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/config.h"
13 #include "util/annotate.h"
14 #include "util/color.h"
16 #include <linux/list.h>
17 #include <linux/rbtree.h>
18 #include <linux/err.h>
19 #include <linux/zalloc.h>
21 #include "util/symbol.h"
22 #include "util/map_symbol.h"
23 #include "util/mem-events.h"
24 #include "util/branch.h"
25 #include "util/callchain.h"
26 #include "util/values.h"
29 #include "util/debug.h"
30 #include "util/evlist.h"
31 #include "util/evsel.h"
32 #include "util/evswitch.h"
33 #include "util/header.h"
34 #include "util/session.h"
35 #include "util/srcline.h"
36 #include "util/tool.h"
38 #include <subcmd/parse-options.h>
39 #include <subcmd/exec-cmd.h>
40 #include "util/parse-events.h"
42 #include "util/thread.h"
43 #include "util/sort.h"
44 #include "util/hist.h"
45 #include "util/data.h"
46 #include "arch/common.h"
47 #include "util/time-utils.h"
48 #include "util/auxtrace.h"
49 #include "util/units.h"
50 #include "util/branch.h"
51 #include "util/util.h" // perf_tip()
53 #include "ui/progress.h"
59 #include <linux/ctype.h>
61 #include <linux/bitmap.h>
62 #include <linux/string.h>
63 #include <linux/stringify.h>
64 #include <linux/time64.h>
65 #include <sys/types.h>
68 #include <linux/mman.h>
71 struct perf_tool tool
;
72 struct perf_session
*session
;
73 struct evswitch evswitch
;
74 bool use_tui
, use_gtk
, use_stdio
;
77 bool inverted_callchain
;
84 bool nonany_branch_mode
;
87 struct perf_read_values show_threads_values
;
88 struct annotation_options annotation_opts
;
89 const char *pretty_printing_style
;
91 const char *symbol_filter_str
;
93 struct perf_time_interval
*ptime_range
;
100 DECLARE_BITMAP(cpu_bitmap
, MAX_NR_CPUS
);
101 struct branch_type_stat brtype_stat
;
105 static int report__config(const char *var
, const char *value
, void *cb
)
107 struct report
*rep
= cb
;
109 if (!strcmp(var
, "report.group")) {
110 symbol_conf
.event_group
= perf_config_bool(var
, value
);
113 if (!strcmp(var
, "report.percent-limit")) {
114 double pcnt
= strtof(value
, NULL
);
116 rep
->min_percent
= pcnt
;
117 callchain_param
.min_percent
= pcnt
;
120 if (!strcmp(var
, "report.children")) {
121 symbol_conf
.cumulate_callchain
= perf_config_bool(var
, value
);
124 if (!strcmp(var
, "report.queue-size"))
125 return perf_config_u64(&rep
->queue_size
, var
, value
);
127 if (!strcmp(var
, "report.sort_order")) {
128 default_sort_order
= strdup(value
);
135 static int hist_iter__report_callback(struct hist_entry_iter
*iter
,
136 struct addr_location
*al
, bool single
,
140 struct report
*rep
= arg
;
141 struct hist_entry
*he
= iter
->he
;
142 struct evsel
*evsel
= iter
->evsel
;
143 struct perf_sample
*sample
= iter
->sample
;
145 struct branch_info
*bi
;
147 if (!ui__has_annotation() && !rep
->symbol_ipc
)
150 if (sort__mode
== SORT_MODE__BRANCH
) {
151 bi
= he
->branch_info
;
152 err
= addr_map_symbol__inc_samples(&bi
->from
, sample
, evsel
);
156 err
= addr_map_symbol__inc_samples(&bi
->to
, sample
, evsel
);
158 } else if (rep
->mem_mode
) {
160 err
= addr_map_symbol__inc_samples(&mi
->daddr
, sample
, evsel
);
164 err
= hist_entry__inc_addr_samples(he
, sample
, evsel
, al
->addr
);
166 } else if (symbol_conf
.cumulate_callchain
) {
168 err
= hist_entry__inc_addr_samples(he
, sample
, evsel
, al
->addr
);
170 err
= hist_entry__inc_addr_samples(he
, sample
, evsel
, al
->addr
);
177 static int hist_iter__branch_callback(struct hist_entry_iter
*iter
,
178 struct addr_location
*al __maybe_unused
,
179 bool single __maybe_unused
,
182 struct hist_entry
*he
= iter
->he
;
183 struct report
*rep
= arg
;
184 struct branch_info
*bi
;
185 struct perf_sample
*sample
= iter
->sample
;
186 struct evsel
*evsel
= iter
->evsel
;
189 if (!ui__has_annotation() && !rep
->symbol_ipc
)
192 bi
= he
->branch_info
;
193 err
= addr_map_symbol__inc_samples(&bi
->from
, sample
, evsel
);
197 err
= addr_map_symbol__inc_samples(&bi
->to
, sample
, evsel
);
199 branch_type_count(&rep
->brtype_stat
, &bi
->flags
,
200 bi
->from
.addr
, bi
->to
.addr
);
206 static void setup_forced_leader(struct report
*report
,
207 struct evlist
*evlist
)
209 if (report
->group_set
)
210 perf_evlist__force_leader(evlist
);
213 static int process_feature_event(struct perf_session
*session
,
214 union perf_event
*event
)
216 struct report
*rep
= container_of(session
->tool
, struct report
, tool
);
218 if (event
->feat
.feat_id
< HEADER_LAST_FEATURE
)
219 return perf_event__process_feature(session
, event
);
221 if (event
->feat
.feat_id
!= HEADER_LAST_FEATURE
) {
222 pr_err("failed: wrong feature ID: %" PRI_lu64
"\n",
223 event
->feat
.feat_id
);
228 * (feat_id = HEADER_LAST_FEATURE) is the end marker which
229 * means all features are received, now 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
,
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 (evswitch__discard(&rep
->evswitch
, evsel
))
260 if (machine__resolve(machine
, &al
, sample
) < 0) {
261 pr_debug("problem processing %d event, skipping it.\n",
266 if (symbol_conf
.hide_unresolved
&& al
.sym
== NULL
)
269 if (rep
->cpu_list
&& !test_bit(sample
->cpu
, rep
->cpu_bitmap
))
272 if (sort__mode
== SORT_MODE__BRANCH
) {
274 * A non-synthesized event might not have a branch stack if
275 * branch stacks have been synthesized (using itrace options).
