builtin-report.c 28 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967
  1. /*
  2. * builtin-report.c
  3. *
  4. * Builtin report command: Analyze the perf.data input file,
  5. * look up and read DSOs and symbol information and display
  6. * a histogram of results, along various sorting keys.
  7. */
  8. #include "builtin.h"
  9. #include "util/util.h"
  10. #include "util/cache.h"
  11. #include "util/annotate.h"
  12. #include "util/color.h"
  13. #include <linux/list.h>
  14. #include <linux/rbtree.h>
  15. #include "util/symbol.h"
  16. #include "util/callchain.h"
  17. #include "util/strlist.h"
  18. #include "util/values.h"
  19. #include "perf.h"
  20. #include "util/debug.h"
  21. #include "util/evlist.h"
  22. #include "util/evsel.h"
  23. #include "util/header.h"
  24. #include "util/session.h"
  25. #include "util/tool.h"
  26. #include "util/parse-options.h"
  27. #include "util/parse-events.h"
  28. #include "util/thread.h"
  29. #include "util/sort.h"
  30. #include "util/hist.h"
  31. #include "util/data.h"
  32. #include "arch/common.h"
  33. #include "util/auxtrace.h"
  34. #include <dlfcn.h>
  35. #include <linux/bitmap.h>
  36. struct report {
  37. struct perf_tool tool;
  38. struct perf_session *session;
  39. bool use_tui, use_gtk, use_stdio;
  40. bool hide_unresolved;
  41. bool dont_use_callchains;
  42. bool show_full_info;
  43. bool show_threads;
  44. bool inverted_callchain;
  45. bool mem_mode;
  46. bool header;
  47. bool header_only;
  48. bool nonany_branch_mode;
  49. int max_stack;
  50. struct perf_read_values show_threads_values;
  51. const char *pretty_printing_style;
  52. const char *cpu_list;
  53. const char *symbol_filter_str;
  54. float min_percent;
  55. u64 nr_entries;
  56. u64 queue_size;
  57. int socket_filter;
  58. DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
  59. };
  60. static int report__config(const char *var, const char *value, void *cb)
  61. {
  62. struct report *rep = cb;
  63. if (!strcmp(var, "report.group")) {
  64. symbol_conf.event_group = perf_config_bool(var, value);
  65. return 0;
  66. }
  67. if (!strcmp(var, "report.percent-limit")) {
  68. rep->min_percent = strtof(value, NULL);
  69. return 0;
  70. }
  71. if (!strcmp(var, "report.children")) {
  72. symbol_conf.cumulate_callchain = perf_config_bool(var, value);
  73. return 0;
  74. }
  75. if (!strcmp(var, "report.queue-size")) {
  76. rep->queue_size = perf_config_u64(var, value);
  77. return 0;
  78. }
  79. return perf_default_config(var, value, cb);
  80. }
  81. static int hist_iter__report_callback(struct hist_entry_iter *iter,
  82. struct addr_location *al, bool single,
  83. void *arg)
  84. {
  85. int err = 0;
  86. struct report *rep = arg;
  87. struct hist_entry *he = iter->he;
  88. struct perf_evsel *evsel = iter->evsel;
  89. struct mem_info *mi;
  90. struct branch_info *bi;
  91. if (!ui__has_annotation())
  92. return 0;
  93. hist__account_cycles(iter->sample->branch_stack, al, iter->sample,
  94. rep->nonany_branch_mode);
  95. if (sort__mode == SORT_MODE__BRANCH) {
  96. bi = he->branch_info;
  97. err = addr_map_symbol__inc_samples(&bi->from, evsel->idx);
  98. if (err)
  99. goto out;
  100. err = addr_map_symbol__inc_samples(&bi->to, evsel->idx);
  101. } else if (rep->mem_mode) {
  102. mi = he->mem_info;
  103. err = addr_map_symbol__inc_samples(&mi->daddr, evsel->idx);
  104. if (err)
  105. goto out;
  106. err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
  107. } else if (symbol_conf.cumulate_callchain) {
  108. if (single)
  109. err = hist_entry__inc_addr_samples(he, evsel->idx,
  110. al->addr);
  111. } else {
  112. err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
  113. }
  114. out:
  115. return err;
  116. }
  117. static int process_sample_event(struct perf_tool *tool,
  118. union perf_event *event,
  119. struct perf_sample *sample,
  120. struct perf_evsel *evsel,
  121. struct machine *machine)
  122. {
  123. struct report *rep = container_of(tool, struct report, tool);
  124. struct addr_location al;
  125. struct hist_entry_iter iter = {
  126. .evsel = evsel,
  127. .sample = sample,
  128. .hide_unresolved = rep->hide_unresolved,
  129. .add_entry_cb = hist_iter__report_callback,
  130. };
  131. int ret = 0;
  132. if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) {
  133. pr_debug("problem processing %d event, skipping it.\n",
  134. event->header.type);
  135. return -1;
  136. }
  137. if (rep->hide_unresolved && al.sym == NULL)
  138. goto out_put;
  139. if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
  140. goto out_put;
  141. if (sort__mode == SORT_MODE__BRANCH) {
  142. /*
  143. * A non-synthesized event might not have a branch stack if
  144. * branch stacks have been synthesized (using itrace options).
