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/color.h"
13 #include "util/list.h"
14 #include "util/cache.h"
15 #include "util/rbtree.h"
16 #include "util/symbol.h"
17 #include "util/string.h"
20 #include "util/header.h"
22 #include "util/parse-options.h"
23 #include "util/parse-events.h"
29 static char const *input_name
= "perf.data";
30 static char *vmlinux
= NULL
;
32 static char default_sort_order
[] = "comm,dso";
33 static char *sort_order
= default_sort_order
;
36 static int show_mask
= SHOW_KERNEL
| SHOW_USER
| SHOW_HV
;
38 static int dump_trace
= 0;
39 #define dprintf(x...) do { if (dump_trace) printf(x); } while (0)
40 #define cdprintf(x...) do { if (dump_trace) color_fprintf(stdout, color, x); } while (0)
43 #define eprintf(x...) do { if (verbose) fprintf(stderr, x); } while (0)
45 static int full_paths
;
47 static unsigned long page_size
;
48 static unsigned long mmap_window
= 32;
50 static char default_parent_pattern
[] = "^sys_|^do_page_fault";
51 static char *parent_pattern
= default_parent_pattern
;
52 static regex_t parent_regex
;
54 static int exclude_other
= 1;
56 static u64 sample_type
;
59 struct perf_event_header header
;
62 unsigned char __more_data
[];
71 struct perf_event_header header
;
76 char filename
[PATH_MAX
];
80 struct perf_event_header header
;
86 struct perf_event_header header
;
91 struct perf_event_header header
;
98 struct perf_event_header header
;
104 struct perf_event_header header
;
110 typedef union event_union
{
111 struct perf_event_header header
;
113 struct mmap_event mmap
;
114 struct comm_event comm
;
115 struct fork_event fork
;
116 struct period_event period
;
117 struct lost_event lost
;
118 struct read_event read
;
121 static LIST_HEAD(dsos
);
122 static struct dso
*kernel_dso
;
123 static struct dso
*vdso
;
125 static void dsos__add(struct dso
*dso
)
127 list_add_tail(&dso
->node
, &dsos
);
130 static struct dso
*dsos__find(const char *name
)
134 list_for_each_entry(pos
, &dsos
, node
)
135 if (strcmp(pos
->name
, name
) == 0)
140 static struct dso
*dsos__findnew(const char *name
)
142 struct dso
*dso
= dsos__find(name
);
148 dso
= dso__new(name
, 0);
152 nr
= dso__load(dso
, NULL
, verbose
);
154 eprintf("Failed to open: %s\n", name
);
158 eprintf("No symbols found in: %s, maybe install a debug package?\n", name
);
169 static void dsos__fprintf(FILE *fp
)
173 list_for_each_entry(pos
, &dsos
, node
)
174 dso__fprintf(pos
, fp
);
177 static struct symbol
*vdso__find_symbol(struct dso
*dso
, u64 ip
)
179 return dso__find_symbol(kernel_dso
, ip
);
182 static int load_kernel(void)
186 kernel_dso
= dso__new("[kernel]", 0);
190 err
= dso__load_kernel(kernel_dso
, vmlinux
, NULL
, verbose
);
192 dso__delete(kernel_dso
);
195 dsos__add(kernel_dso
);
197 vdso
= dso__new("[vdso]", 0);
201 vdso
->find_symbol
= vdso__find_symbol
;
208 static char __cwd
[PATH_MAX
];
209 static char *cwd
= __cwd
;
212 static int strcommon(const char *pathname
)
216 while (pathname
[n
] == cwd
[n
] && n
< cwdlen
)
223 struct list_head node
;
227 u64 (*map_ip
)(struct map
*, u64
);
231 static u64
map__map_ip(struct map
*map
, u64 ip
)
233 return ip
- map
->start
+ map
->pgoff
;
236 static u64
vdso__map_ip(struct map
*map
, u64 ip
)
241 static inline int is_anon_memory(const char *filename
)
243 return strcmp(filename
, "//anon") == 0;
246 static struct map
*map__new(struct mmap_event
*event
)
248 struct map
*self
= malloc(sizeof(*self
));
