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 struct annotation_options annotation_opts
;
75 const char *pretty_printing_style
;
77 const char *symbol_filter_str
;
79 struct perf_time_interval
*ptime_range
;
86 DECLARE_BITMAP(cpu_bitmap
, MAX_NR_CPUS
);
87 struct branch_type_stat brtype_stat
;
90 static int report__config(const char *var
, const char *value
, void *cb
)
92 struct report
*rep
= cb
;
94 if (!strcmp(var
, "report.group")) {
95 symbol_conf
.event_group
= perf_config_bool(var
, value
);
98 if (!strcmp(var
, "report.percent-limit")) {
99 double pcnt
= strtof(value
, NULL
);
101 rep
->min_percent
= pcnt
;
102 callchain_param
.min_percent
= pcnt
;
105 if (!strcmp(var
, "report.children")) {
106 symbol_conf
.cumulate_callchain
= perf_config_bool(var
, value
);
109 if (!strcmp(var
, "report.queue-size"))
110 return perf_config_u64(&rep
->queue_size
, var
, value
);
112 if (!strcmp(var
, "report.sort_order")) {
113 default_sort_order
= strdup(value
);
120 static int hist_iter__report_callback(struct hist_entry_iter
*iter
,
121 struct addr_location
*al
, bool single
,
125 struct report
*rep
= arg
;
126 struct hist_entry
*he
= iter
->he
;
127 struct perf_evsel
*evsel
= iter
->evsel
;
128 struct perf_sample
*sample
= iter
->sample
;
130 struct branch_info
*bi
;
132 if (!ui__has_annotation())
135 hist__account_cycles(sample
->branch_stack
, al
, sample
,
136 rep
->nonany_branch_mode
);
138 if (sort__mode
== SORT_MODE__BRANCH
) {
139 bi
= he
->branch_info
;
140 err
= addr_map_symbol__inc_samples(&bi
->from
, sample
, evsel
);
144 err
= addr_map_symbol__inc_samples(&bi
->to
, sample
, evsel
);
146 } else if (rep
->mem_mode
) {
148 err
= addr_map_symbol__inc_samples(&mi
->daddr
, sample
, evsel
);
152 err
= hist_entry__inc_addr_samples(he
, sample
, evsel
, al
->addr
);
154 } else if (symbol_conf
.cumulate_callchain
) {
156 err
= hist_entry__inc_addr_samples(he
, sample
, evsel
, al
->addr
);
158 err
= hist_entry__inc_addr_samples(he
, sample
, evsel
, 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
);
188 err
= addr_map_symbol__inc_samples(&bi
->to
, sample
, evsel
);
190 branch_type_count(&rep
->brtype_stat
, &bi
->flags
,
191 bi
->from
.addr
, bi
->to
.addr
);
197 static void setup_forced_leader(struct report
*report
,
198 struct perf_evlist
*evlist
)
200 if (report
->group_set
)
201 perf_evlist__force_leader(evlist
);
204 static int process_feature_event(struct perf_tool
*tool
,
205 union perf_event
*event
,
206 struct perf_session
*session __maybe_unused
)
208 struct report
*rep
= container_of(tool
, struct report
, tool
);
210 if (event
->feat
.feat_id
< HEADER_LAST_FEATURE
)
211 return perf_event__process_feature(tool
, event
, session
);
213 if (event
->feat
.feat_id
!= HEADER_LAST_FEATURE
) {
214 pr_err("failed: wrong feature ID: %" PRIu64
"\n",
215 event
->feat
.feat_id
);
220 * (feat_id = HEADER_LAST_FEATURE) is the end marker which
221 * means all features are received, now we can force the
224 setup_forced_leader(rep
, session
->evlist
);
228 static int process_sample_event(struct perf_tool
*tool
,
229 union perf_event
*event
,
230 struct perf_sample
*sample
,
231 struct perf_evsel
*evsel
,
232 struct machine
*machine
)
234 struct report
*rep
= container_of(tool
, struct report
, tool
);
235 struct addr_location al
;
236 struct hist_entry_iter iter
= {
239 .hide_unresolved
= symbol_conf
.hide_unresolved
,
240 .add_entry_cb
= hist_iter__report_callback
,
244 if (perf_time__ranges_skip_sample(rep
->ptime_range
, rep
->range_num
,
249 if (machine__resolve(machine
, &al
, sample
) < 0) {
250 pr_debug("problem processing %d event, skipping it.\n",
255 if (symbol_conf
.hide_unresolved
&& al
.sym
== NULL
)
258 if (rep
->cpu_list
&& !test_bit(sample
->cpu
, rep
->cpu_bitmap
))
261 if (sort__mode
== SORT_MODE__BRANCH
) {
263 * A non-synthesized event might not have a branch stack if
264 * branch stacks have been synthesized (using itrace options).
