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 <linux/list.h>
14 #include "util/cache.h"
15 #include <linux/rbtree.h>
16 #include "util/symbol.h"
17 #include "util/string.h"
18 #include "util/callchain.h"
19 #include "util/strlist.h"
22 #include "util/header.h"
24 #include "util/parse-options.h"
25 #include "util/parse-events.h"
31 static char const *input_name
= "perf.data";
32 static char *vmlinux
= NULL
;
34 static char default_sort_order
[] = "comm,dso";
35 static char *sort_order
= default_sort_order
;
36 static char *dso_list_str
, *comm_list_str
, *sym_list_str
;
37 static struct strlist
*dso_list
, *comm_list
, *sym_list
;
40 static int show_mask
= SHOW_KERNEL
| SHOW_USER
| SHOW_HV
;
42 static int dump_trace
= 0;
43 #define dprintf(x...) do { if (dump_trace) printf(x); } while (0)
44 #define cdprintf(x...) do { if (dump_trace) color_fprintf(stdout, color, x); } while (0)
47 #define eprintf(x...) do { if (verbose) fprintf(stderr, x); } while (0)
51 static int full_paths
;
53 static unsigned long page_size
;
54 static unsigned long mmap_window
= 32;
56 static char default_parent_pattern
[] = "^sys_|^do_page_fault";
57 static char *parent_pattern
= default_parent_pattern
;
58 static regex_t parent_regex
;
60 static int exclude_other
= 1;
62 static enum chain_mode callchain_mode
;
63 static double callchain_min_percent
= 0.0;
65 static u64 sample_type
;
68 struct perf_event_header header
;
71 unsigned char __more_data
[];
75 struct perf_event_header header
;
80 char filename
[PATH_MAX
];
84 struct perf_event_header header
;
90 struct perf_event_header header
;
95 struct perf_event_header header
;
102 struct perf_event_header header
;
108 struct perf_event_header header
;
114 typedef union event_union
{
115 struct perf_event_header header
;
117 struct mmap_event mmap
;
118 struct comm_event comm
;
119 struct fork_event fork
;
120 struct period_event period
;
121 struct lost_event lost
;
122 struct read_event read
;
125 static LIST_HEAD(dsos
);
126 static struct dso
*kernel_dso
;
127 static struct dso
*vdso
;
128 static struct dso
*hypervisor_dso
;
130 static void dsos__add(struct dso
*dso
)
132 list_add_tail(&dso
->node
, &dsos
);
135 static struct dso
*dsos__find(const char *name
)
139 list_for_each_entry(pos
, &dsos
, node
)
140 if (strcmp(pos
->name
, name
) == 0)
145 static struct dso
*dsos__findnew(const char *name
)
147 struct dso
*dso
= dsos__find(name
);
153 dso
= dso__new(name
, 0);
157 nr
= dso__load(dso
, NULL
, verbose
);
159 eprintf("Failed to open: %s\n", name
);
163 eprintf("No symbols found in: %s, maybe install a debug package?\n", name
);
174 static void dsos__fprintf(FILE *fp
)
178 list_for_each_entry(pos
, &dsos
, node
)
179 dso__fprintf(pos
, fp
);
182 static struct symbol
*vdso__find_symbol(struct dso
*dso
, u64 ip
)
184 return dso__find_symbol(dso
, ip
);
187 static int load_kernel(void)
191 kernel_dso
= dso__new("[kernel]", 0);
195 err
= dso__load_kernel(kernel_dso
, vmlinux
, NULL
, verbose
, modules
);
197 dso__delete(kernel_dso
);
200 dsos__add(kernel_dso
);
202 vdso
= dso__new("[vdso]", 0);
206 vdso
->find_symbol
= vdso__find_symbol
;
210 hypervisor_dso
= dso__new("[hypervisor]", 0);
213 dsos__add(hypervisor_dso
);
218 static char __cwd
[PATH_MAX
];
219 static char *cwd
= __cwd
;
222 static int strcommon(const char *pathname
)
226 while (pathname
[n
] == cwd
[n
] && n
< cwdlen
)
233 struct list_head node
;
237 u64 (*map_ip
)(struct map
*, u64
);
241 static u64
map__map_ip(struct map
*map
, u64 ip
)
243 return ip
- map
->start
+ map
->pgoff
;
246 static u64
vdso__map_ip(struct map
*map __used
, u64 ip
)
251 static inline int is_anon_memory(const char *filename
)
253 return strcmp(filename
, "//anon") == 0;
256 static struct map
*map__new(struct mmap_event
*event
)
258 struct map
*self
= malloc(sizeof(*self
));
261 const char *filename
= event
->filename
;
262 char newfilename
[PATH_MAX
];
266 int n
= strcommon(filename
);
269 snprintf(newfilename
, sizeof(newfilename
),
270 ".%s", filename
+ n
);
271 filename
= newfilename
;
275 anon
= is_anon_memory(filename
);
278 snprintf(newfilename
, sizeof(newfilename
), "/tmp/perf-%d.map", event
->pid
);
279 filename
= newfilename
;
282 self
->start
= event
->start
;
283 self
->end
= event
->start
+ event
->len
;
284 self
->pgoff
= event
->pgoff
;
286 self
->dso
