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>
19 #include "util/symbol.h"
20 #include "util/callchain.h"
21 #include "util/values.h"
24 #include "util/debug.h"
25 #include "util/evlist.h"
26 #include "util/evsel.h"
27 #include "util/header.h"
28 #include "util/session.h"
29 #include "util/tool.h"
31 #include <subcmd/parse-options.h>
32 #include <subcmd/exec-cmd.h>
33 #include "util/parse-events.h"
35 #include "util/thread.h"
36 #include "util/sort.h"
37 #include "util/hist.h"
38 #include "util/data.h"
39 #include "arch/common.h"
40 #include "util/time-utils.h"
41 #include "util/auxtrace.h"
42 #include "util/units.h"
43 #include "util/branch.h"
50 #include <linux/bitmap.h>
51 #include <linux/stringify.h>
52 #include <sys/types.h>
55 #include <linux/mman.h>
58 struct perf_tool tool
;
59 struct perf_session
*session
;
60 bool use_tui
, use_gtk
, use_stdio
;
63 bool inverted_callchain
;
70 bool nonany_branch_mode
;
73 struct perf_read_values show_threads_values
;
74 const char *pretty_printing_style
;
76 const char *symbol_filter_str
;
78 struct perf_time_interval
*ptime_range
;
85 DECLARE_BITMAP(cpu_bitmap
, MAX_NR_CPUS
);
86 struct branch_type_stat brtype_stat
;
89 static int report__config(const char *var
, const char *value
, void *cb
)
91 struct report
*rep
= cb
;
93 if (!strcmp(var
, "report.group")) {
94 symbol_conf
.event_group
= perf_config_bool(var
, value
);
97 if (!strcmp(var
, "report.percent-limit")) {
98 double pcnt
= strtof(value
, NULL
);
100 rep
->min_percent
= pcnt
;
101 callchain_param
.min_percent
= pcnt
;
104 if (!strcmp(var
, "report.children")) {
105 symbol_conf
.cumulate_callchain
= perf_config_bool(var
, value
);
108 if (!strcmp(var
, "report.queue-size"))
109 return perf_config_u64(&rep
->queue_size
, var
, value
);
111 if (!strcmp(var
, "report.sort_order")) {
112 default_sort_order
= strdup(value
);
119 static int hist_iter__report_callback(struct hist_entry_iter
*iter
,
120 struct addr_location
*al
, bool single
,
124 struct report
*rep
= arg
;
125 struct hist_entry
*he
= iter
->he
;
126 struct perf_evsel
*evsel
= iter
->evsel
;
127 struct perf_sample
*sample
= iter
->sample
;
129 struct branch_info
*bi
;
131 if (!ui__has_annotation())
134 hist__account_cycles(sample
->branch_stack
, al
, sample
,
135 rep
->nonany_branch_mode
);
137 if (sort__mode
== SORT_MODE__BRANCH
) {
138 bi
= he
->branch_info
;
139 err
= addr_map_symbol__inc_samples(&bi
->from
, sample
, evsel
->idx
);
143 err
= addr_map_symbol__inc_samples(&bi
->to
, sample
, evsel
->idx
);
145 } else if (rep
->mem_mode
) {
147 err
= addr_map_symbol__inc_samples(&mi
->daddr
, sample
, evsel
->idx
);
151 err
= hist_entry__inc_addr_samples(he
, sample
, evsel
->idx
, al
->addr
);
153 } else if (symbol_conf
.cumulate_callchain
) {
155 err
= hist_entry__inc_addr_samples(he
, sample
, evsel
->idx
,
158 err
= hist_entry__inc_addr_samples(he
, sample
, evsel
->idx
, al
->addr
);
165 static int hist_iter__branch_callback(struct hist_entry_iter
*iter
,
166 struct addr_location
*al __maybe_unused
,
167 bool single __maybe_unused
,
170 struct hist_entry
*he
= iter
->he
;
171 struct report
*rep
= arg
;
172 struct branch_info
*bi
;
173 struct perf_sample
*sample
= iter
->sample
;
174 struct perf_evsel
*evsel
= iter
->evsel
;
177 if (!ui__has_annotation())
180 hist__account_cycles(sample
->branch_stack
, al
, sample
,
181 rep
->nonany_branch_mode
);
183 bi
= he
->branch_info
;
184 err
= addr_map_symbol__inc_samples(&bi
->from
, sample
, evsel
->idx
);
188 err
= addr_map_symbol__inc_samples(&bi
->to
, sample
, evsel
->idx
);
190 branch_type_count(&rep
->brtype_stat
, &bi
->flags
,
191 bi
->from
.addr
, bi
->to
.addr
);
198 * Events in data file are not collect in groups, but we still want
199 * the group display. Set the artificial group and set the leader's
200 * forced_leader flag to notify the display code.