251 const char *filename
= event
->filename
;
252 char newfilename
[PATH_MAX
];
256 int n
= strcommon(filename
);
259 snprintf(newfilename
, sizeof(newfilename
),
260 ".%s", filename
+ n
);
261 filename
= newfilename
;
265 anon
= is_anon_memory(filename
);
268 snprintf(newfilename
, sizeof(newfilename
), "/tmp/perf-%d.map", event
->pid
);
269 filename
= newfilename
;
272 self
->start
= event
->start
;
273 self
->end
= event
->start
+ event
->len
;
274 self
->pgoff
= event
->pgoff
;
276 self
->dso
= dsos__findnew(filename
);
277 if (self
->dso
== NULL
)
280 if (self
->dso
== vdso
|| anon
)
281 self
->map_ip
= vdso__map_ip
;
283 self
->map_ip
= map__map_ip
;
291 static struct map
*map__clone(struct map
*self
)
293 struct map
*map
= malloc(sizeof(*self
));
298 memcpy(map
, self
, sizeof(*self
));
303 static int map__overlap(struct map
*l
, struct map
*r
)
305 if (l
->start
> r
->start
) {
311 if (l
->end
> r
->start
)
317 static size_t map__fprintf(struct map
*self
, FILE *fp
)
319 return fprintf(fp
, " %Lx-%Lx %Lx %s\n",
320 self
->start
, self
->end
, self
->pgoff
, self
->dso
->name
);
325 struct rb_node rb_node
;
326 struct list_head maps
;
331 static struct thread
*thread__new(pid_t pid
)
333 struct thread
*self
= malloc(sizeof(*self
));
337 self
->comm
= malloc(32);
339 snprintf(self
->comm
, 32, ":%d", self
->pid
);
340 INIT_LIST_HEAD(&self
->maps
);
346 static int thread__set_comm(struct thread
*self
, const char *comm
)
350 self
->comm
= strdup(comm
);
351 return self
->comm
? 0 : -ENOMEM
;
354 static size_t thread__fprintf(struct thread
*self
, FILE *fp
)
357 size_t ret
= fprintf(fp
, "Thread %d %s\n", self
->pid
, self
->comm
);
359 list_for_each_entry(pos
, &self
->maps
, node
)
360 ret
+= map__fprintf(pos
, fp
);
366 static struct rb_root threads
;
367 static struct thread
*last_match
;
369 static struct thread
*threads__findnew(pid_t pid
)
371 struct rb_node
**p
= &threads
.rb_node
;
372 struct rb_node
*parent
= NULL
;
376 * Font-end cache - PID lookups come in blocks,
377 * so most of the time we dont have to look up
380 if (last_match
&& last_match
->pid
== pid
)
385 th
= rb_entry(parent
, struct thread
, rb_node
);
387 if (th
->pid
== pid
) {
398 th
= thread__new(pid
);
400 rb_link_node(&th
->rb_node
, parent
, p
);
401 rb_insert_color(&th
->rb_node
, &threads
);
408 static void thread__insert_map(struct thread
*self
, struct map
*map
)
410 struct map
*pos
, *tmp
;
412 list_for_each_entry_safe(pos
, tmp
, &self
->maps
, node
) {
413 if (map__overlap(pos
, map
)) {
415 printf("overlapping maps:\n");
416 map__fprintf(map
, stdout
);
417 map__fprintf(pos
, stdout
);
420 if (map
->start
<= pos
->start
&& map
->end
> pos
->start
)
421 pos
->start
= map
->end
;
423 if (map
->end
>= pos
->end
&& map
->start
< pos
->end
)
424 pos
->end
= map
->start
;
427 printf("after collision:\n");
428 map__fprintf(pos
, stdout
);
431 if (pos
->start
>= pos
->end
) {
432 list_del_init(&pos
->node
);
438 list_add_tail(&map
->node
, &self
->maps
);
441 static int thread__fork(struct thread
*self
, struct thread
*parent
)
447 self
->comm
= strdup(parent
->comm
);
451 list_for_each_entry(map
, &parent
->maps
, node
) {
452 struct map
*new = map__clone(map
);
455 thread__insert_map(self
, new);
461 static struct map
*thread__find_map(struct thread
*self
, u64 ip
)
468 list_for_each_entry(pos
, &self
->maps
, node
)
469 if (ip
>= pos
->start
&& ip
<= pos
->end
)
475 static size_t threads__fprintf(FILE *fp
