_handler *trace_handler;
  549. static int evsel__process_lock_acquire(struct evsel *evsel, struct perf_sample *sample)
  550. {
  551. if (trace_handler->acquire_event)
  552. return trace_handler->acquire_event(evsel, sample);
  553. return 0;
  554. }
  555. static int evsel__process_lock_acquired(struct evsel *evsel, struct perf_sample *sample)
  556. {
  557. if (trace_handler->acquired_event)
  558. return trace_handler->acquired_event(evsel, sample);
  559. return 0;
  560. }
  561. static int evsel__process_lock_contended(struct evsel *evsel, struct perf_sample *sample)
  562. {
  563. if (trace_handler->contended_event)
  564. return trace_handler->contended_event(evsel, sample);
  565. return 0;
  566. }
  567. static int evsel__process_lock_release(struct evsel *evsel, struct perf_sample *sample)
  568. {
  569. if (trace_handler->release_event)
  570. return trace_handler->release_event(evsel, sample);
  571. return 0;
  572. }
  573. static void print_bad_events(int bad, int total)
  574. {
  575. /* Output for debug, this have to be removed */
  576. int i;
  577. const char *name[4] =
  578. { "acquire", "acquired", "contended", "release" };
  579. pr_info("\n=== output for debug===\n\n");
  580. pr_info("bad: %d, total: %d\n", bad, total);
  581. pr_info("bad rate: %.2f %%\n", (double)bad / (double)total * 100);
  582. pr_info("histogram of events caused bad sequence\n");
  583. for (i = 0; i < BROKEN_MAX; i++)
  584. pr_info(" %10s: %d\n", name[i], bad_hist[i]);
  585. }
  586. /* TODO: various way to print, coloring, nano or milli sec */
  587. static void print_result(void)
  588. {
  589. struct lock_stat *st;
  590. char cut_name[20];
  591. int bad, total;
  592. pr_info("%20s ", "Name");
  593. pr_info("%10s ", "acquired");
  594. pr_info("%10s ", "contended");
  595. pr_info("%15s ", "avg wait (ns)");
  596. pr_info("%15s ", "total wait (ns)");
  597. pr_info("%15s ", "max wait (ns)");
  598. pr_info("%15s ", "min wait (ns)");
  599. pr_info("\n\n");
  600. bad = total = 0;
  601. while ((st = pop_from_result())) {
  602. total++;
  603. if (st->discard) {
  604. bad++;
  605. continue;
  606. }
  607. bzero(cut_name, 20);
  608. if (strlen(st->name) < 16) {
  609. /* output raw name */
  610. pr_info("%20s ", st->name);
  611. } else {
  612. strncpy(cut_name, st->name, 16);
  613. cut_name[16] = '.';
  614. cut_name[17] = '.';
  615. cut_name[18] = '.';
  616. cut_name[19] = '\0';
  617. /* cut off name for saving output style */
  618. pr_info("%20s ", cut_name);
  619. }
  620. pr_info("%10u ", st->nr_acquired);
  621. pr_info("%10u ", st->nr_contended);
  622. pr_info("%15" PRIu64 " ", st->avg_wait_time);
  623. pr_info("%15" PRIu64 " ", st->wait_time_total);
  624. pr_info("%15" PRIu64 " ", st->wait_time_max);
  625. pr_info("%15" PRIu64 " ", st->wait_time_min == ULLONG_MAX ?