266 if (!sample
->branch_stack
)
269 iter
.add_entry_cb
= hist_iter__branch_callback
;
270 iter
.ops
= &hist_iter_branch
;
271 } else if (rep
->mem_mode
) {
272 iter
.ops
= &hist_iter_mem
;
273 } else if (symbol_conf
.cumulate_callchain
) {
274 iter
.ops
= &hist_iter_cumulative
;
276 iter
.ops
= &hist_iter_normal
;
280 al
.map
->dso
->hit
= 1;
282 ret
= hist_entry_iter__add(&iter
, &al
, rep
->max_stack
, rep
);
284 pr_debug("problem adding hist entry, skipping event\n");
286 addr_location__put(&al
);
290 static int process_read_event(struct perf_tool
*tool
,
291 union perf_event
*event
,
292 struct perf_sample
*sample __maybe_unused
,
293 struct perf_evsel
*evsel
,
294 struct machine
*machine __maybe_unused
)
296 struct report
*rep
= container_of(tool
, struct report
, tool
);
298 if (rep
->show_threads
) {
299 const char *name
= evsel
? perf_evsel__name(evsel
) : "unknown";
300 int err
= perf_read_values_add_value(&rep
->show_threads_values
,
301 event
->read
.pid
, event
->read
.tid
,
313 /* For pipe mode, sample_type is not currently set */
314 static int report__setup_sample_type(struct report
*rep
)
316 struct perf_session
*session
= rep
->session
;
317 u64 sample_type
= perf_evlist__combined_sample_type(session
->evlist
);
318 bool is_pipe
= perf_data__is_pipe(session
->data
);
320 if (session
->itrace_synth_opts
->callchain
||
322 perf_header__has_feat(&session
->header
, HEADER_AUXTRACE
) &&
323 !session
->itrace_synth_opts
->set
))
324 sample_type
|= PERF_SAMPLE_CALLCHAIN
;
326 if (session
->itrace_synth_opts
->last_branch
)
327 sample_type
|= PERF_SAMPLE_BRANCH_STACK
;
329 if (!is_pipe
&& !(sample_type
& PERF_SAMPLE_CALLCHAIN
)) {
330 if (perf_hpp_list
.parent
) {
331 ui__error("Selected --sort parent, but no "
332 "callchain data. Did you call "
333 "'perf record' without -g?\n");
336 if (symbol_conf
.use_callchain
&&
337 !symbol_conf
.show_branchflag_count
) {
338 ui__error("Selected -g or --branch-history.\n"
339 "But no callchain or branch data.\n"
340 "Did you call 'perf record' without -g or -b?\n");
343 } else if (!callchain_param
.enabled
&&
344 callchain_param
.mode
!= CHAIN_NONE
&&
345 !symbol_conf
.use_callchain
) {
346 symbol_conf
.use_callchain
= true;
347 if (callchain_register_param(&callchain_param
) < 0) {
348 ui__error("Can't register callchain params.\n");
353 if (symbol_conf
.cumulate_callchain
) {
354 /* Silently ignore if callchain is missing */
355 if (!(sample_type
& PERF_SAMPLE_CALLCHAIN
)) {
356 symbol_conf
.cumulate_callchain
= false;
357 perf_hpp__cancel_cumulate();
361 if (sort__mode
== SORT_MODE__BRANCH
) {
363 !(sample_type
& PERF_SAMPLE_BRANCH_STACK
)) {
364 ui__error("Selected -b but no branch data. "
365 "Did you call perf record without -b?\n");
370 if (symbol_conf
.use_callchain
|| symbol_conf
.cumulate_callchain
) {
371 if ((sample_type
& PERF_SAMPLE_REGS_USER
) &&
372 (sample_type
& PERF_SAMPLE_STACK_USER
)) {
373 callchain_param
.record_mode
= CALLCHAIN_DWARF
;
374 dwarf_callchain_users
= true;
375 } else if (sample_type
& PERF_SAMPLE_BRANCH_STACK
)
376 callchain_param
.record_mode
= CALLCHAIN_LBR
;
378 callchain_param
.record_mode
= CALLCHAIN_FP
;
381 /* ??? handle more cases than just ANY? */
382 if (!(perf_evlist__combined_branch_type(session
->evlist
) &
383 PERF_SAMPLE_BRANCH_ANY
))
384 rep
->nonany_branch_mode
= true;
386 #if !defined(HAVE_LIBUNWIND_SUPPORT) && !defined(HAVE_DWARF_SUPPORT)
387 if (dwarf_callchain_users
) {
388 ui__warning("Please install libunwind or libdw "
389 "development packages during the perf build.\n");
396 static void sig_handler(int sig __maybe_unused
)
401 static size_t hists__fprintf_nr_sample_events(struct hists
*hists
, struct report
*rep
,
402 const char *evname
, FILE *fp
)
406 unsigned long nr_samples
= hists
->stats
.nr_events
[PERF_RECORD_SAMPLE
];
407 u64 nr_events
= hists
->stats
.total_period
;
408 struct perf_evsel
*evsel
= hists_to_evsel(hists
);
410 size_t size
= sizeof(buf
);
411 int socked_id
= hists
->socket_filter
;
416 if (symbol_conf
.filter_relative
) {
417 nr_samples
= hists
->stats
.nr_non_filtered_samples
;
418 nr_events
= hists
->stats
.total_non_filtered_period
;
421 if (perf_evsel__is_group_event(evsel
)) {
422 struct perf_evsel
