perf tools: Pass tool context in the the perf_event_ops functions
[deliverable/linux.git] / tools / perf / builtin-report.c
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
10 #include "util/util.h"
11
12 #include "util/annotate.h"
13 #include "util/color.h"
14 #include <linux/list.h>
15 #include "util/cache.h"
16 #include <linux/rbtree.h>
17 #include "util/symbol.h"
18 #include "util/callchain.h"
19 #include "util/strlist.h"
20 #include "util/values.h"
21
22 #include "perf.h"
23 #include "util/debug.h"
24 #include "util/evlist.h"
25 #include "util/evsel.h"
26 #include "util/header.h"
27 #include "util/session.h"
28
29 #include "util/parse-options.h"
30 #include "util/parse-events.h"
31
32 #include "util/thread.h"
33 #include "util/sort.h"
34 #include "util/hist.h"
35
36 #include <linux/bitmap.h>
37
38 struct perf_report {
39 struct perf_event_ops ops;
40 struct perf_session *session;
41 char const *input_name;
42 bool force, use_tui, use_stdio;
43 bool hide_unresolved;
44 bool dont_use_callchains;
45 bool show_full_info;
46 bool show_threads;
47 bool inverted_callchain;
48 struct perf_read_values show_threads_values;
49 const char *pretty_printing_style;
50 symbol_filter_t annotate_init;
51 const char *cpu_list;
52 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
53 };
54
55 static int perf_session__add_hist_entry(struct perf_session *session,
56 struct addr_location *al,
57 struct perf_sample *sample,
58 struct perf_evsel *evsel)
59 {
60 struct symbol *parent = NULL;
61 int err = 0;
62 struct hist_entry *he;
63
64 if ((sort__has_parent || symbol_conf.use_callchain) && sample->callchain) {
65 err = perf_session__resolve_callchain(session, evsel, al->thread,
66 sample->callchain, &parent);
67 if (err)
68 return err;
69 }
70
71 he = __hists__add_entry(&evsel->hists, al, parent, sample->period);
72 if (he == NULL)
73 return -ENOMEM;
74
75 if (symbol_conf.use_callchain) {
76 err = callchain_append(he->callchain,
77 &evsel->hists.callchain_cursor,
78 sample->period);
79 if (err)
80 return err;
81 }
82 /*
83 * Only in the newt browser we are doing integrated annotation,
84 * so we don't allocated the extra space needed because the stdio
85 * code will not use it.
86 */
87 if (al->sym != NULL && use_browser > 0) {
88 struct annotation *notes = symbol__annotation(he->ms.sym);
89
90 assert(evsel != NULL);
91
92 err = -ENOMEM;
93 if (notes->src == NULL && symbol__alloc_hist(he->ms.sym) < 0)
94 goto out;
95
96 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
97 }
98
99 evsel->hists.stats.total_period += sample->period;
100 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
101 out:
102 return err;
103 }
104
105
106 static int process_sample_event(struct perf_event_ops *ops,
107 union perf_event *event,
108 struct perf_sample *sample,
109 struct perf_evsel *evsel,
110 struct perf_session *session)
111 {
112 struct perf_report *rep = container_of(ops, struct perf_report, ops);
113 struct addr_location al;
114
115 if (perf_event__preprocess_sample(event, session, &al, sample,
116 rep->annotate_init) < 0) {
117 fprintf(stderr, "problem processing %d event, skipping it.\n",
118 event->header.type);
119 return -1;
120 }
121
122 if (al.filtered || (rep->hide_unresolved && al.sym == NULL))
123 return 0;
124
125 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
126 return 0;
127
128 if (al.map != NULL)
129 al.map->dso->hit = 1;
130
131 if (perf_session__add_hist_entry(session, &al, sample, evsel)) {
132 pr_debug("problem incrementing symbol period, skipping event\n");
133 return -1;
134 }
135
136 return 0;
137 }
138
139 static int process_read_event(struct perf_event_ops *ops,
140 union perf_event *event,
141 struct perf_sample *sample __used,
142 struct perf_session *session)
143 {
144 struct perf_report *rep = container_of(ops, struct perf_report, ops);
145 struct perf_evsel *evsel = perf_evlist__id2evsel(session->evlist,
146 event->read.id);
147 if (rep->show_threads) {