  145. */
  146. if (!sample->branch_stack)
  147. goto out_put;
  148. iter.ops = &hist_iter_branch;
  149. } else if (rep->mem_mode) {
  150. iter.ops = &hist_iter_mem;
  151. } else if (symbol_conf.cumulate_callchain) {
  152. iter.ops = &hist_iter_cumulative;
  153. } else {
  154. iter.ops = &hist_iter_normal;
  155. }
  156. if (al.map != NULL)
  157. al.map->dso->hit = 1;
  158. ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
  159. if (ret < 0)
  160. pr_debug("problem adding hist entry, skipping event\n");
  161. out_put:
  162. addr_location__put(&al);
  163. return ret;
  164. }
  165. static int process_read_event(struct perf_tool *tool,
  166. union perf_event *event,
  167. struct perf_sample *sample __maybe_unused,
  168. struct perf_evsel *evsel,
  169. struct machine *machine __maybe_unused)
  170. {
  171. struct report *rep = container_of(tool, struct report, tool);
  172. if (rep->show_threads) {
  173. const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
  174. perf_read_values_add_value(&rep->show_threads_values,
  175. event->read.pid, event->read.tid,
  176. event->read.id,
  177. name,
  178. event->read.value);
  179. }
  180. dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
  181. evsel ? perf_evsel__name(evsel) : "FAIL",
  182. event->read.value);
  183. return 0;
  184. }
  185. /* For pipe mode, sample_type is not currently set */
  186. static int report__setup_sample_type(struct report *rep)
  187. {
  188. struct perf_session *session = rep->session;
  189. u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
  190. bool is_pipe = perf_data_file__is_pipe(session->file);
  191. if (session->itrace_synth_opts->callchain ||
  192. (!is_pipe &&
  193. perf_header__has_feat(&session->header, HEADER_AUXTRACE) &&
  194. !session->itrace_synth_opts->set))
  195. sample_type |= PERF_SAMPLE_CALLCHAIN;
  196. if (session->itrace_synth_opts->last_branch)
  197. sample_type |= PERF_SAMPLE_BRANCH_STACK;
  198. if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
  199. if (sort__has_parent) {
  200. ui__error("Selected --sort parent, but no "
  201. "callchain data. Did you call "
  202. "'perf record' without -g?\n");
  203. return -EINVAL;
  204. }
  205. if (symbol_conf.use_callchain) {
  206. ui__error("Selected -g or --branch-history but no "
  207. "callchain data. Did\n"
  208. "you call 'perf record' without -g?\n");
  209. return -1;
  210. }
  211. } else if (!rep->dont_use_callchains &&
  212. callchain_param.mode != CHAIN_NONE &&
  213. !symbol_conf.use_callchain) {
  214. symbol_conf.use_callchain = true;
  215. if (callchain_register_param(&callchain_param) < 0) {
  216. ui__error("Can't register callchain params.\n");
  217. return -EINVAL;
  218. }
  219. }
  220. if (symbol_conf.cumulate_callchain) {
  221. /* Silently ignore if callchain is missing */
  222. if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
  223. symbol_conf.cumulate_callchain = false;
  224. perf_hpp__cancel_cumulate();
  225. }
  226. }
  227. if (sort__mode == SORT_MODE__BRANCH) {
  228. if (!is_pipe &&
  229. !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
  230. ui__error("Selected -b but no branch data. "
  231. "Did you call perf record without -b?\n");
  232. return -1;
  233. }
  234. }
  235. if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
  236. if ((sample_type & PERF_SAMPLE_REGS_USER) &&
  237. (sample_type & PERF_SAMPLE_STACK_USER))
  238. callchain_param.record_mode = CALLCHAIN_DWARF;
  239. else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
  240. callchain_param.record_mode = CALLCHAIN_LBR;
  241. else
  242. callchain_param.record_mode = CALLCHAIN_FP;
  243. }
  244. /* ??? handle more cases than just ANY? */
  245. if (!(perf_evlist__combined_branch_type(session->evlist) &
  246. PERF_SAMPLE_BRANCH_ANY))
  247. rep->nonany_branch_mode = true;
  248. return 0;
  249. }
  250. static void sig_handler(int sig __maybe_unused)
  251. {
  252. session_done = 1;
  253. }
  254. static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
  255. const char *evname, FILE *fp)
  256. {
  257. size_t ret;
  258. char unit;
  259. unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
  260. u64 nr_events = hists->stats.total_period;
  261. struct perf_evsel *evsel = hists_to_evsel(hists);
  262. char buf[512];
  263. size_t size = sizeof(buf);
  264. int socked_id = hists->socket_filter;
  265. if (symbol_conf.filter_relative) {
