1 // SPDX-License-Identifier: GPL-2.0
5 * Builtin report command: Analyze the perf.data input file,
6 * look up and read DSOs and symbol information and display
7 * a histogram of results, along various sorting keys.
11 #include "util/util.h"
12 #include "util/config.h"
14 #include "util/annotate.h"
15 #include "util/color.h"
16 #include <linux/list.h>
17 #include <linux/rbtree.h>
18 #include <linux/err.h>
20 #include "util/symbol.h"
21 #include "util/callchain.h"
22 #include "util/values.h"
25 #include "util/debug.h"
26 #include "util/evlist.h"
27 #include "util/evsel.h"
28 #include "util/header.h"
29 #include "util/session.h"
30 #include "util/tool.h"
32 #include <subcmd/parse-options.h>
33 #include <subcmd/exec-cmd.h>
34 #include "util/parse-events.h"
36 #include "util/thread.h"
37 #include "util/sort.h"
38 #include "util/hist.h"
39 #include "util/data.h"
40 #include "arch/common.h"
41 #include "util/time-utils.h"
42 #include "util/auxtrace.h"
43 #include "util/units.h"
44 #include "util/branch.h"
50 #include "sane_ctype.h"
52 #include <linux/bitmap.h>
53 #include <linux/stringify.h>
54 #include <linux/time64.h>
55 #include <sys/types.h>
58 #include <linux/mman.h>
61 struct perf_tool tool
;
62 struct perf_session
*session
;
63 bool use_tui
, use_gtk
, use_stdio
;
66 bool inverted_callchain
;
73 bool nonany_branch_mode
;
76 struct perf_read_values show_threads_values
;
77 struct annotation_options annotation_opts
;
78 const char *pretty_printing_style
;
80 const char *symbol_filter_str
;
82 struct perf_time_interval
*ptime_range
;
89 DECLARE_BITMAP(cpu_bitmap
, MAX_NR_CPUS
);
90 struct branch_type_stat brtype_stat
;
94 static int report__config(const char *var
, const char *value
, void *cb
)
96 struct report
*rep
= cb
;
98 if (!strcmp(var
, "report.group")) {
99 symbol_conf
.event_group
= perf_config_bool(var
, value
);
102 if (!strcmp(var
, "report.percent-limit")) {
103 double pcnt
= strtof(value
, NULL
);
105 rep
->min_percent
= pcnt
;
106 callchain_param
.min_percent
= pcnt
;
109 if (!strcmp(var
, "report.children")) {
110 symbol_conf
.cumulate_callchain
= perf_config_bool(var
, value
);
113 if (!strcmp(var
, "report.queue-size"))
114 return perf_config_u64(&rep
->queue_size
, var
, value
);
116 if (!strcmp(var
, "report.sort_order")) {
117 default_sort_order
= strdup(value
);
124 static int hist_iter__report_callback(struct hist_entry_iter
*iter
,
125 struct addr_location
*al
, bool single
,
129 struct report
*rep
= arg
;
130 struct hist_entry
*he
= iter
->he
;
131 struct perf_evsel
*evsel
= iter
->evsel
;
132 struct perf_sample
*sample
= iter
->sample
;
134 struct branch_info
*bi
;
136 if (!ui__has_annotation() && !rep
->symbol_ipc
)
139 if (sort__mode
== SORT_MODE__BRANCH
) {
140 bi
= he
->branch_info
;
141 err
= addr_map_symbol__inc_samples(&bi
->from
, sample
, evsel
);
145 err
= addr_map_symbol__inc_samples(&bi
->to
, sample
, evsel
);
147 } else if (rep
->mem_mode
) {
149 err
= addr_map_symbol__inc_samples(&mi
->daddr
, sample
, evsel
);
153 err
= hist_entry__inc_addr_samples(he
, sample
, evsel
, al
->addr
);
155 } else if (symbol_conf
.cumulate_callchain
) {
157 err
= hist_entry__inc_addr_samples(he
, sample
, evsel
, al
->addr
);
159 err
= hist_entry__inc_addr_samples(he
, sample
, evsel
, al
->addr
);
166 static int hist_iter__branch_callback(struct hist_entry_iter
*iter
,
167 struct addr_location
*al __maybe_unused
,
168 bool single __maybe_unused
,
171 struct hist_entry
*he
= iter
->he
;
172 struct report
*rep
= arg
;
173 struct branch_info
*bi
;
174 struct perf_sample
*sample
= iter
->sample
;
175 struct perf_evsel
*evsel
= iter
->evsel
;
178 if (!ui__has_annotation() && !rep
->symbol_ipc
)
181 bi
= he
->branch_info
;
182 err
= addr_map_symbol__inc_samples(&bi
->from
, sample
, evsel
);
186 err
= addr_map_symbol__inc_samples(&bi
->to
, sample
, evsel
);
188 branch_type_count(&rep
->brtype_stat
, &bi
->flags
,
189 bi
->from
.addr
, bi
->to
.addr
);
195 static void setup_forced_leader(struct report
*report
,
196 struct perf_evlist
*evlist
)
198 if (report
->group_set
)
199 perf_evlist__force_leader(evlist
);
202 static int process_feature_event(struct perf_session
*session
,
203 union perf_event
*event
)
205 struct report
*rep
= container_of(session
->tool
, struct report
, tool
);
207 if (event
->feat
.feat_id
< HEADER_LAST_FEATURE
)
208 return perf_event__process_feature(session
, event
);
210 if (event
->feat
.feat_id
!= HEADER_LAST_FEATURE
) {
211 pr_err("failed: wrong feature ID: %" PRIu64
"\n",
212 event
->feat
.feat_id
);
217 * (feat_id = HEADER_LAST_FEATURE) is the end marker which
218 * means all features are received, now we can force the
221 setup_forced_leader(rep
, session
->evlist
);
225 static int process_sample_event(struct perf_tool
*tool
,
226 union perf_event
*event
,
227 struct perf_sample
*sample
,
228 struct perf_evsel
*evsel
,
229 struct machine
*machine
)
231 struct report
*rep
= container_of(tool
, struct report
, tool
);
232 struct addr_location al
;
233 struct hist_entry_iter iter
= {
236 .hide_unresolved
= symbol_conf
.hide_unresolved
,
237 .add_entry_cb
= hist_iter__report_callback
,
241 if (perf_time__ranges_skip_sample(rep
->ptime_range
, rep
->range_num
,
246 if (machine__resolve(machine
, &al
, sample
) < 0) {
247 pr_debug("problem processing %d event, skipping it.\n",
252 if (symbol_conf
.hide_unresolved
&& al
.sym
== NULL
)
255 if (rep
->cpu_list
&& !test_bit(sample
->cpu
, rep
->cpu_bitmap
))
258 if (sort__mode
== SORT_MODE__BRANCH
) {
260 * A non-synthesized event might not have a branch stack if
261 * branch stacks have been synthesized (using itrace options).