= dsos__findnew(filename
);
287 if (self
->dso
== NULL
)
290 if (self
->dso
== vdso
|| anon
)
291 self
->map_ip
= vdso__map_ip
;
293 self
->map_ip
= map__map_ip
;
301 static struct map
*map__clone(struct map
*self
)
303 struct map
*map
= malloc(sizeof(*self
));
308 memcpy(map
, self
, sizeof(*self
));
313 static int map__overlap(struct map
*l
, struct map
*r
)
315 if (l
->start
> r
->start
) {
321 if (l
->end
> r
->start
)
327 static size_t map__fprintf(struct map
*self
, FILE *fp
)
329 return fprintf(fp
, " %Lx-%Lx %Lx %s\n",
330 self
->start
, self
->end
, self
->pgoff
, self
->dso
->name
);
335 struct rb_node rb_node
;
336 struct list_head maps
;
341 static struct thread
*thread__new(pid_t pid
)
343 struct thread
*self
= malloc(sizeof(*self
));
347 self
->comm
= malloc(32);
349 snprintf(self
->comm
, 32, ":%d", self
->pid
);
350 INIT_LIST_HEAD(&self
->maps
);
356 static int thread__set_comm(struct thread
*self
, const char *comm
)
360 self
->comm
= strdup(comm
);
361 return self
->comm
? 0 : -ENOMEM
;
364 static size_t thread__fprintf(struct thread
*self
, FILE *fp
)
367 size_t ret
= fprintf(fp
, "Thread %d %s\n", self
->pid
, self
->comm
);
369 list_for_each_entry(pos
, &self
->maps
, node
)
370 ret
+= map__fprintf(pos
, fp
);
376 static struct rb_root threads
;
377 static struct thread
*last_match
;
379 static struct thread
*threads__findnew(pid_t pid
)
381 struct rb_node
**p
= &threads
.rb_node
;
382 struct rb_node
*parent
= NULL
;
386 * Font-end cache - PID lookups come in blocks,
387 * so most of the time we dont have to look up
390 if (last_match
&& last_match
->pid
== pid
)
395 th
= rb_entry(parent
, struct thread
, rb_node
);
397 if (th
->pid
== pid
) {
408 th
= thread__new(pid
);
410 rb_link_node(&th
->rb_node
, parent
, p
);
411 rb_insert_color(&th
->rb_node
, &threads
);
418 static void thread__insert_map(struct thread
*self
, struct map
*map
)
420 struct map
*pos
, *tmp
;
422 list_for_each_entry_safe(pos
, tmp
, &self
->maps
, node
) {
423 if (map__overlap(pos
, map
)) {
425 printf("overlapping maps:\n");
426 map__fprintf(map
, stdout
);
427 map__fprintf(pos
, stdout
);
430 if (map
->start
<= pos
->start
&& map
->end
> pos
->start
)
431 pos
->start
= map
->end
;
433 if (map
->end
>= pos
->end
&& map
->start
< pos
->end
)
434 pos
->end
= map
->start
;
437 printf("after collision:\n");
438 map__fprintf(pos
, stdout
);
441 if (pos
->start
>= pos
->end
) {
442 list_del_init(&pos
->node
);
448 list_add_tail(&map
->node
, &self
->maps
);
451 static int thread__fork(struct thread
*self
, struct thread
*parent
)
457 self
->comm
= strdup(parent
->comm
);
461 list_for_each_entry(map
, &parent
->maps
, node
) {
462 struct map
*new = map__clone(map
);
465 thread__insert_map(self
, new);
471 static struct map
*thread__find_map(struct thread
*self
, u64 ip
)
478 list_for_each_entry(pos
, &self
->maps
, node
)
479 if (ip
>= pos
->start
&& ip
<= pos
->end
)
485 static size_t threads__fprintf(FILE *fp
)
490 for (nd
= rb_first(&threads
); nd
; nd
= rb_next(nd
)) {
491 struct thread
*pos
= rb_entry(nd
, struct thread
, rb_node
);
493 ret
+= thread__fprintf(pos
, fp
);
500 * histogram, sorted on item, collects counts
503 static struct rb_root hist
;
506 struct rb_node rb_node
;
508 struct thread
*thread
;
512 struct symbol
*parent
;
515 struct callchain_node callchain
;
516 struct rb_root sorted_chain
;
522 * configurable sorting bits
526 struct list_head list
;
530 int64_t (*cmp
)(struct hist_entry
*, struct hist_entry
*);
531 int64_t (*collapse
)(struct hist_entry
*, struct hist_entry
*);
532 size_t (*print
)(FILE *fp
, struct hist_entry
*);
535 static int64_t cmp_null(void *l
, void *r
)
548 sort__thread_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
550 return right
->thread
->pid
- left
->thread
->pid
;
554 sort__thread_print(FILE *fp
, struct hist_entry
*self
)
556 return fprintf(fp
, "%16s:%5d", self
->thread
->comm
?: "", self
->thread
->pid
);
559 static struct sort_entry sort_thread
= {
560 .header
= " Command: Pid",
561 .cmp
= sort__thread_cmp
,
562 .print
= sort__thread_print
,
568 sort__comm_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
570 return right
->thread
->pid
- left
->thread
->pid
;
574 sort__comm_collapse(struct hist_entry
*left
, struct hist_entry
*right
)
576 char *comm_l
= left
->thread
->comm
;