148 const char *name = evsel ? event_name(evsel) : "unknown";
149 perf_read_values_add_value(&rep->show_threads_values,
150 event->read.pid, event->read.tid,
151 event->read.id,
152 name,
153 event->read.value);
154 }
155
156 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
157 evsel ? event_name(evsel) : "FAIL",
158 event->read.value);
159
160 return 0;
161 }
162
163 static int perf_report__setup_sample_type(struct perf_report *rep)
164 {
165 struct perf_session *self = rep->session;
166
167 if (!(self->sample_type & PERF_SAMPLE_CALLCHAIN)) {
168 if (sort__has_parent) {
169 ui__warning("Selected --sort parent, but no "
170 "callchain data. Did you call "
171 "'perf record' without -g?\n");
172 return -EINVAL;
173 }
174 if (symbol_conf.use_callchain) {
175 ui__warning("Selected -g but no callchain data. Did "
176 "you call 'perf record' without -g?\n");
177 return -1;
178 }
179 } else if (!rep->dont_use_callchains &&
180 callchain_param.mode != CHAIN_NONE &&
181 !symbol_conf.use_callchain) {
182 symbol_conf.use_callchain = true;
183 if (callchain_register_param(&callchain_param) < 0) {
184 ui__warning("Can't register callchain "
185 "params.\n");
186 return -EINVAL;
187 }
188 }
189
190 return 0;
191 }
192
193 extern volatile int session_done;
194
195 static void sig_handler(int sig __used)
196 {
197 session_done = 1;
198 }
199
200 static size_t hists__fprintf_nr_sample_events(struct hists *self,
201 const char *evname, FILE *fp)
202 {
203 size_t ret;
204 char unit;
205 unsigned long nr_events = self->stats.nr_events[PERF_RECORD_SAMPLE];
206
207 nr_events = convert_unit(nr_events, &unit);
208 ret = fprintf(fp, "# Events: %lu%c", nr_events, unit);
209 if (evname != NULL)
210 ret += fprintf(fp, " %s", evname);
211 return ret + fprintf(fp, "\n#\n");
212 }
213
214 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
215 struct perf_report *rep,
216 const char *help)
217 {
218 struct perf_evsel *pos;
219
220 list_for_each_entry(pos, &evlist->entries, node) {
221 struct hists *hists = &pos->hists;
222 const char *evname = event_name(pos);
223
224 hists__fprintf_nr_sample_events(hists, evname, stdout);
225 hists__fprintf(hists, NULL, false, true, 0, 0, stdout);
226 fprintf(stdout, "\n\n");
227 }
228
229 if (sort_order == default_sort_order &&
230 parent_pattern == default_parent_pattern) {
231 fprintf(stdout, "#\n# (%s)\n#\n", help);
232
233 if (rep->show_threads) {
234 bool style = !strcmp(rep->pretty_printing_style, "raw");
235 perf_read_values_display(stdout, &rep->show_threads_values,
236 style);
237 perf_read_values_destroy(&rep->show_threads_values);
238 }
239 }
240
241 return 0;
242 }
243
244 static int __cmd_report(struct perf_report *rep)
245 {
246 int ret = -EINVAL;
247 u64 nr_samples;
248 struct perf_session *session;
249 struct perf_evsel *pos;
250 struct map *kernel_map;
251 struct kmap *kernel_kmap;
252 const char *help = "For a higher level overview, try: perf report --sort comm,dso";
253
254 signal(SIGINT, sig_handler);
255
256 session = perf_session__new(rep->input_name, O_RDONLY,
257 rep->force, false, &rep->ops);
258 if (session == NULL)
259 return -ENOMEM;
260
261 rep->session = session;
262
263 if (rep->cpu_list) {
264 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
265 rep->cpu_bitmap);
266 if (ret)
267 goto out_delete;
268 }
269
270 if (use_browser <= 0)
271 perf_session__fprintf_info(session, stdout, rep->show_full_info);
272
273 if (rep->show_threads)
274 perf_read_values_init(&rep->show_threads_values);
275
276 ret = perf_report__setup_sample_type(rep);
277 if (ret)
278 goto out_delete;
279
280 ret = perf_session__process_events(session, &rep->ops);
281 if (ret)
282 goto out_delete;
283
284 kernel_map = session->host_machine.vmlinux_maps[MAP__FUNCTION];
285 kernel_kmap = map__kmap(kernel_map);
286 if (kernel_map == NULL ||
287 (kernel_map->dso->hit &&
288 (kernel_kmap->ref_reloc_sym == NULL ||
289 kernel_kmap->ref_reloc_sym->addr == 0))) {
290 const struct dso *kdso = kernel_map->dso;