263 if (!sample
->branch_stack
)
266 iter
.add_entry_cb
= hist_iter__branch_callback
;
267 iter
.ops
= &hist_iter_branch
;
268 } else if (rep
->mem_mode
) {
269 iter
.ops
= &hist_iter_mem
;
270 } else if (symbol_conf
.cumulate_callchain
) {
271 iter
.ops
= &hist_iter_cumulative
;
273 iter
.ops
= &hist_iter_normal
;
277 al
.map
->dso
->hit
= 1;
279 if (ui__has_annotation() || rep
->symbol_ipc
) {
280 hist__account_cycles(sample
->branch_stack
, &al
, sample
,
281 rep
->nonany_branch_mode
);
284 ret
= hist_entry_iter__add(&iter
, &al
, rep
->max_stack
, rep
);
286 pr_debug("problem adding hist entry, skipping event\n");
288 addr_location__put(&al
);
292 static int process_read_event(struct perf_tool
*tool
,
293 union perf_event
*event
,
294 struct perf_sample
*sample __maybe_unused
,
295 struct perf_evsel
*evsel
,
296 struct machine
*machine __maybe_unused
)
298 struct report
*rep
= container_of(tool
, struct report
, tool
);
300 if (rep
->show_threads
) {
301 const char *name
= evsel
? perf_evsel__name(evsel
) : "unknown";
302 int err
= perf_read_values_add_value(&rep
->show_threads_values
,
303 event
->read
.pid
, event
->read
.tid
,
315 /* For pipe mode, sample_type is not currently set */
316 static int report__setup_sample_type(struct report
*rep
)
318 struct perf_session
*session
= rep
->session
;
319 u64 sample_type
= perf_evlist__combined_sample_type(session
->evlist
);
320 bool is_pipe
= perf_data__is_pipe(session
->data
);
322 if (session
->itrace_synth_opts
->callchain
||
324 perf_header__has_feat(&session
->header
, HEADER_AUXTRACE
) &&
325 !session
->itrace_synth_opts
->set
))
326 sample_type
|= PERF_SAMPLE_CALLCHAIN
;
328 if (session
->itrace_synth_opts
->last_branch
)
329 sample_type
|= PERF_SAMPLE_BRANCH_STACK
;
331 if (!is_pipe
&& !(sample_type
& PERF_SAMPLE_CALLCHAIN
)) {
332 if (perf_hpp_list
.parent
) {
333 ui__error("Selected --sort parent, but no "
334 "callchain data. Did you call "
335 "'perf record' without -g?\n");
338 if (symbol_conf
.use_callchain
&&
339 !symbol_conf
.show_branchflag_count
) {
340 ui__error("Selected -g or --branch-history.\n"
341 "But no callchain or branch data.\n"
342 "Did you call 'perf record' without -g or -b?\n");
345 } else if (!callchain_param
.enabled
&&
346 callchain_param
.mode
!= CHAIN_NONE
&&
347 !symbol_conf
.use_callchain
) {
348 symbol_conf
.use_callchain
= true;
349 if (callchain_register_param(&callchain_param
) < 0) {
350 ui__error("Can't register callchain params.\n");
355 if (symbol_conf
.cumulate_callchain
) {
356 /* Silently ignore if callchain is missing */
357 if (!(sample_type
& PERF_SAMPLE_CALLCHAIN
)) {
358 symbol_conf
.cumulate_callchain
= false;
359 perf_hpp__cancel_cumulate();
363 if (sort__mode
== SORT_MODE__BRANCH
) {
365 !(sample_type
& PERF_SAMPLE_BRANCH_STACK
)) {
366 ui__error("Selected -b but no branch data. "
367 "Did you call perf record without -b?\n");
372 if (symbol_conf
.use_callchain
|| symbol_conf
.cumulate_callchain
) {
373 if ((sample_type
& PERF_SAMPLE_REGS_USER
) &&
374 (sample_type
& PERF_SAMPLE_STACK_USER
)) {
375 callchain_param
.record_mode
= CALLCHAIN_DWARF
;
376 dwarf_callchain_users
= true;
377 } else if (sample_type
& PERF_SAMPLE_BRANCH_STACK
)
378 callchain_param
.record_mode
= CALLCHAIN_LBR
;
380 callchain_param
.record_mode
= CALLCHAIN_FP
;
383 /* ??? handle more cases than just ANY? */
384 if (!(perf_evlist__combined_branch_type(session
->evlist
) &
385 PERF_SAMPLE_BRANCH_ANY
))
386 rep
->nonany_branch_mode
= true;
391 static void sig_handler(int sig __maybe_unused
)
396 static size_t hists__fprintf_nr_sample_events(struct hists
*hists
, struct report
*rep
,
397 const char *evname
, FILE *fp
)
401 unsigned long nr_samples
= hists
->stats
.nr_events
[PERF_RECORD_SAMPLE
];
402 u64 nr_events
= hists
->stats
.total_period
;
403 struct perf_evsel
*evsel
= hists_to_evsel(hists
);
405 size_t size
= sizeof(buf
);
406 int socked_id
= hists
->socket_filter
;
411 if (symbol_conf
.filter_relative
) {
412 nr_samples
= hists
->stats
.nr_non_filtered_samples
;
413 nr_events
= hists
->stats
.total_non_filtered_period
;
416 if (perf_evsel__is_group_event(evsel
)) {
417 struct perf_evsel
*pos
;
419 perf_evsel__group_desc(evsel