)
480 for (nd
= rb_first(&threads
); nd
; nd
= rb_next(nd
)) {
481 struct thread
*pos
= rb_entry(nd
, struct thread
, rb_node
);
483 ret
+= thread__fprintf(pos
, fp
);
490 * histogram, sorted on item, collects counts
493 static struct rb_root hist
;
496 struct rb_node rb_node
;
498 struct thread
*thread
;
502 struct symbol
*parent
;
510 * configurable sorting bits
514 struct list_head list
;
518 int64_t (*cmp
)(struct hist_entry
*, struct hist_entry
*);
519 int64_t (*collapse
)(struct hist_entry
*, struct hist_entry
*);
520 size_t (*print
)(FILE *fp
, struct hist_entry
*);
523 static int64_t cmp_null(void *l
, void *r
)
536 sort__thread_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
538 return right
->thread
->pid
- left
->thread
->pid
;
542 sort__thread_print(FILE *fp
, struct hist_entry
*self
)
544 return fprintf(fp
, "%16s:%5d", self
->thread
->comm
?: "", self
->thread
->pid
);
547 static struct sort_entry sort_thread
= {
548 .header
= " Command: Pid",
549 .cmp
= sort__thread_cmp
,
550 .print
= sort__thread_print
,
556 sort__comm_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
558 return right
->thread
->pid
- left
->thread
->pid
;
562 sort__comm_collapse(struct hist_entry
*left
, struct hist_entry
*right
)
564 char *comm_l
= left
->thread
->comm
;
565 char *comm_r
= right
->thread
->comm
;
567 if (!comm_l
|| !comm_r
)
568 return cmp_null(comm_l
, comm_r
);
570 return strcmp(comm_l
, comm_r
);
574 sort__comm_print(FILE *fp
, struct hist_entry
*self
)
576 return fprintf(fp
, "%16s", self
->thread
->comm
);
579 static struct sort_entry sort_comm
= {
580 .header
= " Command",
581 .cmp
= sort__comm_cmp
,
582 .collapse
= sort__comm_collapse
,
583 .print
= sort__comm_print
,
589 sort__dso_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
591 struct dso
*dso_l
= left
->dso
;
592 struct dso
*dso_r
= right
->dso
;
594 if (!dso_l
|| !dso_r
)
595 return cmp_null(dso_l
, dso_r
);
597 return strcmp(dso_l
->name
, dso_r
->name
);
601 sort__dso_print(FILE *fp
, struct hist_entry
*self
)
604 return fprintf(fp
, "%-25s", self
->dso
->name
);
606 return fprintf(fp
, "%016llx ", (u64
)self
->ip
);
609 static struct sort_entry sort_dso
= {
610 .header
= "Shared Object ",
611 .cmp
= sort__dso_cmp
,
612 .print
= sort__dso_print
,
618 sort__sym_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
622 if (left
->sym
== right
->sym
)
625 ip_l
= left
->sym
? left
->sym
->start
: left
->ip
;
626 ip_r
= right
->sym
? right
->sym
->start
: right
->ip
;
628 return (int64_t)(ip_r
- ip_l
);
632 sort__sym_print(FILE *fp
, struct hist_entry
*self
)
637 ret
+= fprintf(fp
, "%#018llx ", (u64
)self
->ip
);
640 ret
+= fprintf(fp
, "[%c] %s",
641 self
->dso
== kernel_dso
? 'k' : '.', self
->sym
->name
);
643 ret
+= fprintf(fp
, "%#016llx", (u64
)self
->ip
);
649 static struct sort_entry sort_sym
= {
651 .cmp
= sort__sym_cmp
,
652 .print
= sort__sym_print
,
658 sort__parent_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
660 struct symbol
*sym_l
= left
->parent
;
661 struct symbol
*sym_r
= right
->parent
;
663 if (!sym_l
|| !sym_r
)
664 return cmp_null(sym_l
, sym_r
);
666 return strcmp(sym_l
->name
, sym_r
->name
);
670 sort__parent_print(FILE *fp
, struct hist_entry
*self
)
674 ret
+= fprintf(fp
, "%-20s", self
->parent
? self
->parent
->name
: "[other]");
679 static struct sort_entry sort_parent
= {
680 .header
= "Parent symbol ",
681 .cmp
= sort__parent_cmp
,
682 .print
= sort__parent_print