291
292 ui__warning(
293 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
294 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
295 "Samples in kernel modules can't be resolved as well.\n\n",
296 RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION]) ?
297 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
298 "can't be resolved." :
299 "If some relocation was applied (e.g. kexec) symbols may be misresolved.");
300 }
301
302 if (dump_trace) {
303 perf_session__fprintf_nr_events(session, stdout);
304 goto out_delete;
305 }
306
307 if (verbose > 3)
308 perf_session__fprintf(session, stdout);
309
310 if (verbose > 2)
311 perf_session__fprintf_dsos(session, stdout);
312
313 nr_samples = 0;
314 list_for_each_entry(pos, &session->evlist->entries, node) {
315 struct hists *hists = &pos->hists;
316
317 hists__collapse_resort(hists);
318 hists__output_resort(hists);
319 nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE];
320 }
321
322 if (nr_samples == 0) {
323 ui__warning("The %s file has no samples!\n",
324 rep->input_name);
325 goto out_delete;
326 }
327
328 if (use_browser > 0) {
329 perf_evlist__tui_browse_hists(session->evlist, help,
330 NULL, NULL, 0);
331 } else
332 perf_evlist__tty_browse_hists(session->evlist, rep, help);
333
334 out_delete:
335 /*
336 * Speed up the exit process, for large files this can
337 * take quite a while.
338 *
339 * XXX Enable this when using valgrind or if we ever
340 * librarize this command.
341 *
342 * Also experiment with obstacks to see how much speed
343 * up we'll get here.
344 *
345 * perf_session__delete(session);
346 */
347 return ret;
348 }
349
350 static int
351 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
352 {
353 struct perf_report *rep = (struct perf_report *)opt->value;
354 char *tok, *tok2;
355 char *endptr;
356
357 /*
358 * --no-call-graph
359 */
360 if (unset) {
361 rep->dont_use_callchains = true;
362 return 0;
363 }
364
365 symbol_conf.use_callchain = true;
366
367 if (!arg)
368 return 0;
369
370 tok = strtok((char *)arg, ",");
371 if (!tok)
372 return -1;
373
374 /* get the output mode */
375 if (!strncmp(tok, "graph", strlen(arg)))
376 callchain_param.mode = CHAIN_GRAPH_ABS;
377
378 else if (!strncmp(tok, "flat", strlen(arg)))
379 callchain_param.mode = CHAIN_FLAT;
380
381 else if (!strncmp(tok, "fractal", strlen(arg)))
382 callchain_param.mode = CHAIN_GRAPH_REL;
383
384 else if (!strncmp(tok, "none", strlen(arg))) {
385 callchain_param.mode = CHAIN_NONE;
386 symbol_conf.use_callchain = false;
387
388 return 0;
389 }
390
391 else
392 return -1;
393
394 /* get the min percentage */
395 tok = strtok(NULL, ",");
396 if (!tok)
397 goto setup;
398
399 callchain_param.min_percent = strtod(tok, &endptr);
400 if (tok == endptr)
401 return -1;
402
403 /* get the print limit */
404 tok2 = strtok(NULL, ",");
405 if (!tok2)
406 goto setup;
407
408 if (tok2[0] != 'c') {
409 callchain_param.print_limit = strtod(tok2, &endptr);
410 tok2 = strtok(NULL, ",");
411 if (!tok2)
412 goto setup;
413 }
414
415 /* get the call chain order */
416 if (!strcmp(tok2, "caller"))
417 callchain_param.order = ORDER_CALLER;
418 else if (!strcmp(tok2, "callee"))
419 callchain_param.order = ORDER_CALLEE;
420 else
421 return -1;
422 setup:
423 if (callchain_register_param(&callchain_param) < 0) {
424 fprintf(stderr, "Can't register callchain params\n");
425 return -1;
426 }
427 return 0;
428 }
429
430 int cmd_report(int argc, const char **argv, const char *prefix __used)
431 {
432 char callchain_default_opt[] = "fractal,0.5,callee";
433 const char * const report_usage[] = {
434 "perf report [<options>] <command>",
435 NULL
436 };
437 struct perf_report report = {
438 .ops = {
439 .sample = process_sample_event,
440 .mmap = perf_event__process_mmap,
441 .comm = perf_event__process_comm,
442 .exit = perf_event__process_task,
443 .fork = perf_event__process_task,
444 .lost = perf_event__process_lost,
445 .read = process_read_event,
446 .attr = perf_event__process_attr,
447 .event_type = perf_event__process_event_type,
448 .tracing_data = perf_event__process_tracing_data,
449 .build_id = perf_event__process_build_id,
450 .ordered_samples = true,
451 .ordering_requires_timestamps = true,
452 },
453 .input_name = "perf.data",
454 .pretty_printing_style = "normal",
455 };
456 const struct option options[] = {
457 OPT_STRING('i', "input", &report.input_name, "file",
458 "input file name"),
459 OPT_INCR('v', "verbose", &verbose,
460 "be more verbose (show symbol address, etc)"),
461 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
462 "dump raw trace in ASCII"),
463 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
464 "file", "vmlinux pathname"),
465 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
466 "file", "kallsyms pathname"),
467 OPT_BOOLEAN('f', "force", &report.force, "don't complain, do it"),
468 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