, buf
, size
);
422 for_each_group_member(pos
, evsel
) {
423 const struct hists
*pos_hists
= evsel__hists(pos
);
425 if (symbol_conf
.filter_relative
) {
426 nr_samples
+= pos_hists
->stats
.nr_non_filtered_samples
;
427 nr_events
+= pos_hists
->stats
.total_non_filtered_period
;
429 nr_samples
+= pos_hists
->stats
.nr_events
[PERF_RECORD_SAMPLE
];
430 nr_events
+= pos_hists
->stats
.total_period
;
435 nr_samples
= convert_unit(nr_samples
, &unit
);
436 ret
= fprintf(fp
, "# Samples: %lu%c", nr_samples
, unit
);
437 if (evname
!= NULL
) {
438 ret
+= fprintf(fp
, " of event%s '%s'",
439 evsel
->nr_members
> 1 ? "s" : "", evname
);
443 ret
+= fprintf(fp
, " (time slices: %s)", rep
->time_str
);
445 if (symbol_conf
.show_ref_callgraph
&&
446 strstr(evname
, "call-graph=no")) {
447 ret
+= fprintf(fp
, ", show reference callgraph");
451 ret
+= fprintf(fp
, "\n# Total weight : %" PRIu64
, nr_events
);
452 ret
+= fprintf(fp
, "\n# Sort order : %s", sort_order
? : default_mem_sort_order
);
454 ret
+= fprintf(fp
, "\n# Event count (approx.): %" PRIu64
, nr_events
);
457 ret
+= fprintf(fp
, "\n# Processor Socket: %d", socked_id
);
459 return ret
+ fprintf(fp
, "\n#\n");
462 static int perf_evlist__tty_browse_hists(struct perf_evlist
*evlist
,
466 struct perf_evsel
*pos
;
469 fprintf(stdout
, "#\n# Total Lost Samples: %" PRIu64
"\n#\n",
470 evlist
->stats
.total_lost_samples
);
473 evlist__for_each_entry(evlist
, pos
) {
474 struct hists
*hists
= evsel__hists(pos
);
475 const char *evname
= perf_evsel__name(pos
);
477 if (symbol_conf
.event_group
&&
478 !perf_evsel__is_group_leader(pos
))
481 hists__fprintf_nr_sample_events(hists
, rep
, evname
, stdout
);
482 hists__fprintf(hists
, !quiet
, 0, 0, rep
->min_percent
, stdout
,
483 !(symbol_conf
.use_callchain
||
484 symbol_conf
.show_branchflag_count
));
485 fprintf(stdout
, "\n\n");
489 fprintf(stdout
, "#\n# (%s)\n#\n", help
);
491 if (rep
->show_threads
) {
492 bool style
= !strcmp(rep
->pretty_printing_style
, "raw");
493 perf_read_values_display(stdout
, &rep
->show_threads_values
,
495 perf_read_values_destroy(&rep
->show_threads_values
);
498 if (sort__mode
== SORT_MODE__BRANCH
)
499 branch_type_stat_display(stdout
, &rep
->brtype_stat
);
504 static void report__warn_kptr_restrict(const struct report
*rep
)
506 struct map
*kernel_map
= machine__kernel_map(&rep
->session
->machines
.host
);
507 struct kmap
*kernel_kmap
= kernel_map
? map__kmap(kernel_map
) : NULL
;
509 if (perf_evlist__exclude_kernel(rep
->session
->evlist
))
512 if (kernel_map
== NULL
||
513 (kernel_map
->dso
->hit
&&
514 (kernel_kmap
->ref_reloc_sym
== NULL
||
515 kernel_kmap
->ref_reloc_sym
->addr
== 0))) {
517 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
518 "can't be resolved.";
520 if (kernel_map
&& map__has_symbols(kernel_map
)) {
521 desc
= "If some relocation was applied (e.g. "
522 "kexec) symbols may be misresolved.";
526 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
527 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
528 "Samples in kernel modules can't be resolved as well.\n\n",
533 static int report__gtk_browse_hists(struct report
*rep
, const char *help
)
535 int (*hist_browser
)(struct perf_evlist
*evlist
, const char *help
,
536 struct hist_browser_timer
*timer
, float min_pcnt
);
538 hist_browser
= dlsym(perf_gtk_handle
, "perf_evlist__gtk_browse_hists");
540 if (hist_browser
== NULL
) {
541 ui__error("GTK browser not found!\n");
545 return hist_browser(rep
->session
->evlist
, help
, NULL
, rep
->min_percent
);
548 static int report__browse_hists(struct report
*rep
)
551 struct perf_session
*session
= rep
->session
;
552 struct perf_evlist
*evlist
= session
->evlist
;
553 const char *help
= perf_tip(system_path(TIPDIR
));
556 /* fallback for people who don't install perf ;-) */
557 help
= perf_tip(DOCDIR
);
559 help
= "Cannot load tips.txt file, please install perf!";
562 switch (use_browser
) {
564 ret
= perf_evlist__tui_browse_hists(evlist
, help
, NULL
,
566 &session
->header
.env
,
567 true, &rep
->annotation_opts
);
569 * Usually "ret" is the last pressed key, and we only
570 * care if the key notifies us to switch data file.
