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
,
38 static struct strlist
*dso_list
, *comm_list
, *sym_list
;
39 static char *field_sep
;
42 static int show_mask
= SHOW_KERNEL
| SHOW_USER
| SHOW_HV
;
44 static int dump_trace
= 0;
45 #define dprintf(x...) do { if (dump_trace) printf(x); } while (0)
46 #define cdprintf(x...) do { if (dump_trace) color_fprintf(stdout, color, x); } while (0)
49 #define eprintf(x...) do { if (verbose) fprintf(stderr, x); } while (0)
53 static int full_paths
;
54 static int show_nr_samples
;
56 static unsigned long page_size
;
57 static unsigned long mmap_window
= 32;
59 static char default_parent_pattern
[] = "^sys_|^do_page_fault";
60 static char *parent_pattern
= default_parent_pattern
;
61 static regex_t parent_regex
;
63 static int exclude_other
= 1;
65 static char callchain_default_opt
[] = "fractal,0.5";
70 struct callchain_param callchain_param
= {
71 .mode
= CHAIN_GRAPH_ABS
,
75 static u64 sample_type
;
78 struct perf_event_header header
;
81 unsigned char __more_data
[];
85 struct perf_event_header header
;
90 char filename
[PATH_MAX
];
94 struct perf_event_header header
;
100 struct perf_event_header header
;
105 struct perf_event_header header
;
111 struct perf_event_header header
;
117 typedef union event_union
{
118 struct perf_event_header header
;
120 struct mmap_event mmap
;
121 struct comm_event comm
;
122 struct fork_event fork
;
123 struct lost_event lost
;
124 struct read_event read
;
127 static int repsep_fprintf(FILE *fp
, const char *fmt
, ...)
134 n
= vfprintf(fp
, fmt
, ap
);
137 n
= vasprintf(&bf
, fmt
, ap
);
141 sep
= strchr(sep
, *field_sep
);
154 static LIST_HEAD(dsos
);
155 static struct dso
*kernel_dso
;
156 static struct dso
*vdso
;
157 static struct dso
*hypervisor_dso
;
159 static void dsos__add(struct dso
*dso
)
161 list_add_tail(&dso
->node
, &dsos
);
164 static struct dso
*dsos__find(const char *name
)
168 list_for_each_entry(pos
, &dsos
, node
)
169 if (strcmp(pos
->name
, name
) == 0)
174 static struct dso
*dsos__findnew(const char *name
)
176 struct dso
*dso
= dsos__find(name
);
182 dso
= dso__new(name
, 0);
186 nr
= dso__load(dso
, NULL
, verbose
);
188 eprintf("Failed to open: %s\n", name
);
192 eprintf("No symbols found in: %s, maybe install a debug package?\n", name
);
203 static void dsos__fprintf(FILE *fp
)
207 list_for_each_entry(pos
, &dsos
, node
)
208 dso__fprintf(pos
, fp
);
211 static struct symbol
*vdso__find_symbol(struct dso
*dso
, u64 ip
)
213 return dso__find_symbol(dso
, ip
);
216 static int load_kernel(void)
220 kernel_dso
= dso__new("[kernel]", 0);
224 err
= dso__load_kernel(kernel_dso
, vmlinux
, NULL
, verbose
, modules
);
226 dso__delete(kernel_dso
);
229 dsos__add(kernel_dso
);
231 vdso
= dso__new("[vdso]", 0);
235 vdso
->find_symbol
= vdso__find_symbol
;
239 hypervisor_dso
= dso__new("[hypervisor]", 0);
242 dsos__add(hypervisor_dso
);
247 static char __cwd
[PATH_MAX
];
248 static char *cwd
= __cwd
;
251 static int strcommon(const char *pathname
)
255 while (pathname
[n
] == cwd
[n
] && n
< cwdlen
)
262 struct list_head node
;
266 u64 (*map_ip
)(struct map
*, u64
);
270 static u64
map__map_ip(struct map
*map
, u64 ip
)
272 return ip
- map
->start
+ map
->pgoff
;
275 static u64
vdso__map_ip(struct map
*map __used
, u64 ip
)
280 static inline int is_anon_memory(const char *filename
)
282 return strcmp(filename
, "//anon") == 0;
285 static struct map
*map__new(struct mmap_event
*event
)
287 struct map
*self
= malloc(sizeof(*self
));
290 const char *filename
= event
->filename
;
291 char newfilename
[PATH_MAX
];
295 int n
= strcommon(filename
);
298 snprintf(newfilename
, sizeof(newfilename
),
299 ".%s", filename
+ n
);
300 filename
= newfilename
;
304 anon
= is_anon_memory(filename
);
307 snprintf(newfilename
, sizeof(newfilename
), "/tmp/perf-%d.map", event
->pid
);
308 filename
= newfilename
;
311 self
->start
= event
->start
;
312 self
->end
= event
->start
+ event
->len
;
313 self
->pgoff
= event
->pgoff
;
315 self
->dso
= dsos__findnew(filename
);
316 if (self
->dso
== NULL