*pos
;
424 perf_evsel__group_desc(evsel
, buf
, size
);
427 for_each_group_member(pos
, evsel
) {
428 const struct hists
*pos_hists
= evsel__hists(pos
);
430 if (symbol_conf
.filter_relative
) {
431 nr_samples
+= pos_hists
->stats
.nr_non_filtered_samples
;
432 nr_events
+= pos_hists
->stats
.total_non_filtered_period
;
434 nr_samples
+= pos_hists
->stats
.nr_events
[PERF_RECORD_SAMPLE
];
435 nr_events
+= pos_hists
->stats
.total_period
;
440 nr_samples
= convert_unit(nr_samples
, &unit
);
441 ret
= fprintf(fp
, "# Samples: %lu%c", nr_samples
, unit
);
442 if (evname
!= NULL
) {
443 ret
+= fprintf(fp
, " of event%s '%s'",
444 evsel
->nr_members
> 1 ? "s" : "", evname
);
448 ret
+= fprintf(fp
, " (time slices: %s)", rep
->time_str
);
450 if (symbol_conf
.show_ref_callgraph
&& evname
&& strstr(evname
, "call-graph=no")) {
451 ret
+= fprintf(fp
, ", show reference callgraph");
455 ret
+= fprintf(fp
, "\n# Total weight : %" PRIu64
, nr_events
);
456 ret
+= fprintf(fp
, "\n# Sort order : %s", sort_order
? : default_mem_sort_order
);
458 ret
+= fprintf(fp
, "\n# Event count (approx.): %" PRIu64
, nr_events
);
461 ret
+= fprintf(fp
, "\n# Processor Socket: %d", socked_id
);
463 return ret
+ fprintf(fp
, "\n#\n");
466 static int perf_evlist__tty_browse_hists(struct perf_evlist
*evlist
,
470 struct perf_evsel
*pos
;
473 fprintf(stdout
, "#\n# Total Lost Samples: %" PRIu64
"\n#\n",
474 evlist
->stats
.total_lost_samples
);
477 evlist__for_each_entry(evlist
, pos
) {
478 struct hists
*hists
= evsel__hists(pos
);
479 const char *evname
= perf_evsel__name(pos
);
481 if (symbol_conf
.event_group
&&
482 !perf_evsel__is_group_leader(pos
))
485 hists__fprintf_nr_sample_events(hists
, rep
, evname
, stdout
);
486 hists__fprintf(hists
, !quiet
, 0, 0, rep
->min_percent
, stdout
,
487 !(symbol_conf
.use_callchain
||
488 symbol_conf
.show_branchflag_count
));
489 fprintf(stdout
, "\n\n");
493 fprintf(stdout
, "#\n# (%s)\n#\n", help
);
495 if (rep
->show_threads
) {
496 bool style
= !strcmp(rep
->pretty_printing_style
, "raw");
497 perf_read_values_display(stdout
, &rep
->show_threads_values
,
499 perf_read_values_destroy(&rep
->show_threads_values
);
502 if (sort__mode
== SORT_MODE__BRANCH
)
503 branch_type_stat_display(stdout
, &rep
->brtype_stat
);
508 static void report__warn_kptr_restrict(const struct report
*rep
)
510 struct map
*kernel_map
= machine__kernel_map(&rep
->session
->machines
.host
);
511 struct kmap
*kernel_kmap
= kernel_map
? map__kmap(kernel_map
) : NULL
;
513 if (perf_evlist__exclude_kernel(rep
->session
->evlist
))
516 if (kernel_map
== NULL
||
517 (kernel_map
->dso
->hit
&&
518 (kernel_kmap
->ref_reloc_sym
== NULL
||
519 kernel_kmap
->ref_reloc_sym
->addr
== 0))) {
521 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
522 "can't be resolved.";
524 if (kernel_map
&& map__has_symbols(kernel_map
)) {
525 desc
= "If some relocation was applied (e.g. "
526 "kexec) symbols may be misresolved.";
530 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
531 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
532 "Samples in kernel modules can't be resolved as well.\n\n",
537 static int report__gtk_browse_hists(struct report
*rep
, const char *help
)
539 int (*hist_browser
)(struct perf_evlist
*evlist
, const char *help
,
540 struct hist_browser_timer
*timer
, float min_pcnt
);
542 hist_browser
= dlsym(perf_gtk_handle
, "perf_evlist__gtk_browse_hists");
544 if (hist_browser
== NULL
) {
545 ui__error("GTK browser not found!\n");
549 return hist_browser(rep
->session
->evlist
, help
, NULL
, rep
->min_percent
);
552 static int report__browse_hists(struct report
*rep
)
555 struct perf_session
*session
= rep
->session
;
556 struct perf_evlist
*evlist
= session
->evlist
;
557 const char *help
= perf_tip(system_path(TIPDIR
));
560 /* fallback for people who don't install perf ;-) */
561 help
= perf_tip(DOCDIR
);
563 help
= "Cannot load tips.txt file, please install perf!";
566 switch (use_browser
) {
568 ret
= perf_evlist__tui_browse_hists(evlist
, help
, NULL
,
570 &session
->header
.env
,
571 true, &rep
->annotation_opts
);
573 * Usually "ret" is the last pressed key, and we only
574 * care if the key notifies us to switch data file.