202 static void setup_forced_leader(struct report
*report
,
203 struct perf_evlist
*evlist
)
205 if (report
->group_set
&& !evlist
->nr_groups
) {
206 struct perf_evsel
*leader
= perf_evlist__first(evlist
);
208 perf_evlist__set_leader(evlist
);
209 leader
->forced_leader
= true;
213 static int process_feature_event(struct perf_tool
*tool
,
214 union perf_event
*event
,
215 struct perf_session
*session __maybe_unused
)
217 struct report
*rep
= container_of(tool
, struct report
, tool
);
219 if (event
->feat
.feat_id
< HEADER_LAST_FEATURE
)
220 return perf_event__process_feature(tool
, event
, session
);
222 if (event
->feat
.feat_id
!= HEADER_LAST_FEATURE
) {
223 pr_err("failed: wrong feature ID: %" PRIu64
"\n",
224 event
->feat
.feat_id
);
229 * All features are received, we can force the
232 setup_forced_leader(rep
, session
->evlist
);
236 static int process_sample_event(struct perf_tool
*tool
,
237 union perf_event
*event
,
238 struct perf_sample
*sample
,
239 struct perf_evsel
*evsel
,
240 struct machine
*machine
)
242 struct report
*rep
= container_of(tool
, struct report
, tool
);
243 struct addr_location al
;
244 struct hist_entry_iter iter
= {
247 .hide_unresolved
= symbol_conf
.hide_unresolved
,
248 .add_entry_cb
= hist_iter__report_callback
,
252 if (perf_time__ranges_skip_sample(rep
->ptime_range
, rep
->range_num
,
257 if (machine__resolve(machine
, &al
, sample
) < 0) {
258 pr_debug("problem processing %d event, skipping it.\n",
263 if (symbol_conf
.hide_unresolved
&& al
.sym
== NULL
)
266 if (rep
->cpu_list
&& !test_bit(sample
->cpu
, rep
->cpu_bitmap
))
269 if (sort__mode
== SORT_MODE__BRANCH
) {
271 * A non-synthesized event might not have a branch stack if
272 * branch stacks have been synthesized (using itrace options).
274 if (!sample
->branch_stack
)
277 iter
.add_entry_cb
= hist_iter__branch_callback
;
278 iter
.ops
= &hist_iter_branch
;
279 } else if (rep
->mem_mode
) {
280 iter
.ops
= &hist_iter_mem
;
281 } else if (symbol_conf
.cumulate_callchain
) {
282 iter
.ops
= &hist_iter_cumulative
;
284 iter
.ops
= &hist_iter_normal
;
288 al
.map
->dso
->hit
= 1;
290 ret
= hist_entry_iter__add(&iter
, &al
, rep
->max_stack
, rep
);
292 pr_debug("problem adding hist entry, skipping event\n");
294 addr_location__put(&al
);
298 static int process_read_event(struct perf_tool
*tool
,
299 union perf_event
*event
,
300 struct perf_sample
*sample __maybe_unused
,
301 struct perf_evsel
*evsel
,
302 struct machine
*machine __maybe_unused
)
304 struct report
*rep
= container_of(tool
, struct report
, tool
);
306 if (rep
->show_threads
) {
307 const char *name
= evsel
? perf_evsel__name(evsel
) : "unknown";
308 int err
= perf_read_values_add_value(&rep
->show_threads_values
,
309 event
->read
.pid
, event
->read
.tid
,
321 /* For pipe mode, sample_type is not currently set */
322 static int report__setup_sample_type(struct report
*rep
)
324 struct perf_session
*session
= rep
->session
;
325 u64 sample_type
= perf_evlist__combined_sample_type(session
->evlist
);
326 bool is_pipe
= perf_data__is_pipe(session
->data
);
328 if (session
->itrace_synth_opts
->callchain
||
330 perf_header__has_feat(&session
->header
, HEADER_AUXTRACE
) &&
331 !session
->itrace_synth_opts
->set
))
332 sample_type
|= PERF_SAMPLE_CALLCHAIN
;
334 if (session
->itrace_synth_opts
->last_branch
)
335 sample_type
|= PERF_SAMPLE_BRANCH_STACK
;
337 if (!is_pipe
&& !(sample_type
& PERF_SAMPLE_CALLCHAIN
)) {
338 if (perf_hpp_list
.parent
) {
339 ui__error("Selected --sort parent, but no "
340 "callchain data. Did you call "
341 "'perf record' without -g?\n");
344 if (symbol_conf
.use_callchain
&&
345 !symbol_conf
.show_branchflag_count
) {
346 ui__error("Selected -g or --branch-history.\n"
347 "But no callchain or branch data.\n"
348 "Did you call 'perf record' without -g or -b?\n");
351 } else if (!callchain_param
.enabled
&&
352 callchain_param
.mode
!= CHAIN_NONE
&&
353 !symbol_conf
.use_callchain
) {
354 symbol_conf
.use_callchain
= true;
355 if (callchain_register_param(&callchain_param
) < 0) {
356 ui__error("Can't register callchain params.\n");
361 if (symbol_conf
.cumulate_callchain
) {
362 /* Silently ignore if callchain is missing */
363 if (!(sample_type
& PERF_SAMPLE_CALLCHAIN
)) {
364 symbol_conf
.cumulate_callchain
= false;
365 perf_hpp__cancel_cumulate();
369 if (sort__mode
== SORT_MODE__BRANCH
) {
371 !(sample_type
& PERF_SAMPLE_BRANCH_STACK
)) {
372 ui__error("Selected -b but no branch data. "
373 "Did you call perf record without -b?\n");
378 if (symbol_conf
.use_callchain
|| symbol_conf
.cumulate_callchain
) {
379 if ((sample_type
& PERF_SAMPLE_REGS_USER
) &&
380 (sample_type
& PERF_SAMPLE_STACK_USER
)) {
381 callchain_param
.record_mode
= CALLCHAIN_DWARF
;
382 dwarf_callchain_users
= true;
383 } else if (sample_type
& PERF_SAMPLE_BRANCH_STACK
)
384 callchain_param
.record_mode
= CALLCHAIN_LBR
;
386 callchain_param
.record_mode
= CALLCHAIN_FP
;
389 /* ??? handle more cases than just ANY? */