,
685 static int sort__need_collapse
= 0;
686 static int sort__has_parent
= 0;
688 struct sort_dimension
{
690 struct sort_entry
*entry
;
694 static struct sort_dimension sort_dimensions
[] = {
695 { .name
= "pid", .entry
= &sort_thread
, },
696 { .name
= "comm", .entry
= &sort_comm
, },
697 { .name
= "dso", .entry
= &sort_dso
, },
698 { .name
= "symbol", .entry
= &sort_sym
, },
699 { .name
= "parent", .entry
= &sort_parent
, },
702 static LIST_HEAD(hist_entry__sort_list
);
704 static int sort_dimension__add(char *tok
)
708 for (i
= 0; i
< ARRAY_SIZE(sort_dimensions
); i
++) {
709 struct sort_dimension
*sd
= &sort_dimensions
[i
];
714 if (strncasecmp(tok
, sd
->name
, strlen(tok
)))
717 if (sd
->entry
->collapse
)
718 sort__need_collapse
= 1;
720 if (sd
->entry
== &sort_parent
) {
721 int ret
= regcomp(&parent_regex
, parent_pattern
, REG_EXTENDED
);
725 regerror(ret
, &parent_regex
, err
, sizeof(err
));
726 fprintf(stderr
, "Invalid regex: %s\n%s",
727 parent_pattern
, err
);
730 sort__has_parent
= 1;
733 list_add_tail(&sd
->entry
->list
, &hist_entry__sort_list
);
743 hist_entry__cmp(struct hist_entry
*left
, struct hist_entry
*right
)
745 struct sort_entry
*se
;
748 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
749 cmp
= se
->cmp(left
, right
);
758 hist_entry__collapse(struct hist_entry
*left
, struct hist_entry
*right
)
760 struct sort_entry
*se
;
763 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
764 int64_t (*f
)(struct hist_entry
*, struct hist_entry
*);
766 f
= se
->collapse
?: se
->cmp
;
768 cmp
= f(left
, right
);
777 hist_entry__fprintf(FILE *fp
, struct hist_entry
*self
, u64 total_samples
)
779 struct sort_entry
*se
;
782 if (exclude_other
&& !self
->parent
)
786 double percent
= self
->count
* 100.0 / total_samples
;
787 char *color
= PERF_COLOR_NORMAL
;
790 * We color high-overhead entries in red, mid-overhead
791 * entries in green - and keep the low overhead places
794 if (percent
>= 5.0) {
795 color
= PERF_COLOR_RED
;
798 color
= PERF_COLOR_GREEN
;
801 ret
= color_fprintf(fp
, color
, " %6.2f%%",
802 (self
->count
* 100.0) / total_samples
);
804 ret
= fprintf(fp
, "%12Ld ", self
->count
);
806 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
807 if (exclude_other
&& (se
== &sort_parent
))
811 ret
+= se
->print(fp
, self
);
814 ret
+= fprintf(fp
, "\n");
823 static struct symbol
*
824 resolve_symbol(struct thread
*thread
, struct map
**mapp
,
825 struct dso
**dsop
, u64
*ipp
)
827 struct dso
*dso
= dsop
? *dsop
: NULL
;
828 struct map
*map
= mapp
? *mapp
: NULL
;
840 map
= thread__find_map(thread
, ip
);
845 ip
= map
->map_ip(map
, ip
);
850 * If this is outside of all known maps,
851 * and is a negative address, try to look it
852 * up in the kernel dso, as it might be a
853 * vsyscall (which executes in user-mode):
855 if ((long long)ip
< 0)
858 dprintf(" ...... dso: %s\n", dso
? dso
->name
: "<not found>");
859 dprintf(" ...... map: %Lx -> %Lx\n", *ipp
, ip
);
868 return dso
->find_symbol(dso
, ip
);
871 static int call__match(struct symbol
*sym
)
873 if (sym
->name
&& !regexec(&parent_regex
, sym
->name
, 0, NULL
, 0))
880 * collect histogram counts
884 hist_entry__add(struct thread
*thread
, struct map
*map
, struct dso
*dso
,
885 struct symbol
*sym
, u64 ip
, struct ip_callchain
*chain
,
886 char level
, u64 count
)
888 struct rb_node
**p
= &hist
.rb_node
;
889 struct rb_node
*parent
= NULL
;
890 struct hist_entry
*he
;