576 if (ret
!= K_SWITCH_INPUT_DATA
)
580 ret
= report__gtk_browse_hists(rep
, help
);
583 ret
= perf_evlist__tty_browse_hists(evlist
, rep
, help
);
590 static int report__collapse_hists(struct report
*rep
)
592 struct ui_progress prog
;
593 struct perf_evsel
*pos
;
596 ui_progress__init(&prog
, rep
->nr_entries
, "Merging related events...");
598 evlist__for_each_entry(rep
->session
->evlist
, pos
) {
599 struct hists
*hists
= evsel__hists(pos
);
602 hists
->symbol_filter_str
= rep
->symbol_filter_str
;
604 hists
->socket_filter
= rep
->socket_filter
;
606 ret
= hists__collapse_resort(hists
, &prog
);
610 /* Non-group events are considered as leader */
611 if (symbol_conf
.event_group
&&
612 !perf_evsel__is_group_leader(pos
)) {
613 struct hists
*leader_hists
= evsel__hists(pos
->leader
);
615 hists__match(leader_hists
, hists
);
616 hists__link(leader_hists
, hists
);
620 ui_progress__finish();
624 static void report__output_resort(struct report
*rep
)
626 struct ui_progress prog
;
627 struct perf_evsel
*pos
;
629 ui_progress__init(&prog
, rep
->nr_entries
, "Sorting events for output...");
631 evlist__for_each_entry(rep
->session
->evlist
, pos
)
632 perf_evsel__output_resort(pos
, &prog
);
634 ui_progress__finish();
637 static void stats_setup(struct report
*rep
)
639 memset(&rep
->tool
, 0, sizeof(rep
->tool
));
640 rep
->tool
.no_warn
= true;
643 static int stats_print(struct report
*rep
)
645 struct perf_session
*session
= rep
->session
;
647 perf_session__fprintf_nr_events(session
, stdout
);
651 static void tasks_setup(struct report
*rep
)
653 memset(&rep
->tool
, 0, sizeof(rep
->tool
));
654 rep
->tool
.ordered_events
= true;
655 if (rep
->mmaps_mode
) {
656 rep
->tool
.mmap
= perf_event__process_mmap
;
657 rep
->tool
.mmap2
= perf_event__process_mmap2
;
659 rep
->tool
.comm
= perf_event__process_comm
;
660 rep
->tool
.exit
= perf_event__process_exit
;
661 rep
->tool
.fork
= perf_event__process_fork
;
662 rep
->tool
.no_warn
= true;
666 struct thread
*thread
;
667 struct list_head list
;
668 struct list_head children
;
671 static struct task
*tasks_list(struct task
*task
, struct machine
*machine
)
673 struct thread
*parent_thread
, *thread
= task
->thread
;
674 struct task
*parent_task
;
676 /* Already listed. */
677 if (!list_empty(&task
->list
))
680 /* Last one in the chain. */
681 if (thread
->ppid
== -1)
684 parent_thread
= machine__find_thread(machine
, -1, thread
->ppid
);
686 return ERR_PTR(-ENOENT
);
688 parent_task
= thread__priv(parent_thread
);
689 list_add_tail(&task
->list
, &parent_task
->children
);
690 return tasks_list(parent_task
, machine
);
693 static size_t maps__fprintf_task(struct maps
*maps
, int indent
, FILE *fp
)
698 for (nd
= rb_first(&maps
->entries
); nd
; nd
= rb_next(nd
)) {
699 struct map
*map
= rb_entry(nd
, struct map
, rb_node
);
701 printed
+= fprintf(fp
, "%*s %" PRIx64
"-%" PRIx64
" %c%c%c%c %08" PRIx64
" %" PRIu64
" %s\n",
702 indent
, "", map
->start
, map
->end
,
703 map
->prot
& PROT_READ
? 'r' : '-',
704 map
->prot
& PROT_WRITE
? 'w' : '-',
705 map
->prot
& PROT_EXEC
? 'x' : '-',
706 map
->flags
& MAP_SHARED
? 's' : 'p',
708 map
->ino
, map
->dso
->name
);
714 static int map_groups__fprintf_task(struct map_groups
*mg
, int indent
, FILE *fp
)
716 return maps__fprintf_task(&mg
->maps
, indent
, fp
);
719 static void task__print_level(struct task
*task
, FILE *fp
, int level
)
721 struct thread
*thread
= task
->thread
;
723 int comm_indent
= fprintf(fp
, " %8d %8d %8d |%*s",
724 thread
->pid_
, thread
->tid
, thread
->ppid
,
727 fprintf(fp
, "%s\n", thread__comm_str(thread
));
729 map_groups__fprintf_task(thread
->mg
, comm_indent
, fp
);
731 if (!list_empty(&task
->children
)) {
732 list_for_each_entry(child
, &task
->children
, list
)
733 task__print_level(child
, fp
, level
+ 1);
737 static int tasks_print(struct report
*rep
, FILE *fp
)
739 struct perf_session
*session
= rep
->session
;
740 struct machine
*machine
= &session
->machines
.host
;
741 struct task
*tasks
, *task
;
742 unsigned int nr
= 0, itask
= 0, i
;
747 * No locking needed while accessing machine->threads,
748 * because --tasks is single threaded command.
751 /* Count all the threads. */
752 for (i
= 0; i
< THREADS__TABLE_SIZE
; i
++)
753 nr
+= machine
->threads
[i
].nr
;
755 tasks
= malloc(sizeof(*tasks
) * nr
);
759 for (i
= 0; i
< THREADS__TABLE_SIZE
; i
++) {
760 struct threads
*threads
= &machine
->threads
[i
];
762 for (nd
= rb_first(&threads
->entries
); nd
; nd
= rb_next(nd
)) {
763 task
= tasks
+ itask
++;
765 task
->thread
= rb_entry(nd
, struct thread
, rb_node
);
766 INIT_LIST_HEAD(&task
->children
);
767 INIT_LIST_HEAD(&task
->list
);
768 thread__set_priv(task
->thread
, task
);
773 * Iterate every task down to the unprocessed parent
774 * and link all in task children list. Task with no
775 * parent is added into 'list'.
