4 * Builtin report command: Analyze the perf.data input file,
5 * look up and read DSOs and symbol information and display
6 * a histogram of results, along various sorting keys.
10 #include "util/util.h"
12 #include "util/annotate.h"
13 #include "util/color.h"
14 #include <linux/list.h>
15 #include "util/cache.h"
16 #include <linux/rbtree.h>
17 #include "util/symbol.h"
18 #include "util/callchain.h"
19 #include "util/strlist.h"
20 #include "util/values.h"
23 #include "util/debug.h"
24 #include "util/evlist.h"
25 #include "util/evsel.h"
26 #include "util/header.h"
27 #include "util/session.h"
28 #include "util/tool.h"
30 #include "util/parse-options.h"
31 #include "util/parse-events.h"
33 #include "util/thread.h"
34 #include "util/sort.h"
35 #include "util/hist.h"
37 #include <linux/bitmap.h>
40 struct perf_tool tool
;
41 struct perf_session
*session
;
42 char const *input_name
;
43 bool force
, use_tui
, use_gtk
, use_stdio
;
45 bool dont_use_callchains
;
48 bool inverted_callchain
;
49 struct perf_read_values show_threads_values
;
50 const char *pretty_printing_style
;
51 symbol_filter_t annotate_init
;
53 const char *symbol_filter_str
;
54 DECLARE_BITMAP(cpu_bitmap
, MAX_NR_CPUS
);
57 static int perf_report__add_branch_hist_entry(struct perf_tool
*tool
,
58 struct addr_location
*al
,
59 struct perf_sample
*sample
,
60 struct perf_evsel
*evsel
,
61 struct machine
*machine
)
63 struct perf_report
*rep
= container_of(tool
, struct perf_report
, tool
);
64 struct symbol
*parent
= NULL
;
67 struct hist_entry
*he
;
68 struct branch_info
*bi
, *bx
;
70 if ((sort__has_parent
|| symbol_conf
.use_callchain
)
71 && sample
->callchain
) {
72 err
= machine__resolve_callchain(machine
, evsel
, al
->thread
,
73 sample
->callchain
, &parent
);
78 bi
= machine__resolve_bstack(machine
, al
->thread
,
79 sample
->branch_stack
);
83 for (i
= 0; i
< sample
->branch_stack
->nr
; i
++) {
84 if (rep
->hide_unresolved
&& !(bi
[i
].from
.sym
&& bi
[i
].to
.sym
))
87 * The report shows the percentage of total branches captured
88 * and not events sampled. Thus we use a pseudo period of 1.
90 he
= __hists__add_branch_entry(&evsel
->hists
, al
, parent
,
93 struct annotation
*notes
;
96 if (bx
->from
.sym
&& use_browser
> 0) {
97 notes
= symbol__annotation(bx
->from
.sym
);
99 && symbol__alloc_hist(bx
->from
.sym
) < 0)
102 err
= symbol__inc_addr_samples(bx
->from
.sym
,
110 if (bx
->to
.sym
&& use_browser
> 0) {
111 notes
= symbol__annotation(bx
->to
.sym
);
113 && symbol__alloc_hist(bx
->to
.sym
) < 0)
116 err
= symbol__inc_addr_samples(bx
->to
.sym
,
123 evsel
->hists
.stats
.total_period
+= 1;
124 hists__inc_nr_events(&evsel
->hists
, PERF_RECORD_SAMPLE
);
133 static int perf_evsel__add_hist_entry(struct perf_evsel
*evsel
,
134 struct addr_location
*al
,
135 struct perf_sample
*sample
,
136 struct machine
*machine
)
138 struct symbol
*parent
= NULL
;
140 struct hist_entry
*he
;
142 if ((sort__has_parent
|| symbol_conf
.use_callchain
) && sample
->callchain
) {
143 err
= machine__resolve_callchain(machine
, evsel
, al
->thread
,
144 sample
->callchain
, &parent
);
149 he
= __hists__add_entry(&evsel
->hists
, al
, parent
, sample
->period
);
153 if (symbol_conf
.use_callchain
) {
154 err
= callchain_append(he
->callchain
,
155 &evsel
->hists
.callchain_cursor
,
161 * Only in the newt browser we are doing integrated annotation,
162 * so we don't allocated the extra space needed because the stdio
163 * code will not use it.