577 char *comm_r
= right
->thread
->comm
;
579 if (!comm_l
|| !comm_r
)
580 return cmp_null(comm_l
, comm_r
);
582 return strcmp(comm_l
, comm_r
);
586 sort__comm_print(FILE *fp
, struct hist_entry
*self
)
588 return fprintf(fp
, "%16s", self
->thread
->comm
);
591 static struct sort_entry sort_comm
= {
592 .header
= " Command",
593 .cmp
= sort__comm_cmp
,
594 .collapse
= sort__comm_collapse
,
595 .print
= sort__comm_print
,
601 sort__dso_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
603 struct dso
*dso_l
= left
->dso
;
604 struct dso
*dso_r
= right
->dso
;
606 if (!dso_l
|| !dso_r
)
607 return cmp_null(dso_l
, dso_r
);
609 return strcmp(dso_l
->name
, dso_r
->name
);
613 sort__dso_print(FILE *fp
, struct hist_entry
*self
)
616 return fprintf(fp
, "%-25s", self
->dso
->name
);
618 return fprintf(fp
, "%016llx ", (u64
)self
->ip
);
621 static struct sort_entry sort_dso
= {
622 .header
= "Shared Object ",
623 .cmp
= sort__dso_cmp
,
624 .print
= sort__dso_print
,
630 sort__sym_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
634 if (left
->sym
== right
->sym
)
637 ip_l
= left
->sym
? left
->sym
->start
: left
->ip
;
638 ip_r
= right
->sym
? right
->sym
->start
: right
->ip
;
640 return (int64_t)(ip_r
- ip_l
);
644 sort__sym_print(FILE *fp
, struct hist_entry
*self
)
649 ret
+= fprintf(fp
, "%#018llx ", (u64
)self
->ip
);
652 ret
+= fprintf(fp
, "[%c] %s",
653 self
->dso
== kernel_dso
? 'k' :
654 self
->dso
== hypervisor_dso
? 'h' : '.', self
->sym
->name
);
656 if (self
->sym
->module
)
657 ret
+= fprintf(fp
, "\t[%s]", self
->sym
->module
->name
);
659 ret
+= fprintf(fp
, "%#016llx", (u64
)self
->ip
);
665 static struct sort_entry sort_sym
= {
667 .cmp
= sort__sym_cmp
,
668 .print
= sort__sym_print
,
674 sort__parent_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
676 struct symbol
*sym_l
= left
->parent
;
677 struct symbol
*sym_r
= right
->parent
;
679 if (!sym_l
|| !sym_r
)
680 return cmp_null(sym_l
, sym_r
);
682 return strcmp(sym_l
->name
, sym_r
->name
);
686 sort__parent_print(FILE *fp
, struct hist_entry
*self
)
690 ret
+= fprintf(fp
, "%-20s", self
->parent
? self
->parent
->name
: "[other]");
695 static struct sort_entry sort_parent
= {
696 .header
= "Parent symbol ",
697 .cmp
= sort__parent_cmp
,
698 .print
= sort__parent_print
,
701 static int sort__need_collapse
= 0;
702 static int sort__has_parent
= 0;
704 struct sort_dimension
{
706 struct sort_entry
*entry
;
710 static struct sort_dimension sort_dimensions
[] = {
711 { .name
= "pid", .entry
= &sort_thread
, },
712 { .name
= "comm", .entry
= &sort_comm
, },
713 { .name
= "dso", .entry
= &sort_dso
, },
714 { .name
= "symbol", .entry
= &sort_sym
, },
715 { .name
= "parent", .entry
= &sort_parent
, },
718 static LIST_HEAD(hist_entry__sort_list
);
720 static int sort_dimension__add(char *tok
)
724 for (i
= 0; i
< ARRAY_SIZE(sort_dimensions
); i
++) {
725 struct sort_dimension
*sd
= &sort_dimensions
[i
];
730 if (strncasecmp(tok
, sd
->name
, strlen(tok
)))
733 if (sd
->entry
->collapse
)
734 sort__need_collapse
= 1;
736 if (sd
->entry
== &sort_parent
) {
737 int ret
= regcomp(&parent_regex
, parent_pattern
, REG_EXTENDED
);
741 regerror(ret
, &parent_regex
, err
, sizeof(err
));
742 fprintf(stderr
, "Invalid regex: %s\n%s",
743 parent_pattern
, err
);
746 sort__has_parent
= 1;
749 list_add_tail(&sd
->entry
->list
, &hist_entry__sort_list
);
759 hist_entry__cmp(struct hist_entry
*left
, struct hist_entry
*right
)
761 struct sort_entry
*se
;
764 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
765 cmp
= se
->cmp(left
, right
);
774 hist_entry__collapse(struct hist_entry
*left
, struct hist_entry
*right
)
776 struct sort_entry
*se
;
779 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
780 int64_t (*f
)(struct hist_entry
*, struct hist_entry
*);
782 f
= se
->collapse
?: se
->cmp
;
784 cmp
= f(left
, right
);
792 static size_t ipchain__fprintf_graph_line(FILE *fp
, int depth
, int depth_mask
)
797 ret
+= fprintf(fp
, "%s", " ");
799 for (i
= 0; i
< depth
; i
++)
800 if (depth_mask
& (1 << i
))
801 ret
+= fprintf(fp
, "| ");
803 ret
+= fprintf(fp
, " ");
805 ret
+= fprintf(fp
, "\n");
810 ipchain__fprintf_graph(FILE *fp
, struct callchain_list
*chain
, int depth
,
811 int depth_mask
, int count
, u64 total_samples
,
817 ret
+= fprintf(fp
, "%s", " ");
818 for (i