)
319 if (self
->dso
== vdso
|| anon
)
320 self
->map_ip
= vdso__map_ip
;
322 self
->map_ip
= map__map_ip
;
330 static struct map
*map__clone(struct map
*self
)
332 struct map
*map
= malloc(sizeof(*self
));
337 memcpy(map
, self
, sizeof(*self
));
342 static int map__overlap(struct map
*l
, struct map
*r
)
344 if (l
->start
> r
->start
) {
350 if (l
->end
> r
->start
)
356 static size_t map__fprintf(struct map
*self
, FILE *fp
)
358 return fprintf(fp
, " %Lx-%Lx %Lx %s\n",
359 self
->start
, self
->end
, self
->pgoff
, self
->dso
->name
);
364 struct rb_node rb_node
;
365 struct list_head maps
;
370 static struct thread
*thread__new(pid_t pid
)
372 struct thread
*self
= malloc(sizeof(*self
));
376 self
->comm
= malloc(32);
378 snprintf(self
->comm
, 32, ":%d", self
->pid
);
379 INIT_LIST_HEAD(&self
->maps
);
385 static unsigned int dsos__col_width
,
389 static int thread__set_comm(struct thread
*self
, const char *comm
)
393 self
->comm
= strdup(comm
);
397 if (!col_width_list_str
&& !field_sep
&&
398 (!comm_list
|| strlist__has_entry(comm_list
, comm
))) {
399 unsigned int slen
= strlen(comm
);
400 if (slen
> comms__col_width
) {
401 comms__col_width
= slen
;
402 threads__col_width
= slen
+ 6;
409 static size_t thread__fprintf(struct thread
*self
, FILE *fp
)
412 size_t ret
= fprintf(fp
, "Thread %d %s\n", self
->pid
, self
->comm
);
414 list_for_each_entry(pos
, &self
->maps
, node
)
415 ret
+= map__fprintf(pos
, fp
);
421 static struct rb_root threads
;
422 static struct thread
*last_match
;
424 static struct thread
*threads__findnew(pid_t pid
)
426 struct rb_node
**p
= &threads
.rb_node
;
427 struct rb_node
*parent
= NULL
;
431 * Font-end cache - PID lookups come in blocks,
432 * so most of the time we dont have to look up
435 if (last_match
&& last_match
->pid
== pid
)
440 th
= rb_entry(parent
, struct thread
, rb_node
);
442 if (th
->pid
== pid
) {
453 th
= thread__new(pid
);
455 rb_link_node(&th
->rb_node
, parent
, p
);
456 rb_insert_color(&th
->rb_node
, &threads
);
463 static void thread__insert_map(struct thread
*self
, struct map
*map
)
465 struct map
*pos
, *tmp
;
467 list_for_each_entry_safe(pos
, tmp
, &self
->maps
, node
) {
468 if (map__overlap(pos
, map
)) {
470 printf("overlapping maps:\n");
471 map__fprintf(map
, stdout
);
472 map__fprintf(pos
, stdout
);
475 if (map
->start
<= pos
->start
&& map
->end
> pos
->start
)
476 pos
->start
= map
->end
;
478 if (map
->end
>= pos
->end
&& map
->start
< pos
->end
)
479 pos
->end
= map
->start
;
482 printf("after collision:\n");
483 map__fprintf(pos
, stdout
);
486 if (pos
->start
>= pos
->end
) {
487 list_del_init(&pos
->node
);
493 list_add_tail(&map
->node
, &self
->maps
);
496 static int thread__fork(struct thread
*self
, struct thread
*parent
)
502 self
->comm
= strdup(parent
->comm
);
506 list_for_each_entry(map
, &parent
->maps
, node
) {
507 struct map
*new = map__clone(map
);
510 thread__insert_map(self
, new);
516 static struct map
*thread__find_map(struct thread
*self
, u64 ip
)
523 list_for_each_entry(pos
, &self
->maps
, node
)
524 if (ip
>= pos
->start
&& ip
<= pos
->end
)
530 static size_t threads__fprintf(FILE *fp
)
535 for (nd
= rb_first(&threads
); nd
; nd
= rb_next(nd
)) {
536 struct thread
*pos
= rb_entry(nd
, struct thread
, rb_node
);
538 ret
+= thread__fprintf(pos
, fp
);
545 * histogram, sorted on item, collects counts
548 static struct rb_root hist
;
551 struct rb_node rb_node
;
553 struct thread
*thread
;
557 struct symbol
*parent
;
560 struct callchain_node callchain
;
561 struct rb_root sorted_chain
;
567 * configurable sorting bits
571 struct list_head list
;
575 int64_t (*cmp
)(struct hist_entry
*, struct hist_entry
*);
576 int64_t (*collapse
)(struct hist_entry
*, struct hist_entry
*);
577 size_t (*print
)(FILE *fp
, struct hist_entry
*, unsigned int width
);
582 static int64_t cmp_null(void *l
, void *r
)
595 sort__thread_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
597 return right
->thread
->pid
- left
->thread
->pid
;
601 sort__thread_print(FILE *fp
, struct hist_entry
*self