165 if (al
->sym
!= NULL
&& use_browser
> 0) {
166 struct annotation
*notes
= symbol__annotation(he
->ms
.sym
);
168 assert(evsel
!= NULL
);
171 if (notes
->src
== NULL
&& symbol__alloc_hist(he
->ms
.sym
) < 0)
174 err
= hist_entry__inc_addr_samples(he
, evsel
->idx
, al
->addr
);
177 evsel
->hists
.stats
.total_period
+= sample
->period
;
178 hists__inc_nr_events(&evsel
->hists
, PERF_RECORD_SAMPLE
);
184 static int process_sample_event(struct perf_tool
*tool
,
185 union perf_event
*event
,
186 struct perf_sample
*sample
,
187 struct perf_evsel
*evsel
,
188 struct machine
*machine
)
190 struct perf_report
*rep
= container_of(tool
, struct perf_report
, tool
);
191 struct addr_location al
;
193 if (perf_event__preprocess_sample(event
, machine
, &al
, sample
,
194 rep
->annotate_init
) < 0) {
195 fprintf(stderr
, "problem processing %d event, skipping it.\n",
200 if (al
.filtered
|| (rep
->hide_unresolved
&& al
.sym
== NULL
))
203 if (rep
->cpu_list
&& !test_bit(sample
->cpu
, rep
->cpu_bitmap
))
206 if (sort__branch_mode
== 1) {
207 if (perf_report__add_branch_hist_entry(tool
, &al
, sample
,
209 pr_debug("problem adding lbr entry, skipping event\n");
214 al
.map
->dso
->hit
= 1;
216 if (perf_evsel__add_hist_entry(evsel
, &al
, sample
, machine
)) {
217 pr_debug("problem incrementing symbol period, skipping event\n");
224 static int process_read_event(struct perf_tool
*tool
,
225 union perf_event
*event
,
226 struct perf_sample
*sample __used
,
227 struct perf_evsel
*evsel
,
228 struct machine
*machine __used
)
230 struct perf_report
*rep
= container_of(tool
, struct perf_report
, tool
);
232 if (rep
->show_threads
) {
233 const char *name
= evsel
? event_name(evsel
) : "unknown";
234 perf_read_values_add_value(&rep
->show_threads_values
,
235 event
->read
.pid
, event
->read
.tid
,
241 dump_printf(": %d %d %s %" PRIu64
"\n", event
->read
.pid
, event
->read
.tid
,
242 evsel
? event_name(evsel
) : "FAIL",
248 static int perf_report__setup_sample_type(struct perf_report
*rep
)
250 struct perf_session
*self
= rep
->session
;
252 if (!(self
->sample_type
& PERF_SAMPLE_CALLCHAIN
)) {
253 if (sort__has_parent
) {
254 ui__warning("Selected --sort parent, but no "
255 "callchain data. Did you call "
256 "'perf record' without -g?\n");
259 if (symbol_conf
.use_callchain
) {
260 ui__warning("Selected -g but no callchain data. Did "
261 "you call 'perf record' without -g?\n");
264 } else if (!rep
->dont_use_callchains
&&
265 callchain_param
.mode
!= CHAIN_NONE
&&
266 !symbol_conf
.use_callchain
) {
267 symbol_conf
.use_callchain
= true;
268 if (callchain_register_param(&callchain_param
) < 0) {
269 ui__warning("Can't register callchain "
275 if (sort__branch_mode
== 1) {
276 if (!(self
->sample_type
& PERF_SAMPLE_BRANCH_STACK
)) {
277 fprintf(stderr
, "selected -b but no branch data."