277 if (!sample
->branch_stack
)
280 iter
.add_entry_cb
= hist_iter__branch_callback
;
281 iter
.ops
= &hist_iter_branch
;
282 } else if (rep
->mem_mode
) {
283 iter
.ops
= &hist_iter_mem
;
284 } else if (symbol_conf
.cumulate_callchain
) {
285 iter
.ops
= &hist_iter_cumulative
;
287 iter
.ops
= &hist_iter_normal
;
291 al
.map
->dso
->hit
= 1;
293 if (ui__has_annotation() || rep
->symbol_ipc
) {
294 hist__account_cycles(sample
->branch_stack
, &al
, sample
,
295 rep
->nonany_branch_mode
);
298 ret
= hist_entry_iter__add(&iter
, &al
, rep
->max_stack
, rep
);
300 pr_debug("problem adding hist entry, skipping event\n");
302 addr_location__put(&al
);
306 static int process_read_event(struct perf_tool
*tool
,
307 union perf_event
*event
,
308 struct perf_sample
*sample __maybe_unused
,
310 struct machine
*machine __maybe_unused
)
312 struct report
*rep
= container_of(tool
, struct report
, tool
);
314 if (rep
->show_threads
) {
315 const char *name
= perf_evsel__name(evsel
);
316 int err
= perf_read_values_add_value(&rep
->show_threads_values
,
317 event
->read
.pid
, event
->read
.tid
,
329 /* For pipe mode, sample_type is not currently set */
330 static int report__setup_sample_type(struct report
*rep
)
332 struct perf_session
*session
= rep
->session
;
333 u64 sample_type
= perf_evlist__combined_sample_type(session
->evlist
);
334 bool is_pipe
= perf_data__is_pipe(session
->data
);
336 if (session
->itrace_synth_opts
->callchain
||
338 perf_header__has_feat(&session
->header
, HEADER_AUXTRACE
) &&
339 !session
->itrace_synth_opts
->set
))
340 sample_type
|= PERF_SAMPLE_CALLCHAIN
;
342 if (session
->itrace_synth_opts
->last_branch
)
343 sample_type
|= PERF_SAMPLE_BRANCH_STACK
;
345 if (!is_pipe
&& !(sample_type
& PERF_SAMPLE_CALLCHAIN
)) {
346 if (perf_hpp_list
.parent
) {
347 ui__error("Selected --sort parent, but no "
348 "callchain data. Did you call "
349 "'perf record' without -g?\n");
352 if (symbol_conf
.use_callchain
&&
353 !symbol_conf
.show_branchflag_count
) {
354 ui__error("Selected -g or --branch-history.\n"
355 "But no callchain or branch data.\n"
356 "Did you call 'perf record' without -g or -b?\n");
359 } else if (!callchain_param
.enabled
&&
360 callchain_param
.mode
!= CHAIN_NONE
&&
361 !symbol_conf
.use_callchain
) {
362 symbol_conf
.use_callchain
= true;
363 if (callchain_register_param(&callchain_param
) < 0) {
364 ui__error("Can't register callchain params.\n");
369 if (symbol_conf
.cumulate_callchain
) {
370 /* Silently ignore if callchain is missing */
371 if (!(sample_type
& PERF_SAMPLE_CALLCHAIN
)) {
372 symbol_conf
.cumulate_callchain
= false;
373 perf_hpp__cancel_cumulate();
377 if (sort__mode
== SORT_MODE__BRANCH
) {
379 !(sample_type
& PERF_SAMPLE_BRANCH_STACK
)) {
380 ui__error("Selected -b but no branch data. "
381 "Did you call perf record without -b?\n");
386 if (symbol_conf
.use_callchain
|| symbol_conf
.cumulate_callchain
) {
387 if ((sample_type
& PERF_SAMPLE_REGS_USER
) &&
388 (sample_type
& PERF_SAMPLE_STACK_USER
)) {
389 callchain_param
.record_mode
= CALLCHAIN_DWARF
;
390 dwarf_callchain_users
= true;
391 } else if (sample_type
& PERF_SAMPLE_BRANCH_STACK
)
392 callchain_param
.record_mode
= CALLCHAIN_LBR
;
394 callchain_param
.record_mode
= CALLCHAIN_FP
;
397 /* ??? handle more cases than just ANY? */
398 if (!(perf_evlist__combined_branch_type(session
->evlist
) &
399 PERF_SAMPLE_BRANCH_ANY
))
400 rep
->nonany_branch_mode
= true;
405 static void sig_handler(int sig __maybe_unused
)
410 static size_t hists__fprintf_nr_sample_events(struct hists
*hists
, struct report
*rep
,
411 const char *evname
, FILE *fp
)
415 unsigned long nr_samples
= hists
->stats
.nr_events
[PERF_RECORD_SAMPLE
];
416 u64 nr_events
= hists
->stats
.total_period
;
417 struct evsel
*evsel
= hists_to_evsel(hists
);
419 size_t size
= sizeof(buf
);
420 int socked_id
= hists
->socket_filter
;
425 if (symbol_conf
.filter_relative
) {
426 nr_samples
= hists
->stats
.nr_non_filtered_samples
;
427 nr_events
= hists
->stats
.total_non_filtered_period
;
430 if (perf_evsel__is_group_event(evsel
)) {
433 perf_evsel__group_desc(evsel
, buf