572 if (ret
!= K_SWITCH_INPUT_DATA
)
576 ret
= report__gtk_browse_hists(rep
, help
);
579 ret
= perf_evlist__tty_browse_hists(evlist
, rep
, help
);
586 static int report__collapse_hists(struct report
*rep
)
588 struct ui_progress prog
;
589 struct perf_evsel
*pos
;
592 ui_progress__init(&prog
, rep
->nr_entries
, "Merging related events...");
594 evlist__for_each_entry(rep
->session
->evlist
, pos
) {
595 struct hists
*hists
= evsel__hists(pos
);
598 hists
->symbol_filter_str
= rep
->symbol_filter_str
;
600 hists
->socket_filter
= rep
->socket_filter
;
602 ret
= hists__collapse_resort(hists
, &prog
);
606 /* Non-group events are considered as leader */
607 if (symbol_conf
.event_group
&&
608 !perf_evsel__is_group_leader(pos
)) {
609 struct hists
*leader_hists
= evsel__hists(pos
->leader
);
611 hists__match(leader_hists
, hists
);
612 hists__link(leader_hists
, hists
);
616 ui_progress__finish();
620 static int hists__resort_cb(struct hist_entry
*he
, void *arg
)
622 struct report
*rep
= arg
;
623 struct symbol
*sym
= he
->ms
.sym
;
625 if (rep
->symbol_ipc
&& sym
&& !sym
->annotate2
) {
626 struct perf_evsel
*evsel
= hists_to_evsel(he
->hists
);
628 symbol__annotate2(sym
, he
->ms
.map
, evsel
,
629 &annotation__default_options
, NULL
);
635 static void report__output_resort(struct report
*rep
)
637 struct ui_progress prog
;
638 struct perf_evsel
*pos
;
640 ui_progress__init(&prog
, rep
->nr_entries
, "Sorting events for output...");
642 evlist__for_each_entry(rep
->session
->evlist
, pos
) {
643 perf_evsel__output_resort_cb(pos
, &prog
,
644 hists__resort_cb
, rep
);
647 ui_progress__finish();
650 static void stats_setup(struct report
*rep
)
652 memset(&rep
->tool
, 0, sizeof(rep
->tool
));
653 rep
->tool
.no_warn
= true;
656 static int stats_print(struct report
*rep
)
658 struct perf_session
*session
= rep
->session
;
660 perf_session__fprintf_nr_events(session
, stdout
);
664 static void tasks_setup(struct report
*rep
)
666 memset(&rep
->tool
, 0, sizeof(rep
->tool
));
667 rep
->tool
.ordered_events
= true;
668 if (rep
->mmaps_mode
) {
669 rep
->tool
.mmap
= perf_event__process_mmap
;
670 rep
->tool
.mmap2
= perf_event__process_mmap2
;
672 rep
->tool
.comm
= perf_event__process_comm
;
673 rep
->tool
.exit
= perf_event__process_exit
;
674 rep
->tool
.fork
= perf_event__process_fork
;
675 rep
->tool
.no_warn
= true;
679 struct thread
*thread
;
680 struct list_head list
;
681 struct list_head children
;
684 static struct task
*tasks_list(struct task
*task
, struct machine
*machine
)
686 struct thread
*parent_thread
, *thread
= task
->thread
;
687 struct task
*parent_task
;
689 /* Already listed. */
690 if (!list_empty(&task
->list
))
693 /* Last one in the chain. */
694 if (thread
->ppid
== -1)
697 parent_thread
= machine__find_thread(machine
, -1, thread
->ppid
);
699 return ERR_PTR(-ENOENT
);
701 parent_task
= thread__priv(parent_thread
);
702 list_add_tail(&task
->list
, &parent_task
->children
);
703 return tasks_list(parent_task
, machine
);
706 static size_t maps__fprintf_task(struct maps
*maps
, int indent
, FILE *fp
)
711 for (nd
= rb_first(&maps
->entries
); nd
; nd
= rb_next(nd
)) {
712 struct map
*map
= rb_entry(nd
, struct map
, rb_node
);
714 printed
+= fprintf(fp
, "%*s %" PRIx64
"-%" PRIx64
" %c%c%c%c %08" PRIx64
" %" PRIu64
" %s\n",
715 indent
, "", map
->start
, map
->end
,
716 map
->prot
& PROT_READ
? 'r' : '-',
717 map
->prot
& PROT_WRITE
? 'w' : '-',
718 map
->prot
& PROT_EXEC
? 'x' : '-',
719 map
->flags
& MAP_SHARED
? 's' : 'p',
721 map
->ino
, map
->dso
->name
);
727 static int map_groups__fprintf_task(struct map_groups
*mg
, int indent
, FILE *fp
)
729 return maps__fprintf_task(&mg
->maps
, indent
, fp
);
732 static void task__print_level(struct task
*task
, FILE *fp
, int level
)
734 struct thread
*thread
= task
->thread
;
736 int comm_indent
= fprintf(fp
, " %8d %8d %8d |%*s",
737 thread
->pid_
, thread
->tid
, thread
->ppid
,
740 fprintf(fp
, "%s\n", thread__comm_str(thread
));
742 map_groups__fprintf_task(thread
->mg
, comm_indent
, fp
);
744 if (!list_empty(&task
->children
)) {
745 list_for_each_entry(child
, &task
->children
, list
)
746 task__print_level(child
, fp
, level
+ 1);
750 static int tasks_print(struct report
*rep
, FILE *fp
)
752 struct perf_session
*session
= rep
->session
;
753 struct machine
*machine
= &session
->machines
.host
;
754 struct task
*tasks
, *task
;
755 unsigned int nr
= 0, itask
= 0, i
;
760 * No locking needed while accessing machine->threads,
761 * because --tasks is single threaded command.