= 0; i
< depth
; i
++) {
819 if (depth_mask
& (1 << i
))
820 ret
+= fprintf(fp
, "|");
822 ret
+= fprintf(fp
, " ");
823 if (!count
&& i
== depth
- 1) {
826 percent
= hits
* 100.0 / total_samples
;
827 ret
+= percent_color_fprintf(fp
, "--%2.2f%%-- ", percent
);
829 ret
+= fprintf(fp
, "%s", " ");
832 ret
+= fprintf(fp
, "%s\n", chain
->sym
->name
);
834 ret
+= fprintf(fp
, "%p\n", (void *)(long)chain
->ip
);
840 callchain__fprintf_graph(FILE *fp
, struct callchain_node
*self
,
841 u64 total_samples
, int depth
, int depth_mask
)
843 struct rb_node
*node
, *next
;
844 struct callchain_node
*child
;
845 struct callchain_list
*chain
;
846 int new_depth_mask
= depth_mask
;
850 node
= rb_first(&self
->rb_root
);
852 child
= rb_entry(node
, struct callchain_node
, rb_node
);
855 * The depth mask manages the output of pipes that show
856 * the depth. We don't want to keep the pipes of the current
857 * level for the last child of this depth
859 next
= rb_next(node
);
861 new_depth_mask
&= ~(1 << (depth
- 1));
864 * But we keep the older depth mask for the line seperator
865 * to keep the level link until we reach the last child
867 ret
+= ipchain__fprintf_graph_line(fp
, depth
, depth_mask
);
869 list_for_each_entry(chain
, &child
->val
, list
) {
870 if (chain
->ip
>= PERF_CONTEXT_MAX
)
872 ret
+= ipchain__fprintf_graph(fp
, chain
, depth
,
877 ret
+= callchain__fprintf_graph(fp
, child
, total_samples
,
879 new_depth_mask
| (1 << depth
));
887 callchain__fprintf_flat(FILE *fp
, struct callchain_node
*self
,
890 struct callchain_list
*chain
;
896 ret
+= callchain__fprintf_flat(fp
, self
->parent
, total_samples
);
899 list_for_each_entry(chain
, &self
->val
, list
) {
900 if (chain
->ip
>= PERF_CONTEXT_MAX
)
903 ret
+= fprintf(fp
, " %s\n", chain
->sym
->name
);
905 ret
+= fprintf(fp
, " %p\n",
906 (void *)(long)chain
->ip
);
913 hist_entry_callchain__fprintf(FILE *fp
, struct hist_entry
*self
,
916 struct rb_node
*rb_node
;
917 struct callchain_node
*chain
;
920 rb_node
= rb_first(&self
->sorted_chain
);
924 chain
= rb_entry(rb_node
, struct callchain_node
, rb_node
);
925 percent
= chain
->hit
* 100.0 / total_samples
;
926 if (callchain_mode
== FLAT
) {
927 ret
+= percent_color_fprintf(fp
, " %6.2f%%\n",
929 ret
+= callchain__fprintf_flat(fp
, chain
, total_samples
);
930 } else if (callchain_mode
== GRAPH
) {
931 ret
+= callchain__fprintf_graph(fp
, chain
,
932 total_samples
, 1, 1);
934 ret
+= fprintf(fp
, "\n");
935 rb_node
= rb_next(rb_node
);
943 hist_entry__fprintf(FILE *fp
, struct hist_entry
*self
, u64 total_samples
)
945 struct sort_entry
*se
;
948 if (exclude_other
&& !self
->parent
)
952 ret
= percent_color_fprintf(fp
, " %6.2f%%",
953 (self
->count
* 100.0) / total_samples
);
955 ret
= fprintf(fp
, "%12Ld ", self
->count
);
957 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
958 if (exclude_other
&& (se
== &sort_parent
))
962 ret
+= se
->print(fp
, self
);
965 ret
+= fprintf(fp
, "\n");
968 hist_entry_callchain__fprintf(fp
, self
, total_samples
);
977 static struct symbol
*
978 resolve_symbol(struct thread
*thread
, struct map
**mapp
,
979 struct dso
**dsop
, u64
*ipp
)
981 struct dso
*dso
= dsop
? *dsop
: NULL
;
982 struct map
*map
= mapp
? *mapp
: NULL
;
994 map
= thread__find_map(thread
, ip
);
999 ip
= map
->map_ip(map
, ip
);
1004 * If this is outside of all known maps,
1005 * and is a negative address, try to look it
1006 * up in the kernel dso, as it might be a
1007 * vsyscall (which executes in user-mode):
1009 if ((long long)ip
< 0)
1012 dprintf(" ...... dso: %s\n", dso
? dso
->name
: "<not found>");
1013 dprintf(" ...... map: %Lx -> %Lx\n", *ipp
, ip
);
1022 return dso
->find_symbol(dso
, ip
);
1025 static int call__match(struct symbol
*sym
)
1027 if (sym
->name
&& !regexec(&parent_regex
, sym
->name
, 0, NULL
, 0))
1033 static struct symbol
**
1034 resolve_callchain(struct thread
*thread
, struct map
*map __used
,
1035 struct ip_callchain
*chain
, struct hist_entry
*entry
)
1037 u64 context
= PERF_CONTEXT_MAX
;
1038 struct symbol
**syms
= NULL
;
1042 syms
= calloc(chain
->nr
, sizeof(*syms
));
1044 fprintf(stderr
, "Can't allocate memory for symbols\n");
1049 for (i
= 0; i
< chain
->nr
; i
++) {
1050 u64 ip
= chain
->ips
[i
];
1051 struct dso
*dso
= NULL
;
1054 if (ip
>= PERF_CONTEXT_MAX
) {
1060 case PERF_CONTEXT_HV