278 " Did you call perf record without"
287 extern volatile int session_done
;
289 static void sig_handler(int sig __used
)
294 static size_t hists__fprintf_nr_sample_events(struct hists
*self
,
295 const char *evname
, FILE *fp
)
299 unsigned long nr_events
= self
->stats
.nr_events
[PERF_RECORD_SAMPLE
];
301 nr_events
= convert_unit(nr_events
, &unit
);
302 ret
= fprintf(fp
, "# Events: %lu%c", nr_events
, unit
);
304 ret
+= fprintf(fp
, " %s", evname
);
305 return ret
+ fprintf(fp
, "\n#\n");
308 static int perf_evlist__tty_browse_hists(struct perf_evlist
*evlist
,
309 struct perf_report
*rep
,
312 struct perf_evsel
*pos
;
314 list_for_each_entry(pos
, &evlist
->entries
, node
) {
315 struct hists
*hists
= &pos
->hists
;
316 const char *evname
= event_name(pos
);
318 hists__fprintf_nr_sample_events(hists
, evname
, stdout
);
319 hists__fprintf(hists
, NULL
, false, true, 0, 0, stdout
);
320 fprintf(stdout
, "\n\n");
323 if (sort_order
== default_sort_order
&&
324 parent_pattern
== default_parent_pattern
) {
325 fprintf(stdout
, "#\n# (%s)\n#\n", help
);
327 if (rep
->show_threads
) {
328 bool style
= !strcmp(rep
->pretty_printing_style
, "raw");
329 perf_read_values_display(stdout
, &rep
->show_threads_values
,
331 perf_read_values_destroy(&rep
->show_threads_values
);
338 static int __cmd_report(struct perf_report
*rep
)
342 struct perf_session
*session
= rep
->session
;
343 struct perf_evsel
*pos
;
344 struct map
*kernel_map
;
345 struct kmap
*kernel_kmap
;
346 const char *help
= "For a higher level overview, try: perf report --sort comm,dso";
348 signal(SIGINT
, sig_handler
);
351 ret
= perf_session__cpu_bitmap(session
, rep
->cpu_list
,
357 if (use_browser
<= 0)
358 perf_session__fprintf_info(session
, stdout
, rep
->show_full_info
);
360 if (rep
->show_threads
)
361 perf_read_values_init(&rep
->show_threads_values
);
363 ret
= perf_report__setup_sample_type(rep
);
367 ret
= perf_session__process_events(session
, &rep
->tool
);
371 kernel_map
= session
->host_machine
.vmlinux_maps
[MAP__FUNCTION
];
372 kernel_kmap
= map__kmap(kernel_map
);
373 if (kernel_map
== NULL
||
374 (kernel_map
->dso
->hit
&&
375 (kernel_kmap
->ref_reloc_sym
== NULL
||
376 kernel_kmap
->ref_reloc_sym
->addr
== 0))) {
377 const struct dso
*kdso
= kernel_map
->dso
;
380 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
381 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
382 "Samples in kernel modules can't be resolved as well.\n\n",
383 RB_EMPTY_ROOT(&kdso
->symbols
[MAP__FUNCTION
]) ?
384 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
385 "can't be resolved." :
386 "If some relocation was applied (e.g. kexec) symbols may be misresolved.");
390 perf_session__fprintf_nr_events(session
, stdout
);
395 perf_session__fprintf(session
, stdout
);
398 perf_session__fprintf_dsos(session
, stdout
);
401 list_for_each_entry(pos
, &session
->evlist
->entries
, node
) {
402 struct hists
*hists
= &pos
->hists
;
405 hists
->symbol_filter_str
= rep
->symbol_filter_str
;
407 hists__collapse_resort(hists
);
408 hists__output_resort(hists
);
409 nr_samples
+= hists
->stats
.nr_events
[PERF_RECORD_SAMPLE
];
412 if (nr_samples
== 0) {
413 ui__warning("The %s file has no samples!\n", session
->filename
);
417 if (use_browser
> 0) {
418 if (use_browser
== 1) {
419 perf_evlist__tui_browse_hists(session
->evlist
, help
,
421 } else if (use_browser
== 2) {
422 perf_evlist__gtk_browse_hists(session
->evlist
, help
,
426 perf_evlist__tty_browse_hists(session
->evlist
, rep
, help
);
430 * Speed up the exit process, for large files this can
431 * take quite a while.