  266. nr_samples = hists->stats.nr_non_filtered_samples;
  267. nr_events = hists->stats.total_non_filtered_period;
  268. }
  269. if (perf_evsel__is_group_event(evsel)) {
  270. struct perf_evsel *pos;
  271. perf_evsel__group_desc(evsel, buf, size);
  272. evname = buf;
  273. for_each_group_member(pos, evsel) {
  274. const struct hists *pos_hists = evsel__hists(pos);
  275. if (symbol_conf.filter_relative) {
  276. nr_samples += pos_hists->stats.nr_non_filtered_samples;
  277. nr_events += pos_hists->stats.total_non_filtered_period;
  278. } else {
  279. nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE];
  280. nr_events += pos_hists->stats.total_period;
  281. }
  282. }
  283. }
  284. nr_samples = convert_unit(nr_samples, &unit);
  285. ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
  286. if (evname != NULL)
  287. ret += fprintf(fp, " of event '%s'", evname);
  288. if (symbol_conf.show_ref_callgraph &&
  289. strstr(evname, "call-graph=no")) {
  290. ret += fprintf(fp, ", show reference callgraph");
  291. }
  292. if (rep->mem_mode) {
  293. ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
  294. ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order);
  295. } else
  296. ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
  297. if (socked_id > -1)
  298. ret += fprintf(fp, "\n# Processor Socket: %d", socked_id);
  299. return ret + fprintf(fp, "\n#\n");
  300. }
  301. static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
  302. struct report *rep,
  303. const char *help)
  304. {
  305. struct perf_evsel *pos;
  306. fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n", evlist->stats.total_lost_samples);
  307. evlist__for_each(evlist, pos) {
  308. struct hists *hists = evsel__hists(pos);
  309. const char *evname = perf_evsel__name(pos);
  310. if (symbol_conf.event_group &&
  311. !perf_evsel__is_group_leader(pos))
  312. continue;
  313. hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
  314. hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout);
  315. fprintf(stdout, "\n\n");
  316. }
  317. if (sort_order == NULL &&
  318. parent_pattern == default_parent_pattern)
  319. fprintf(stdout, "#\n# (%s)\n#\n", help);
  320. if (rep->show_threads) {
  321. bool style = !strcmp(rep->pretty_printing_style, "raw");
  322. perf_read_values_display(stdout, &rep->show_threads_values,
  323. style);
  324. perf_read_values_destroy(&rep->show_threads_values);
  325. }
  326. return 0;
  327. }
  328. static void report__warn_kptr_restrict(const struct report *rep)
  329. {
  330. struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
  331. struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
  332. if (kernel_map == NULL ||
  333. (kernel_map->dso->hit &&
  334. (kernel_kmap->ref_reloc_sym == NULL ||
  335. kernel_kmap->ref_reloc_sym->addr == 0))) {
  336. const char *desc =
  337. "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
  338. "can't be resolved.";
  339. if (kernel_map) {
  340. const struct dso *kdso = kernel_map->dso;
  341. if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) {
  342. desc = "If some relocation was applied (e.g. "
  343. "kexec) symbols may be misresolved.";
  344. }
  345. }
  346. ui__warning(
  347. "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
  348. "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
  349. "Samples in kernel modules can't be resolved as well.\n\n",
  350. desc);
  351. }
  352. }
  353. static int report__gtk_browse_hists(struct report *rep, const char *help)
  354. {
  355. int (*hist_browser)(struct perf_evlist *evlist, const char *help,
  356. struct hist_browser_timer *timer, float min_pcnt);
  357. hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists");
  358. if (hist_browser == NULL) {
  359. ui__error("GTK browser not found!\n");
  360. return -1;
  361. }
  362. return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
  363. }
  364. static int report__browse_hists(struct report *rep)
  365. {
  366. int ret;
  367. struct perf_session *session = rep->session;
  368. struct perf_evlist *evlist = session->evlist;
  369. const char *help = "For a higher level overview, try: perf report --sort comm,dso";
  370. switch (use_browser) {
  371. case 1:
  372. ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
  373. rep->min_percent,
  374. &session->header.env);
  375. /*
  376. * Usually "ret" is the last pressed key, and we only
  377. * care if the key notifies us to switch data file.