777 for (itask
= 0; itask
< nr
; itask
++) {
778 task
= tasks
+ itask
;
780 if (!list_empty(&task
->list
))
783 task
= tasks_list(task
, machine
);
785 pr_err("Error: failed to process tasks\n");
787 return PTR_ERR(task
);
791 list_add_tail(&task
->list
, &list
);
794 fprintf(fp
, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm");
796 list_for_each_entry(task
, &list
, list
)
797 task__print_level(task
, fp
, 0);
803 static int __cmd_report(struct report
*rep
)
806 struct perf_session
*session
= rep
->session
;
807 struct perf_evsel
*pos
;
808 struct perf_data
*data
= session
->data
;
810 signal(SIGINT
, sig_handler
);
813 ret
= perf_session__cpu_bitmap(session
, rep
->cpu_list
,
816 ui__error("failed to set cpu bitmap\n");
819 session
->itrace_synth_opts
->cpu_bitmap
= rep
->cpu_bitmap
;
822 if (rep
->show_threads
) {
823 ret
= perf_read_values_init(&rep
->show_threads_values
);
828 ret
= report__setup_sample_type(rep
);
830 /* report__setup_sample_type() already showed error message */
840 ret
= perf_session__process_events(session
);
842 ui__error("failed to process sample\n");
847 return stats_print(rep
);
850 return tasks_print(rep
, stdout
);
852 report__warn_kptr_restrict(rep
);
854 evlist__for_each_entry(session
->evlist
, pos
)
855 rep
->nr_entries
+= evsel__hists(pos
)->nr_entries
;
857 if (use_browser
== 0) {
859 perf_session__fprintf(session
, stdout
);
862 perf_session__fprintf_dsos(session
, stdout
);
865 perf_session__fprintf_nr_events(session
, stdout
);
866 perf_evlist__fprintf_nr_events(session
->evlist
, stdout
);
871 ret
= report__collapse_hists(rep
);
873 ui__error("failed to process hist entry\n");
881 * recalculate number of entries after collapsing since it
882 * might be changed during the collapse phase.
885 evlist__for_each_entry(session
->evlist
, pos
)
886 rep
->nr_entries
+= evsel__hists(pos
)->nr_entries
;
888 if (rep
->nr_entries
== 0) {
889 ui__error("The %s file has no samples!\n", data
->file
.path
);
893 report__output_resort(rep
);
895 return report__browse_hists(rep
);
899 report_parse_callchain_opt(const struct option
*opt
, const char *arg
, int unset
)
901 struct callchain_param
*callchain
= opt
->value
;
903 callchain
->enabled
= !unset
;
908 symbol_conf
.use_callchain
= false;
909 callchain
->mode
= CHAIN_NONE
;
913 return parse_callchain_report_opt(arg
);
917 report_parse_ignore_callees_opt(const struct option
*opt __maybe_unused
,
918 const char *arg
, int unset __maybe_unused
)
921 int err
= regcomp(&ignore_callees_regex
, arg
, REG_EXTENDED
);
924 regerror(err
, &ignore_callees_regex
, buf
, sizeof(buf
));
925 pr_err("Invalid --ignore-callees regex: %s\n%s", arg
, buf
);
928 have_ignore_callees
= 1;
935 parse_branch_mode(const struct option
*opt
,
936 const char *str __maybe_unused
, int unset
)
938 int *branch_mode
= opt
->value
;
940 *branch_mode
= !unset
;
945 parse_percent_limit(const struct option
*opt
, const char *str
,
946 int unset __maybe_unused
)
948 struct report
*rep
= opt
->value
;
949 double pcnt
= strtof(str
, NULL
);
951 rep
->min_percent
= pcnt
;
952 callchain_param
.min_percent
= pcnt
;
956 int cmd_report(int argc
, const char **argv
)
958 struct perf_session
*session
;
959 struct itrace_synth_opts itrace_synth_opts
= { .set
= 0, };
961 bool has_br_stack
= false;
962 int branch_mode
= -1;
964 bool branch_call_mode
= false;
965 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
966 const char report_callchain_help
[] = "Display call graph (stack chain/backtrace):\n\n"
967 CALLCHAIN_REPORT_HELP
968 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT
;
969 char callchain_default_opt
[] = CALLCHAIN_DEFAULT_OPT
;
970 const char * const report_usage
[] = {
971 "perf report [<options>]",
974 struct report report
= {
976 .sample
= process_sample_event
,
977 .mmap
= perf_event__process_mmap
,
978 .mmap2
= perf_event__process_mmap2
,
979 .comm
= perf_event__process_comm
,
980 .namespaces
= perf_event__process_namespaces
,
981 .exit
= perf_event__process_exit
,
982 .fork
= perf_event__process_fork
,
983 .lost
= perf_event__process_lost
,
984 .read
= process_read_event
,
985 .attr
= perf_event__process_attr
,
986 .tracing_data
= perf_event__process_tracing_data
,
987 .build_id
= perf_event__process_build_id
,
988 .id_index
= perf_event__process_id_index
,
989 .auxtrace_info
= perf_event__process_auxtrace_info
,
990 .auxtrace
= perf_event__process_auxtrace
,
991 .event_update
= perf_event__process_event_update
,
992 .feature
= process_feature_event
,
993 .ordered_events
= true,
994 .ordering_requires_timestamps
= true,
996 .max_stack
= PERF_MAX_STACK_DEPTH
,
997 .pretty_printing_style
= "normal",
999 .annotation_opts
= annotation__default_options
,
1001 const struct option options
[] = {
1002 OPT_STRING('i', "input", &input_name
, "file",
1004 OPT_INCR('v', "verbose", &verbose
,
1005 "be more verbose (show symbol address, etc)"),
1006 OPT_BOOLEAN('q', "quiet", &quiet
