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/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"
23 #include "util/debug.h"
24 #include "util/header.h"
25 #include "util/session.h"
27 #include "util/parse-options.h"
28 #include "util/parse-events.h"
30 #include "util/thread.h"
31 #include "util/sort.h"
32 #include "util/hist.h"
34 static char const *input_name
= "perf.data";
36 static bool force
, use_tui
, use_stdio
;
37 static bool hide_unresolved
;
38 static bool dont_use_callchains
;
40 static bool show_threads
;
41 static struct perf_read_values show_threads_values
;
43 static const char default_pretty_printing_style
[] = "normal";
44 static const char *pretty_printing_style
= default_pretty_printing_style
;
46 static char callchain_default_opt
[] = "fractal,0.5";
48 static struct hists
*perf_session__hists_findnew(struct perf_session
*self
,
49 u64 event_stream
, u32 type
,
52 struct rb_node
**p
= &self
->hists_tree
.rb_node
;
53 struct rb_node
*parent
= NULL
;
54 struct hists
*iter
, *new;
58 iter
= rb_entry(parent
, struct hists
, rb_node
);
59 if (iter
->config
== config
)
63 if (config
> iter
->config
)
69 new = malloc(sizeof(struct hists
));
72 memset(new, 0, sizeof(struct hists
));
73 new->event_stream
= event_stream
;
76 rb_link_node(&new->rb_node
, parent
, p
);
77 rb_insert_color(&new->rb_node
, &self
->hists_tree
);
81 static int perf_session__add_hist_entry(struct perf_session
*session
,
82 struct addr_location
*al
,
83 struct perf_sample
*sample
)
85 struct symbol
*parent
= NULL
;
87 struct hist_entry
*he
;
89 struct perf_event_attr
*attr
;
91 if ((sort__has_parent
|| symbol_conf
.use_callchain
) && sample
->callchain
) {
92 err
= perf_session__resolve_callchain(session
, al
->thread
,
93 sample
->callchain
, &parent
);
98 attr
= perf_header__find_attr(sample
->id
, &session
->header
);
100 hists
= perf_session__hists_findnew(session
, sample
->id
, attr
->type
, attr
->config
);
102 hists
= perf_session__hists_findnew(session
, sample
->id
, 0, 0);
106 he
= __hists__add_entry(hists
, al
, parent
, sample
->period
);
110 if (symbol_conf
.use_callchain
) {
111 err
= callchain_append(he
->callchain
, &session
->callchain_cursor
,
117 * Only in the newt browser we are doing integrated annotation,
118 * so we don't allocated the extra space needed because the stdio
119 * code will not use it.
121 if (al
->sym
!= NULL
&& use_browser
> 0) {
123 * All aggregated on the first sym_hist.
125 struct annotation
*notes
= symbol__annotation(he
->ms
.sym
);
126 if (notes
->src
== NULL
&&
127 symbol__alloc_hist(he
->ms
.sym
, 1) < 0)
130 err
= hist_entry__inc_addr_samples(he
, 0, al
->addr
);
136 static int add_event_total(struct perf_session
*session
,
137 struct perf_sample
*sample
,
138 struct perf_event_attr
*attr
)
143 hists
= perf_session__hists_findnew(session
, sample
->id
,
144 attr
->type
, attr
->config
);
146 hists
= perf_session__hists_findnew(session
, sample
->id
, 0, 0);
151 hists
->stats
.total_period
+= sample
->period
;
153 * FIXME: add_event_total should be moved from here to
154 * perf_session__process_event so that the proper hist is passed to
155 * the event_op methods.