  626. 0 : st->wait_time_min);
  627. pr_info("\n");
  628. }
  629. print_bad_events(bad, total);
  630. }
  631. static bool info_threads, info_map;
  632. static void dump_threads(void)
  633. {
  634. struct thread_stat *st;
  635. struct rb_node *node;
  636. struct thread *t;
  637. pr_info("%10s: comm\n", "Thread ID");
  638. node = rb_first(&thread_stats);
  639. while (node) {
  640. st = container_of(node, struct thread_stat, rb);
  641. t = perf_session__findnew(session, st->tid);
  642. pr_info("%10d: %s\n", st->tid, thread__comm_str(t));
  643. node = rb_next(node);
  644. thread__put(t);
  645. }
  646. }
  647. static void dump_map(void)
  648. {
  649. unsigned int i;
  650. struct lock_stat *st;
  651. pr_info("Address of instance: name of class\n");
  652. for (i = 0; i < LOCKHASH_SIZE; i++) {
  653. list_for_each_entry(st, &lockhash_table[i], hash_entry) {
  654. pr_info(" %p: %s\n", st->addr, st->name);
  655. }
  656. }
  657. }
  658. static int dump_info(void)
  659. {
  660. int rc = 0;
  661. if (info_threads)
  662. dump_threads();
  663. else if (info_map)
  664. dump_map();
  665. else {
  666. rc = -1;
  667. pr_err("Unknown type of information\n");
  668. }
  669. return rc;
  670. }
  671. typedef int (*tracepoint_handler)(struct evsel *evsel,
  672. struct perf_sample *sample);
  673. static int process_sample_event(struct perf_tool *tool __maybe_unused,
  674. union perf_event *event,
  675. struct perf_sample *sample,
  676. struct evsel *evsel,
  677. struct machine *machine)
  678. {
  679. int err = 0;
  680. struct thread *thread = machine__findnew_thread(machine, sample->pid,
  681. sample->tid);
  682. if (thread == NULL) {
  683. pr_debug("problem processing %d event, skipping it.\n",
  684. event->header.type);
  685. return -1;
  686. }
  687. if (evsel->handler != NULL) {
  688. tracepoint_handler f = evsel->handler;
  689. err = f(evsel, sample);
  690. }
  691. thread__put(thread);
  692. return err;
  693. }
  694. static void sort_result(void)
  695. {
  696. unsigned int i;
  697. struct lock_stat *st;
  698. for (i = 0; i < LOCKHASH_SIZE; i++) {
  699. list_for_each_entry(st, &lockhash_table[i], hash_entry) {
  700. insert_to_result(st, compare);
  701. }
  702. }
  703. }
  704. static const struct evsel_str_handler lock_tracepoints[] = {
  705. { "lock:lock_acquire", evsel__process_lock_acquire, }, /* CONFIG_LOCKDEP */
  706. { "lock:lock_acquired", evsel__process_lock_acquired, }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
  707. { "lock:lock_contended", evsel__process_lock_contended, }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
  708. { "lock:lock_release", evsel__process_lock_release, }, /* CONFIG_LOCKDEP */
  709. };
  710. static bool force;
  711. static int __cmd_report(bool display_info)
  712. {
  713. int err = -EINVAL;
  714. struct perf_tool eops = {
  715. .sample = process_sample_event,
  716. .comm = perf_event__process_comm,
  717. .namespaces = perf_event__process_namespaces,
  718. .ordered_events = true,
  719. };
  720. struct perf_data data = {
  721. .path = input_name,
  722. .mode = PERF_DATA_MODE_READ,
  723. .force = force,
  724. };
  725. session = perf_session__new(&data, false, &eops);
  726. if (IS_ERR(session)) {
  727. pr_err("Initializing perf session failed\n");
  728. return PTR_ERR(session);
  729. }
  730. symbol__init(&session->header.env);
  731. if (!perf_session__has_traces(session, "lock record"))
  732. goto out_delete;
  733. if (perf_session__set_tracepoints_handlers(session, lock_tracepoints)) {
  734. pr_err("Initializing perf session tracepoint handlers failed\n");
  735. goto out_delete;
  736. }
  737. if (select_key())
  738. goto out_delete;
  739. err = perf_session__process_events(session);
  740. if (err)
  741. goto out_delete;
  742. setup_pager();
  743. if (display_info) /* used for info subcommand */
  744. err = dump_info();
  745. else {
  746. sort_result();
  747. print_result();
  748. }