764 /* Count all the threads. */
765 for (i
= 0; i
< THREADS__TABLE_SIZE
; i
++)
766 nr
+= machine
->threads
[i
].nr
;
768 tasks
= malloc(sizeof(*tasks
) * nr
);
772 for (i
= 0; i
< THREADS__TABLE_SIZE
; i
++) {
773 struct threads
*threads
= &machine
->threads
[i
];
775 for (nd
= rb_first_cached(&threads
->entries
); nd
;
777 task
= tasks
+ itask
++;
779 task
->thread
= rb_entry(nd
, struct thread
, rb_node
);
780 INIT_LIST_HEAD(&task
->children
);
781 INIT_LIST_HEAD(&task
->list
);
782 thread__set_priv(task
->thread
, task
);
787 * Iterate every task down to the unprocessed parent
788 * and link all in task children list. Task with no
789 * parent is added into 'list'.
791 for (itask
= 0; itask
< nr
; itask
++) {
792 task
= tasks
+ itask
;
794 if (!list_empty(&task
->list
))
797 task
= tasks_list(task
, machine
);
799 pr_err("Error: failed to process tasks\n");
801 return PTR_ERR(task
);
805 list_add_tail(&task
->list
, &list
);
808 fprintf(fp
, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm");
810 list_for_each_entry(task
, &list
, list
)
811 task__print_level(task
, fp
, 0);
817 static int __cmd_report(struct report
*rep
)
820 struct perf_session
*session
= rep
->session
;
821 struct perf_evsel
*pos
;
822 struct perf_data
*data
= session
->data
;
824 signal(SIGINT
, sig_handler
);
827 ret
= perf_session__cpu_bitmap(session
, rep
->cpu_list
,
830 ui__error("failed to set cpu bitmap\n");
833 session
->itrace_synth_opts
->cpu_bitmap
= rep
->cpu_bitmap
;
836 if (rep
->show_threads
) {
837 ret
= perf_read_values_init(&rep
->show_threads_values
);
842 ret
= report__setup_sample_type(rep
);
844 /* report__setup_sample_type() already showed error message */
854 ret
= perf_session__process_events(session
);
856 ui__error("failed to process sample\n");
861 return stats_print(rep
);
864 return tasks_print(rep
, stdout
);
866 report__warn_kptr_restrict(rep
);
868 evlist__for_each_entry(session
->evlist
, pos
)
869 rep
->nr_entries
+= evsel__hists(pos
)->nr_entries
;
871 if (use_browser
== 0) {
873 perf_session__fprintf(session
, stdout
);
876 perf_session__fprintf_dsos(session
, stdout
);
879 perf_session__fprintf_nr_events(session
, stdout
);
880 perf_evlist__fprintf_nr_events(session
->evlist
, stdout
);
885 ret
= report__collapse_hists(rep
);
887 ui__error("failed to process hist entry\n");
895 * recalculate number of entries after collapsing since it
896 * might be changed during the collapse phase.
899 evlist__for_each_entry(session
->evlist
, pos
)
900 rep
->nr_entries
+= evsel__hists(pos
)->nr_entries
;
902 if (rep
->nr_entries
== 0) {
903 ui__error("The %s data has no samples!\n", data
->path
);
907 report__output_resort(rep
);
909 return report__browse_hists(rep
);
913 report_parse_callchain_opt(const struct option
*opt
, const char *arg
, int unset
)
915 struct callchain_param
*callchain
= opt
->value
;
917 callchain
->enabled
= !unset
;
922 symbol_conf
.use_callchain
= false;
923 callchain
->mode
= CHAIN_NONE
;
927 return parse_callchain_report_opt(arg
);
931 parse_time_quantum(const struct option
*opt
, const char *arg
,
932 int unset __maybe_unused
)
934 unsigned long *time_q
= opt
->value
;
937 *time_q
= strtoul(arg
, &end
, 0);
941 pr_err("time quantum cannot be 0");
944 while (isspace(*end
))
948 if (!strcmp(end
, "s")) {
949 *time_q
*= NSEC_PER_SEC
;
952 if (!strcmp(end
, "ms")) {
953 *time_q
*= NSEC_PER_MSEC
;
956 if (!strcmp(end
, "us")) {
957 *time_q
*= NSEC_PER_USEC
;
960 if (!strcmp(end
, "ns"))
963 pr_err("Cannot parse time quantum `%s'\n", arg
);
968 report_parse_ignore_callees_opt(const struct option
*opt __maybe_unused
,
969 const char *arg
, int unset __maybe_unused
)
972 int err
= regcomp(&ignore_callees_regex
, arg
, REG_EXTENDED
);
975 regerror(err
, &ignore_callees_regex
, buf
, sizeof(buf
));
976 pr_err("Invalid --ignore-callees regex: %s\n%s", arg
, buf
);
979 have_ignore_callees
= 1;
986 parse_branch_mode(const struct option
*opt
,
987 const char *str __maybe_unused
, int unset
)
989 int *branch_mode
= opt
->value
;
991 *branch_mode
= !unset
;
996 parse_percent_limit(const struct option
*opt
, const char *str
,
997 int unset __maybe_unused
)
999 struct report
*rep
= opt
->value
;
1000 double pcnt
= strtof(str
, NULL
);
1002 rep
->min_percent
= pcnt
;
1003 callchain_param
.min_percent
= pcnt
;
1007 int cmd_report(int argc
, const char **argv
)
1009 struct perf_session
*session
;
1010 struct itrace_synth_opts itrace_synth_opts
= { .set
= 0, };
1012 bool has_br_stack
= false;
1013 int branch_mode
= -1;
1014 bool branch_call_mode
= false;
1015 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
1016 static const char report_callchain_help
[] = "Display call graph (stack chain/backtrace):\n\n"
1017 CALLCHAIN_REPORT_HELP
1018 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT
;
1019 char callchain_default_opt
[] = CALLCHAIN_DEFAULT_OPT
;
1020 const char * const report_usage
[] = {
1021 "perf report [<options>]",
1024 struct report report
= {
1026 .sample
= process_sample_event
,
1027 .mmap
= perf_event__process_mmap
,
1028 .mmap2
= perf_event__process_mmap2
,
1029 .comm
= perf_event__process_comm
,
1030 .namespaces
= perf_event__process_namespaces
,
1031 .exit
= perf_event__process_exit
,
1032 .fork
= perf_event__process_fork
,
1033 .lost
= perf_event__process_lost
,
1034 .read
= process_read_event
,
1035 .attr
= perf_event__process_attr
,
1036 .tracing_data
= perf_event__process_tracing_data
,
1037 .build_id
= perf_event__process_build_id
,
1038 .id_index