  378. */
  379. if (ret != K_SWITCH_INPUT_DATA)
  380. ret = 0;
  381. break;
  382. case 2:
  383. ret = report__gtk_browse_hists(rep, help);
  384. break;
  385. default:
  386. ret = perf_evlist__tty_browse_hists(evlist, rep, help);
  387. break;
  388. }
  389. return ret;
  390. }
  391. static void report__collapse_hists(struct report *rep)
  392. {
  393. struct ui_progress prog;
  394. struct perf_evsel *pos;
  395. ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
  396. evlist__for_each(rep->session->evlist, pos) {
  397. struct hists *hists = evsel__hists(pos);
  398. if (pos->idx == 0)
  399. hists->symbol_filter_str = rep->symbol_filter_str;
  400. hists->socket_filter = rep->socket_filter;
  401. hists__collapse_resort(hists, &prog);
  402. /* Non-group events are considered as leader */
  403. if (symbol_conf.event_group &&
  404. !perf_evsel__is_group_leader(pos)) {
  405. struct hists *leader_hists = evsel__hists(pos->leader);
  406. hists__match(leader_hists, hists);
  407. hists__link(leader_hists, hists);
  408. }
  409. }
  410. ui_progress__finish();
  411. }
  412. static void report__output_resort(struct report *rep)
  413. {
  414. struct ui_progress prog;
  415. struct perf_evsel *pos;
  416. ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
  417. evlist__for_each(rep->session->evlist, pos)
  418. hists__output_resort(evsel__hists(pos), &prog);
  419. ui_progress__finish();
  420. }
  421. static int __cmd_report(struct report *rep)
  422. {
  423. int ret;
  424. struct perf_session *session = rep->session;
  425. struct perf_evsel *pos;
  426. struct perf_data_file *file = session->file;
  427. signal(SIGINT, sig_handler);
  428. if (rep->cpu_list) {
  429. ret = perf_session__cpu_bitmap(session, rep->cpu_list,
  430. rep->cpu_bitmap);
  431. if (ret)
  432. return ret;
  433. }
  434. if (rep->show_threads)
  435. perf_read_values_init(&rep->show_threads_values);
  436. ret = report__setup_sample_type(rep);
  437. if (ret)
  438. return ret;
  439. ret = perf_session__process_events(session);
  440. if (ret)
  441. return ret;
  442. report__warn_kptr_restrict(rep);
  443. evlist__for_each(session->evlist, pos)
  444. rep->nr_entries += evsel__hists(pos)->nr_entries;
  445. if (use_browser == 0) {
  446. if (verbose > 3)
  447. perf_session__fprintf(session, stdout);
  448. if (verbose > 2)
  449. perf_session__fprintf_dsos(session, stdout);
  450. if (dump_trace) {
  451. perf_session__fprintf_nr_events(session, stdout);
  452. perf_evlist__fprintf_nr_events(session->evlist, stdout);
  453. return 0;
  454. }
  455. }
  456. report__collapse_hists(rep);
  457. if (session_done())
  458. return 0;
  459. /*
  460. * recalculate number of entries after collapsing since it
  461. * might be changed during the collapse phase.