  749. out_delete:
  750. perf_session__delete(session);
  751. return err;
  752. }
  753. static int __cmd_record(int argc, const char **argv)
  754. {
  755. const char *record_args[] = {
  756. "record", "-R", "-m", "1024", "-c", "1",
  757. };
  758. unsigned int rec_argc, i, j, ret;
  759. const char **rec_argv;
  760. for (i = 0; i < ARRAY_SIZE(lock_tracepoints); i++) {
  761. if (!is_valid_tracepoint(lock_tracepoints[i].name)) {
  762. pr_err("tracepoint %s is not enabled. "
  763. "Are CONFIG_LOCKDEP and CONFIG_LOCK_STAT enabled?\n",
  764. lock_tracepoints[i].name);
  765. return 1;
  766. }
  767. }
  768. rec_argc = ARRAY_SIZE(record_args) + argc - 1;
  769. /* factor of 2 is for -e in front of each tracepoint */
  770. rec_argc += 2 * ARRAY_SIZE(lock_tracepoints);
  771. rec_argv = calloc(rec_argc + 1, sizeof(char *));
  772. if (!rec_argv)
  773. return -ENOMEM;
  774. for (i = 0; i < ARRAY_SIZE(record_args); i++)
  775. rec_argv[i] = strdup(record_args[i]);
  776. for (j = 0; j < ARRAY_SIZE(lock_tracepoints); j++) {
  777. rec_argv[i++] = "-e";
  778. rec_argv[i++] = strdup(lock_tracepoints[j].name);
  779. }
  780. for (j = 1; j < (unsigned int)argc; j++, i++)
  781. rec_argv[i] = argv[j];
  782. BUG_ON(i != rec_argc);
  783. ret = cmd_record(i, rec_argv);
  784. free(rec_argv);
  785. return ret;
  786. }
  787. int cmd_lock(int argc, const char **argv)
  788. {
  789. const struct option lock_options[] = {
  790. OPT_STRING('i', "input", &input_name, "file", "input file name"),
  791. OPT_INCR('v', "verbose", &verbose, "be more verbose (show symbol address, etc)"),
  792. OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, "dump raw trace in ASCII"),
  793. OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
  794. OPT_END()
  795. };
  796. const struct option info_options[] = {
  797. OPT_BOOLEAN('t', "threads", &info_threads,
  798. "dump thread list in perf.data"),
  799. OPT_BOOLEAN('m', "map", &info_map,
  800. "map of lock instances (address:name table)"),
  801. OPT_PARENT(lock_options)
  802. };
  803. const struct option report_options[] = {
  804. OPT_STRING('k', "key", &sort_key, "acquired",
  805. "key for sorting (acquired / contended / avg_wait / wait_total / wait_max / wait_min)"),
  806. /* TODO: type */
  807. OPT_PARENT(lock_options)
  808. };
  809. const char * const info_usage[] = {
  810. "perf lock info [<options>]",
  811. NULL
  812. };
  813. const char *const lock_subcommands[] = { "record", "report", "script",
  814. "info", NULL };
  815. const char *lock_usage[] = {
  816. NULL,
  817. NULL
  818. };
  819. const char * const report_usage[] = {
  820. "perf lock report [<options>]",
  821. NULL
  822. };
  823. unsigned int i;
  824. int rc = 0;
  825. for (i = 0; i < LOCKHASH_SIZE; i++)
  826. INIT_LIST_HEAD(lockhash_table + i);
  827. argc = parse_options_subcommand(argc, argv, lock_options, lock_subcommands,
  828. lock_usage, PARSE_OPT_STOP_AT_NON_OPTION);
  829. if (!argc)
  830. usage_with_options(lock_usage, lock_options);
  831. if (!strncmp(argv[0], "rec", 3)) {
  832. return __cmd_record(argc, argv);
  833. } else if (!strncmp(argv[0], "report", 6)) {
  834. trace_handler = &report_lock_ops;
  835. if (argc) {
  836. argc = parse_options(argc, argv,
  837. report_options, report_usage, 0);
  838. if (argc)
  839. usage_with_options(report_usage, report_options);
  840. }
  841. rc = __cmd_report(false);
  842. } else if (!strcmp(argv[0], "script")) {
  843. /* Aliased to 'perf script' */
  844. return cmd_script(argc, argv);
  845. } else if (!strcmp(argv[0], "info")) {
  846. if (argc) {
  847. argc = parse_options(argc, argv,
  848. info_options, info_usage, 0);
  849. if (argc)
  850. usage_with_options(info_usage, info_options);
  851. }
  852. /* recycling report_lock_ops */
  853. trace_handler = &report_lock_ops;
  854. rc = __cmd_report(true);
  855. } else {
  856. usage_with_options(lock_usage, lock_options);
  857. }
  858. return rc;
  859. }