= perf_event__process_id_index
,
1039 .auxtrace_info
= perf_event__process_auxtrace_info
,
1040 .auxtrace
= perf_event__process_auxtrace
,
1041 .event_update
= perf_event__process_event_update
,
1042 .feature
= process_feature_event
,
1043 .ordered_events
= true,
1044 .ordering_requires_timestamps
= true,
1046 .max_stack
= PERF_MAX_STACK_DEPTH
,
1047 .pretty_printing_style
= "normal",
1048 .socket_filter
= -1,
1049 .annotation_opts
= annotation__default_options
,
1051 const struct option options
[] = {
1052 OPT_STRING('i', "input", &input_name
, "file",
1054 OPT_INCR('v', "verbose", &verbose
,
1055 "be more verbose (show symbol address, etc)"),
1056 OPT_BOOLEAN('q', "quiet", &quiet
, "Do not show any message"),
1057 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace
,
1058 "dump raw trace in ASCII"),
1059 OPT_BOOLEAN(0, "stats", &report
.stats_mode
, "Display event stats"),
1060 OPT_BOOLEAN(0, "tasks", &report
.tasks_mode
, "Display recorded tasks"),
1061 OPT_BOOLEAN(0, "mmaps", &report
.mmaps_mode
, "Display recorded tasks memory maps"),
1062 OPT_STRING('k', "vmlinux", &symbol_conf
.vmlinux_name
,
1063 "file", "vmlinux pathname"),
1064 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf
.ignore_vmlinux
,
1065 "don't load vmlinux even if found"),
1066 OPT_STRING(0, "kallsyms", &symbol_conf
.kallsyms_name
,
1067 "file", "kallsyms pathname"),
1068 OPT_BOOLEAN('f', "force", &symbol_conf
.force
, "don't complain, do it"),
1069 OPT_BOOLEAN('m', "modules", &symbol_conf
.use_modules
,
1070 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1071 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf
.show_nr_samples
,
1072 "Show a column with the number of samples"),
1073 OPT_BOOLEAN('T', "threads", &report
.show_threads
,
1074 "Show per-thread event counters"),
1075 OPT_STRING(0, "pretty", &report
.pretty_printing_style
, "key",
1076 "pretty printing style key: normal raw"),
1077 OPT_BOOLEAN(0, "tui", &report
.use_tui
, "Use the TUI interface"),
1078 OPT_BOOLEAN(0, "gtk", &report
.use_gtk
, "Use the GTK2 interface"),
1079 OPT_BOOLEAN(0, "stdio", &report
.use_stdio
,
1080 "Use the stdio interface"),
1081 OPT_BOOLEAN(0, "header", &report
.header
, "Show data header."),
1082 OPT_BOOLEAN(0, "header-only", &report
.header_only
,
1083 "Show only data header."),
1084 OPT_STRING('s', "sort", &sort_order
, "key[,key2...]",
1085 sort_help("sort by key(s):")),
1086 OPT_STRING('F', "fields", &field_order
, "key[,keys...]",
1087 sort_help("output field(s): overhead period sample ")),
1088 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf
.show_cpu_utilization
,
1089 "Show sample percentage for different cpu modes"),
1090 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf
.show_cpu_utilization
,
1091 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN
),
1092 OPT_STRING('p', "parent", &parent_pattern
, "regex",
1093 "regex filter to identify parent, see: '--sort parent'"),
1094 OPT_BOOLEAN('x', "exclude-other", &symbol_conf
.exclude_other
,
1095 "Only display entries with parent-match"),
1096 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param
,
1097 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
1098 report_callchain_help
, &report_parse_callchain_opt
,
1099 callchain_default_opt
),
1100 OPT_BOOLEAN(0, "children", &symbol_conf
.cumulate_callchain
,
1101 "Accumulate callchains of children and show total overhead as well"),
1102 OPT_INTEGER(0, "max-stack", &report
.max_stack
,
1103 "Set the maximum stack depth when parsing the callchain, "
1104 "anything beyond the specified depth will be ignored. "
1105 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH
)),
1106 OPT_BOOLEAN('G', "inverted", &report
.inverted_callchain
,
1107 "alias for inverted call graph"),
1108 OPT_CALLBACK(0, "ignore-callees", NULL
, "regex",
1109 "ignore callees of these functions in call graphs",
1110 report_parse_ignore_callees_opt
),
1111 OPT_STRING('d', "dsos", &symbol_conf
.dso_list_str
, "dso[,dso...]",
1112 "only consider symbols in these dsos"),
1113 OPT_STRING('c', "comms", &symbol_conf
.comm_list_str
, "comm[,comm...]",
1114 "only consider symbols in these comms"),
1115 OPT_STRING(0, "pid", &symbol_conf
.pid_list_str
, "pid[,pid...]",
1116 "only consider symbols in these pids"),
1117 OPT_STRING(0, "tid", &symbol_conf
.tid_list_str
, "tid[,tid...]",
1118 "only consider symbols in these tids"),
1119 OPT_STRING('S', "symbols", &symbol_conf
.sym_list_str
, "symbol[,symbol...]",
1120 "only consider these symbols"),
1121 OPT_STRING(0, "symbol-filter", &report
.symbol_filter_str
, "filter",
1122 "only show symbols that (partially) match with this filter"),
1123 OPT_STRING('w', "column-widths", &symbol_conf
.col_width_list_str
,
1125 "don't try to adjust column width, use these fixed values"),
1126 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf
.field_sep
, "separator",
1127 "separator for columns, no spaces will be added between "
1128 "columns '.' is reserved."),
1129 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf
.hide_unresolved
,
1130 "Only display entries resolved to a symbol"),
1131 OPT_CALLBACK(0, "symfs", NULL
, "directory",
1132 "Look for files with symbols relative to this directory",
1133 symbol__config_symfs
),
1134 OPT_STRING('C', "cpu", &report
.cpu_list
, "cpu",
1135 "list of cpus to profile"),
1136 OPT_BOOLEAN('I', "show-info", &report
.show_full_info
,
1137 "Display extended information about perf.data file"),
1138 OPT_BOOLEAN(0, "source", &report
.annotation_opts
.annotate_src
,