390 if (!(perf_evlist__combined_branch_type(session
->evlist
) &
391 PERF_SAMPLE_BRANCH_ANY
))
392 rep
->nonany_branch_mode
= true;
397 static void sig_handler(int sig __maybe_unused
)
402 static size_t hists__fprintf_nr_sample_events(struct hists
*hists
, struct report
*rep
,
403 const char *evname
, FILE *fp
)
407 unsigned long nr_samples
= hists
->stats
.nr_events
[PERF_RECORD_SAMPLE
];
408 u64 nr_events
= hists
->stats
.total_period
;
409 struct perf_evsel
*evsel
= hists_to_evsel(hists
);
411 size_t size
= sizeof(buf
);
412 int socked_id
= hists
->socket_filter
;
417 if (symbol_conf
.filter_relative
) {
418 nr_samples
= hists
->stats
.nr_non_filtered_samples
;
419 nr_events
= hists
->stats
.total_non_filtered_period
;
422 if (perf_evsel__is_group_event(evsel
)) {
423 struct perf_evsel
*pos
;
425 perf_evsel__group_desc(evsel
, buf
, size
);
428 for_each_group_member(pos
, evsel
) {
429 const struct hists
*pos_hists
= evsel__hists(pos
);
431 if (symbol_conf
.filter_relative
) {
432 nr_samples
+= pos_hists
->stats
.nr_non_filtered_samples
;
433 nr_events
+= pos_hists
->stats
.total_non_filtered_period
;
435 nr_samples
+= pos_hists
->stats
.nr_events
[PERF_RECORD_SAMPLE
];
436 nr_events
+= pos_hists
->stats
.total_period
;
441 nr_samples
= convert_unit(nr_samples
, &unit
);
442 ret
= fprintf(fp
, "# Samples: %lu%c", nr_samples
, unit
);
443 if (evname
!= NULL
) {
444 ret
+= fprintf(fp
, " of event%s '%s'",
445 evsel
->nr_members
> 1 ? "s" : "", evname
);
449 ret
+= fprintf(fp
, " (time slices: %s)", rep
->time_str
);
451 if (symbol_conf
.show_ref_callgraph
&&
452 strstr(evname
, "call-graph=no")) {
453 ret
+= fprintf(fp
, ", show reference callgraph");
457 ret
+= fprintf(fp
, "\n# Total weight : %" PRIu64
, nr_events
);
458 ret
+= fprintf(fp
, "\n# Sort order : %s", sort_order
? : default_mem_sort_order
);
460 ret
+= fprintf(fp
, "\n# Event count (approx.): %" PRIu64
, nr_events
);
463 ret
+= fprintf(fp
, "\n# Processor Socket: %d", socked_id
);
465 return ret
+ fprintf(fp
, "\n#\n");
468 static int perf_evlist__tty_browse_hists(struct perf_evlist
*evlist
,
472 struct perf_evsel
*pos
;
475 fprintf(stdout
, "#\n# Total Lost Samples: %" PRIu64
"\n#\n",
476 evlist
->stats
.total_lost_samples
);
479 evlist__for_each_entry(evlist
, pos
) {
480 struct hists
*hists
= evsel__hists(pos
);
481 const char *evname
= perf_evsel__name(pos
);
483 if (symbol_conf
.event_group
&&
484 !perf_evsel__is_group_leader(pos
))
487 hists__fprintf_nr_sample_events(hists
, rep
, evname
, stdout
);
488 hists__fprintf(hists
, !quiet
, 0, 0, rep
->min_percent
, stdout
,
489 symbol_conf
.use_callchain
||
490 symbol_conf
.show_branchflag_count
);
491 fprintf(stdout
, "\n\n");
495 fprintf(stdout
, "#\n# (%s)\n#\n", help
);
497 if (rep
->show_threads
) {
498 bool style
= !strcmp(rep
->pretty_printing_style
, "raw");
499 perf_read_values_display(stdout
, &rep
->show_threads_values
,
501 perf_read_values_destroy(&rep
->show_threads_values
);
504 if (sort__mode
== SORT_MODE__BRANCH
)
505 branch_type_stat_display(stdout
, &rep
->brtype_stat
);
510 static void report__warn_kptr_restrict(const struct report
*rep
)
512 struct map
*kernel_map
= machine__kernel_map(&rep
->session
->machines
.host
);
513 struct kmap
*kernel_kmap
= kernel_map
? map__kmap(kernel_map
) : NULL
;
515 if (perf_evlist__exclude_kernel(rep
->session
->evlist
))
518 if (kernel_map
== NULL
||
519 (kernel_map
->dso
->hit
&&
520 (kernel_kmap
->ref_reloc_sym
== NULL
||
521 kernel_kmap
->ref_reloc_sym
->addr
== 0))) {
523 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
524 "can't be resolved.";
526 if (kernel_map
&& map__has_symbols(kernel_map
)) {
527 desc
= "If some relocation was applied (e.g. "
528 "kexec) symbols may be misresolved.";
532 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
533 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
534 "Samples in kernel modules can't be resolved as well.\n\n",
539 static int report__gtk_browse_hists(struct report
*rep
, const char *help
)
541 int (*hist_browser
)(struct perf_evlist
*evlist
, const char *help
,
542 struct hist_browser_timer
*timer
, float min_pcnt
);
544 hist_browser
= dlsym(perf_gtk_handle
, "perf_evlist__gtk_browse_hists");
546 if (hist_browser
== NULL
) {
547 ui__error("GTK browser not found!\n");
551 return hist_browser(rep
->session
->evlist
, help
, NULL
, rep
->min_percent
);
554 static int report__browse_hists(struct report
*rep
)
557 struct perf_session
*session
= rep
->session
;
558 struct perf_evlist
*evlist
= session
->evlist
;
559 const char *help
= perf_tip(system_path(TIPDIR
));
562 /* fallback for people who don't install perf ;-) */
563 help
= perf_tip(DOCDIR
);
565 help
= "Cannot load tips.txt file, please install perf!";
568 switch (use_browser
) {
570 ret
= perf_evlist__tui_browse_hists(evlist
, help
, NULL
,
572 &session
->header
.env
,
575 * Usually "ret" is the last pressed key, and we only
576 * care if the key notifies us to switch data file.