  462. */
  463. rep->nr_entries = 0;
  464. evlist__for_each(session->evlist, pos)
  465. rep->nr_entries += evsel__hists(pos)->nr_entries;
  466. if (rep->nr_entries == 0) {
  467. ui__error("The %s file has no samples!\n", file->path);
  468. return 0;
  469. }
  470. report__output_resort(rep);
  471. return report__browse_hists(rep);
  472. }
  473. static int
  474. report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
  475. {
  476. struct report *rep = (struct report *)opt->value;
  477. /*
  478. * --no-call-graph
  479. */
  480. if (unset) {
  481. rep->dont_use_callchains = true;
  482. return 0;
  483. }
  484. return parse_callchain_report_opt(arg);
  485. }
  486. int
  487. report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
  488. const char *arg, int unset __maybe_unused)
  489. {
  490. if (arg) {
  491. int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
  492. if (err) {
  493. char buf[BUFSIZ];
  494. regerror(err, &ignore_callees_regex, buf, sizeof(buf));
  495. pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
  496. return -1;
  497. }
  498. have_ignore_callees = 1;
  499. }
  500. return 0;
  501. }
  502. static int
  503. parse_branch_mode(const struct option *opt __maybe_unused,
  504. const char *str __maybe_unused, int unset)
  505. {
  506. int *branch_mode = opt->value;
  507. *branch_mode = !unset;
  508. return 0;
  509. }
  510. static int
  511. parse_percent_limit(const struct option *opt, const char *str,
  512. int unset __maybe_unused)
  513. {
  514. struct report *rep = opt->value;
  515. rep->min_percent = strtof(str, NULL);
  516. return 0;
  517. }
  518. #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function"
  519. const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
  520. CALLCHAIN_REPORT_HELP
  521. "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
  522. int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
  523. {
  524. struct perf_session *session;
  525. struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
  526. struct stat st;
  527. bool has_br_stack = false;
  528. int branch_mode = -1;
  529. bool branch_call_mode = false;
  530. char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
  531. const char * const report_usage[] = {
  532. "perf report [<options>]",
  533. NULL
  534. };
  535. struct report report = {
  536. .tool = {
  537. .sample = process_sample_event,
  538. .mmap = perf_event__process_mmap,
  539. .mmap2 = perf_event__process_mmap2,
  540. .comm = perf_event__process_comm,
  541. .exit = perf_event__process_exit,
  542. .fork = perf_event__process_fork,
  543. .lost = perf_event__process_lost,
  544. .read = process_read_event,
  545. .attr = perf_event__process_attr,
  546. .tracing_data = perf_event__process_tracing_data,
  547. .build_id = perf_event__process_build_id,
  548. .id_index = perf_event__process_id_index,
  549. .auxtrace_info = perf_event__process_auxtrace_info,
  550. .auxtrace = perf_event__process_auxtrace,
  551. .ordered_events = true,
  552. .ordering_requires_timestamps = true,
  553. },
  554. .max_stack = PERF_MAX_STACK_DEPTH,
  555. .pretty_printing_style = "normal",
  556. .socket_filter = -1,
  557. };
  558. const struct option options[] = {
  559. OPT_STRING('i', "input", &input_name, "file",
  560. "input file name"),
  561. OPT_INCR('v', "verbose", &verbose,
  562. "be more verbose (show symbol address, etc)"),
  563. OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
  564. "dump raw trace in ASCII"),
  565. OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
  566. "file", "vmlinux pathname"),
  567. OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
  568. "file", "kallsyms pathname"),
  569. OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
  570. OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
  571. "load module symbols - WARNING: use only with -k and LIVE kernel"),
  572. OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
  573. "Show a column with the number of samples"),
  574. OPT_BOOLEAN('T', "threads", &report.show_threads,
  575. "Show per-thread event counters"),
  576. OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
  577. "pretty printing style key: normal raw"),
  578. OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
  579. OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
  580. OPT_BOOLEAN(0, "stdio", &report.use_stdio,
  581. "Use the stdio interface"),
  582. OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
  583. OPT_BOOLEAN(0, "header-only", &report.header_only,
  584. "Show only data header."),
  585. OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
  586. "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
  587. " Please refer the man page for the complete list."),
  588. OPT_STRING('F', "fields", &field_order, "key[,keys...]",
  589. "output field(s): overhead, period, sample plus all of sort keys"),
  590. OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
  591. "Show sample percentage for different cpu modes"),
  592. OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
  593. "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
  594. OPT_STRING('p', "parent", &parent_pattern, "regex",
  595. "regex filter to identify parent, see: '--sort parent'"),
  596. OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
  597. "Only display entries with parent-match"),
  598. OPT_CALLBACK_DEFAULT('g', "call-graph", &report,
  599. "print_type,threshold[,print_limit],order,sort_key[,branch]",
  600. report_callchain_help, &report_parse_callchain_opt,
  601. callchain_default_opt),
  602. OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
  603. "Accumulate callchains of children and show total overhead as well"),
  604. OPT_INTEGER(0, "max-stack", &report.max_stack,
  605. "Set the maximum stack depth when parsing the callchain, "