1139 "Interleave source code with assembly code (default)"),
1140 OPT_BOOLEAN(0, "asm-raw", &report
.annotation_opts
.show_asm_raw
,
1141 "Display raw encoding of assembly instructions (default)"),
1142 OPT_STRING('M', "disassembler-style", &report
.annotation_opts
.disassembler_style
, "disassembler style",
1143 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1144 OPT_BOOLEAN(0, "show-total-period", &symbol_conf
.show_total_period
,
1145 "Show a column with the sum of periods"),
1146 OPT_BOOLEAN_SET(0, "group", &symbol_conf
.event_group
, &report
.group_set
,
1147 "Show event group information together"),
1148 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode
, "",
1149 "use branch records for per branch histogram filling",
1151 OPT_BOOLEAN(0, "branch-history", &branch_call_mode
,
1152 "add last branch records to call history"),
1153 OPT_STRING(0, "objdump", &report
.annotation_opts
.objdump_path
, "path",
1154 "objdump binary to use for disassembly and annotations"),
1155 OPT_BOOLEAN(0, "demangle", &symbol_conf
.demangle
,
1156 "Disable symbol demangling"),
1157 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf
.demangle_kernel
,
1158 "Enable kernel symbol demangling"),
1159 OPT_BOOLEAN(0, "mem-mode", &report
.mem_mode
, "mem access profile"),
1160 OPT_INTEGER(0, "samples", &symbol_conf
.res_sample
,
1161 "Number of samples to save per histogram entry for individual browsing"),
1162 OPT_CALLBACK(0, "percent-limit", &report
, "percent",
1163 "Don't show entries under that percent", parse_percent_limit
),
1164 OPT_CALLBACK(0, "percentage", NULL
, "relative|absolute",
1165 "how to display percentage of filtered entries", parse_filter_percentage
),
1166 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts
, NULL
, "opts",
1167 "Instruction Tracing options\n" ITRACE_HELP
,
1168 itrace_parse_synth_opts
),
1169 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename
,
1170 "Show full source file name path for source lines"),
1171 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf
.show_ref_callgraph
,
1172 "Show callgraph from reference event"),
1173 OPT_INTEGER(0, "socket-filter", &report
.socket_filter
,
1174 "only show processor socket that match with this filter"),
1175 OPT_BOOLEAN(0, "raw-trace", &symbol_conf
.raw_trace
,
1176 "Show raw trace event output (do not use print fmt or plugins)"),
1177 OPT_BOOLEAN(0, "hierarchy", &symbol_conf
.report_hierarchy
,
1178 "Show entries in a hierarchy"),
1179 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL
, "mode",
1180 "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
1181 stdio__config_color
, "always"),
1182 OPT_STRING(0, "time", &report
.time_str
, "str",
1183 "Time span of interest (start,stop)"),
1184 OPT_BOOLEAN(0, "inline", &symbol_conf
.inline_name
,
1185 "Show inline function"),
1186 OPT_CALLBACK(0, "percent-type", &report
.annotation_opts
, "local-period",
1187 "Set percent type local/global-period/hits",
1188 annotate_parse_percent_type
),
1189 OPT_BOOLEAN(0, "ns", &symbol_conf
.nanosecs
, "Show times in nanosecs"),
1190 OPT_CALLBACK(0, "time-quantum", &symbol_conf
.time_quantum
, "time (ms|us|ns|s)",
1191 "Set time quantum for time sort key (default 100ms)",
1192 parse_time_quantum
),
1195 struct perf_data data
= {
1196 .mode
= PERF_DATA_MODE_READ
,
1198 int ret
= hists__init();
1204 ret
= perf_config(report__config
, &report
);
1208 argc
= parse_options(argc
, argv
, options
, report_usage
, 0);
1211 * Special case: if there's an argument left then assume that
1212 * it's a symbol filter:
1215 usage_with_options(report_usage
, options
);
1217 report
.symbol_filter_str
= argv
[0];
1220 if (report
.mmaps_mode
)
1221 report
.tasks_mode
= true;
1224 perf_quiet_option();
1226 if (symbol_conf
.vmlinux_name
&&
1227 access(symbol_conf
.vmlinux_name
, R_OK
)) {
1228 pr_err("Invalid file: %s\n", symbol_conf
.vmlinux_name
);
1231 if (symbol_conf
.kallsyms_name
&&
1232 access(symbol_conf
.kallsyms_name
, R_OK
)) {
1233 pr_err("Invalid file: %s\n", symbol_conf
.kallsyms_name
);
1237 if (report
.inverted_callchain
)
1238 callchain_param
.order
= ORDER_CALLER
;
1239 if (symbol_conf
.cumulate_callchain
&& !callchain_param
.order_set
)
1240 callchain_param
.order
= ORDER_CALLER
;
1242 if (itrace_synth_opts
.callchain
&&
1243 (int)itrace_synth_opts
.callchain_sz
> report
.max_stack
)
1244 report
.max_stack
= itrace_synth_opts
.callchain_sz
;
1246 if (!input_name
|| !strlen(input_name
)) {
1247 if (!fstat(STDIN_FILENO
, &st
) && S_ISFIFO(st
.st_mode
))
1250 input_name
= "perf.data";
1253 data
.path
= input_name
;
1254 data
.force
= symbol_conf
.force
;
1257 session
= perf_session__new(&data
, false, &report
.tool
);
1258 if (session
== NULL
)
1261 if (zstd_init(&(session
->zstd_data
), 0) < 0)
1262 pr_warning("Decompression initialization failed. Reported data may be incomplete.\n");
1264 if (report
.queue_size
) {
1265 ordered_events__set_alloc_size(&session
->ordered_events
,
1269 session
->itrace_synth_opts
= &itrace_synth_opts
;
1271 report
.session
= session
;
1273 has_br_stack
= perf_header__has_feat(&session
->header
,
1274 HEADER_BRANCH_STACK
);
1276 setup_forced_leader(&report
, session
->evlist
);
1278 if (itrace_synth_opts
.last_branch
)
1279 has_br_stack
= true;
1281 if (has_br_stack
&& branch_call_mode
)
1282 symbol_conf
.show_branchflag_count
= true;
1284 memset(&report
.brtype_stat
, 0, sizeof(struct branch_type_stat
));
1287 * Branch mode is a tristate:
1288 * -1 means default, so decide based on the file having branch data.