433 * XXX Enable this when using valgrind or if we ever
434 * librarize this command.
436 * Also experiment with obstacks to see how much speed
439 * perf_session__delete(session);
445 parse_callchain_opt(const struct option
*opt
, const char *arg
, int unset
)
447 struct perf_report
*rep
= (struct perf_report
*)opt
->value
;
455 rep
->dont_use_callchains
= true;
459 symbol_conf
.use_callchain
= true;
464 tok
= strtok((char *)arg
, ",");
468 /* get the output mode */
469 if (!strncmp(tok
, "graph", strlen(arg
)))
470 callchain_param
.mode
= CHAIN_GRAPH_ABS
;
472 else if (!strncmp(tok
, "flat", strlen(arg
)))
473 callchain_param
.mode
= CHAIN_FLAT
;
475 else if (!strncmp(tok
, "fractal", strlen(arg
)))
476 callchain_param
.mode
= CHAIN_GRAPH_REL
;
478 else if (!strncmp(tok
, "none", strlen(arg
))) {
479 callchain_param
.mode
= CHAIN_NONE
;
480 symbol_conf
.use_callchain
= false;
488 /* get the min percentage */
489 tok
= strtok(NULL
, ",");
493 callchain_param
.min_percent
= strtod(tok
, &endptr
);
497 /* get the print limit */
498 tok2
= strtok(NULL
, ",");
502 if (tok2
[0] != 'c') {
503 callchain_param
.print_limit
= strtoul(tok2
, &endptr
, 0);
504 tok2
= strtok(NULL
, ",");
509 /* get the call chain order */
510 if (!strcmp(tok2
, "caller"))
511 callchain_param
.order
= ORDER_CALLER
;
512 else if (!strcmp(tok2
, "callee"))
513 callchain_param
.order
= ORDER_CALLEE
;
517 if (callchain_register_param(&callchain_param
) < 0) {
518 fprintf(stderr
, "Can't register callchain params\n");
525 parse_branch_mode(const struct option
*opt __used
, const char *str __used
, int unset
)
527 sort__branch_mode
= !unset
;
531 int cmd_report(int argc
, const char **argv
, const char *prefix __used
)
533 struct perf_session
*session
;
535 bool has_br_stack
= false;
537 char callchain_default_opt
[] = "fractal,0.5,callee";
538 const char * const report_usage
[] = {
539 "perf report [<options>]",
542 struct perf_report report
= {
544 .sample
= process_sample_event
,
545 .mmap
= perf_event__process_mmap
,
546 .comm
= perf_event__process_comm
,
547 .exit
= perf_event__process_task
,
548 .fork
= perf_event__process_task
,
549 .lost
= perf_event__process_lost
,
550 .read
= process_read_event
,
551 .attr
= perf_event__process_attr
,
552 .event_type
= perf_event__process_event_type
,
553 .tracing_data
= perf_event__process_tracing_data
,
554 .build_id
= perf_event__process_build_id
,
555 .ordered_samples
= true,
556 .ordering_requires_timestamps
= true,
558 .pretty_printing_style
= "normal",
560 const struct option options
[] = {
561 OPT_STRING('i', "input", &report
.input_name
, "file",
563 OPT_INCR('v', "verbose", &verbose
,
564 "be more verbose (show symbol address, etc)"),
565 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace
,
566 "dump raw trace in ASCII"),
567 OPT_STRING('k', "vmlinux", &symbol_conf
.vmlinux_name
,
568 "file", "vmlinux pathname"),
569 OPT_STRING(0, "kallsyms", &symbol_conf
.kallsyms_name
,
570 "file", "kallsyms pathname"),
571 OPT_BOOLEAN('f', "force", &report
.force
, "don't complain, do it"),
572 OPT_BOOLEAN('m', "modules", &symbol_conf
.use_modules
,
573 "load module symbols - WARNING: use only with -k and LIVE kernel"),
574 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf
.show_nr_samples
,
575 "Show a column with the number of samples"),
576 OPT_BOOLEAN('T', "threads", &report
.show_threads
,
577 "Show per-thread event counters"),
578 OPT_STRING(0, "pretty", &report
.pretty_printing_style
, "key",
579 "pretty printing style key: normal raw"),
580 OPT_BOOLEAN(0, "tui", &report
.use_tui
, "Use the TUI interface"),
581 OPT_BOOLEAN(0, "gtk", &report
.use_gtk
, "Use the GTK2 interface"),
582 OPT_BOOLEAN(0, "stdio", &report
.use_stdio
,
583 "Use the stdio interface"),
584 OPT_STRING('s', "sort", &sort_order
, "key[,key2...]",
585 "sort by key(s): pid, comm, dso, symbol, parent, dso_to,"
586 " dso_from, symbol_to, symbol_from, mispredict"),
587 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf
.show_cpu_utilization
,
588 "Show sample percentage for different cpu modes"),
589 OPT_STRING('p', "parent", &parent_pattern
, "regex",
590 "regex filter to identify parent, see: '--sort parent'"),
591 OPT_BOOLEAN('x', "exclude-other", &symbol_conf
.exclude_other
,
592 "Only display entries with parent-match"),
593 OPT_CALLBACK_DEFAULT('g', "call-graph", &report
, "output_type,min_percent[,print_limit],call_order",
594 "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold, optional print limit and callchain order. "
595 "Default: fractal,0.5,callee", &parse_callchain_opt
, callchain_default_opt
),
596 OPT_BOOLEAN('G', "inverted", &report
.inverted_callchain
,
597 "alias for inverted call graph"),
598 OPT_STRING('d', "dsos", &symbol_conf
.dso_list_str
, "dso[,dso...]",
599 "only consider symbols in these dsos"),
600 OPT_STRING('c', "comms", &symbol_conf
.comm_list_str
, "comm[,comm...]",
601 "only consider symbols in these comms"),
602 OPT_STRING('S', "symbols", &symbol_conf
.sym_list_str
, "symbol[,symbol...]",
603 "only consider these symbols"),
604 OPT_STRING(0, "symbol-filter", &report
.symbol_filter_str
, "filter",
605 "only show symbols that (partially) match with this filter"),
606 OPT_STRING('w', "column-widths", &symbol_conf
.col_width_list_str
,
608 "don't try to adjust column width, use these fixed values"),
609 OPT_STRING('t', "field-separator", &symbol_conf
.field_sep
, "separator",
610 "separator for columns, no spaces will be added between "
611 "columns '.' is reserved."),
612 OPT_BOOLEAN('U', "hide-unresolved", &report
.hide_unresolved
,
613 "Only display entries resolved to a symbol"),
614 OPT_STRING(0, "symfs", &symbol_conf
.symfs
, "directory",
615 "Look for files with symbols relative to this directory"),
616 OPT_STRING('C', "cpu", &report
.cpu_list
, "cpu",
617 "list of cpus to profile"),
618 OPT_BOOLEAN('I', "show-info", &report
.show_full_info
,
619 "Display extended information about perf.data file"),
620 OPT_BOOLEAN(0, "source", &symbol_conf
.annotate_src
,
621 "Interleave source code with assembly code (default)"),
622 OPT_BOOLEAN(0, "asm-raw", &symbol_conf
.annotate_asm_raw
,
623 "Display raw encoding of assembly instructions (default)"),
624 OPT_STRING('M', "disassembler-style", &disassembler_style
, "disassembler style",
625 "Specify disassembler style (e.g. -M intel for intel syntax)"),
626 OPT_BOOLEAN(0, "show-total-period", &symbol_conf
.show_total_period
,
627 "Show a column with the sum of periods"),