, "Do not show any message"),
1007 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace
,
1008 "dump raw trace in ASCII"),
1009 OPT_BOOLEAN(0, "stats", &report
.stats_mode
, "Display event stats"),
1010 OPT_BOOLEAN(0, "tasks", &report
.tasks_mode
, "Display recorded tasks"),
1011 OPT_BOOLEAN(0, "mmaps", &report
.mmaps_mode
, "Display recorded tasks memory maps"),
1012 OPT_STRING('k', "vmlinux", &symbol_conf
.vmlinux_name
,
1013 "file", "vmlinux pathname"),
1014 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf
.ignore_vmlinux
,
1015 "don't load vmlinux even if found"),
1016 OPT_STRING(0, "kallsyms", &symbol_conf
.kallsyms_name
,
1017 "file", "kallsyms pathname"),
1018 OPT_BOOLEAN('f', "force", &symbol_conf
.force
, "don't complain, do it"),
1019 OPT_BOOLEAN('m', "modules", &symbol_conf
.use_modules
,
1020 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1021 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf
.show_nr_samples
,
1022 "Show a column with the number of samples"),
1023 OPT_BOOLEAN('T', "threads", &report
.show_threads
,
1024 "Show per-thread event counters"),
1025 OPT_STRING(0, "pretty", &report
.pretty_printing_style
, "key",
1026 "pretty printing style key: normal raw"),
1027 OPT_BOOLEAN(0, "tui", &report
.use_tui
, "Use the TUI interface"),
1028 OPT_BOOLEAN(0, "gtk", &report
.use_gtk
, "Use the GTK2 interface"),
1029 OPT_BOOLEAN(0, "stdio", &report
.use_stdio
,
1030 "Use the stdio interface"),
1031 OPT_BOOLEAN(0, "header", &report
.header
, "Show data header."),
1032 OPT_BOOLEAN(0, "header-only", &report
.header_only
,
1033 "Show only data header."),
1034 OPT_STRING('s', "sort", &sort_order
, "key[,key2...]",
1035 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
1036 " Please refer the man page for the complete list."),
1037 OPT_STRING('F', "fields", &field_order
, "key[,keys...]",
1038 "output field(s): overhead, period, sample plus all of sort keys"),
1039 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf
.show_cpu_utilization
,
1040 "Show sample percentage for different cpu modes"),
1041 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf
.show_cpu_utilization
,
1042 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN
),
1043 OPT_STRING('p', "parent", &parent_pattern
, "regex",
1044 "regex filter to identify parent, see: '--sort parent'"),
1045 OPT_BOOLEAN('x', "exclude-other", &symbol_conf
.exclude_other
,
1046 "Only display entries with parent-match"),
1047 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param
,
1048 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
1049 report_callchain_help
, &report_parse_callchain_opt
,
1050 callchain_default_opt
),
1051 OPT_BOOLEAN(0, "children", &symbol_conf
.cumulate_callchain
,
1052 "Accumulate callchains of children and show total overhead as well"),
1053 OPT_INTEGER(0, "max-stack", &report
.max_stack
,
1054 "Set the maximum stack depth when parsing the callchain, "
1055 "anything beyond the specified depth will be ignored. "
1056 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH
)),
1057 OPT_BOOLEAN('G', "inverted", &report
.inverted_callchain
,
1058 "alias for inverted call graph"),
1059 OPT_CALLBACK(0, "ignore-callees", NULL
, "regex",
1060 "ignore callees of these functions in call graphs",
1061 report_parse_ignore_callees_opt
),
1062 OPT_STRING('d', "dsos", &symbol_conf
.dso_list_str
, "dso[,dso...]",
1063 "only consider symbols in these dsos"),
1064 OPT_STRING('c', "comms", &symbol_conf
.comm_list_str
, "comm[,comm...]",
1065 "only consider symbols in these comms"),
1066 OPT_STRING(0, "pid", &symbol_conf
.pid_list_str
, "pid[,pid...]",
1067 "only consider symbols in these pids"),
1068 OPT_STRING(0, "tid", &symbol_conf
.tid_list_str
, "tid[,tid...]",
1069 "only consider symbols in these tids"),
1070 OPT_STRING('S', "symbols", &symbol_conf
.sym_list_str
, "symbol[,symbol...]",
1071 "only consider these symbols"),
1072 OPT_STRING(0, "symbol-filter", &report
.symbol_filter_str
, "filter",
1073 "only show symbols that (partially) match with this filter"),
1074 OPT_STRING('w', "column-widths", &symbol_conf
.col_width_list_str
,
1076 "don't try to adjust column width, use these fixed values"),
1077 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf
.field_sep
, "separator",
1078 "separator for columns, no spaces will be added between "
1079 "columns '.' is reserved."),
1080 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf
.hide_unresolved
,
1081 "Only display entries resolved to a symbol"),
1082 OPT_CALLBACK(0, "symfs", NULL
, "directory",
1083 "Look for files with symbols relative to this directory",
1084 symbol__config_symfs
),
1085 OPT_STRING('C', "cpu", &report
.cpu_list
, "cpu",
1086 "list of cpus to profile"),
1087 OPT_BOOLEAN('I', "show-info", &report
.show_full_info
,
1088 "Display extended information about perf.data file"),
1089 OPT_BOOLEAN(0, "source", &report
.annotation_opts
.annotate_src
,
1090 "Interleave source code with assembly code (default)"),