891 struct hist_entry entry
= {
903 if (sort__has_parent
&& chain
) {
904 u64 context
= PERF_CONTEXT_MAX
;
907 for (i
= 0; i
< chain
->nr
; i
++) {
908 u64 ip
= chain
->ips
[i
];
909 struct dso
*dso
= NULL
;
912 if (ip
>= PERF_CONTEXT_MAX
) {
918 case PERF_CONTEXT_KERNEL
:
925 sym
= resolve_symbol(thread
, NULL
, &dso
, &ip
);
927 if (sym
&& call__match(sym
)) {
936 he
= rb_entry(parent
, struct hist_entry
, rb_node
);
938 cmp
= hist_entry__cmp(&entry
, he
);
951 he
= malloc(sizeof(*he
));
955 rb_link_node(&he
->rb_node
, parent
, p
);
956 rb_insert_color(&he
->rb_node
, &hist
);
961 static void hist_entry__free(struct hist_entry
*he
)
967 * collapse the histogram
970 static struct rb_root collapse_hists
;
972 static void collapse__insert_entry(struct hist_entry
*he
)
974 struct rb_node
**p
= &collapse_hists
.rb_node
;
975 struct rb_node
*parent
= NULL
;
976 struct hist_entry
*iter
;
981 iter
= rb_entry(parent
, struct hist_entry
, rb_node
);
983 cmp
= hist_entry__collapse(iter
, he
);
986 iter
->count
+= he
->count
;
987 hist_entry__free(he
);
997 rb_link_node(&he
->rb_node
, parent
, p
);
998 rb_insert_color(&he
->rb_node
, &collapse_hists
);
1001 static void collapse__resort(void)
1003 struct rb_node
*next
;
1004 struct hist_entry
*n
;
1006 if (!sort__need_collapse
)
1009 next
= rb_first(&hist
);
1011 n
= rb_entry(next
, struct hist_entry
, rb_node
);
1012 next
= rb_next(&n
->rb_node
);
1014 rb_erase(&n
->rb_node
, &hist
);
1015 collapse__insert_entry(n
);
1020 * reverse the map, sort on count.
1023 static struct rb_root output_hists
;
1025 static void output__insert_entry(struct hist_entry
*he
)
1027 struct rb_node
**p
= &output_hists
.rb_node
;
1028 struct rb_node
*parent
= NULL
;
1029 struct hist_entry
*iter
;
1031 while (*p
!= NULL
) {
1033 iter
= rb_entry(parent
, struct hist_entry
, rb_node
);
1035 if (he
->count
> iter
->count
)
1038 p
= &(*p
)->rb_right
;
1041 rb_link_node(&he
->rb_node
, parent
, p
);
1042 rb_insert_color(&he
->rb_node
, &output_hists
);
1045 static void output__resort(void)
1047 struct rb_node
*next
;
1048 struct hist_entry
*n
;
1049 struct rb_root
*tree
= &hist
;
1051 if (sort__need_collapse
)
1052 tree
= &collapse_hists
;
1054 next
= rb_first(tree
);
1057 n
= rb_entry(next
, struct hist_entry
, rb_node
);
1058 next
= rb_next(&n
->rb_node
);
1060 rb_erase(&n
->rb_node
, tree
);
1061 output__insert_entry(n
);
1065 static size_t output__fprintf(FILE *fp
, u64 total_samples
)
1067 struct hist_entry
*pos
;
1068 struct sort_entry
*se
;
1074 fprintf(fp
, "# (%Ld samples)\n", (u64
)total_samples
);
1077 fprintf(fp
, "# Overhead");
1078 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
1079 if (exclude_other
&& (se
== &sort_parent
))
1081 fprintf(fp
, " %s", se
->header
);
1085 fprintf(fp
, "# ........");
1086 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
1089 if (exclude_other
&& (se
== &sort_parent
))
1093 for (i
= 0; i
< strlen(se
->header
); i
++)
1100 for (nd
= rb_first(&output_hists
); nd
; nd
= rb_next(nd
)) {
1101 pos
= rb_entry(nd
, struct hist_entry
, rb_node
);
1102 ret
+= hist_entry__fprintf(fp
, pos
, total_samples
);
1105 if (sort_order
== default_sort_order
&&
1106 parent_pattern
== default_parent_pattern
) {
1108 fprintf(fp
, "# (For more details, try: perf report --sort comm,dso,symbol)\n");
1116 static void register_idle_thread(void)
1118 struct thread
*thread
= threads__findnew(0);
1120 if (thread
== NULL
||
1121 thread__set_comm(thread