:
1061 dso
= hypervisor_dso
;
1063 case PERF_CONTEXT_KERNEL
:
1070 sym
= resolve_symbol(thread
, NULL
, &dso
, &ip
);
1073 if (sort__has_parent
&& call__match(sym
) &&
1075 entry
->parent
= sym
;
1086 * collect histogram counts
1090 hist_entry__add(struct thread
*thread
, struct map
*map
, struct dso
*dso
,
1091 struct symbol
*sym
, u64 ip
, struct ip_callchain
*chain
,
1092 char level
, u64 count
)
1094 struct rb_node
**p
= &hist
.rb_node
;
1095 struct rb_node
*parent
= NULL
;
1096 struct hist_entry
*he
;
1097 struct symbol
**syms
= NULL
;
1098 struct hist_entry entry
= {
1107 .sorted_chain
= RB_ROOT
1111 if ((sort__has_parent
|| callchain
) && chain
)
1112 syms
= resolve_callchain(thread
, map
, chain
, &entry
);
1114 while (*p
!= NULL
) {
1116 he
= rb_entry(parent
, struct hist_entry
, rb_node
);
1118 cmp
= hist_entry__cmp(&entry
, he
);
1123 append_chain(&he
->callchain
, chain
, syms
);
1132 p
= &(*p
)->rb_right
;
1135 he
= malloc(sizeof(*he
));
1140 callchain_init(&he
->callchain
);
1141 append_chain(&he
->callchain
, chain
, syms
);
1144 rb_link_node(&he
->rb_node
, parent
, p
);
1145 rb_insert_color(&he
->rb_node
, &hist
);
1150 static void hist_entry__free(struct hist_entry
*he
)
1156 * collapse the histogram
1159 static struct rb_root collapse_hists
;
1161 static void collapse__insert_entry(struct hist_entry
*he
)
1163 struct rb_node
**p
= &collapse_hists
.rb_node
;
1164 struct rb_node
*parent
= NULL
;
1165 struct hist_entry
*iter
;
1168 while (*p
!= NULL
) {
1170 iter
= rb_entry(parent
, struct hist_entry
, rb_node
);
1172 cmp
= hist_entry__collapse(iter
, he
);
1175 iter
->count
+= he
->count
;
1176 hist_entry__free(he
);
1183 p
= &(*p
)->rb_right
;
1186 rb_link_node(&he
->rb_node
, parent
, p
);
1187 rb_insert_color(&he
->rb_node
, &collapse_hists
);
1190 static void collapse__resort(void)
1192 struct rb_node
*next
;
1193 struct hist_entry
*n
;
1195 if (!sort__need_collapse
)
1198 next
= rb_first(&hist
);
1200 n
= rb_entry(next
, struct hist_entry
, rb_node
);
1201 next
= rb_next(&n
->rb_node
);
1203 rb_erase(&n
->rb_node
, &hist
);
1204 collapse__insert_entry(n
);
1209 * reverse the map, sort on count.
1212 static struct rb_root output_hists
;
1214 static void output__insert_entry(struct hist_entry
*he
, u64 min_callchain_hits
)
1216 struct rb_node
**p
= &output_hists
.rb_node
;
1217 struct rb_node
*parent
= NULL
;
1218 struct hist_entry
*iter
;
1221 if (callchain_mode
== FLAT
)
1222 sort_chain_flat(&he
->sorted_chain
, &he
->callchain
,
1223 min_callchain_hits
);
1224 else if (callchain_mode
== GRAPH
)
1225 sort_chain_graph(&he
->sorted_chain
, &he
->callchain
,
1226 min_callchain_hits
);
1229 while (*p
!= NULL
) {
1231 iter
= rb_entry(parent
, struct hist_entry
, rb_node
);
1233 if (he
->count
> iter
->count
)
1236 p
= &(*p
)->rb_right
;
1239 rb_link_node(&he
->rb_node
, parent
, p
);
1240 rb_insert_color(&he
->rb_node
, &output_hists
);
1243 static void output__resort(u64 total_samples
)
1245 struct rb_node
*next
;
1246 struct hist_entry
*n
;
1247 struct rb_root
*tree
= &hist
;
1248 u64 min_callchain_hits
;
1250 min_callchain_hits
= total_samples
* (callchain_min_percent
/ 100);
1252 if (sort__need_collapse
)
1253 tree
= &collapse_hists
;
1255 next
= rb_first(tree
);
1258 n
= rb_entry(next
, struct hist_entry
, rb_node
);
1259 next
= rb_next(&n
->rb_node
);
1261 rb_erase(&n
->rb_node
, tree
);
1262 output__insert_entry(n
, min_callchain_hits
);
1266 static size_t output__fprintf(FILE *fp
, u64 total_samples
)
1268 struct hist_entry
*pos
;
1269 struct sort_entry
*se
;
1275 fprintf(fp
, "# (%Ld samples)\n", (u64
)total_samples
);
1278 fprintf(fp
, "# Overhead");
1279 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
1280 if (exclude_other
&& (se
== &sort_parent
))
1282 fprintf(fp
, " %s", se
->header
);
1286 fprintf(fp
, "# ........");
1287 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
1290 if (exclude_other
&& (se
== &sort_parent
))
1294 for (i
= 0; i
< strlen(se
->header
); i
++)
1301 for (nd
= rb_first(&output_hists
); nd
; nd
= rb_next(nd
)) {
1302 pos
= rb_entry(nd
, struct hist_entry
, rb_node
);
1303 ret
+= hist_entry__fprintf(fp
, pos
, total_samples
);
1306 if (sort_order
== default_sort_order
&&
1307 parent_pattern
== default_parent_pattern
) {
1309 fprintf(fp