578 if (ret
!= K_SWITCH_INPUT_DATA
)
582 ret
= report__gtk_browse_hists(rep
, help
);
585 ret
= perf_evlist__tty_browse_hists(evlist
, rep
, help
);
592 static int report__collapse_hists(struct report
*rep
)
594 struct ui_progress prog
;
595 struct perf_evsel
*pos
;
598 ui_progress__init(&prog
, rep
->nr_entries
, "Merging related events...");
600 evlist__for_each_entry(rep
->session
->evlist
, pos
) {
601 struct hists
*hists
= evsel__hists(pos
);
604 hists
->symbol_filter_str
= rep
->symbol_filter_str
;
606 hists
->socket_filter
= rep
->socket_filter
;
608 ret
= hists__collapse_resort(hists
, &prog
);
612 /* Non-group events are considered as leader */
613 if (symbol_conf
.event_group
&&
614 !perf_evsel__is_group_leader(pos
)) {
615 struct hists
*leader_hists
= evsel__hists(pos
->leader
);
617 hists__match(leader_hists
, hists
);
618 hists__link(leader_hists
, hists
);
622 ui_progress__finish();
626 static void report__output_resort(struct report
*rep
)
628 struct ui_progress prog
;
629 struct perf_evsel
*pos
;
631 ui_progress__init(&prog
, rep
->nr_entries
, "Sorting events for output...");
633 evlist__for_each_entry(rep
->session
->evlist
, pos
)
634 perf_evsel__output_resort(pos
, &prog
);
636 ui_progress__finish();
639 static void stats_setup(struct report
*rep
)
641 memset(&rep
->tool
, 0, sizeof(rep
->tool
));
642 rep
->tool
.no_warn
= true;
645 static int stats_print(struct report
*rep
)
647 struct perf_session
*session
= rep
->session
;
649 perf_session__fprintf_nr_events(session
, stdout
);
653 static void tasks_setup(struct report
*rep
)
655 memset(&rep
->tool
, 0, sizeof(rep
->tool
));
656 rep
->tool
.ordered_events
= true;
657 if (rep
->mmaps_mode
) {
658 rep
->tool
.mmap
= perf_event__process_mmap
;
659 rep
->tool
.mmap2
= perf_event__process_mmap2
;
661 rep
->tool
.comm
= perf_event__process_comm
;
662 rep
->tool
.exit
= perf_event__process_exit
;
663 rep
->tool
.fork
= perf_event__process_fork
;
664 rep
->tool
.no_warn
= true;
668 struct thread
*thread
;
669 struct list_head list
;
670 struct list_head children
;
673 static struct task
*tasks_list(struct task
*task
, struct machine
*machine
)
675 struct thread
*parent_thread
, *thread
= task
->thread
;
676 struct task
*parent_task
;
678 /* Already listed. */
679 if (!list_empty(&task
->list
))
682 /* Last one in the chain. */
683 if (thread
->ppid
== -1)
686 parent_thread
= machine__find_thread(machine
, -1, thread
->ppid
);
688 return ERR_PTR(-ENOENT
);
690 parent_task
= thread__priv(parent_thread
);
691 list_add_tail(&task
->list
, &parent_task
->children
);
692 return tasks_list(parent_task
, machine
);
695 static size_t maps__fprintf_task(struct maps
*maps
, int indent
, FILE *fp
)
700 for (nd
= rb_first(&maps
->entries
); nd
; nd
= rb_next(nd
)) {
701 struct map
*map
= rb_entry(nd
, struct map
, rb_node
);
703 printed
+= fprintf(fp
, "%*s %" PRIx64
"-%" PRIx64
" %c%c%c%c %08" PRIx64
" %" PRIu64
" %s\n",
704 indent
, "", map
->start
, map
->end
,
705 map
->prot
& PROT_READ
? 'r' : '-',
706 map
->prot
& PROT_WRITE
? 'w' : '-',
707 map
->prot
& PROT_EXEC
? 'x' : '-',
708 map
->flags
& MAP_SHARED
? 's' : 'p',
710 map
->ino
, map
->dso
->name
);
716 static int map_groups__fprintf_task(struct map_groups
*mg
, int indent
, FILE *fp
)
718 return maps__fprintf_task(&mg
->maps
, indent
, fp
);
721 static void task__print_level(struct task
*task
, FILE *fp
, int level
)
723 struct thread
*thread
= task
->thread
;
725 int comm_indent
= fprintf(fp
, " %8d %8d %8d |%*s",
726 thread
->pid_
, thread
->tid
, thread
->ppid
,
729 fprintf(fp
, "%s\n", thread__comm_str(thread
));
731 map_groups__fprintf_task(thread
->mg
, comm_indent
, fp
);
733 if (!list_empty(&task
->children
)) {
734 list_for_each_entry(child
, &task
->children
, list
)
735 task__print_level(child
, fp
, level
+ 1);
739 static int tasks_print(struct report
*rep
, FILE *fp
)
741 struct perf_session
*session
= rep
->session
;
742 struct machine
*machine
= &session
->machines
.host
;
743 struct task
*tasks
, *task
;
744 unsigned int nr
= 0, itask
= 0, i
;
749 * No locking needed while accessing machine->threads,
750 * because --tasks is single threaded command.
