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.
10 #include "util/util.h"
12 #include "util/color.h"
13 #include <linux/list.h>
14 #include "util/cache.h"
15 #include <linux/rbtree.h>
16 #include "util/symbol.h"
17 #include "util/callchain.h"
18 #include "util/strlist.h"
19 #include "util/values.h"
22 #include "util/debug.h"
23 #include "util/header.h"
24 #include "util/session.h"
26 #include "util/parse-options.h"
27 #include "util/parse-events.h"
29 #include "util/thread.h"
30 #include "util/sort.h"
31 #include "util/hist.h"
33 static char const *input_name
= "perf.data";
35 static bool force
, use_tui
, use_stdio
;
36 static bool hide_unresolved
;
37 static bool dont_use_callchains
;
39 static bool show_threads
;
40 static struct perf_read_values show_threads_values
;
42 static const char default_pretty_printing_style
[] = "normal";
43 static const char *pretty_printing_style
= default_pretty_printing_style
;
45 static char callchain_default_opt
[] = "fractal,0.5";
47 static struct hists
*perf_session__hists_findnew(struct perf_session
*self
,
48 u64 event_stream
, u32 type
,
51 struct rb_node
**p
= &self
->hists_tree
.rb_node
;
52 struct rb_node
*parent
= NULL
;
53 struct hists
*iter
, *new;
57 iter
= rb_entry(parent
, struct hists
, rb_node
);
58 if (iter
->config
== config
)
62 if (config
> iter
->config
)
68 new = malloc(sizeof(struct hists
));
71 memset(new, 0, sizeof(struct hists
));
72 new->event_stream
= event_stream
;
75 rb_link_node(&new->rb_node
, parent
, p
);
76 rb_insert_color(&new->rb_node
, &self
->hists_tree
);
80 static int perf_session__add_hist_entry(struct perf_session
*session
,
81 struct addr_location
*al
,
82 struct perf_sample
*sample
)
84 struct symbol
*parent
= NULL
;
86 struct hist_entry
*he
;
88 struct perf_event_attr
*attr
;
90 if ((sort__has_parent
|| symbol_conf
.use_callchain
) && sample
->callchain
) {
91 err
= perf_session__resolve_callchain(session
, al
->thread
,
92 sample
->callchain
, &parent
);
97 attr
= perf_header__find_attr(sample
->id
, &session
->header
);
99 hists
= perf_session__hists_findnew(session
, sample
->id
, attr
->type
, attr
->config
);
101 hists
= perf_session__hists_findnew(session
, sample
->id
, 0, 0);
105 he
= __hists__add_entry(hists
, al
, parent
, sample
->period
);
109 if (symbol_conf
.use_callchain
) {
110 err
= callchain_append(he
->callchain
, &session
->callchain_cursor
,
116 * Only in the newt browser we are doing integrated annotation,
117 * so we don't allocated the extra space needed because the stdio
118 * code will not use it.
121 err
= hist_entry__inc_addr_samples(he
, al
->addr
);
126 static int add_event_total(struct perf_session
*session
,
127 struct perf_sample
*sample
,
128 struct perf_event_attr
*attr
)
133 hists
= perf_session__hists_findnew(session
, sample
->id
,
134 attr
->type
, attr
->config
);
136 hists
= perf_session__hists_findnew(session
, sample
->id
, 0, 0);
141 hists
->stats
.total_period
+= sample
->period
;
143 * FIXME: add_event_total should be moved from here to
144 * perf_session__process_event so that the proper hist is passed to
145 * the event_op methods.