, size
);
436 for_each_group_member(pos
, evsel
) {
437 const struct hists
*pos_hists
= evsel__hists(pos
);
439 if (symbol_conf
.filter_relative
) {
440 nr_samples
+= pos_hists
->stats
.nr_non_filtered_samples
;
441 nr_events
+= pos_hists
->stats
.total_non_filtered_period
;
443 nr_samples
+= pos_hists
->stats
.nr_events
[PERF_RECORD_SAMPLE
];
444 nr_events
+= pos_hists
->stats
.total_period
;
449 nr_samples
= convert_unit(nr_samples
, &unit
);
450 ret
= fprintf(fp
, "# Samples: %lu%c", nr_samples
, unit
);
451 if (evname
!= NULL
) {
452 ret
+= fprintf(fp
, " of event%s '%s'",
453 evsel
->core
.nr_members
> 1 ? "s" : "", evname
);
457 ret
+= fprintf(fp
, " (time slices: %s)", rep
->time_str
);
459 if (symbol_conf
.show_ref_callgraph
&&
460 strstr(evname
, "call-graph=no")) {
461 ret
+= fprintf(fp
, ", show reference callgraph");
465 ret
+= fprintf(fp
, "\n# Total weight : %" PRIu64
, nr_events
);
466 ret
+= fprintf(fp
, "\n# Sort order : %s", sort_order
? : default_mem_sort_order
);
468 ret
+= fprintf(fp
, "\n# Event count (approx.): %" PRIu64
, nr_events
);
471 ret
+= fprintf(fp
, "\n# Processor Socket: %d", socked_id
);
473 return ret
+ fprintf(fp
, "\n#\n");
476 static int perf_evlist__tty_browse_hists(struct evlist
*evlist
,
483 fprintf(stdout
, "#\n# Total Lost Samples: %" PRIu64
"\n#\n",
484 evlist
->stats
.total_lost_samples
);
487 evlist__for_each_entry(evlist
, pos
) {
488 struct hists
*hists
= evsel__hists(pos
);
489 const char *evname
= perf_evsel__name(pos
);
491 if (symbol_conf
.event_group
&&
492 !perf_evsel__is_group_leader(pos
))
495 hists__fprintf_nr_sample_events(hists
, rep
, evname
, stdout
);
496 hists__fprintf(hists
, !quiet
, 0, 0, rep
->min_percent
, stdout
,
497 !(symbol_conf
.use_callchain
||
498 symbol_conf
.show_branchflag_count
));
499 fprintf(stdout
, "\n\n");
503 fprintf(stdout
, "#\n# (%s)\n#\n", help
);
505 if (rep
->show_threads
) {
506 bool style
= !strcmp(rep
->pretty_printing_style
, "raw");
507 perf_read_values_display(stdout
, &rep
->show_threads_values
,
509 perf_read_values_destroy(&rep
->show_threads_values
);
512 if (sort__mode
== SORT_MODE__BRANCH
)
513 branch_type_stat_display(stdout
, &rep
->brtype_stat
);
518 static void report__warn_kptr_restrict(const struct report
*rep
)
520 struct map
*kernel_map
= machine__kernel_map(&rep
->session
->machines
.host
);
521 struct kmap
*kernel_kmap
= kernel_map
? map__kmap(kernel_map
) : NULL
;
523 if (perf_evlist__exclude_kernel(rep
->session
->evlist
))
526 if (kernel_map
== NULL
||
527 (kernel_map
->dso
->hit
&&
528 (kernel_kmap
->ref_reloc_sym
== NULL
||
529 kernel_kmap
->ref_reloc_sym
->addr
== 0))) {
531 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
532 "can't be resolved.";
534 if (kernel_map
&& map__has_symbols(kernel_map
)) {
535 desc
= "If some relocation was applied (e.g. "
536 "kexec) symbols may be misresolved.";
540 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
541 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
542 "Samples in kernel modules can't be resolved as well.\n\n",
547 static int report__gtk_browse_hists(struct report
*rep
, const char *help
)
549 int (*hist_browser
)(struct evlist
*evlist
, const char *help
,
550 struct hist_browser_timer
*timer
, float min_pcnt
);
552 hist_browser
= dlsym(perf_gtk_handle
, "perf_evlist__gtk_browse_hists");
554 if (hist_browser
== NULL
) {
555 ui__error("GTK browser not found!\n");
559 return hist_browser(rep
->session
->evlist
, help
, NULL
, rep
->min_percent
);
562 static int report__browse_hists(struct report
*rep
)
565 struct perf_session
*session
= rep
->session
;
566 struct evlist
*evlist
= session
->evlist
;
567 const char *help
= perf_tip(system_path(TIPDIR
));
570 /* fallback for people who don't install perf ;-) */
571 help
= perf_tip(DOCDIR
);
573 help
= "Cannot load tips.txt file, please install perf!";
576 switch (use_browser
) {
578 ret
= perf_evlist__tui_browse_hists(evlist
, help
, NULL
,
580 &session
->header
.env
,
581 true, &rep
->annotation_opts
);
583 * Usually "ret" is the last pressed key, and we only
584 * care if the key notifies us to switch data file.