  606. "anything beyond the specified depth will be ignored. "
  607. "Default: " __stringify(PERF_MAX_STACK_DEPTH)),
  608. OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
  609. "alias for inverted call graph"),
  610. OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
  611. "ignore callees of these functions in call graphs",
  612. report_parse_ignore_callees_opt),
  613. OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
  614. "only consider symbols in these dsos"),
  615. OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
  616. "only consider symbols in these comms"),
  617. OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
  618. "only consider symbols in these pids"),
  619. OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
  620. "only consider symbols in these tids"),
  621. OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
  622. "only consider these symbols"),
  623. OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
  624. "only show symbols that (partially) match with this filter"),
  625. OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
  626. "width[,width...]",
  627. "don't try to adjust column width, use these fixed values"),
  628. OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
  629. "separator for columns, no spaces will be added between "
  630. "columns '.' is reserved."),
  631. OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved,
  632. "Only display entries resolved to a symbol"),
  633. OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
  634. "Look for files with symbols relative to this directory"),
  635. OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
  636. "list of cpus to profile"),
  637. OPT_BOOLEAN('I', "show-info", &report.show_full_info,
  638. "Display extended information about perf.data file"),
  639. OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
  640. "Interleave source code with assembly code (default)"),
  641. OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
  642. "Display raw encoding of assembly instructions (default)"),
  643. OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
  644. "Specify disassembler style (e.g. -M intel for intel syntax)"),
  645. OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
  646. "Show a column with the sum of periods"),
  647. OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
  648. "Show event group information together"),
  649. OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
  650. "use branch records for per branch histogram filling",
  651. parse_branch_mode),
  652. OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
  653. "add last branch records to call history"),
  654. OPT_STRING(0, "objdump", &objdump_path, "path",
  655. "objdump binary to use for disassembly and annotations"),
  656. OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
  657. "Disable symbol demangling"),
  658. OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
  659. "Enable kernel symbol demangling"),
  660. OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
  661. OPT_CALLBACK(0, "percent-limit", &report, "percent",
  662. "Don't show entries under that percent", parse_percent_limit),
  663. OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
  664. "how to display percentage of filtered entries", parse_filter_percentage),
  665. OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
  666. "Instruction Tracing options",
  667. itrace_parse_synth_opts),
  668. OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
  669. "Show full source file name path for source lines"),
  670. OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
  671. "Show callgraph from reference event"),
  672. OPT_INTEGER(0, "socket-filter", &report.socket_filter,
  673. "only show processor socket that match with this filter"),
  674. OPT_END()
  675. };
  676. struct perf_data_file file = {
  677. .mode = PERF_DATA_MODE_READ,
  678. };
  679. int ret = hists__init();
  680. if (ret < 0)
  681. return ret;
  682. perf_config(report__config, &report);
  683. argc = parse_options(argc, argv, options, report_usage, 0);
  684. if (symbol_conf.vmlinux_name &&
  685. access(symbol_conf.vmlinux_name, R_OK)) {
  686. pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
  687. return -EINVAL;
  688. }
  689. if (symbol_conf.kallsyms_name &&
  690. access(symbol_conf.kallsyms_name, R_OK)) {
  691. pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
  692. return -EINVAL;
  693. }
  694. if (report.use_stdio)
  695. use_browser = 0;
  696. else if (report.use_tui)
  697. use_browser = 1;
  698. else if (report.use_gtk)
  699. use_browser = 2;
  700. if (report.inverted_callchain)
  701. callchain_param.order = ORDER_CALLER;
  702. if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
  703. callchain_param.order = ORDER_CALLER;
  704. if (itrace_synth_opts.callchain &&
  705. (int)itrace_synth_opts.callchain_sz > report.max_stack)
  706. report.max_stack = itrace_synth_opts.callchain_sz;
  707. if (!input_name || !strlen(input_name)) {
  708. if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
  709. input_name = "-";
  710. else
  711. input_name = "perf.data";
  712. }
  713. file.path = input_name;
  714. file.force = symbol_conf.force;
  715. repeat:
  716. session = perf_session__new(&file, false, &report.tool);
  717. if (session == NULL)
  718. return -1;
  719. if (report.queue_size) {
  720. ordered_events__set_alloc_size(&session->ordered_events,
  721. report.queue_size);
  722. }
  723. session->itrace_synth_opts = &itrace_synth_opts;
  724. report.session = session;
  725. has_br_stack = perf_header__has_feat(&session->header,
  726. HEADER_BRANCH_STACK);
  727. if (itrace_synth_opts.last_branch)
  728. has_br_stack = true;
  729. /*
  730. * Branch mode is a tristate:
  731. * -1 means default, so decide based on the file having branch data.