157 hists__inc_nr_events(hists
, PERF_RECORD_SAMPLE
);
158 session
->hists
.stats
.total_period
+= sample
->period
;
162 static int process_sample_event(union perf_event
*event
,
163 struct perf_sample
*sample
,
164 struct perf_session
*session
)
166 struct addr_location al
;
167 struct perf_event_attr
*attr
;
169 if (perf_event__preprocess_sample(event
, session
, &al
, sample
,
170 symbol__annotate_init
) < 0) {
171 fprintf(stderr
, "problem processing %d event, skipping it.\n",
176 if (al
.filtered
|| (hide_unresolved
&& al
.sym
== NULL
))
179 if (perf_session__add_hist_entry(session
, &al
, sample
)) {
180 pr_debug("problem incrementing symbol period, skipping event\n");
184 attr
= perf_header__find_attr(sample
->id
, &session
->header
);
186 if (add_event_total(session
, sample
, attr
)) {
187 pr_debug("problem adding event period\n");
194 static int process_read_event(union perf_event
*event
,
195 struct perf_sample
*sample __used
,
196 struct perf_session
*session __used
)
198 struct perf_event_attr
*attr
;
200 attr
= perf_header__find_attr(event
->read
.id
, &session
->header
);
203 const char *name
= attr
? __event_name(attr
->type
, attr
->config
)
205 perf_read_values_add_value(&show_threads_values
,
206 event
->read
.pid
, event
->read
.tid
,
212 dump_printf(": %d %d %s %" PRIu64
"\n", event
->read
.pid
, event
->read
.tid
,
213 attr
? __event_name(attr
->type
, attr
->config
) : "FAIL",
219 static int perf_session__setup_sample_type(struct perf_session
*self
)
221 if (!(self
->sample_type
& PERF_SAMPLE_CALLCHAIN
)) {
222 if (sort__has_parent
) {
223 fprintf(stderr
, "selected --sort parent, but no"
224 " callchain data. Did you call"
225 " perf record without -g?\n");
228 if (symbol_conf
.use_callchain
) {
229 fprintf(stderr
, "selected -g but no callchain data."
230 " Did you call perf record without"
234 } else if (!dont_use_callchains
&& callchain_param
.mode
!= CHAIN_NONE
&&
235 !symbol_conf
.use_callchain
) {
236 symbol_conf
.use_callchain
= true;
237 if (callchain_register_param(&callchain_param
) < 0) {
238 fprintf(stderr
, "Can't register callchain"
247 static struct perf_event_ops event_ops
= {
248 .sample
= process_sample_event
,
249 .mmap
= perf_event__process_mmap
,
250 .comm
= perf_event__process_comm
,
251 .exit
= perf_event__process_task
,
252 .fork
= perf_event__process_task
,
253 .lost
= perf_event__process_lost
,
254 .read
= process_read_event
,
255 .attr
= perf_event__process_attr
,
256 .event_type
= perf_event__process_event_type
,
257 .tracing_data
= perf_event__process_tracing_data
,
258 .build_id
= perf_event__process_build_id
,
259 .ordered_samples
= true,
260 .ordering_requires_timestamps
= true,
263 extern volatile int session_done
;
265 static void sig_handler(int sig __used
)
270 static size_t hists__fprintf_nr_sample_events(struct hists
*self
,
271 const char *evname
, FILE *fp
)
275 unsigned long nr_events
= self
->stats
.nr_events
[PERF_RECORD_SAMPLE
];
277 nr_events
= convert_unit(nr_events
, &unit
);
278 ret
= fprintf(fp
, "# Events: %lu%c", nr_events
, unit
);
280 ret
+= fprintf(fp
, " %s", evname
);
281 return ret
+ fprintf(fp
, "\n#\n");
284 static int hists__tty_browse_tree(struct rb_root
*tree
, const char *help
)
286 struct rb_node
*next
= rb_first(tree
);
289 struct hists
*hists
= rb_entry(next
, struct hists
, rb_node
);
290 const char *evname
= NULL
;
292 if (rb_first(&hists
->entries
) != rb_last(&hists
->entries
))
293 evname
= __event_name(hists