586 if (ret
!= K_SWITCH_INPUT_DATA
)
590 ret
= report__gtk_browse_hists(rep
, help
);
593 ret
= perf_evlist__tty_browse_hists(evlist
, rep
, help
);
600 static int report__collapse_hists(struct report
*rep
)
602 struct ui_progress prog
;
606 ui_progress__init(&prog
, rep
->nr_entries
, "Merging related events...");
608 evlist__for_each_entry(rep
->session
->evlist
, pos
) {
609 struct hists
*hists
= evsel__hists(pos
);
612 hists
->symbol_filter_str
= rep
->symbol_filter_str
;
614 hists
->socket_filter
= rep
->socket_filter
;
616 ret
= hists__collapse_resort(hists
, &prog
);
620 /* Non-group events are considered as leader */
621 if (symbol_conf
.event_group
&&
622 !perf_evsel__is_group_leader(pos
)) {
623 struct hists
*leader_hists
= evsel__hists(pos
->leader
);
625 hists__match(leader_hists
, hists
);
626 hists__link(leader_hists
, hists
);
630 ui_progress__finish();
634 static int hists__resort_cb(struct hist_entry
*he
, void *arg
)
636 struct report
*rep
= arg
;
637 struct symbol
*sym
= he
->ms
.sym
;
639 if (rep
->symbol_ipc
&& sym
&& !sym
->annotate2
) {
640 struct evsel
*evsel
= hists_to_evsel(he
->hists
);
642 symbol__annotate2(sym
, he
->ms
.map
, evsel
,
643 &annotation__default_options
, NULL
);
649 static void report__output_resort(struct report
*rep
)
651 struct ui_progress prog
;
654 ui_progress__init(&prog
, rep
->nr_entries
, "Sorting events for output...");
656 evlist__for_each_entry(rep
->session
->evlist
, pos
) {
657 perf_evsel__output_resort_cb(pos
, &prog
,
658 hists__resort_cb
, rep
);
661 ui_progress__finish();
664 static void stats_setup(struct report
*rep
)
666 memset(&rep
->tool
, 0, sizeof(rep
->tool
));
667 rep
->tool
.no_warn
= true;
670 static int stats_print(struct report
*rep
)
672 struct perf_session
*session
= rep
->session
;
674 perf_session__fprintf_nr_events(session
, stdout
);
678 static void tasks_setup(struct report
*rep
)
680 memset(&rep
->tool
, 0, sizeof(rep
->tool
));
681 rep
->tool
.ordered_events
= true;
682 if (rep
->mmaps_mode
) {
683 rep
->tool
.mmap
= perf_event__process_mmap
;
684 rep
->tool
.mmap2
= perf_event__process_mmap2
;
686 rep
->tool
.comm
= perf_event__process_comm
;
687 rep
->tool
.exit
= perf_event__process_exit
;
688 rep
->tool
.fork
= perf_event__process_fork
;
689 rep
->tool
.no_warn
= true;
693 struct thread
*thread
;
694 struct list_head list
;
695 struct list_head children
;
698 static struct task
*tasks_list(struct task
*task
, struct machine
*machine
)
700 struct thread
*parent_thread
, *thread
= task
->thread
;
701 struct task
*parent_task
;
703 /* Already listed. */
704 if (!list_empty(&task
->list
))
707 /* Last one in the chain. */
708 if (thread
->ppid
== -1)
711 parent_thread
= machine__find_thread(machine
, -1, thread
->ppid
);
713 return ERR_PTR(-ENOENT
);
715 parent_task
= thread__priv(parent_thread
);
716 list_add_tail(&task
->list
, &parent_task
->children
);
717 return tasks_list(parent_task
, machine
);
720 static size_t maps__fprintf_task(struct maps
*maps
, int indent
, FILE *fp
)
725 for (nd
= rb_first(&maps
->entries
); nd
; nd
= rb_next(nd
)) {
726 struct map
*map
= rb_entry(nd
, struct map
, rb_node
);
728 printed
+= fprintf(fp
, "%*s %" PRIx64
"-%" PRIx64
" %c%c%c%c %08" PRIx64
" %" PRIu64
" %s\n",
729 indent
, "", map
->start
, map
->end
,
730 map
->prot
& PROT_READ
? 'r' : '-',
731 map
->prot
& PROT_WRITE
? 'w' : '-',
732 map
->prot
& PROT_EXEC
? 'x' : '-',
733 map
->flags
& MAP_SHARED
? 's' : 'p',
735 map
->ino
, map
->dso
->name
);
741 static int map_groups__fprintf_task(struct map_groups
*mg
, int indent
, FILE *fp
)
743 return maps__fprintf_task(&mg
->maps
, indent
, fp
);
746 static void task__print_level(struct task
*task
, FILE *fp
, int level
)
748 struct thread
*thread
= task
->thread
;
750 int comm_indent
= fprintf(fp
, " %8d %8d %8d |%*s",
751 thread
->pid_
, thread
->tid
, thread
->ppid
,
754 fprintf(fp
, "%s\n", thread__comm_str(thread
));
756 map_groups__fprintf_task(thread
->mg
, comm_indent
, fp
);
758 if (!list_empty(&task
->children
)) {
759 list_for_each_entry(child
, &task
->children
, list
)
760 task__print_level(child
, fp
, level
+ 1);
764 static int tasks_print(struct report
*rep
, FILE *fp
)
766 struct perf_session
*session
= rep
->session
;
767 struct machine
*machine
= &session
->machines
.host
;
768 struct task
*tasks
, *task
;
769 unsigned int nr
= 0, itask
= 0, i
;
774 * No locking needed while accessing machine->threads,
775 * because --tasks is single threaded command.
778 /* Count all the threads. */
779 for (i
= 0; i
< THREADS__TABLE_SIZE
; i
++)
780 nr
+= machine
->threads
[i
].nr
;
782 tasks
= malloc(sizeof(*tasks
) * nr
);
786 for (i
= 0; i
< THREADS__TABLE_SIZE
; i
++) {
787 struct threads
*threads
= &machine
->threads
[i
];
789 for (nd
= rb_first_cached(&threads
->entries
); nd
;
791 task
= tasks
+ itask
++;
793 task
->thread
= rb_entry(nd
, struct thread
, rb_node
);
794 INIT_LIST_HEAD(&task
->children
);
795 INIT_LIST_HEAD(&task
->list
);
796 thread__set_priv(task
->thread
, task
);
801 * Iterate every task down to the unprocessed parent
802 * and link all in task children list. Task with no
803 * parent is added into 'list'.