  732. * 0/1 means the user chose a mode.
  733. */
  734. if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
  735. !branch_call_mode) {
  736. sort__mode = SORT_MODE__BRANCH;
  737. symbol_conf.cumulate_callchain = false;
  738. }
  739. if (branch_call_mode) {
  740. callchain_param.key = CCKEY_ADDRESS;
  741. callchain_param.branch_callstack = 1;
  742. symbol_conf.use_callchain = true;
  743. callchain_register_param(&callchain_param);
  744. if (sort_order == NULL)
  745. sort_order = "srcline,symbol,dso";
  746. }
  747. if (report.mem_mode) {
  748. if (sort__mode == SORT_MODE__BRANCH) {
  749. pr_err("branch and mem mode incompatible\n");
  750. goto error;
  751. }
  752. sort__mode = SORT_MODE__MEMORY;
  753. symbol_conf.cumulate_callchain = false;
  754. }
  755. if (setup_sorting() < 0) {
  756. if (sort_order)
  757. parse_options_usage(report_usage, options, "s", 1);
  758. if (field_order)
  759. parse_options_usage(sort_order ? NULL : report_usage,
  760. options, "F", 1);
  761. goto error;
  762. }
  763. /* Force tty output for header output and per-thread stat. */
  764. if (report.header || report.header_only || report.show_threads)
  765. use_browser = 0;
  766. if (strcmp(input_name, "-") != 0)
  767. setup_browser(true);
  768. else
  769. use_browser = 0;
  770. if (report.header || report.header_only) {
  771. perf_session__fprintf_info(session, stdout,
  772. report.show_full_info);
  773. if (report.header_only) {
  774. ret = 0;
  775. goto error;
  776. }
  777. } else if (use_browser == 0) {
  778. fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
  779. stdout);
  780. }
  781. /*
  782. * Only in the TUI browser we are doing integrated annotation,
  783. * so don't allocate extra space that won't be used in the stdio
  784. * implementation.
  785. */
  786. if (ui__has_annotation()) {
  787. symbol_conf.priv_size = sizeof(struct annotation);
  788. machines__set_symbol_filter(&session->machines,
  789. symbol__annotate_init);
  790. /*
  791. * For searching by name on the "Browse map details".
  792. * providing it only in verbose mode not to bloat too
  793. * much struct symbol.
  794. */
  795. if (verbose) {
  796. /*
  797. * XXX: Need to provide a less kludgy way to ask for
  798. * more space per symbol, the u32 is for the index on
  799. * the ui browser.
  800. * See symbol__browser_index.
  801. */
  802. symbol_conf.priv_size += sizeof(u32);
  803. symbol_conf.sort_by_name = true;
  804. }
  805. }
  806. if (symbol__init(&session->header.env) < 0)
  807. goto error;
  808. if (argc) {
  809. /*
  810. * Special case: if there's an argument left then assume that
  811. * it's a symbol filter:
  812. */
  813. if (argc > 1)
  814. usage_with_options(report_usage, options);
  815. report.symbol_filter_str = argv[0];
  816. }
  817. sort__setup_elide(stdout);
  818. ret = __cmd_report(&report);
  819. if (ret == K_SWITCH_INPUT_DATA) {
  820. perf_session__delete(session);
  821. goto repeat;
  822. } else
  823. ret = 0;
  824. error:
  825. perf_session__delete(session);
  826. return ret;
  827. }