->type
, hists
->config
);
295 hists__fprintf_nr_sample_events(hists
, evname
, stdout
);
296 hists__fprintf(hists
, NULL
, false, stdout
);
297 fprintf(stdout
, "\n\n");
298 next
= rb_next(&hists
->rb_node
);
301 if (sort_order
== default_sort_order
&&
302 parent_pattern
== default_parent_pattern
) {
303 fprintf(stdout
, "#\n# (%s)\n#\n", help
);
306 bool style
= !strcmp(pretty_printing_style
, "raw");
307 perf_read_values_display(stdout
, &show_threads_values
,
309 perf_read_values_destroy(&show_threads_values
);
316 static int __cmd_report(void)
319 struct perf_session
*session
;
320 struct rb_node
*next
;
321 const char *help
= "For a higher level overview, try: perf report --sort comm,dso";
323 signal(SIGINT
, sig_handler
);
325 session
= perf_session__new(input_name
, O_RDONLY
, force
, false, &event_ops
);
330 perf_read_values_init(&show_threads_values
);
332 ret
= perf_session__setup_sample_type(session
);
336 ret
= perf_session__process_events(session
, &event_ops
);
341 perf_session__fprintf_nr_events(session
, stdout
);
346 perf_session__fprintf(session
, stdout
);
349 perf_session__fprintf_dsos(session
, stdout
);
351 next
= rb_first(&session
->hists_tree
);
355 hists
= rb_entry(next
, struct hists
, rb_node
);
356 hists__collapse_resort(hists
);
357 hists__output_resort(hists
);
358 next
= rb_next(&hists
->rb_node
);
362 hists__tui_browse_tree(&session
->hists_tree
, help
, 0);
364 hists__tty_browse_tree(&session
->hists_tree
, help
);
368 * Speed up the exit process, for large files this can
369 * take quite a while.
371 * XXX Enable this when using valgrind or if we ever
372 * librarize this command.
374 * Also experiment with obstacks to see how much speed
377 * perf_session__delete(session);
383 parse_callchain_opt(const struct option
*opt __used
, const char *arg
,
393 dont_use_callchains
= true;
397 symbol_conf
.use_callchain
= true;
402 tok
= strtok((char *)arg
, ",");
406 /* get the output mode */
407 if (!strncmp(tok
, "graph", strlen(arg
)))
408 callchain_param
.mode
= CHAIN_GRAPH_ABS
;
410 else if (!strncmp(tok
, "flat", strlen(arg
)))
411 callchain_param
.mode
= CHAIN_FLAT
;
413 else if (!strncmp(tok
, "fractal", strlen(arg
)))
414 callchain_param
.mode
= CHAIN_GRAPH_REL
;
416 else if (!strncmp(tok
, "none", strlen(arg
))) {
417 callchain_param
.mode
= CHAIN_NONE
;
418 symbol_conf
.use_callchain
= false;
426 /* get the min percentage */
427 tok
= strtok(NULL
, ",");
431 tok2
= strtok(NULL
, ",");
432 callchain_param
.min_percent
= strtod(tok
, &endptr
);
437 callchain_param
.print_limit
= strtod(tok2
, &endptr
);
439 if (callchain_register_param(&callchain_param
) < 0) {
440 fprintf(stderr
, "Can't register callchain params\n");
446 static const char * const report_usage
[] = {
447 "perf report [<options>] <command>",
451 static const struct option options
[] = {
452 OPT_STRING('i', "input", &input_name
, "file",
454 OPT_INCR('v', "verbose", &verbose
,
455 "be more verbose (show symbol address, etc)"),
456 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace
,
457 "dump raw trace in ASCII"),
458 OPT_STRING('k', "vmlinux", &symbol_conf
.vmlinux_name
,
459 "file", "vmlinux pathname"),
460 OPT_STRING(0, "kallsyms", &symbol_conf
.kallsyms_name
,
461 "file", "kallsyms pathname"),
462 OPT_BOOLEAN('f', "force", &force
, "don't complain, do it"),
463 OPT_BOOLEAN('m', "modules", &symbol_conf
.use_modules
,
464 "load module symbols - WARNING: use only with -k and LIVE kernel"),