147 hists__inc_nr_events(hists
, PERF_RECORD_SAMPLE
);
148 session
->hists
.stats
.total_period
+= sample
->period
;
152 static int process_sample_event(event_t
*event
, struct perf_sample
*sample
,
153 struct perf_session
*session
)
155 struct addr_location al
;
156 struct perf_event_attr
*attr
;
158 if (event__preprocess_sample(event
, session
, &al
, sample
, NULL
) < 0) {
159 fprintf(stderr
, "problem processing %d event, skipping it.\n",
164 if (al
.filtered
|| (hide_unresolved
&& al
.sym
== NULL
))
167 if (perf_session__add_hist_entry(session
, &al
, sample
)) {
168 pr_debug("problem incrementing symbol period, skipping event\n");
172 attr
= perf_header__find_attr(sample
->id
, &session
->header
);
174 if (add_event_total(session
, sample
, attr
)) {
175 pr_debug("problem adding event period\n");
182 static int process_read_event(event_t
*event
, struct perf_sample
*sample __used
,
183 struct perf_session
*session __used
)
185 struct perf_event_attr
*attr
;
187 attr
= perf_header__find_attr(event
->read
.id
, &session
->header
);
190 const char *name
= attr
? __event_name(attr
->type
, attr
->config
)
192 perf_read_values_add_value(&show_threads_values
,
193 event
->read
.pid
, event
->read
.tid
,
199 dump_printf(": %d %d %s %" PRIu64
"\n", event
->read
.pid
, event
->read
.tid
,
200 attr
? __event_name(attr
->type
, attr
->config
) : "FAIL",
206 static int perf_session__setup_sample_type(struct perf_session
*self
)
208 if (!(self
->sample_type
& PERF_SAMPLE_CALLCHAIN
)) {
209 if (sort__has_parent
) {
210 fprintf(stderr
, "selected --sort parent, but no"
211 " callchain data. Did you call"
212 " perf record without -g?\n");
215 if (symbol_conf
.use_callchain
) {
216 fprintf(stderr
, "selected -g but no callchain data."
217 " Did you call perf record without"
221 } else if (!dont_use_callchains
&& callchain_param
.mode
!= CHAIN_NONE
&&
222 !symbol_conf
.use_callchain
) {
223 symbol_conf
.use_callchain
= true;
224 if (callchain_register_param(&callchain_param
) < 0) {
225 fprintf(stderr
, "Can't register callchain"
234 static struct perf_event_ops event_ops
= {
235 .sample
= process_sample_event
,
236 .mmap
= event__process_mmap
,
237 .comm
= event__process_comm
,
238 .exit
= event__process_task
,
239 .fork
= event__process_task
,
240 .lost
= event__process_lost
,
241 .read
= process_read_event
,
242 .attr
= event__process_attr
,
243 .event_type
= event__process_event_type
,
244 .tracing_data
= event__process_tracing_data
,
245 .build_id
= event__process_build_id
,
246 .ordered_samples
= true,
247 .ordering_requires_timestamps
= true,
250 extern volatile int session_done
;
252 static void sig_handler(int sig __used
)
257 static size_t hists__fprintf_nr_sample_events(struct hists
*self
,
258 const char *evname
, FILE *fp
)
262 unsigned long nr_events
= self
->stats
.nr_events
[PERF_RECORD_SAMPLE
];
264 nr_events
= convert_unit(nr_events
, &unit
);
265 ret
= fprintf(fp
, "# Events: %lu%c", nr_events
, unit
);
267 ret
+= fprintf(fp
, " %s", evname
);
268 return ret
+ fprintf(fp
, "\n#\n");
271 static int hists__tty_browse_tree(struct rb_root
*tree
, const char *help
)
273 struct rb_node
*next
= rb_first(tree
);
276 struct hists
*hists
= rb_entry(next
, struct hists
, rb_node
);
277 const char *evname
= NULL
;
279 if (rb_first(&hists
->entries
) != rb_last(&hists
->entries
))
280 evname
= __event_name(hists
->type
, hists
->config
);
282 hists__fprintf_nr_sample_events(hists