753 /* Count all the threads. */
754 for (i
= 0; i
< THREADS__TABLE_SIZE
; i
++)
755 nr
+= machine
->threads
[i
].nr
;
757 tasks
= malloc(sizeof(*tasks
) * nr
);
761 for (i
= 0; i
< THREADS__TABLE_SIZE
; i
++) {
762 struct threads
*threads
= &machine
->threads
[i
];
764 for (nd
= rb_first(&threads
->entries
); nd
; nd
= rb_next(nd
)) {
765 task
= tasks
+ itask
++;
767 task
->thread
= rb_entry(nd
, struct thread
, rb_node
);
768 INIT_LIST_HEAD(&task
->children
);
769 INIT_LIST_HEAD(&task
->list
);
770 thread__set_priv(task
->thread
, task
);
775 * Iterate every task down to the unprocessed parent
776 * and link all in task children list. Task with no
777 * parent is added into 'list'.
779 for (itask
= 0; itask
< nr
; itask
++) {
780 task
= tasks
+ itask
;
782 if (!list_empty(&task
->list
))
785 task
= tasks_list(task
, machine
);
787 pr_err("Error: failed to process tasks\n");
789 return PTR_ERR(task
);
793 list_add_tail(&task
->list
, &list
);
796 fprintf(fp
, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm");
798 list_for_each_entry(task
, &list
, list
)
799 task__print_level(task
, fp
, 0);
805 static int __cmd_report(struct report
*rep
)
808 struct perf_session
*session
= rep
->session
;
809 struct perf_evsel
*pos
;
810 struct perf_data
*data
= session
->data
;
812 signal(SIGINT
, sig_handler
);
815 ret
= perf_session__cpu_bitmap(session
, rep
->cpu_list
,
818 ui__error("failed to set cpu bitmap\n");
821 session
->itrace_synth_opts
->cpu_bitmap
= rep
->cpu_bitmap
;
824 if (rep
->show_threads
) {
825 ret
= perf_read_values_init(&rep
->show_threads_values
);
830 ret
= report__setup_sample_type(rep
);
832 /* report__setup_sample_type() already showed error message */
842 ret
= perf_session__process_events(session
);
844 ui__error("failed to process sample\n");
849 return stats_print(rep
);
852 return tasks_print(rep
, stdout
);
854 report__warn_kptr_restrict(rep
);
856 evlist__for_each_entry(session
->evlist
, pos
)
857 rep
->nr_entries
+= evsel__hists(pos
)->nr_entries
;
859 if (use_browser
== 0) {
861 perf_session__fprintf(session
, stdout
);
864 perf_session__fprintf_dsos(session
, stdout
);
867 perf_session__fprintf_nr_events(session
, stdout
);
868 perf_evlist__fprintf_nr_events(session
->evlist
, stdout
);
873 ret
= report__collapse_hists(rep
);
875 ui__error("failed to process hist entry\n");
883 * recalculate number of entries after collapsing since it
884 * might be changed during the collapse phase.
887 evlist__for_each_entry(session
->evlist
, pos
)
888 rep
->nr_entries
+= evsel__hists(pos
)->nr_entries
;
890 if (rep
->nr_entries
== 0) {
891 ui__error("The %s file has no samples!\n", data
->file
.path
);
895 report__output_resort(rep
);
897 return report__browse_hists(rep
);
901 report_parse_callchain_opt(const struct option
*opt
, const char *arg
, int unset
)
903 struct callchain_param
*callchain
= opt
->value
;
905 callchain
->enabled
= !unset
;
910 symbol_conf
.use_callchain
= false;
911 callchain
->mode
= CHAIN_NONE
;
915 return parse_callchain_report_opt(arg
);
919 report_parse_ignore_callees_opt(const struct option
*opt __maybe_unused
,
920 const char *arg
, int unset __maybe_unused
)
923 int err
= regcomp(&ignore_callees_regex
, arg
, REG_EXTENDED
);
926 regerror(err
, &ignore_callees_regex
, buf
, sizeof(buf
));
927 pr_err("Invalid --ignore-callees regex: %s\n%s", arg
, buf
);
930 have_ignore_callees
= 1;
937 parse_branch_mode(const struct option
*opt
,
938 const char *str __maybe_unused
, int unset
)
940 int *branch_mode
= opt
->value
;
942 *branch_mode
= !unset
;
947 parse_percent_limit(const struct option
*opt
, const char *str
,
948 int unset __maybe_unused
)
950 struct report
*rep
= opt
->value
;
951 double pcnt
= strtof(str
, NULL
);
953 rep
->min_percent
= pcnt
;
954 callchain_param
.min_percent
= pcnt
;
958 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
960 const char report_callchain_help
[] = "Display call graph (stack chain/backtrace):\n\n"
961 CALLCHAIN_REPORT_HELP
962 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT
;
964 int cmd_report(int argc
, const char **argv
)
966 struct perf_session
*session
;
967 struct itrace_synth_opts itrace_synth_opts
= { .set
= 0, };
969 bool has_br_stack
= false;
970 int branch_mode
= -1;
971 bool branch_call_mode
= false;
972 char callchain_default_opt
[] = CALLCHAIN_DEFAULT_OPT
;
973 const char * const report_usage
[] = {
974 "perf report [<options>]",
977 struct report report
= {
979 .sample
= process_sample_event
,
980 .mmap
= perf_event__process_mmap
,
981 .mmap2
= perf_event__process_mmap2
,
982 .comm
= perf_event__process_comm
,
983 .namespaces
= perf_event__process_namespaces
,
984 .exit
= perf_event__process_exit
,
985 .fork
= perf_event__process_fork
,
986 .lost
= perf_event__process_lost
,
987 .read
= process_read_event
,
988 .attr
= perf_event__process_attr
,
989 .tracing_data
= perf_event__process_tracing_data
,
990 .build_id
= perf_event__process_build_id
,
991 .id_index
= perf_event__process_id_index
,
992 .auxtrace_info
= perf_event__process_auxtrace_info
,
993 .auxtrace