465 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf
.show_nr_samples
,
466 "Show a column with the number of samples"),
467 OPT_BOOLEAN('T', "threads", &show_threads
,
468 "Show per-thread event counters"),
469 OPT_STRING(0, "pretty", &pretty_printing_style
, "key",
470 "pretty printing style key: normal raw"),
471 OPT_BOOLEAN(0, "tui", &use_tui
, "Use the TUI interface"),
472 OPT_BOOLEAN(0, "stdio", &use_stdio
, "Use the stdio interface"),
473 OPT_STRING('s', "sort", &sort_order
, "key[,key2...]",
474 "sort by key(s): pid, comm, dso, symbol, parent"),
475 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf
.show_cpu_utilization
,
476 "Show sample percentage for different cpu modes"),
477 OPT_STRING('p', "parent", &parent_pattern
, "regex",
478 "regex filter to identify parent, see: '--sort parent'"),
479 OPT_BOOLEAN('x', "exclude-other", &symbol_conf
.exclude_other
,
480 "Only display entries with parent-match"),
481 OPT_CALLBACK_DEFAULT('g', "call-graph", NULL
, "output_type,min_percent",
482 "Display callchains using output_type (graph, flat, fractal, or none) and min percent threshold. "
483 "Default: fractal,0.5", &parse_callchain_opt
, callchain_default_opt
),
484 OPT_STRING('d', "dsos", &symbol_conf
.dso_list_str
, "dso[,dso...]",
485 "only consider symbols in these dsos"),
486 OPT_STRING('C', "comms", &symbol_conf
.comm_list_str
, "comm[,comm...]",
487 "only consider symbols in these comms"),
488 OPT_STRING('S', "symbols", &symbol_conf
.sym_list_str
, "symbol[,symbol...]",
489 "only consider these symbols"),
490 OPT_STRING('w', "column-widths", &symbol_conf
.col_width_list_str
,
492 "don't try to adjust column width, use these fixed values"),
493 OPT_STRING('t', "field-separator", &symbol_conf
.field_sep
, "separator",
494 "separator for columns, no spaces will be added between "
495 "columns '.' is reserved."),
496 OPT_BOOLEAN('U', "hide-unresolved", &hide_unresolved
,
497 "Only display entries resolved to a symbol"),
498 OPT_STRING(0, "symfs", &symbol_conf
.symfs
, "directory",
499 "Look for files with symbols relative to this directory"),
503 int cmd_report(int argc
, const char **argv
, const char *prefix __used
)
505 argc
= parse_options(argc
, argv
, options
, report_usage
, 0);
512 if (strcmp(input_name
, "-") != 0)
517 * Only in the newt browser we are doing integrated annotation,
518 * so don't allocate extra space that won't be used in the stdio
521 if (use_browser
> 0) {
522 symbol_conf
.priv_size
= sizeof(struct annotation
);
524 * For searching by name on the "Browse map details".
525 * providing it only in verbose mode not to bloat too
526 * much struct symbol.
530 * XXX: Need to provide a less kludgy way to ask for
531 * more space per symbol, the u32 is for the index on
533 * See symbol__browser_index.
535 symbol_conf
.priv_size
+= sizeof(u32
);
536 symbol_conf
.sort_by_name
= true;
540 if (symbol__init() < 0)
543 setup_sorting(report_usage
, options
);
545 if (parent_pattern
!= default_parent_pattern
) {
546 if (sort_dimension__add("parent") < 0)
548 sort_parent
.elide
= 1;
550 symbol_conf
.exclude_other
= false;
553 * Any (unrecognized) arguments left?
556 usage_with_options(report_usage
, options
);
558 sort_entry__setup_elide(&sort_dso
, symbol_conf
.dso_list
, "dso", stdout
);
559 sort_entry__setup_elide(&sort_comm
, symbol_conf
.comm_list
, "comm", stdout
);
560 sort_entry__setup_elide(&sort_sym
, symbol_conf
.sym_list
, "symbol", stdout
);
562 return __cmd_report();