469 "load module symbols - WARNING: use only with -k and LIVE kernel"),
470 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
471 "Show a column with the number of samples"),
472 OPT_BOOLEAN('T', "threads", &report.show_threads,
473 "Show per-thread event counters"),
474 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
475 "pretty printing style key: normal raw"),
476 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
477 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
478 "Use the stdio interface"),
479 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
480 "sort by key(s): pid, comm, dso, symbol, parent"),
481 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
482 "Show sample percentage for different cpu modes"),
483 OPT_STRING('p', "parent", &parent_pattern, "regex",
484 "regex filter to identify parent, see: '--sort parent'"),
485 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
486 "Only display entries with parent-match"),
487 OPT_CALLBACK_DEFAULT('g', "call-graph", &report, "output_type,min_percent, call_order",
488 "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold and callchain order. "
489 "Default: fractal,0.5,callee", &parse_callchain_opt, callchain_default_opt),
490 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
491 "alias for inverted call graph"),
492 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
493 "only consider symbols in these dsos"),
494 OPT_STRING('C', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
495 "only consider symbols in these comms"),
496 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
497 "only consider these symbols"),
498 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
499 "width[,width...]",
500 "don't try to adjust column width, use these fixed values"),
501 OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
502 "separator for columns, no spaces will be added between "
503 "columns '.' is reserved."),
504 OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved,
505 "Only display entries resolved to a symbol"),
506 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
507 "Look for files with symbols relative to this directory"),
508 OPT_STRING('c', "cpu", &report.cpu_list, "cpu",
509 "list of cpus to profile"),
510 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
511 "Display extended information about perf.data file"),
512 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
513 "Interleave source code with assembly code (default)"),
514 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
515 "Display raw encoding of assembly instructions (default)"),
516 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
517 "Specify disassembler style (e.g. -M intel for intel syntax)"),
518 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
519 "Show a column with the sum of periods"),
520 OPT_END()
521 };
522
523 argc = parse_options(argc, argv, options, report_usage, 0);
524
525 if (report.use_stdio)
526 use_browser = 0;
527 else if (report.use_tui)
528 use_browser = 1;
529
530 if (report.inverted_callchain)
531 callchain_param.order = ORDER_CALLER;
532
533 if (strcmp(report.input_name, "-") != 0)
534 setup_browser(true);
535 else
536 use_browser = 0;
537 /*
538 * Only in the newt browser we are doing integrated annotation,
539 * so don't allocate extra space that won't be used in the stdio
540 * implementation.
541 */
542 if (use_browser > 0) {
543 symbol_conf.priv_size = sizeof(struct annotation);
544 report.annotate_init = symbol__annotate_init;
545 /*
546 * For searching by name on the "Browse map details".
547 * providing it only in verbose mode not to bloat too
548 * much struct symbol.
549 */
550 if (verbose) {
551 /*
552 * XXX: Need to provide a less kludgy way to ask for
553 * more space per symbol, the u32 is for the index on
554 * the ui browser.
555 * See symbol__browser_index.
556 */
557 symbol_conf.priv_size += sizeof(u32);
558 symbol_conf.sort_by_name = true;
559 }
560 }
561
562 if (symbol__init() < 0)
563 return -1;
564
565 setup_sorting(report_usage, options);
566
567 if (parent_pattern != default_parent_pattern) {
568 if (sort_dimension__add("parent") < 0)
569 return -1;
570
571 /*
572 * Only show the parent fields if we explicitly
573 * sort that way. If we only use parent machinery
574 * for filtering, we don't want it.
575 */
576 if (!strstr(sort_order, "parent"))
577 sort_parent.elide = 1;
578 } else
579 symbol_conf.exclude_other = false;
580
581 /*
582 * Any (unrecognized) arguments left?
583 */
584 if (argc)
585 usage_with_options(report_usage, options);
586
587 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "dso", stdout);
588 sort_entry__setup_elide(&sort_comm, symbol_conf.comm_list, "comm", stdout);
589 sort_entry__setup_elide(&sort_sym, symbol_conf.sym_list, "symbol", stdout);
590
591 return __cmd_report(&report);
592 }
This page took 0.059385 seconds and 6 git commands to generate.