= perf_event__process_auxtrace
,
994 .feature
= process_feature_event
,
995 .ordered_events
= true,
996 .ordering_requires_timestamps
= true,
998 .max_stack
= PERF_MAX_STACK_DEPTH
,
999 .pretty_printing_style
= "normal",
1000 .socket_filter
= -1,
1002 const struct option options
[] = {
1003 OPT_STRING('i', "input", &input_name
, "file",
1005 OPT_INCR('v', "verbose", &verbose
,
1006 "be more verbose (show symbol address, etc)"),
1007 OPT_BOOLEAN('q', "quiet", &quiet
, "Do not show any message"),
1008 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace
,
1009 "dump raw trace in ASCII"),
1010 OPT_BOOLEAN(0, "stats", &report
.stats_mode
, "Display event stats"),
1011 OPT_BOOLEAN(0, "tasks", &report
.tasks_mode
, "Display recorded tasks"),
1012 OPT_BOOLEAN(0, "mmaps", &report
.mmaps_mode
, "Display recorded tasks memory maps"),
1013 OPT_STRING('k', "vmlinux", &symbol_conf
.vmlinux_name
,
1014 "file", "vmlinux pathname"),
1015 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf
.ignore_vmlinux
,
1016 "don't load vmlinux even if found"),
1017 OPT_STRING(0, "kallsyms", &symbol_conf
.kallsyms_name
,
1018 "file", "kallsyms pathname"),
1019 OPT_BOOLEAN('f', "force", &symbol_conf
.force
, "don't complain, do it"),
1020 OPT_BOOLEAN('m', "modules", &symbol_conf
.use_modules
,
1021 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1022 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf
.show_nr_samples
,
1023 "Show a column with the number of samples"),
1024 OPT_BOOLEAN('T', "threads", &report
.show_threads
,
1025 "Show per-thread event counters"),
1026 OPT_STRING(0, "pretty", &report
.pretty_printing_style
, "key",
1027 "pretty printing style key: normal raw"),
1028 OPT_BOOLEAN(0, "tui", &report
.use_tui
, "Use the TUI interface"),
1029 OPT_BOOLEAN(0, "gtk", &report
.use_gtk
, "Use the GTK2 interface"),
1030 OPT_BOOLEAN(0, "stdio", &report
.use_stdio
,
1031 "Use the stdio interface"),
1032 OPT_BOOLEAN(0, "header", &report
.header
, "Show data header."),
1033 OPT_BOOLEAN(0, "header-only", &report
.header_only
,
1034 "Show only data header."),
1035 OPT_STRING('s', "sort", &sort_order
, "key[,key2...]",
1036 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
1037 " Please refer the man page for the complete list."),
1038 OPT_STRING('F', "fields", &field_order
, "key[,keys...]",
1039 "output field(s): overhead, period, sample plus all of sort keys"),
1040 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf
.show_cpu_utilization
,
1041 "Show sample percentage for different cpu modes"),
1042 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf
.show_cpu_utilization
,
1043 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN
),
1044 OPT_STRING('p', "parent", &parent_pattern
, "regex",
1045 "regex filter to identify parent, see: '--sort parent'"),
1046 OPT_BOOLEAN('x', "exclude-other", &symbol_conf
.exclude_other
,
1047 "Only display entries with parent-match"),
1048 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param
,
1049 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
1050 report_callchain_help
, &report_parse_callchain_opt
,
1051 callchain_default_opt
),
1052 OPT_BOOLEAN(0, "children", &symbol_conf
.cumulate_callchain
,
1053 "Accumulate callchains of children and show total overhead as well"),
1054 OPT_INTEGER(0, "max-stack", &report
.max_stack
,
1055 "Set the maximum stack depth when parsing the callchain, "
1056 "anything beyond the specified depth will be ignored. "
1057 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH
)),
1058 OPT_BOOLEAN('G', "inverted", &report
.inverted_callchain
,
1059 "alias for inverted call graph"),
1060 OPT_CALLBACK(0, "ignore-callees", NULL
, "regex",
1061 "ignore callees of these functions in call graphs",
1062 report_parse_ignore_callees_opt
),
1063 OPT_STRING('d', "dsos", &symbol_conf
.dso_list_str
, "dso[,dso...]",
1064 "only consider symbols in these dsos"),
1065 OPT_STRING('c', "comms", &symbol_conf
.comm_list_str
, "comm[,comm...]",
1066 "only consider symbols in these comms"),
1067 OPT_STRING(0, "pid", &symbol_conf
.pid_list_str
, "pid[,pid...]",
1068 "only consider symbols in these pids"),
1069 OPT_STRING(0, "tid", &symbol_conf
.tid_list_str
, "tid[,tid...]",
1070 "only consider symbols in these tids"),
1071 OPT_STRING('S', "symbols", &symbol_conf
.sym_list_str
, "symbol[,symbol...]",
1072 "only consider these symbols"),
1073 OPT_STRING(0, "symbol-filter", &report
.symbol_filter_str
, "filter",
1074 "only show symbols that (partially) match with this filter"),
1075 OPT_STRING('w', "column-widths", &symbol_conf
.col_width_list_str
,
1077 "don't try to adjust column width, use these fixed values"),
1078 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf
.field_sep
, "separator",
1079 "separator for columns, no spaces will be added between "
1080 "columns '.' is reserved."),
1081 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf
.hide_unresolved
,
1082 "Only display entries resolved to a symbol"),
1083 OPT_CALLBACK(0, "symfs", NULL
, "directory",
1084 "Look for files with symbols relative to this directory",