, evname
, stdout
);
283 hists__fprintf(hists
, NULL
, false, stdout
);
284 fprintf(stdout
, "\n\n");
285 next
= rb_next(&hists
->rb_node
);
288 if (sort_order
== default_sort_order
&&
289 parent_pattern
== default_parent_pattern
) {
290 fprintf(stdout
, "#\n# (%s)\n#\n", help
);
293 bool style
= !strcmp(pretty_printing_style
, "raw");
294 perf_read_values_display(stdout
, &show_threads_values
,
296 perf_read_values_destroy(&show_threads_values
);
303 static int __cmd_report(void)
306 struct perf_session
*session
;
307 struct rb_node
*next
;
308 const char *help
= "For a higher level overview, try: perf report --sort comm,dso";
310 signal(SIGINT
, sig_handler
);
312 session
= perf_session__new(input_name
, O_RDONLY
, force
, false, &event_ops
);
317 perf_read_values_init(&show_threads_values
);
319 ret
= perf_session__setup_sample_type(session
);
323 ret
= perf_session__process_events(session
, &event_ops
);
328 perf_session__fprintf_nr_events(session
, stdout
);
333 perf_session__fprintf(session
, stdout
);
336 perf_session__fprintf_dsos(session
, stdout
);
338 next
= rb_first(&session
->hists_tree
);
342 hists
= rb_entry(next
, struct hists
, rb_node
);
343 hists__collapse_resort(hists
);
344 hists__output_resort(hists
);
345 next
= rb_next(&hists
->rb_node
);
349 hists__tui_browse_tree(&session
->hists_tree
, help
);
351 hists__tty_browse_tree(&session
->hists_tree
, help
);
355 * Speed up the exit process, for large files this can
356 * take quite a while.
358 * XXX Enable this when using valgrind or if we ever
359 * librarize this command.
361 * Also experiment with obstacks to see how much speed
364 * perf_session__delete(session);
370 parse_callchain_opt(const struct option
*opt __used
, const char *arg
,
380 dont_use_callchains
= true;
384 symbol_conf
.use_callchain
= true;
389 tok
= strtok((char *)arg
, ",");
393 /* get the output mode */
394 if (!strncmp(tok
, "graph", strlen(arg
)))
395 callchain_param
.mode
= CHAIN_GRAPH_ABS
;
397 else if (!strncmp(tok
, "flat", strlen(arg
)))
398 callchain_param
.mode
= CHAIN_FLAT
;
400 else if (!strncmp(tok
, "fractal", strlen(arg
)))
401 callchain_param
.mode
= CHAIN_GRAPH_REL
;
403 else if (!strncmp(tok
, "none", strlen(arg
))) {
404 callchain_param
.mode
= CHAIN_NONE
;
405 symbol_conf
.use_callchain
= false;
413 /* get the min percentage */
414 tok
= strtok(NULL
, ",");
418 tok2
= strtok(NULL
, ",");
419 callchain_param
.min_percent
= strtod(tok
, &endptr
);
424 callchain_param
.print_limit
= strtod(tok2
, &endptr
);
426 if (callchain_register_param(&callchain_param
) < 0) {
427 fprintf(stderr
, "Can't register callchain params\n");
433 static const char * const report_usage
[] = {
434 "perf report [<options>] <command>",
438 static const struct option options
[] = {
439 OPT_STRING('i', "input", &input_name
, "file",
441 OPT_INCR('v', "verbose", &verbose
,
442 "be more verbose (show symbol address, etc)"),
443 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace
,
444 "dump raw trace in ASCII"),
445 OPT_STRING('k', "vmlinux", &symbol_conf
.vmlinux_name
,
446 "file", "vmlinux pathname"),
447 OPT_STRING(0, "kallsyms", &symbol_conf
.kallsyms_name
,
448 "file", "kallsyms pathname"),
449 OPT_BOOLEAN('f', "force", &force
, "don't complain, do it"),
450 OPT_BOOLEAN('m', "modules", &symbol_conf
.use_modules
,
451 "load module symbols - WARNING: use only with -k and LIVE kernel"),
452 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf
.show_nr_samples
,
453 "Show a column with the number of samples"),
454 OPT_BOOLEAN('T', "threads", &show_threads
,
455 "Show per-thread event counters"),
456 OPT_STRING(0, "pretty", &pretty_printing_style
, "key",
457 "pretty printing style key: normal raw"),
458 OPT_BOOLEAN(0, "tui", &use_tui
, "Use the TUI interface"),
459 OPT_BOOLEAN(0, "stdio", &use_stdio
, "Use the stdio interface"),
460 OPT_STRING('s', "sort", &sort_order
, "key[,key2...]",
461 "sort by key(s): pid, comm, dso, symbol, parent"),
462 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf
.show_cpu_utilization
,
463 "Show sample percentage for different cpu modes"),
464 OPT_STRING('p', "parent", &parent_pattern
, "regex",
465 "regex filter to identify parent, see: '--sort parent'"),
466 OPT_BOOLEAN('x', "exclude-other", &symbol_conf
.exclude_other
,
467 "Only display entries with parent-match"),
468 OPT_CALLBACK_DEFAULT('g', "call-graph", NULL
, "output_type,min_percent",
469 "Display callchains using output_type (graph, flat, fractal, or none) and min percent threshold. "
470 "Default: fractal,0.5", &parse_callchain_opt
, callchain_default_opt
),
471 OPT_STRING('d', "dsos", &symbol_conf
.dso_list_str
, "dso[,dso...]",
472 "only consider symbols in these dsos"),
473 OPT_STRING('C', "comms", &symbol_conf
.comm_list_str
, "comm[,comm...]",
474 "only consider symbols in these comms"),
475 OPT_STRING('S', "symbols", &symbol_conf
.sym_list_str
, "symbol[,symbol...]",
476 "only consider these symbols"),
477 OPT_STRING('w', "column-widths", &symbol_conf
.col_width_list_str
,
479 "don't try to adjust column width, use these fixed values"),
480 OPT_STRING('t', "field-separator", &symbol_conf
.field_sep
, "separator",
481 "separator for columns, no spaces will be added between "
482 "columns '.' is reserved."),
483 OPT_BOOLEAN('U', "hide-unresolved", &hide_unresolved
,
484 "Only display entries resolved to a symbol"),
485 OPT_STRING(0, "symfs", &symbol_conf
.symfs
, "directory",
486 "Look for files with symbols relative to this directory"),
490 int cmd_report(int argc
, const char **argv
, const char *prefix __used
)
492 argc
= parse_options(argc
, argv
, options
, report_usage
, 0);
499 if (strcmp(input_name
, "-") != 0)
504 * Only in the newt browser we are doing integrated annotation,
505 * so don't allocate extra space that won't be used in the stdio
508 if (use_browser
> 0) {
509 symbol_conf
.priv_size
= sizeof(struct sym_priv
);
511 * For searching by name on the "Browse map details".
512 * providing it only in verbose mode not to bloat too
513 * much struct symbol.
517 * XXX: Need to provide a less kludgy way to ask for
518 * more space per symbol, the u32 is for the index on
520 * See symbol__browser_index.
522 symbol_conf
.priv_size
+= sizeof(u32
);
523 symbol_conf
.sort_by_name
= true;
527 if (symbol__init() < 0)
530 setup_sorting(report_usage
, options
);
532 if (parent_pattern
!= default_parent_pattern
) {
533 if (sort_dimension__add("parent") < 0)
535 sort_parent
.elide
= 1;
537 symbol_conf
.exclude_other
= false;
540 * Any (unrecognized) arguments left?
543 usage_with_options(report_usage
, options
);
545 sort_entry__setup_elide(&sort_dso
, symbol_conf
.dso_list
, "dso", stdout
);
546 sort_entry__setup_elide(&sort_comm
, symbol_conf
.comm_list
, "comm", stdout
);
547 sort_entry__setup_elide(&sort_sym
, symbol_conf
.sym_list
, "symbol", stdout
);
549 return __cmd_report();