805 for (itask
= 0; itask
< nr
; itask
++) {
806 task
= tasks
+ itask
;
808 if (!list_empty(&task
->list
))
811 task
= tasks_list(task
, machine
);
813 pr_err("Error: failed to process tasks\n");
815 return PTR_ERR(task
);
819 list_add_tail(&task
->list
, &list
);
822 fprintf(fp
, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm");
824 list_for_each_entry(task
, &list
, list
)
825 task__print_level(task
, fp
, 0);
831 static int __cmd_report(struct report
*rep
)
834 struct perf_session
*session
= rep
->session
;
836 struct perf_data
*data
= session
->data
;
838 signal(SIGINT
, sig_handler
);
841 ret
= perf_session__cpu_bitmap(session
, rep
->cpu_list
,
844 ui__error("failed to set cpu bitmap\n");
847 session
->itrace_synth_opts
->cpu_bitmap
= rep
->cpu_bitmap
;
850 if (rep
->show_threads
) {
851 ret
= perf_read_values_init(&rep
->show_threads_values
);
856 ret
= report__setup_sample_type(rep
);
858 /* report__setup_sample_type() already showed error message */
868 ret
= perf_session__process_events(session
);
870 ui__error("failed to process sample\n");
875 return stats_print(rep
);
878 return tasks_print(rep
, stdout
);
880 report__warn_kptr_restrict(rep
);
882 evlist__for_each_entry(session
->evlist
, pos
)
883 rep
->nr_entries
+= evsel__hists(pos
)->nr_entries
;
885 if (use_browser
== 0) {
887 perf_session__fprintf(session
, stdout
);
890 perf_session__fprintf_dsos(session
, stdout
);
893 perf_session__fprintf_nr_events(session
, stdout
);
894 perf_evlist__fprintf_nr_events(session
->evlist
, stdout
);
899 ret
= report__collapse_hists(rep
);
901 ui__error("failed to process hist entry\n");
909 * recalculate number of entries after collapsing since it
910 * might be changed during the collapse phase.
913 evlist__for_each_entry(session
->evlist
, pos
)
914 rep
->nr_entries
+= evsel__hists(pos
)->nr_entries
;
916 if (rep
->nr_entries
== 0) {
917 ui__error("The %s data has no samples!\n", data
->path
);
921 report__output_resort(rep
);
923 return report__browse_hists(rep
);
927 report_parse_callchain_opt(const struct option
*opt
, const char *arg
, int unset
)
929 struct callchain_param
*callchain
= opt
->value
;
931 callchain
->enabled
= !unset
;
936 symbol_conf
.use_callchain
= false;
937 callchain
->mode
= CHAIN_NONE
;
941 return parse_callchain_report_opt(arg
);
945 parse_time_quantum(const struct option
*opt
, const char *arg
,
946 int unset __maybe_unused
)
948 unsigned long *time_q
= opt
->value
;
951 *time_q
= strtoul(arg
, &end
, 0);
955 pr_err("time quantum cannot be 0");
958 end
= skip_spaces(end
);
961 if (!strcmp(end
, "s")) {
962 *time_q
*= NSEC_PER_SEC
;
965 if (!strcmp(end
, "ms")) {
966 *time_q
*= NSEC_PER_MSEC
;
969 if (!strcmp(end
, "us")) {
970 *time_q
*= NSEC_PER_USEC
;
973 if (!strcmp(end
, "ns"))
976 pr_err("Cannot parse time quantum `%s'\n", arg
);
981 report_parse_ignore_callees_opt(const struct option
*opt __maybe_unused
,
982 const char *arg
, int unset __maybe_unused
)
985 int err
= regcomp(&ignore_callees_regex
, arg
, REG_EXTENDED
);
988 regerror(err
, &ignore_callees_regex
, buf
, sizeof(buf
));
989 pr_err("Invalid --ignore-callees regex: %s\n%s", arg
, buf
);
992 have_ignore_callees
= 1;
999 parse_branch_mode(const struct option
*opt
,
1000 const char *str __maybe_unused
, int unset
)
1002 int *branch_mode
= opt
->value
;
1004 *branch_mode
= !unset
;
1009 parse_percent_limit(const struct option
*opt
, const char *str
,
1010 int unset __maybe_unused
)
1012 struct report
*rep
= opt
->value
;
1013 double pcnt
= strtof(str
, NULL
);
1015 rep
->min_percent
= pcnt
;
1016 callchain_param
.min_percent
= pcnt
;
1020 int cmd_report(int argc
, const char **argv
)
1022 struct perf_session
*session
;
1023 struct itrace_synth_opts itrace_synth_opts
= { .set
= 0, };
1025 bool has_br_stack
= false;
1026 int branch_mode
= -1;
1027 bool branch_call_mode
= false;
1028 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
1029 static const char report_callchain_help
[] = "Display call graph (stack chain/backtrace):\n\n"
1030 CALLCHAIN_REPORT_HELP
1031 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT
;
1032 char callchain_default_opt
[] = CALLCHAIN_DEFAULT_OPT
;
1033 const char * const report_usage
[] = {
1034 "perf report [<options>]",
1037 struct report report
= {
1039 .sample
= process_sample_event
,
1040 .mmap
= perf_event__process_mmap
,
1041 .mmap2
= perf_event__process_mmap2
,
1042 .comm
= perf_event__process_comm
,
1043 .namespaces
= perf_event__process_namespaces
,
1044 .exit
= perf_event__process_exit
,
1045 .fork
= perf_event__process_fork
,
1046 .lost
= perf_event__process_lost
,
1047 .read
= process_read_event
,
1048 .attr
= perf_event__process_attr
,
1049 .tracing_data
= perf_event__process_tracing_data
,
1050 .build_id
= perf_event__process_build_id
,
1051 .id_index
= perf_event__process_id_index
,
1052 .auxtrace_info
= perf_event__process_auxtrace_info
,
1053 .auxtrace
= perf_event__process_auxtrace
,
1054 .event_update
= perf_event__process_event_update
,
1055 .feature
= process_feature_event
,
1056 .ordered_events
= true,
1057 .ordering_requires_timestamps
= true,
1059 .max_stack
= PERF_MAX_STACK_DEPTH
,
1060 .pretty_printing_style
= "normal",
1061 .socket_filter
= -1,
1062 .annotation_opts
= annotation__default_options
,
1064 const struct option options
[] = {
1065 OPT_STRING('i', "input", &input_name
, "file",