1085 symbol__config_symfs
),
1086 OPT_STRING('C', "cpu", &report
.cpu_list
, "cpu",
1087 "list of cpus to profile"),
1088 OPT_BOOLEAN('I', "show-info", &report
.show_full_info
,
1089 "Display extended information about perf.data file"),
1090 OPT_BOOLEAN(0, "source", &symbol_conf
.annotate_src
,
1091 "Interleave source code with assembly code (default)"),
1092 OPT_BOOLEAN(0, "asm-raw", &symbol_conf
.annotate_asm_raw
,
1093 "Display raw encoding of assembly instructions (default)"),
1094 OPT_STRING('M', "disassembler-style", &disassembler_style
, "disassembler style",
1095 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1096 OPT_BOOLEAN(0, "show-total-period", &symbol_conf
.show_total_period
,
1097 "Show a column with the sum of periods"),
1098 OPT_BOOLEAN_SET(0, "group", &symbol_conf
.event_group
, &report
.group_set
,
1099 "Show event group information together"),
1100 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode
, "",
1101 "use branch records for per branch histogram filling",
1103 OPT_BOOLEAN(0, "branch-history", &branch_call_mode
,
1104 "add last branch records to call history"),
1105 OPT_STRING(0, "objdump", &objdump_path
, "path",
1106 "objdump binary to use for disassembly and annotations"),
1107 OPT_BOOLEAN(0, "demangle", &symbol_conf
.demangle
,
1108 "Disable symbol demangling"),
1109 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf
.demangle_kernel
,
1110 "Enable kernel symbol demangling"),
1111 OPT_BOOLEAN(0, "mem-mode", &report
.mem_mode
, "mem access profile"),
1112 OPT_CALLBACK(0, "percent-limit", &report
, "percent",
1113 "Don't show entries under that percent", parse_percent_limit
),
1114 OPT_CALLBACK(0, "percentage", NULL
, "relative|absolute",
1115 "how to display percentage of filtered entries", parse_filter_percentage
),
1116 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts
, NULL
, "opts",
1117 "Instruction Tracing options",
1118 itrace_parse_synth_opts
),
1119 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename
,
1120 "Show full source file name path for source lines"),
1121 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf
.show_ref_callgraph
,
1122 "Show callgraph from reference event"),
1123 OPT_INTEGER(0, "socket-filter", &report
.socket_filter
,
1124 "only show processor socket that match with this filter"),
1125 OPT_BOOLEAN(0, "raw-trace", &symbol_conf
.raw_trace
,
1126 "Show raw trace event output (do not use print fmt or plugins)"),
1127 OPT_BOOLEAN(0, "hierarchy", &symbol_conf
.report_hierarchy
,
1128 "Show entries in a hierarchy"),
1129 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL
, "mode",
1130 "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
1131 stdio__config_color
, "always"),
1132 OPT_STRING(0, "time", &report
.time_str
, "str",
1133 "Time span of interest (start,stop)"),
1134 OPT_BOOLEAN(0, "inline", &symbol_conf
.inline_name
,
1135 "Show inline function"),
1138 struct perf_data data
= {
1139 .mode
= PERF_DATA_MODE_READ
,
1141 int ret
= hists__init();
1146 ret
= perf_config(report__config
, &report
);
1150 argc
= parse_options(argc
, argv
, options
, report_usage
, 0);
1153 * Special case: if there's an argument left then assume that
1154 * it's a symbol filter:
1157 usage_with_options(report_usage
, options
);
1159 report
.symbol_filter_str
= argv
[0];
1162 if (report
.mmaps_mode
)
1163 report
.tasks_mode
= true;
1166 perf_quiet_option();
1168 if (symbol_conf
.vmlinux_name
&&
1169 access(symbol_conf
.vmlinux_name
, R_OK
)) {
1170 pr_err("Invalid file: %s\n", symbol_conf
.vmlinux_name
);
1173 if (symbol_conf
.kallsyms_name
&&
1174 access(symbol_conf
.kallsyms_name
, R_OK
)) {
1175 pr_err("Invalid file: %s\n", symbol_conf
.kallsyms_name
);
1179 if (report
.inverted_callchain
)
1180 callchain_param
.order
= ORDER_CALLER
;
1181 if (symbol_conf
.cumulate_callchain
&& !callchain_param
.order_set
)
1182 callchain_param
.order
= ORDER_CALLER
;
1184 if (itrace_synth_opts
.callchain
&&
1185 (int)itrace_synth_opts
.callchain_sz
> report
.max_stack
)
1186 report
.max_stack
= itrace_synth_opts
.callchain_sz
;
1188 if (!input_name
|| !strlen(input_name
)) {
1189 if (!fstat(STDIN_FILENO
, &st
) && S_ISFIFO(st
.st_mode
))
1192 input_name
= "perf.data";
1195 data
.file
.path
= input_name
;
1196 data
.force
= symbol_conf
.force
;
1199 session
= perf_session__new(&data
, false, &report
.tool
);
1200 if (session
== NULL
)
1203 if (report
.queue_size
) {
1204 ordered_events__set_alloc_size(&session
->ordered_events
,
1208 session
->itrace_synth_opts
= &itrace_synth_opts
;
1210 report
.session
= session
;
1212 has_br_stack
= perf_header__has_feat(&session
->header
,
1213 HEADER_BRANCH_STACK
);
1215 setup_forced_leader(&report
, session
->evlist
);
1217 if (itrace_synth_opts
.last_branch
)
1218 has_br_stack
= true;
1220 if (has_br_stack
&& branch_call_mode
)
1221 symbol_conf
.show_branchflag_count
= true;
1223 memset(&report
.brtype_stat
, 0, sizeof(struct branch_type_stat
));
1226 * Branch mode is a tristate:
1227 * -1 means default, so decide based on the file having branch data.