, "# (For more details, try: perf report --sort comm,dso,symbol)\n");
1317 static void register_idle_thread(void)
1319 struct thread
*thread
= threads__findnew(0);
1321 if (thread
== NULL
||
1322 thread__set_comm(thread
, "[idle]")) {
1323 fprintf(stderr
, "problem inserting idle task.\n");
1328 static unsigned long total
= 0,
1335 static int validate_chain(struct ip_callchain
*chain
, event_t
*event
)
1337 unsigned int chain_size
;
1339 chain_size
= event
->header
.size
;
1340 chain_size
-= (unsigned long)&event
->ip
.__more_data
- (unsigned long)event
;
1342 if (chain
->nr
*sizeof(u64
) > chain_size
)
1349 process_sample_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1353 struct dso
*dso
= NULL
;
1354 struct thread
*thread
= threads__findnew(event
->ip
.pid
);
1355 u64 ip
= event
->ip
.ip
;
1357 struct map
*map
= NULL
;
1358 void *more_data
= event
->ip
.__more_data
;
1359 struct ip_callchain
*chain
= NULL
;
1362 if (sample_type
& PERF_SAMPLE_PERIOD
) {
1363 period
= *(u64
*)more_data
;
1364 more_data
+= sizeof(u64
);
1367 dprintf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d: %p period: %Ld\n",
1368 (void *)(offset
+ head
),
1369 (void *)(long)(event
->header
.size
),
1375 if (sample_type
& PERF_SAMPLE_CALLCHAIN
) {
1378 chain
= (void *)more_data
;
1380 dprintf("... chain: nr:%Lu\n", chain
->nr
);
1382 if (validate_chain(chain
, event
) < 0) {
1383 eprintf("call-chain problem with event, skipping it.\n");
1388 for (i
= 0; i
< chain
->nr
; i
++)
1389 dprintf("..... %2d: %016Lx\n", i
, chain
->ips
[i
]);
1393 dprintf(" ... thread: %s:%d\n", thread
->comm
, thread
->pid
);
1395 if (thread
== NULL
) {
1396 eprintf("problem processing %d event, skipping it.\n",
1397 event
->header
.type
);
1401 if (comm_list
&& !strlist__has_entry(comm_list
, thread
->comm
))
1404 cpumode
= event
->header
.misc
& PERF_EVENT_MISC_CPUMODE_MASK
;
1406 if (cpumode
== PERF_EVENT_MISC_KERNEL
) {
1412 dprintf(" ...... dso: %s\n", dso
->name
);
1414 } else if (cpumode
== PERF_EVENT_MISC_USER
) {
1423 dso
= hypervisor_dso
;
1425 dprintf(" ...... dso: [hypervisor]\n");
1428 if (show
& show_mask
) {
1429 struct symbol
*sym
= resolve_symbol(thread
, &map
, &dso
, &ip
);
1431 if (dso_list
&& dso
&& dso
->name
&& !strlist__has_entry(dso_list
, dso
->name
))
1434 if (sym_list
&& sym
&& !strlist__has_entry(sym_list
, sym
->name
))
1437 if (hist_entry__add(thread
, map
, dso
, sym
, ip
, chain
, level
, period
)) {
1438 eprintf("problem incrementing symbol count, skipping event\n");
1448 process_mmap_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1450 struct thread
*thread
= threads__findnew(event
->mmap
.pid
);
1451 struct map
*map
= map__new(&event
->mmap
);
1453 dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n",
1454 (void *)(offset
+ head
),
1455 (void *)(long)(event
->header
.size
),
1457 (void *)(long)event
->mmap
.start
,
1458 (void *)(long)event
->mmap
.len
,
1459 (void *)(long)event
->mmap
.pgoff
,
1460 event
->mmap
.filename
);
1462 if (thread
== NULL
|| map
== NULL
) {
1463 dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1467 thread__insert_map(thread
, map
);
1474 process_comm_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1476 struct thread
*thread
= threads__findnew(event
->comm
.pid
);
1478 dprintf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
1479 (void *)(offset
+ head
),
1480 (void *)(long)(event
->header
.size
),
1481 event
->comm
.comm
, event
->comm
.pid
);
1483 if (thread
== NULL
||
1484 thread__set_comm(thread
, event
->comm
.comm
)) {
1485 dprintf("problem processing PERF_EVENT_COMM, skipping event.\n");
1494 process_fork_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1496 struct thread
*thread
= threads__findnew(event
->fork
.pid
);
1497 struct thread
*parent
= threads__findnew(event
->fork
.ppid
);
1499 dprintf("%p [%p]: PERF_EVENT_FORK: %d:%d\n",
1500 (void *)(offset
+ head
),
1501 (void *)(long)(event
->header
.size
),
1502 event
->fork
.pid
, event
->fork
.ppid
);
1504 if (!thread
|| !parent
|| thread__fork(thread
, parent
)) {
1505 dprintf("problem processing PERF_EVENT_FORK, skipping event.\n");
1514 process_period_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1516 dprintf("%p [%p]: PERF_EVENT_PERIOD: time:%Ld, id:%Ld: period:%Ld\n",
1517 (void *)(offset
+ head
),
1518 (void *)(long)(event
->header
.size
),
1521 event