1067 OPT_INCR('v', "verbose", &verbose
,
1068 "be more verbose (show symbol address, etc)"),
1069 OPT_BOOLEAN('q', "quiet", &quiet
, "Do not show any message"),
1070 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace
,
1071 "dump raw trace in ASCII"),
1072 OPT_BOOLEAN(0, "stats", &report
.stats_mode
, "Display event stats"),
1073 OPT_BOOLEAN(0, "tasks", &report
.tasks_mode
, "Display recorded tasks"),
1074 OPT_BOOLEAN(0, "mmaps", &report
.mmaps_mode
, "Display recorded tasks memory maps"),
1075 OPT_STRING('k', "vmlinux", &symbol_conf
.vmlinux_name
,
1076 "file", "vmlinux pathname"),
1077 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf
.ignore_vmlinux
,
1078 "don't load vmlinux even if found"),
1079 OPT_STRING(0, "kallsyms", &symbol_conf
.kallsyms_name
,
1080 "file", "kallsyms pathname"),
1081 OPT_BOOLEAN('f', "force", &symbol_conf
.force
, "don't complain, do it"),
1082 OPT_BOOLEAN('m', "modules", &symbol_conf
.use_modules
,
1083 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1084 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf
.show_nr_samples
,
1085 "Show a column with the number of samples"),
1086 OPT_BOOLEAN('T', "threads", &report
.show_threads
,
1087 "Show per-thread event counters"),
1088 OPT_STRING(0, "pretty", &report
.pretty_printing_style
, "key",
1089 "pretty printing style key: normal raw"),
1090 OPT_BOOLEAN(0, "tui", &report
.use_tui
, "Use the TUI interface"),
1091 OPT_BOOLEAN(0, "gtk", &report
.use_gtk
, "Use the GTK2 interface"),
1092 OPT_BOOLEAN(0, "stdio", &report
.use_stdio
,
1093 "Use the stdio interface"),
1094 OPT_BOOLEAN(0, "header", &report
.header
, "Show data header."),
1095 OPT_BOOLEAN(0, "header-only", &report
.header_only
,
1096 "Show only data header."),
1097 OPT_STRING('s', "sort", &sort_order
, "key[,key2...]",
1098 sort_help("sort by key(s):")),
1099 OPT_STRING('F', "fields", &field_order
, "key[,keys...]",
1100 sort_help("output field(s): overhead period sample ")),
1101 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf
.show_cpu_utilization
,
1102 "Show sample percentage for different cpu modes"),
1103 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf
.show_cpu_utilization
,
1104 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN
),
1105 OPT_STRING('p', "parent", &parent_pattern
, "regex",
1106 "regex filter to identify parent, see: '--sort parent'"),
1107 OPT_BOOLEAN('x', "exclude-other", &symbol_conf
.exclude_other
,
1108 "Only display entries with parent-match"),
1109 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param
,
1110 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
1111 report_callchain_help
, &report_parse_callchain_opt
,
1112 callchain_default_opt
),
1113 OPT_BOOLEAN(0, "children", &symbol_conf
.cumulate_callchain
,
1114 "Accumulate callchains of children and show total overhead as well"),
1115 OPT_INTEGER(0, "max-stack", &report
.max_stack
,
1116 "Set the maximum stack depth when parsing the callchain, "
1117 "anything beyond the specified depth will be ignored. "
1118 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH
)),
1119 OPT_BOOLEAN('G', "inverted", &report
.inverted_callchain
,
1120 "alias for inverted call graph"),
1121 OPT_CALLBACK(0, "ignore-callees", NULL
, "regex",
1122 "ignore callees of these functions in call graphs",
1123 report_parse_ignore_callees_opt
),
1124 OPT_STRING('d', "dsos", &symbol_conf
.dso_list_str
, "dso[,dso...]",
1125 "only consider symbols in these dsos"),
1126 OPT_STRING('c', "comms", &symbol_conf
.comm_list_str
, "comm[,comm...]",
1127 "only consider symbols in these comms"),
1128 OPT_STRING(0, "pid", &symbol_conf
.pid_list_str
, "pid[,pid...]",
1129 "only consider symbols in these pids"),
1130 OPT_STRING(0, "tid", &symbol_conf
.tid_list_str
, "tid[,tid...]",
1131 "only consider symbols in these tids"),
1132 OPT_STRING('S', "symbols", &symbol_conf
.sym_list_str
, "symbol[,symbol...]",
1133 "only consider these symbols"),
1134 OPT_STRING(0, "symbol-filter", &report
.symbol_filter_str
, "filter",
1135 "only show symbols that (partially) match with this filter"),
1136 OPT_STRING('w', "column-widths", &symbol_conf
.col_width_list_str
,
1138 "don't try to adjust column width, use these fixed values"),
1139 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf
.field_sep
, "separator",
1140 "separator for columns, no spaces will be added between "
1141 "columns '.' is reserved."),
1142 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf
.hide_unresolved
,
1143 "Only display entries resolved to a symbol"),
1144 OPT_CALLBACK(0, "symfs", NULL
, "directory",
1145 "Look for files with symbols relative to this directory",
1146 symbol__config_symfs
),
1147 OPT_STRING('C', "cpu", &report
.cpu_list
, "cpu",
1148 "list of cpus to profile"),
1149 OPT_BOOLEAN('I', "show-info", &report
.show_full_info
,
1150 "Display extended information about perf.data file"),
1151 OPT_BOOLEAN(0, "source", &report
.annotation_opts
.annotate_src
,
1152 "Interleave source code with assembly code (default)"),
1153 OPT_BOOLEAN(0, "asm-raw", &report
.annotation_opts
.show_asm_raw
,
1154 "Display raw encoding of assembly instructions (default)"),
1155 OPT_STRING('M', "disassembler-style", &report
.annotation_opts
.disassembler_style
, "disassembler style",
1156 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1157 OPT_BOOLEAN(0, "show-total-period", &symbol_conf