1289 * 0/1 means the user chose a mode.
1291 if (((branch_mode
== -1 && has_br_stack
) || branch_mode
== 1) &&
1292 !branch_call_mode
) {
1293 sort__mode
= SORT_MODE__BRANCH
;
1294 symbol_conf
.cumulate_callchain
= false;
1296 if (branch_call_mode
) {
1297 callchain_param
.key
= CCKEY_ADDRESS
;
1298 callchain_param
.branch_callstack
= 1;
1299 symbol_conf
.use_callchain
= true;
1300 callchain_register_param(&callchain_param
);
1301 if (sort_order
== NULL
)
1302 sort_order
= "srcline,symbol,dso";
1305 if (report
.mem_mode
) {
1306 if (sort__mode
== SORT_MODE__BRANCH
) {
1307 pr_err("branch and mem mode incompatible\n");
1310 sort__mode
= SORT_MODE__MEMORY
;
1311 symbol_conf
.cumulate_callchain
= false;
1314 if (symbol_conf
.report_hierarchy
) {
1315 /* disable incompatible options */
1316 symbol_conf
.cumulate_callchain
= false;
1319 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1320 parse_options_usage(report_usage
, options
, "F", 1);
1321 parse_options_usage(NULL
, options
, "hierarchy", 0);
1325 perf_hpp_list
.need_collapse
= true;
1328 if (report
.use_stdio
)
1330 else if (report
.use_tui
)
1332 else if (report
.use_gtk
)
1335 /* Force tty output for header output and per-thread stat. */
1336 if (report
.header
|| report
.header_only
|| report
.show_threads
)
1338 if (report
.header
|| report
.header_only
)
1339 report
.tool
.show_feat_hdr
= SHOW_FEAT_HEADER
;
1340 if (report
.show_full_info
)
1341 report
.tool
.show_feat_hdr
= SHOW_FEAT_HEADER_FULL_INFO
;
1342 if (report
.stats_mode
|| report
.tasks_mode
)
1344 if (report
.stats_mode
&& report
.tasks_mode
) {
1345 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n");
1349 if (strcmp(input_name
, "-") != 0)
1350 setup_browser(true);
1354 if (sort_order
&& strstr(sort_order
, "ipc")) {
1355 parse_options_usage(report_usage
, options
, "s", 1);
1359 if (sort_order
&& strstr(sort_order
, "symbol")) {
1360 if (sort__mode
== SORT_MODE__BRANCH
) {
1361 snprintf(sort_tmp
, sizeof(sort_tmp
), "%s,%s",
1362 sort_order
, "ipc_lbr");
1363 report
.symbol_ipc
= true;
1365 snprintf(sort_tmp
, sizeof(sort_tmp
), "%s,%s",
1366 sort_order
, "ipc_null");
1369 sort_order
= sort_tmp
;
1372 if (setup_sorting(session
->evlist
) < 0) {
1374 parse_options_usage(report_usage
, options
, "s", 1);
1376 parse_options_usage(sort_order
? NULL
: report_usage
,
1381 if ((report
.header
|| report
.header_only
) && !quiet
) {
1382 perf_session__fprintf_info(session
, stdout
,
1383 report
.show_full_info
);
1384 if (report
.header_only
) {
1388 } else if (use_browser
== 0 && !quiet
&&
1389 !report
.stats_mode
&& !report
.tasks_mode
) {
1390 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
1395 * Only in the TUI browser we are doing integrated annotation,
1396 * so don't allocate extra space that won't be used in the stdio
1399 if (ui__has_annotation() || report
.symbol_ipc
) {
1400 ret
= symbol__annotation_init();
1404 * For searching by name on the "Browse map details".
1405 * providing it only in verbose mode not to bloat too
1406 * much struct symbol.
1410 * XXX: Need to provide a less kludgy way to ask for
1411 * more space per symbol, the u32 is for the index on
1413 * See symbol__browser_index.
1415 symbol_conf
.priv_size
+= sizeof(u32
);
1416 symbol_conf
.sort_by_name
= true;
1418 annotation_config__init();
1421 if (symbol__init(&session
->header
.env
) < 0)
1424 if (report
.time_str
) {
1425 ret
= perf_time__parse_for_ranges(report
.time_str
, session
,
1426 &report
.ptime_range
,
1433 if (session
->tevent
.pevent
&&
1434 tep_set_function_resolver(session
->tevent
.pevent
,
1435 machine__resolve_kernel_addr
,
1436 &session
->machines
.host
) < 0) {
1437 pr_err("%s: failed to set libtraceevent function resolver\n",
1442 sort__setup_elide(stdout
);
1444 ret
= __cmd_report(&report
);
1445 if (ret
== K_SWITCH_INPUT_DATA
) {
1446 perf_session__delete(session
);
1452 if (report
.ptime_range
)
1453 zfree(&report
.ptime_range
);
1454 zstd_fini(&(session
->zstd_data
));
1455 perf_session__delete(session
);