, "[idle]")) {
1122 fprintf(stderr
, "problem inserting idle task.\n");
1127 static unsigned long total
= 0,
1134 static int validate_chain(struct ip_callchain
*chain
, event_t
*event
)
1136 unsigned int chain_size
;
1138 chain_size
= event
->header
.size
;
1139 chain_size
-= (unsigned long)&event
->ip
.__more_data
- (unsigned long)event
;
1141 if (chain
->nr
*sizeof(u64
) > chain_size
)
1148 process_sample_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1152 struct dso
*dso
= NULL
;
1153 struct thread
*thread
= threads__findnew(event
->ip
.pid
);
1154 u64 ip
= event
->ip
.ip
;
1156 struct map
*map
= NULL
;
1157 void *more_data
= event
->ip
.__more_data
;
1158 struct ip_callchain
*chain
= NULL
;
1160 if (sample_type
& PERF_SAMPLE_PERIOD
) {
1161 period
= *(u64
*)more_data
;
1162 more_data
+= sizeof(u64
);
1165 dprintf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d: %p period: %Ld\n",
1166 (void *)(offset
+ head
),
1167 (void *)(long)(event
->header
.size
),
1173 if (sample_type
& PERF_SAMPLE_CALLCHAIN
) {
1176 chain
= (void *)more_data
;
1178 dprintf("... chain: nr:%Lu\n", chain
->nr
);
1180 if (validate_chain(chain
, event
) < 0) {
1181 eprintf("call-chain problem with event, skipping it.\n");
1186 for (i
= 0; i
< chain
->nr
; i
++)
1187 dprintf("..... %2d: %016Lx\n", i
, chain
->ips
[i
]);
1191 dprintf(" ... thread: %s:%d\n", thread
->comm
, thread
->pid
);
1193 if (thread
== NULL
) {
1194 eprintf("problem processing %d event, skipping it.\n",
1195 event
->header
.type
);
1199 if (event
->header
.misc
& PERF_EVENT_MISC_KERNEL
) {
1205 dprintf(" ...... dso: %s\n", dso
->name
);
1207 } else if (event
->header
.misc
& PERF_EVENT_MISC_USER
) {
1215 dprintf(" ...... dso: [hypervisor]\n");
1218 if (show
& show_mask
) {
1219 struct symbol
*sym
= resolve_symbol(thread
, &map
, &dso
, &ip
);
1221 if (hist_entry__add(thread
, map
, dso
, sym
, ip
, chain
, level
, period
)) {
1222 eprintf("problem incrementing symbol count, skipping event\n");
1232 process_mmap_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1234 struct thread
*thread
= threads__findnew(event
->mmap
.pid
);
1235 struct map
*map
= map__new(&event
->mmap
);
1237 dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n",
1238 (void *)(offset
+ head
),
1239 (void *)(long)(event
->header
.size
),
1241 (void *)(long)event
->mmap
.start
,
1242 (void *)(long)event
->mmap
.len
,
1243 (void *)(long)event
->mmap
.pgoff
,
1244 event
->mmap
.filename
);
1246 if (thread
== NULL
|| map
== NULL
) {
1247 dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1251 thread__insert_map(thread
, map
);
1258 process_comm_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1260 struct thread
*thread
= threads__findnew(event
->comm
.pid
);
1262 dprintf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
1263 (void *)(offset
+ head
),
1264 (void *)(long)(event
->header
.size
),
1265 event
->comm
.comm
, event
->comm
.pid
);
1267 if (thread
== NULL
||
1268 thread__set_comm(thread
, event
->comm
.comm
)) {
1269 dprintf("problem processing PERF_EVENT_COMM, skipping event.\n");
1278 process_fork_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1280 struct thread
*thread
= threads__findnew(event
->fork
.pid
);
1281 struct thread
*parent
= threads__findnew(event
->fork
.ppid
);
1283 dprintf("%p [%p]: PERF_EVENT_FORK: %d:%d\n",
1284 (void *)(offset
+ head
),
1285 (void *)(long)(event
->header
.size
),
1286 event
->fork