.show_total_period
,
1158 "Show a column with the sum of periods"),
1159 OPT_BOOLEAN_SET(0, "group", &symbol_conf
.event_group
, &report
.group_set
,
1160 "Show event group information together"),
1161 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode
, "",
1162 "use branch records for per branch histogram filling",
1164 OPT_BOOLEAN(0, "branch-history", &branch_call_mode
,
1165 "add last branch records to call history"),
1166 OPT_STRING(0, "objdump", &report
.annotation_opts
.objdump_path
, "path",
1167 "objdump binary to use for disassembly and annotations"),
1168 OPT_BOOLEAN(0, "demangle", &symbol_conf
.demangle
,
1169 "Disable symbol demangling"),
1170 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf
.demangle_kernel
,
1171 "Enable kernel symbol demangling"),
1172 OPT_BOOLEAN(0, "mem-mode", &report
.mem_mode
, "mem access profile"),
1173 OPT_INTEGER(0, "samples", &symbol_conf
.res_sample
,
1174 "Number of samples to save per histogram entry for individual browsing"),
1175 OPT_CALLBACK(0, "percent-limit", &report
, "percent",
1176 "Don't show entries under that percent", parse_percent_limit
),
1177 OPT_CALLBACK(0, "percentage", NULL
, "relative|absolute",
1178 "how to display percentage of filtered entries", parse_filter_percentage
),
1179 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts
, NULL
, "opts",
1180 "Instruction Tracing options\n" ITRACE_HELP
,
1181 itrace_parse_synth_opts
),
1182 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename
,
1183 "Show full source file name path for source lines"),
1184 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf
.show_ref_callgraph
,
1185 "Show callgraph from reference event"),
1186 OPT_INTEGER(0, "socket-filter", &report
.socket_filter
,
1187 "only show processor socket that match with this filter"),
1188 OPT_BOOLEAN(0, "raw-trace", &symbol_conf
.raw_trace
,
1189 "Show raw trace event output (do not use print fmt or plugins)"),
1190 OPT_BOOLEAN(0, "hierarchy", &symbol_conf
.report_hierarchy
,
1191 "Show entries in a hierarchy"),
1192 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL
, "mode",
1193 "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
1194 stdio__config_color
, "always"),
1195 OPT_STRING(0, "time", &report
.time_str
, "str",
1196 "Time span of interest (start,stop)"),
1197 OPT_BOOLEAN(0, "inline", &symbol_conf
.inline_name
,
1198 "Show inline function"),
1199 OPT_CALLBACK(0, "percent-type", &report
.annotation_opts
, "local-period",
1200 "Set percent type local/global-period/hits",
1201 annotate_parse_percent_type
),
1202 OPT_BOOLEAN(0, "ns", &symbol_conf
.nanosecs
, "Show times in nanosecs"),
1203 OPT_CALLBACK(0, "time-quantum", &symbol_conf
.time_quantum
, "time (ms|us|ns|s)",
1204 "Set time quantum for time sort key (default 100ms)",
1205 parse_time_quantum
),
1206 OPTS_EVSWITCH(&report
.evswitch
),
1209 struct perf_data data
= {
1210 .mode
= PERF_DATA_MODE_READ
,
1212 int ret
= hists__init();
1218 ret
= perf_config(report__config
, &report
);
1222 argc
= parse_options(argc
, argv
, options
, report_usage
, 0);
1225 * Special case: if there's an argument left then assume that
1226 * it's a symbol filter:
1229 usage_with_options(report_usage
, options
);
1231 report
.symbol_filter_str
= argv
[0];
1234 if (report
.mmaps_mode
)
1235 report
.tasks_mode
= true;
1238 perf_quiet_option();
1240 if (symbol_conf
.vmlinux_name
&&
1241 access(symbol_conf
.vmlinux_name
, R_OK
)) {
1242 pr_err("Invalid file: %s\n", symbol_conf
.vmlinux_name
);
1245 if (symbol_conf
.kallsyms_name
&&
1246 access(symbol_conf
.kallsyms_name
, R_OK
)) {
1247 pr_err("Invalid file: %s\n", symbol_conf
.kallsyms_name
);
1251 if (report
.inverted_callchain
)
1252 callchain_param
.order
= ORDER_CALLER
;
1253 if (symbol_conf
.cumulate_callchain
&& !callchain_param
.order_set
)
1254 callchain_param
.order
= ORDER_CALLER
;
1256 if (itrace_synth_opts
.callchain
&&
1257 (int)itrace_synth_opts
.callchain_sz
> report
.max_stack
)
1258 report
.max_stack
= itrace_synth_opts
.callchain_sz
;
1260 if (!input_name
|| !strlen(input_name
)) {
1261 if (!fstat(STDIN_FILENO
, &st
) && S_ISFIFO(st
.st_mode
))
1264 input_name
= "perf.data";
1267 data
.path
= input_name
;
1268 data
.force
= symbol_conf
.force
;
1271 session
= perf_session__new(&data
, false, &report
.tool
);
1272 if (IS_ERR(session
))
1273 return PTR_ERR(session
);
1275 ret
= evswitch__init(&report
.evswitch
, session
->evlist
, stderr
);
1279 if (zstd_init(&(session
->zstd_data
), 0) < 0)
1280 pr_warning("Decompression initialization failed. Reported data may be incomplete.\n");
1282 if (report
.queue_size
) {
1283 ordered_events__set_alloc_size(&session
->ordered_events
,
1287 session
->itrace_synth_opts
= &itrace_synth_opts
;
1289 report
.session
= session
;
1291 has_br_stack
= perf_header__has_feat(&session
->header
,
1292 HEADER_BRANCH_STACK
);
1293 if (perf_evlist__combined_sample_type(session
->evlist
) & PERF_SAMPLE_STACK_USER
)
1294 has_br_stack
= false;
1296 setup_forced_leader(&report
, session
->evlist
);
1298 if (itrace_synth_opts
.last_branch
)
1299 has_br_stack
= true;
1301 if (has_br_stack
&& branch_call_mode
)
1302 symbol_conf
.show_branchflag_count
= true;
1304 memset(&report
.brtype_stat
, 0, sizeof(struct branch_type_stat
));
1307 * Branch mode is a tristate:
1308 * -1 means default, so decide based on the file having branch data.