->period
.sample_period
);
1527 process_lost_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1529 dprintf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n",
1530 (void *)(offset
+ head
),
1531 (void *)(long)(event
->header
.size
),
1535 total_lost
+= event
->lost
.lost
;
1540 static void trace_event(event_t
*event
)
1542 unsigned char *raw_event
= (void *)event
;
1543 char *color
= PERF_COLOR_BLUE
;
1550 cdprintf("\n. ... raw event: size %d bytes\n", event
->header
.size
);
1552 for (i
= 0; i
< event
->header
.size
; i
++) {
1553 if ((i
& 15) == 0) {
1555 cdprintf(" %04x: ", i
);
1558 cdprintf(" %02x", raw_event
[i
]);
1560 if (((i
& 15) == 15) || i
== event
->header
.size
-1) {
1562 for (j
= 0; j
< 15-(i
& 15); j
++)
1564 for (j
= 0; j
< (i
& 15); j
++) {
1565 if (isprint(raw_event
[i
-15+j
]))
1566 cdprintf("%c", raw_event
[i
-15+j
]);
1577 process_read_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1579 dprintf("%p [%p]: PERF_EVENT_READ: %d %d %Lu\n",
1580 (void *)(offset
+ head
),
1581 (void *)(long)(event
->header
.size
),
1590 process_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1594 switch (event
->header
.type
) {
1595 case PERF_EVENT_SAMPLE
:
1596 return process_sample_event(event
, offset
, head
);
1598 case PERF_EVENT_MMAP
:
1599 return process_mmap_event(event
, offset
, head
);
1601 case PERF_EVENT_COMM
:
1602 return process_comm_event(event
, offset
, head
);
1604 case PERF_EVENT_FORK
:
1605 return process_fork_event(event
, offset
, head
);
1607 case PERF_EVENT_PERIOD
:
1608 return process_period_event(event
, offset
, head
);
1610 case PERF_EVENT_LOST
:
1611 return process_lost_event(event
, offset
, head
);
1613 case PERF_EVENT_READ
:
1614 return process_read_event(event
, offset
, head
);
1617 * We dont process them right now but they are fine:
1620 case PERF_EVENT_THROTTLE
:
1621 case PERF_EVENT_UNTHROTTLE
:
1631 static struct perf_header
*header
;
1633 static u64
perf_header__sample_type(void)
1635 u64 sample_type
= 0;
1638 for (i
= 0; i
< header
->attrs
; i
++) {
1639 struct perf_header_attr
*attr
= header
->attr
[i
];
1642 sample_type
= attr
->attr
.sample_type
;
1643 else if (sample_type
!= attr
->attr
.sample_type
)
1644 die("non matching sample_type");
1650 static int __cmd_report(void)
1652 int ret
, rc
= EXIT_FAILURE
;
1653 unsigned long offset
= 0;
1654 unsigned long head
, shift
;
1660 register_idle_thread();
1662 input
= open(input_name
, O_RDONLY
);
1664 fprintf(stderr
, " failed to open file: %s", input_name
);
1665 if (!strcmp(input_name
, "perf.data"))
1666 fprintf(stderr
, " (try 'perf record' first)");
1667 fprintf(stderr
, "\n");
1671 ret
= fstat(input
, &stat
);
1673 perror("failed to stat file");
1677 if (!stat
.st_size
) {
1678 fprintf(stderr
, "zero-sized file, nothing to do!\n");
1682 header
= perf_header__read(input
);
1683 head
= header
->data_offset
;
1685 sample_type
= perf_header__sample_type();
1687 if (sort__has_parent
&& !(sample_type
& PERF_SAMPLE_CALLCHAIN
)) {
1688 fprintf(stderr
, "selected --sort parent, but no callchain data\n");
1692 if (load_kernel() < 0) {
1693 perror("failed to load kernel symbols");
1694 return EXIT_FAILURE
;
1698 if (getcwd(__cwd
, sizeof(__cwd
)) == NULL
) {
1699 perror("failed to get the current directory");
1700 return EXIT_FAILURE
;
1702 cwdlen
= strlen(cwd
);
1708 shift
= page_size
* (head
/ page_size
);
1713 buf
= (char *)mmap(NULL
, page_size
* mmap_window
, PROT_READ
,
1714 MAP_SHARED
, input
, offset
);
1715 if (buf
== MAP_FAILED
) {
1716 perror("failed to mmap file");
1721 event
= (event_t
*)(buf
+ head
);
1723 size
= event
->header
.size
;
1727 if (head
+ event
->header
.size
>= page_size
* mmap_window
) {
1730 shift
= page_size
* (head
/ page_size
);
1732 ret
= munmap(buf
, page_size
* mmap_window
);
1740 size
= event
->header
.size
;
1742 dprintf("\n%p [%p]: event: %d\n",
1743 (void *)(offset
+ head
),
1744 (void *)(long)event
->header
.size
,
1745 event
->header
.type
);
1747 if (!size
|| process_event(event
, offset
, head
) < 0) {
1749 dprintf("%p [%p]: skipping unknown header type: %d\n",
1750 (void *)(offset
+ head
),
1751 (void *)(long)(event
->header
.size
),
1752 event
->header
.type
);
1757 * assume we lost track of the stream, check alignment, and
1758 * increment a single u64 in the hope to catch on again 'soon'.