.pid
, event
->fork
.ppid
);
1288 if (!thread
|| !parent
|| thread__fork(thread
, parent
)) {
1289 dprintf("problem processing PERF_EVENT_FORK, skipping event.\n");
1298 process_period_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1300 dprintf("%p [%p]: PERF_EVENT_PERIOD: time:%Ld, id:%Ld: period:%Ld\n",
1301 (void *)(offset
+ head
),
1302 (void *)(long)(event
->header
.size
),
1305 event
->period
.sample_period
);
1311 process_lost_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1313 dprintf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n",
1314 (void *)(offset
+ head
),
1315 (void *)(long)(event
->header
.size
),
1319 total_lost
+= event
->lost
.lost
;
1324 static void trace_event(event_t
*event
)
1326 unsigned char *raw_event
= (void *)event
;
1327 char *color
= PERF_COLOR_BLUE
;
1334 cdprintf("\n. ... raw event: size %d bytes\n", event
->header
.size
);
1336 for (i
= 0; i
< event
->header
.size
; i
++) {
1337 if ((i
& 15) == 0) {
1339 cdprintf(" %04x: ", i
);
1342 cdprintf(" %02x", raw_event
[i
]);
1344 if (((i
& 15) == 15) || i
== event
->header
.size
-1) {
1346 for (j
= 0; j
< 15-(i
& 15); j
++)
1348 for (j
= 0; j
< (i
& 15); j
++) {
1349 if (isprint(raw_event
[i
-15+j
]))
1350 cdprintf("%c", raw_event
[i
-15+j
]);
1361 process_read_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1363 dprintf("%p [%p]: PERF_EVENT_READ: %d %d %Lu\n",
1364 (void *)(offset
+ head
),
1365 (void *)(long)(event
->header
.size
),
1374 process_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1378 switch (event
->header
.type
) {
1379 case PERF_EVENT_SAMPLE
:
1380 return process_sample_event(event
, offset
, head
);
1382 case PERF_EVENT_MMAP
:
1383 return process_mmap_event(event
, offset
, head
);
1385 case PERF_EVENT_COMM
:
1386 return process_comm_event(event
, offset
, head
);
1388 case PERF_EVENT_FORK
:
1389 return process_fork_event(event
, offset
, head
);
1391 case PERF_EVENT_PERIOD
:
1392 return process_period_event(event
, offset
, head
);
1394 case PERF_EVENT_LOST
:
1395 return process_lost_event(event
, offset
, head
);
1397 case PERF_EVENT_READ
:
1398 return process_read_event(event
, offset
, head
);
1401 * We dont process them right now but they are fine:
1404 case PERF_EVENT_THROTTLE
:
1405 case PERF_EVENT_UNTHROTTLE
:
1415 static struct perf_header
*header
;
1417 static u64
perf_header__sample_type(void)
1419 u64 sample_type
= 0;
1422 for (i
= 0; i
< header
->attrs
; i
++) {
1423 struct perf_header_attr
*attr
= header
->attr
[i
];
1426 sample_type
= attr
->attr
.sample_type
;
1427 else if (sample_type
!= attr
->attr
.sample_type
)
1428 die("non matching sample_type");
1434 static int __cmd_report(void)
1436 int ret
, rc
= EXIT_FAILURE
;
1437 unsigned long offset
= 0;
1438 unsigned long head
, shift
;
1444 register_idle_thread();
1446 input
= open(input_name
, O_RDONLY
);
1448 fprintf(stderr
, " failed to open file: %s", input_name
);
1449 if (!strcmp(input_name
, "perf.data"))
1450 fprintf(stderr
, " (try 'perf record' first)");
1451 fprintf(stderr
, "\n");
1455 ret
= fstat(input
, &stat
);
1457 perror("failed to stat file");
1461 if (!stat
.st_size
) {
1462 fprintf(stderr
, "zero-sized file, nothing to do!\n");
1466 header
= perf_header__read(input
);
1467 head
= header
->data_offset
;
1469 sample_type
= perf_header__sample_type();
1471 if (sort__has_parent
&& !(sample_type
& PERF_SAMPLE_CALLCHAIN
)) {
1472 fprintf(stderr
, "selected --sort parent, but no callchain data\n");