1309 * 0/1 means the user chose a mode.
1311 if (((branch_mode
== -1 && has_br_stack
) || branch_mode
== 1) &&
1312 !branch_call_mode
) {
1313 sort__mode
= SORT_MODE__BRANCH
;
1314 symbol_conf
.cumulate_callchain
= false;
1316 if (branch_call_mode
) {
1317 callchain_param
.key
= CCKEY_ADDRESS
;
1318 callchain_param
.branch_callstack
= 1;
1319 symbol_conf
.use_callchain
= true;
1320 callchain_register_param(&callchain_param
);
1321 if (sort_order
== NULL
)
1322 sort_order
= "srcline,symbol,dso";
1325 if (report
.mem_mode
) {
1326 if (sort__mode
== SORT_MODE__BRANCH
) {
1327 pr_err("branch and mem mode incompatible\n");
1330 sort__mode
= SORT_MODE__MEMORY
;
1331 symbol_conf
.cumulate_callchain
= false;
1334 if (symbol_conf
.report_hierarchy
) {
1335 /* disable incompatible options */
1336 symbol_conf
.cumulate_callchain
= false;
1339 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1340 parse_options_usage(report_usage
, options
, "F", 1);
1341 parse_options_usage(NULL
, options
, "hierarchy", 0);
1345 perf_hpp_list
.need_collapse
= true;
1348 if (report
.use_stdio
)
1350 else if (report
.use_tui
)
1352 else if (report
.use_gtk
)
1355 /* Force tty output for header output and per-thread stat. */
1356 if (report
.header
|| report
.header_only
|| report
.show_threads
)
1358 if (report
.header
|| report
.header_only
)
1359 report
.tool
.show_feat_hdr
= SHOW_FEAT_HEADER
;
1360 if (report
.show_full_info
)
1361 report
.tool
.show_feat_hdr
= SHOW_FEAT_HEADER_FULL_INFO
;
1362 if (report
.stats_mode
|| report
.tasks_mode
)
1364 if (report
.stats_mode
&& report
.tasks_mode
) {
1365 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n");
1369 if (strcmp(input_name
, "-") != 0)
1370 setup_browser(true);
1374 if (sort_order
&& strstr(sort_order
, "ipc")) {
1375 parse_options_usage(report_usage
, options
, "s", 1);
1379 if (sort_order
&& strstr(sort_order
, "symbol")) {
1380 if (sort__mode
== SORT_MODE__BRANCH
) {
1381 snprintf(sort_tmp
, sizeof(sort_tmp
), "%s,%s",
1382 sort_order
, "ipc_lbr");
1383 report
.symbol_ipc
= true;
1385 snprintf(sort_tmp
, sizeof(sort_tmp
), "%s,%s",
1386 sort_order
, "ipc_null");
1389 sort_order
= sort_tmp
;
1392 if (setup_sorting(session
->evlist
) < 0) {
1394 parse_options_usage(report_usage
, options
, "s", 1);
1396 parse_options_usage(sort_order
? NULL
: report_usage
,
1401 if ((report
.header
|| report
.header_only
) && !quiet
) {
1402 perf_session__fprintf_info(session
, stdout
,
1403 report
.show_full_info
);
1404 if (report
.header_only
) {
1408 } else if (use_browser
== 0 && !quiet
&&
1409 !report
.stats_mode
&& !report
.tasks_mode
) {
1410 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
1415 * Only in the TUI browser we are doing integrated annotation,
1416 * so don't allocate extra space that won't be used in the stdio
1419 if (ui__has_annotation() || report
.symbol_ipc
) {
1420 ret
= symbol__annotation_init();
1424 * For searching by name on the "Browse map details".
1425 * providing it only in verbose mode not to bloat too
1426 * much struct symbol.
1430 * XXX: Need to provide a less kludgy way to ask for
1431 * more space per symbol, the u32 is for the index on
1433 * See symbol__browser_index.
1435 symbol_conf
.priv_size
+= sizeof(u32
);
1436 symbol_conf
.sort_by_name
= true;
1438 annotation_config__init();
1441 if (symbol__init(&session
->header
.env
) < 0)
1444 if (report
.time_str
) {
1445 ret
= perf_time__parse_for_ranges(report
.time_str
, session
,
1446 &report
.ptime_range
,
1452 itrace_synth_opts__set_time_range(&itrace_synth_opts
,
1457 if (session
->tevent
.pevent
&&
1458 tep_set_function_resolver(session
->tevent
.pevent
,
1459 machine__resolve_kernel_addr
,
1460 &session
->machines
.host
) < 0) {
1461 pr_err("%s: failed to set libtraceevent function resolver\n",
1466 sort__setup_elide(stdout
);
1468 ret
= __cmd_report(&report
);
1469 if (ret
== K_SWITCH_INPUT_DATA
) {
1470 perf_session__delete(session
);
1476 if (report
.ptime_range
) {
1477 itrace_synth_opts__clear_time_range(&itrace_synth_opts
);
1478 zfree(&report
.ptime_range
);
1480 zstd_fini(&(session
->zstd_data
));
1481 perf_session__delete(session
);