628 OPT_CALLBACK_NOOPT('b', "branch-stack", &sort__branch_mode
, "",
629 "use branch records for histogram filling", parse_branch_mode
),
633 argc
= parse_options(argc
, argv
, options
, report_usage
, 0);
635 if (report
.use_stdio
)
637 else if (report
.use_tui
)
639 else if (report
.use_gtk
)
642 if (report
.inverted_callchain
)
643 callchain_param
.order
= ORDER_CALLER
;
645 if (!report
.input_name
|| !strlen(report
.input_name
)) {
646 if (!fstat(STDIN_FILENO
, &st
) && S_ISFIFO(st
.st_mode
))
647 report
.input_name
= "-";
649 report
.input_name
= "perf.data";
651 session
= perf_session__new(report
.input_name
, O_RDONLY
,
652 report
.force
, false, &report
.tool
);
656 report
.session
= session
;
658 has_br_stack
= perf_header__has_feat(&session
->header
,
659 HEADER_BRANCH_STACK
);
661 if (sort__branch_mode
== -1 && has_br_stack
)
662 sort__branch_mode
= 1;
664 /* sort__branch_mode could be 0 if --no-branch-stack */
665 if (sort__branch_mode
== 1) {
667 * if no sort_order is provided, then specify
668 * branch-mode specific order
670 if (sort_order
== default_sort_order
)
671 sort_order
= "comm,dso_from,symbol_from,"
676 if (strcmp(report
.input_name
, "-") != 0) {
678 perf_gtk_setup_browser(argc
, argv
, true);
686 * Only in the newt browser we are doing integrated annotation,
687 * so don't allocate extra space that won't be used in the stdio
690 if (use_browser
> 0) {
691 symbol_conf
.priv_size
= sizeof(struct annotation
);
692 report
.annotate_init
= symbol__annotate_init
;
694 * For searching by name on the "Browse map details".
695 * providing it only in verbose mode not to bloat too
696 * much struct symbol.
700 * XXX: Need to provide a less kludgy way to ask for
701 * more space per symbol, the u32 is for the index on
703 * See symbol__browser_index.
705 symbol_conf
.priv_size
+= sizeof(u32
);
706 symbol_conf
.sort_by_name
= true;
710 if (symbol__init() < 0)
713 setup_sorting(report_usage
, options
);
715 if (parent_pattern
!= default_parent_pattern
) {
716 if (sort_dimension__add("parent") < 0)
720 * Only show the parent fields if we explicitly
721 * sort that way. If we only use parent machinery
722 * for filtering, we don't want it.
724 if (!strstr(sort_order
, "parent"))
725 sort_parent
.elide
= 1;
727 symbol_conf
.exclude_other
= false;
731 * Special case: if there's an argument left then assume that
732 * it's a symbol filter:
735 usage_with_options(report_usage
, options
);
737 report
.symbol_filter_str
= argv
[0];
740 sort_entry__setup_elide(&sort_comm
, symbol_conf
.comm_list
, "comm", stdout
);
742 if (sort__branch_mode
== 1) {
743 sort_entry__setup_elide(&sort_dso_from
, symbol_conf
.dso_from_list
, "dso_from", stdout
);
744 sort_entry__setup_elide(&sort_dso_to
, symbol_conf
.dso_to_list
, "dso_to", stdout
);
745 sort_entry__setup_elide(&sort_sym_from
, symbol_conf
.sym_from_list
, "sym_from", stdout
);
746 sort_entry__setup_elide(&sort_sym_to
, symbol_conf
.sym_to_list
, "sym_to", stdout
);
748 sort_entry__setup_elide(&sort_dso
, symbol_conf
.dso_list
, "dso", stdout
);
749 sort_entry__setup_elide(&sort_sym
, symbol_conf
.sym_list
, "symbol", stdout
);
752 ret
= __cmd_report(&report
);
754 perf_session__delete(session
);