1091 OPT_BOOLEAN(0, "asm-raw", &report
.annotation_opts
.show_asm_raw
,
1092 "Display raw encoding of assembly instructions (default)"),
1093 OPT_STRING('M', "disassembler-style", &report
.annotation_opts
.disassembler_style
, "disassembler style",
1094 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1095 OPT_BOOLEAN(0, "show-total-period", &symbol_conf
.show_total_period
,
1096 "Show a column with the sum of periods"),
1097 OPT_BOOLEAN_SET(0, "group", &symbol_conf
.event_group
, &report
.group_set
,
1098 "Show event group information together"),
1099 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode
, "",
1100 "use branch records for per branch histogram filling",
1102 OPT_BOOLEAN(0, "branch-history", &branch_call_mode
,
1103 "add last branch records to call history"),
1104 OPT_STRING(0, "objdump", &report
.annotation_opts
.objdump_path
, "path",
1105 "objdump binary to use for disassembly and annotations"),
1106 OPT_BOOLEAN(0, "demangle", &symbol_conf
.demangle
,
1107 "Disable symbol demangling"),
1108 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf
.demangle_kernel
,
1109 "Enable kernel symbol demangling"),
1110 OPT_BOOLEAN(0, "mem-mode", &report
.mem_mode
, "mem access profile"),
1111 OPT_CALLBACK(0, "percent-limit", &report
, "percent",
1112 "Don't show entries under that percent", parse_percent_limit
),
1113 OPT_CALLBACK(0, "percentage", NULL
, "relative|absolute",
1114 "how to display percentage of filtered entries", parse_filter_percentage
),
1115 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts
, NULL
, "opts",
1116 "Instruction Tracing options",
1117 itrace_parse_synth_opts
),
1118 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename
,
1119 "Show full source file name path for source lines"),
1120 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf
.show_ref_callgraph
,
1121 "Show callgraph from reference event"),
1122 OPT_INTEGER(0, "socket-filter", &report
.socket_filter
,
1123 "only show processor socket that match with this filter"),
1124 OPT_BOOLEAN(0, "raw-trace", &symbol_conf
.raw_trace
,
1125 "Show raw trace event output (do not use print fmt or plugins)"),
1126 OPT_BOOLEAN(0, "hierarchy", &symbol_conf
.report_hierarchy
,
1127 "Show entries in a hierarchy"),
1128 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL
, "mode",
1129 "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
1130 stdio__config_color
, "always"),
1131 OPT_STRING(0, "time", &report
.time_str
, "str",
1132 "Time span of interest (start,stop)"),
1133 OPT_BOOLEAN(0, "inline", &symbol_conf
.inline_name
,
1134 "Show inline function"),
1135 OPT_CALLBACK(0, "percent-type", &report
.annotation_opts
, "local-period",
1136 "Set percent type local/global-period/hits",
1137 annotate_parse_percent_type
),
1140 struct perf_data data
= {
1141 .mode
= PERF_DATA_MODE_READ
,
1143 int ret
= hists__init();
1148 ret
= perf_config(report__config
, &report
);
1152 argc
= parse_options(argc
, argv
, options
, report_usage
, 0);
1155 * Special case: if there's an argument left then assume that
1156 * it's a symbol filter:
1159 usage_with_options(report_usage
, options
);
1161 report
.symbol_filter_str
= argv
[0];
1164 if (report
.mmaps_mode
)
1165 report
.tasks_mode
= true;
1168 perf_quiet_option();
1170 if (symbol_conf
.vmlinux_name
&&
1171 access(symbol_conf
.vmlinux_name
, R_OK
)) {
1172 pr_err("Invalid file: %s\n", symbol_conf
.vmlinux_name
);
1175 if (symbol_conf
.kallsyms_name
&&
1176 access(symbol_conf
.kallsyms_name
, R_OK
)) {
1177 pr_err("Invalid file: %s\n", symbol_conf
.kallsyms_name
);
1181 if (report
.inverted_callchain
)
1182 callchain_param
.order
= ORDER_CALLER
;
1183 if (symbol_conf
.cumulate_callchain
&& !callchain_param
.order_set
)
1184 callchain_param
.order
= ORDER_CALLER
;
1186 if (itrace_synth_opts
.callchain
&&
1187 (int)itrace_synth_opts
.callchain_sz
> report
.max_stack
)
1188 report
.max_stack
= itrace_synth_opts
.callchain_sz
;
1190 if (!input_name
|| !strlen(input_name
)) {
1191 if (!fstat(STDIN_FILENO
, &st
) && S_ISFIFO(st
.st_mode
))
1194 input_name
= "perf.data";
1197 data
.file
.path
= input_name
;
1198 data
.force
= symbol_conf
.force
;
1201 session
= perf_session__new(&data
, false, &report
.tool
);
1202 if (session
== NULL
)
1205 if (report
.queue_size
) {
1206 ordered_events__set_alloc_size(&session
->ordered_events
,
1210 session
->itrace_synth_opts
= &itrace_synth_opts
;
1212 report
.session
= session
;
1214 has_br_stack
= perf_header__has_feat(&session
->header
,
1215 HEADER_BRANCH_STACK
);
1217 setup_forced_leader(&report
, session
->evlist
);
1219 if (itrace_synth_opts
.last_branch
)
1220 has_br_stack
= true;
1222 if (has_br_stack
&& branch_call_mode
)
1223 symbol_conf
.show_branchflag_count
= true;
1225 memset(&report
.brtype_stat
, 0, sizeof(struct branch_type_stat
));
1228 * Branch mode is a tristate:
1229 * -1 means default, so decide based on the file having branch data.