1761 if (unlikely(head
& 7))
1769 if (offset
+ head
>= header
->data_offset
+ header
->data_size
)
1772 if (offset
+ head
< (unsigned long)stat
.st_size
)
1779 dprintf(" IP events: %10ld\n", total
);
1780 dprintf(" mmap events: %10ld\n", total_mmap
);
1781 dprintf(" comm events: %10ld\n", total_comm
);
1782 dprintf(" fork events: %10ld\n", total_fork
);
1783 dprintf(" lost events: %10ld\n", total_lost
);
1784 dprintf(" unknown events: %10ld\n", total_unknown
);
1790 threads__fprintf(stdout
);
1793 dsos__fprintf(stdout
);
1796 output__resort(total
);
1797 output__fprintf(stdout
, total
);
1803 parse_callchain_opt(const struct option
*opt __used
, const char *arg
,
1814 tok
= strtok((char *)arg
, ",");
1818 /* get the output mode */
1819 if (!strncmp(tok
, "graph", strlen(arg
)))
1820 callchain_mode
= GRAPH
;
1822 else if (!strncmp(tok
, "flat", strlen(arg
)))
1823 callchain_mode
= FLAT
;
1827 /* get the min percentage */
1828 tok
= strtok(NULL
, ",");
1832 callchain_min_percent
= strtod(tok
, &endptr
);
1839 static const char * const report_usage
[] = {
1840 "perf report [<options>] <command>",
1844 static const struct option options
[] = {
1845 OPT_STRING('i', "input", &input_name
, "file",
1847 OPT_BOOLEAN('v', "verbose", &verbose
,
1848 "be more verbose (show symbol address, etc)"),
1849 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace
,
1850 "dump raw trace in ASCII"),
1851 OPT_STRING('k', "vmlinux", &vmlinux
, "file", "vmlinux pathname"),
1852 OPT_BOOLEAN('m', "modules", &modules
,
1853 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1854 OPT_STRING('s', "sort", &sort_order
, "key[,key2...]",
1855 "sort by key(s): pid, comm, dso, symbol, parent"),
1856 OPT_BOOLEAN('P', "full-paths", &full_paths
,
1857 "Don't shorten the pathnames taking into account the cwd"),
1858 OPT_STRING('p', "parent", &parent_pattern
, "regex",
1859 "regex filter to identify parent, see: '--sort parent'"),
1860 OPT_BOOLEAN('x', "exclude-other", &exclude_other
,
1861 "Only display entries with parent-match"),
1862 OPT_CALLBACK_DEFAULT('c', "callchain", NULL
, "output_type,min_percent",
1863 "Display callchains using output_type and min percent threshold. "
1864 "Default: flat,0", &parse_callchain_opt
, "flat,100"),
1865 OPT_STRING('d', "dsos", &dso_list_str
, "dso[,dso...]",
1866 "only consider symbols in these dsos"),
1867 OPT_STRING('C', "comms", &comm_list_str
, "comm[,comm...]",
1868 "only consider symbols in these comms"),
1869 OPT_STRING('S', "symbols", &sym_list_str
, "symbol[,symbol...]",
1870 "only consider these symbols"),
1874 static void setup_sorting(void)
1876 char *tmp
, *tok
, *str
= strdup(sort_order
);
1878 for (tok
= strtok_r(str
, ", ", &tmp
);
1879 tok
; tok
= strtok_r(NULL
, ", ", &tmp
)) {
1880 if (sort_dimension__add(tok
) < 0) {
1881 error("Unknown --sort key: `%s'", tok
);
1882 usage_with_options(report_usage
, options
);
1889 static void setup_list(struct strlist
**list
, const char *list_str
,
1890 const char *list_name
)
1893 *list
= strlist__new(true, list_str
);
1895 fprintf(stderr
, "problems parsing %s list\n",
1902 int cmd_report(int argc
, const char **argv
, const char *prefix __used
)
1906 page_size
= getpagesize();
1908 argc
= parse_options(argc
, argv
, options
, report_usage
, 0);
1912 if (parent_pattern
!= default_parent_pattern
)
1913 sort_dimension__add("parent");
1918 * Any (unrecognized) arguments left?
1921 usage_with_options(report_usage
, options
);
1923 setup_list(&dso_list
, dso_list_str
, "dso");
1924 setup_list(&comm_list
, comm_list_str
, "comm");
1925 setup_list(&sym_list
, sym_list_str
, "symbol");
1929 return __cmd_report();