1228 * 0/1 means the user chose a mode.
1230 if (((branch_mode
== -1 && has_br_stack
) || branch_mode
== 1) &&
1231 !branch_call_mode
) {
1232 sort__mode
= SORT_MODE__BRANCH
;
1233 symbol_conf
.cumulate_callchain
= false;
1235 if (branch_call_mode
) {
1236 callchain_param
.key
= CCKEY_ADDRESS
;
1237 callchain_param
.branch_callstack
= 1;
1238 symbol_conf
.use_callchain
= true;
1239 callchain_register_param(&callchain_param
);
1240 if (sort_order
== NULL
)
1241 sort_order
= "srcline,symbol,dso";
1244 if (report
.mem_mode
) {
1245 if (sort__mode
== SORT_MODE__BRANCH
) {
1246 pr_err("branch and mem mode incompatible\n");
1249 sort__mode
= SORT_MODE__MEMORY
;
1250 symbol_conf
.cumulate_callchain
= false;
1253 if (symbol_conf
.report_hierarchy
) {
1254 /* disable incompatible options */
1255 symbol_conf
.cumulate_callchain
= false;
1258 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1259 parse_options_usage(report_usage
, options
, "F", 1);
1260 parse_options_usage(NULL
, options
, "hierarchy", 0);
1264 perf_hpp_list
.need_collapse
= true;
1267 if (report
.use_stdio
)
1269 else if (report
.use_tui
)
1271 else if (report
.use_gtk
)
1274 /* Force tty output for header output and per-thread stat. */
1275 if (report
.header
|| report
.header_only
|| report
.show_threads
)
1277 if (report
.header
|| report
.header_only
)
1278 report
.tool
.show_feat_hdr
= SHOW_FEAT_HEADER
;
1279 if (report
.show_full_info
)
1280 report
.tool
.show_feat_hdr
= SHOW_FEAT_HEADER_FULL_INFO
;
1281 if (report
.stats_mode
|| report
.tasks_mode
)
1283 if (report
.stats_mode
&& report
.tasks_mode
) {
1284 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n");
1288 if (strcmp(input_name
, "-") != 0)
1289 setup_browser(true);
1293 if (setup_sorting(session
->evlist
) < 0) {
1295 parse_options_usage(report_usage
, options
, "s", 1);
1297 parse_options_usage(sort_order
? NULL
: report_usage
,
1302 if ((report
.header
|| report
.header_only
) && !quiet
) {
1303 perf_session__fprintf_info(session
, stdout
,
1304 report
.show_full_info
);
1305 if (report
.header_only
) {
1309 } else if (use_browser
== 0 && !quiet
&&
1310 !report
.stats_mode
&& !report
.tasks_mode
) {
1311 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
1316 * Only in the TUI browser we are doing integrated annotation,
1317 * so don't allocate extra space that won't be used in the stdio
1320 if (ui__has_annotation()) {
1321 ret
= symbol__annotation_init();
1325 * For searching by name on the "Browse map details".
1326 * providing it only in verbose mode not to bloat too
1327 * much struct symbol.
1331 * XXX: Need to provide a less kludgy way to ask for
1332 * more space per symbol, the u32 is for the index on
1334 * See symbol__browser_index.
1336 symbol_conf
.priv_size
+= sizeof(u32
);
1337 symbol_conf
.sort_by_name
= true;
1339 annotation_config__init();
1342 if (symbol__init(&session
->header
.env
) < 0)
1345 report
.ptime_range
= perf_time__range_alloc(report
.time_str
,
1346 &report
.range_size
);
1347 if (!report
.ptime_range
) {
1352 if (perf_time__parse_str(report
.ptime_range
, report
.time_str
) != 0) {
1353 if (session
->evlist
->first_sample_time
== 0 &&
1354 session
->evlist
->last_sample_time
== 0) {
1355 pr_err("HINT: no first/last sample time found in perf data.\n"
1356 "Please use latest perf binary to execute 'perf record'\n"
1357 "(if '--buildid-all' is enabled, please set '--timestamp-boundary').\n");
1362 report
.range_num
= perf_time__percent_parse_str(
1363 report
.ptime_range
, report
.range_size
,
1365 session
->evlist
->first_sample_time
,
1366 session
->evlist
->last_sample_time
);
1368 if (report
.range_num
< 0) {
1369 pr_err("Invalid time string\n");
1374 report
.range_num
= 1;
1377 if (session
->tevent
.pevent
&&
1378 pevent_set_function_resolver(session
->tevent
.pevent
,
1379 machine__resolve_kernel_addr
,
1380 &session
->machines
.host
) < 0) {
1381 pr_err("%s: failed to set libtraceevent function resolver\n",
1386 sort__setup_elide(stdout
);
1388 ret
= __cmd_report(&report
);
1389 if (ret
== K_SWITCH_INPUT_DATA
) {
1390 perf_session__delete(session
);
1396 zfree(&report
.ptime_range
);
1398 perf_session__delete(session
);