1230 * 0/1 means the user chose a mode.
1232 if (((branch_mode
== -1 && has_br_stack
) || branch_mode
== 1) &&
1233 !branch_call_mode
) {
1234 sort__mode
= SORT_MODE__BRANCH
;
1235 symbol_conf
.cumulate_callchain
= false;
1237 if (branch_call_mode
) {
1238 callchain_param
.key
= CCKEY_ADDRESS
;
1239 callchain_param
.branch_callstack
= 1;
1240 symbol_conf
.use_callchain
= true;
1241 callchain_register_param(&callchain_param
);
1242 if (sort_order
== NULL
)
1243 sort_order
= "srcline,symbol,dso";
1246 if (report
.mem_mode
) {
1247 if (sort__mode
== SORT_MODE__BRANCH
) {
1248 pr_err("branch and mem mode incompatible\n");
1251 sort__mode
= SORT_MODE__MEMORY
;
1252 symbol_conf
.cumulate_callchain
= false;
1255 if (symbol_conf
.report_hierarchy
) {
1256 /* disable incompatible options */
1257 symbol_conf
.cumulate_callchain
= false;
1260 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1261 parse_options_usage(report_usage
, options
, "F", 1);
1262 parse_options_usage(NULL
, options
, "hierarchy", 0);
1266 perf_hpp_list
.need_collapse
= true;
1269 if (report
.use_stdio
)
1271 else if (report
.use_tui
)
1273 else if (report
.use_gtk
)
1276 /* Force tty output for header output and per-thread stat. */
1277 if (report
.header
|| report
.header_only
|| report
.show_threads
)
1279 if (report
.header
|| report
.header_only
)
1280 report
.tool
.show_feat_hdr
= SHOW_FEAT_HEADER
;
1281 if (report
.show_full_info
)
1282 report
.tool
.show_feat_hdr
= SHOW_FEAT_HEADER_FULL_INFO
;
1283 if (report
.stats_mode
|| report
.tasks_mode
)
1285 if (report
.stats_mode
&& report
.tasks_mode
) {
1286 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n");
1290 if (strcmp(input_name
, "-") != 0)
1291 setup_browser(true);
1295 if ((last_key
!= K_SWITCH_INPUT_DATA
) &&
1296 (setup_sorting(session
->evlist
) < 0)) {
1298 parse_options_usage(report_usage
, options
, "s", 1);
1300 parse_options_usage(sort_order
? NULL
: report_usage
,
1305 if ((report
.header
|| report
.header_only
) && !quiet
) {
1306 perf_session__fprintf_info(session
, stdout
,
1307 report
.show_full_info
);
1308 if (report
.header_only
) {
1312 } else if (use_browser
== 0 && !quiet
&&
1313 !report
.stats_mode
&& !report
.tasks_mode
) {
1314 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
1319 * Only in the TUI browser we are doing integrated annotation,
1320 * so don't allocate extra space that won't be used in the stdio
1323 if (ui__has_annotation()) {
1324 ret
= symbol__annotation_init();
1328 * For searching by name on the "Browse map details".
1329 * providing it only in verbose mode not to bloat too
1330 * much struct symbol.
1334 * XXX: Need to provide a less kludgy way to ask for
1335 * more space per symbol, the u32 is for the index on
1337 * See symbol__browser_index.
1339 symbol_conf
.priv_size
+= sizeof(u32
);
1340 symbol_conf
.sort_by_name
= true;
1342 annotation_config__init();
1345 if (symbol__init(&session
->header
.env
) < 0)
1348 report
.ptime_range
= perf_time__range_alloc(report
.time_str
,
1349 &report
.range_size
);
1350 if (!report
.ptime_range
) {
1355 if (perf_time__parse_str(report
.ptime_range
, report
.time_str
) != 0) {
1356 if (session
->evlist
->first_sample_time
== 0 &&
1357 session
->evlist
->last_sample_time
== 0) {
1358 pr_err("HINT: no first/last sample time found in perf data.\n"
1359 "Please use latest perf binary to execute 'perf record'\n"
1360 "(if '--buildid-all' is enabled, please set '--timestamp-boundary').\n");
1365 report
.range_num
= perf_time__percent_parse_str(
1366 report
.ptime_range
, report
.range_size
,
1368 session
->evlist
->first_sample_time
,
1369 session
->evlist
->last_sample_time
);
1371 if (report
.range_num
< 0) {
1372 pr_err("Invalid time string\n");
1377 report
.range_num
= 1;
1380 if (session
->tevent
.pevent
&&
1381 tep_set_function_resolver(session
->tevent
.pevent
,
1382 machine__resolve_kernel_addr
,
1383 &session
->machines
.host
) < 0) {
1384 pr_err("%s: failed to set libtraceevent function resolver\n",
1389 sort__setup_elide(stdout
);
1391 ret
= __cmd_report(&report
);
1392 if (ret
== K_SWITCH_INPUT_DATA
) {
1393 perf_session__delete(session
);
1394 last_key
= K_SWITCH_INPUT_DATA
;
1400 zfree(&report
.ptime_range
);
1402 perf_session__delete(session
);