1476 if (load_kernel() < 0) {
1477 perror("failed to load kernel symbols");
1478 return EXIT_FAILURE
;
1482 if (getcwd(__cwd
, sizeof(__cwd
)) == NULL
) {
1483 perror("failed to get the current directory");
1484 return EXIT_FAILURE
;
1486 cwdlen
= strlen(cwd
);
1492 shift
= page_size
* (head
/ page_size
);
1497 buf
= (char *)mmap(NULL
, page_size
* mmap_window
, PROT_READ
,
1498 MAP_SHARED
, input
, offset
);
1499 if (buf
== MAP_FAILED
) {
1500 perror("failed to mmap file");
1505 event
= (event_t
*)(buf
+ head
);
1507 size
= event
->header
.size
;
1511 if (head
+ event
->header
.size
>= page_size
* mmap_window
) {
1514 shift
= page_size
* (head
/ page_size
);
1516 ret
= munmap(buf
, page_size
* mmap_window
);
1524 size
= event
->header
.size
;
1526 dprintf("\n%p [%p]: event: %d\n",
1527 (void *)(offset
+ head
),
1528 (void *)(long)event
->header
.size
,
1529 event
->header
.type
);
1531 if (!size
|| process_event(event
, offset
, head
) < 0) {
1533 dprintf("%p [%p]: skipping unknown header type: %d\n",
1534 (void *)(offset
+ head
),
1535 (void *)(long)(event
->header
.size
),
1536 event
->header
.type
);
1541 * assume we lost track of the stream, check alignment, and
1542 * increment a single u64 in the hope to catch on again 'soon'.
1545 if (unlikely(head
& 7))
1553 if (offset
+ head
>= header
->data_offset
+ header
->data_size
)
1556 if (offset
+ head
< stat
.st_size
)
1563 dprintf(" IP events: %10ld\n", total
);
1564 dprintf(" mmap events: %10ld\n", total_mmap
);
1565 dprintf(" comm events: %10ld\n", total_comm
);
1566 dprintf(" fork events: %10ld\n", total_fork
);
1567 dprintf(" lost events: %10ld\n", total_lost
);
1568 dprintf(" unknown events: %10ld\n", total_unknown
);
1574 threads__fprintf(stdout
);
1577 dsos__fprintf(stdout
);
1581 output__fprintf(stdout
, total
);
1586 static const char * const report_usage
[] = {
1587 "perf report [<options>] <command>",
1591 static const struct option options
[] = {
1592 OPT_STRING('i', "input", &input_name
, "file",
1594 OPT_BOOLEAN('v', "verbose", &verbose
,
1595 "be more verbose (show symbol address, etc)"),
1596 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace
,
1597 "dump raw trace in ASCII"),
1598 OPT_STRING('k', "vmlinux", &vmlinux
, "file", "vmlinux pathname"),
1599 OPT_STRING('s', "sort", &sort_order
, "key[,key2...]",
1600 "sort by key(s): pid, comm, dso, symbol, parent"),
1601 OPT_BOOLEAN('P', "full-paths", &full_paths
,
1602 "Don't shorten the pathnames taking into account the cwd"),
1603 OPT_STRING('p', "parent", &parent_pattern
, "regex",
1604 "regex filter to identify parent, see: '--sort parent'"),
1605 OPT_BOOLEAN('x', "exclude-other", &exclude_other
,
1606 "Only display entries with parent-match"),
1610 static void setup_sorting(void)
1612 char *tmp
, *tok
, *str
= strdup(sort_order
);
1614 for (tok
= strtok_r(str
, ", ", &tmp
);
1615 tok
; tok
= strtok_r(NULL
, ", ", &tmp
)) {
1616 if (sort_dimension__add(tok
) < 0) {
1617 error("Unknown --sort key: `%s'", tok
);
1618 usage_with_options(report_usage
, options
);
1625 int cmd_report(int argc
, const char **argv
, const char *prefix
)
1629 page_size
= getpagesize();
1631 argc
= parse_options(argc
, argv
, options
, report_usage
, 0);
1635 if (parent_pattern
!= default_parent_pattern
)
1636 sort_dimension__add("parent");
1641 * Any (unrecognized) arguments left?
1644 usage_with_options(report_usage
, options
);
1648 return __cmd_report();