, unsigned int width
)
603 return repsep_fprintf(fp
, "%*s:%5d", width
- 6,
604 self
->thread
->comm
?: "", self
->thread
->pid
);
607 static struct sort_entry sort_thread
= {
608 .header
= "Command: Pid",
609 .cmp
= sort__thread_cmp
,
610 .print
= sort__thread_print
,
611 .width
= &threads__col_width
,
617 sort__comm_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
619 return right
->thread
->pid
- left
->thread
->pid
;
623 sort__comm_collapse(struct hist_entry
*left
, struct hist_entry
*right
)
625 char *comm_l
= left
->thread
->comm
;
626 char *comm_r
= right
->thread
->comm
;
628 if (!comm_l
|| !comm_r
)
629 return cmp_null(comm_l
, comm_r
);
631 return strcmp(comm_l
, comm_r
);
635 sort__comm_print(FILE *fp
, struct hist_entry
*self
, unsigned int width
)
637 return repsep_fprintf(fp
, "%*s", width
, self
->thread
->comm
);
640 static struct sort_entry sort_comm
= {
642 .cmp
= sort__comm_cmp
,
643 .collapse
= sort__comm_collapse
,
644 .print
= sort__comm_print
,
645 .width
= &comms__col_width
,
651 sort__dso_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
653 struct dso
*dso_l
= left
->dso
;
654 struct dso
*dso_r
= right
->dso
;
656 if (!dso_l
|| !dso_r
)
657 return cmp_null(dso_l
, dso_r
);
659 return strcmp(dso_l
->name
, dso_r
->name
);
663 sort__dso_print(FILE *fp
, struct hist_entry
*self
, unsigned int width
)
666 return repsep_fprintf(fp
, "%-*s", width
, self
->dso
->name
);
668 return repsep_fprintf(fp
, "%*llx", width
, (u64
)self
->ip
);
671 static struct sort_entry sort_dso
= {
672 .header
= "Shared Object",
673 .cmp
= sort__dso_cmp
,
674 .print
= sort__dso_print
,
675 .width
= &dsos__col_width
,
681 sort__sym_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
685 if (left
->sym
== right
->sym
)
688 ip_l
= left
->sym
? left
->sym
->start
: left
->ip
;
689 ip_r
= right
->sym
? right
->sym
->start
: right
->ip
;
691 return (int64_t)(ip_r
- ip_l
);
695 sort__sym_print(FILE *fp
, struct hist_entry
*self
, unsigned int width __used
)
700 ret
+= repsep_fprintf(fp
, "%#018llx ", (u64
)self
->ip
);
702 ret
+= repsep_fprintf(fp
, "[%c] ", self
->level
);
704 ret
+= repsep_fprintf(fp
, "%s", self
->sym
->name
);
706 if (self
->sym
->module
)
707 ret
+= repsep_fprintf(fp
, "\t[%s]",
708 self
->sym
->module
->name
);
710 ret
+= repsep_fprintf(fp
, "%#016llx", (u64
)self
->ip
);
716 static struct sort_entry sort_sym
= {
718 .cmp
= sort__sym_cmp
,
719 .print
= sort__sym_print
,
725 sort__parent_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
727 struct symbol
*sym_l
= left
->parent
;
728 struct symbol
*sym_r
= right
->parent
;
730 if (!sym_l
|| !sym_r
)
731 return cmp_null(sym_l
, sym_r
);
733 return strcmp(sym_l
->name
, sym_r
->name
);
737 sort__parent_print(FILE *fp
, struct hist_entry
*self
, unsigned int width
)
739 return repsep_fprintf(fp
, "%-*s", width
,
740 self
->parent
? self
->parent
->name
: "[other]");
743 static unsigned int parent_symbol__col_width
;
745 static struct sort_entry sort_parent
= {
746 .header
= "Parent symbol",
747 .cmp
= sort__parent_cmp
,
748 .print
= sort__parent_print
,
749 .width
= &parent_symbol__col_width
,
752 static int sort__need_collapse
= 0;
753 static int sort__has_parent
= 0;
755 struct sort_dimension
{
757 struct sort_entry
*entry
;
761 static struct sort_dimension sort_dimensions
[] = {
762 { .name
= "pid", .entry
= &sort_thread
, },
763 { .name
= "comm", .entry
= &sort_comm
, },
764 { .name
= "dso", .entry
= &sort_dso
, },
765 { .name
= "symbol", .entry
= &sort_sym
, },
766 { .name
= "parent", .entry
= &sort_parent
, },
769 static LIST_HEAD(hist_entry__sort_list
);
771 static int sort_dimension__add(char *tok
)
775 for (i
= 0; i
< ARRAY_SIZE(sort_dimensions
); i
++) {
776 struct sort_dimension
*sd
= &sort_dimensions
[i
];
781 if (strncasecmp(tok
, sd
->name
, strlen(tok
)))
784 if (sd
->entry
->collapse
)
785 sort__need_collapse
= 1;
787 if (sd
->entry
== &sort_parent
) {
788 int ret
= regcomp(&parent_regex
, parent_pattern
, REG_EXTENDED
);
792 regerror(ret
, &parent_regex
, err
, sizeof(err
));
793 fprintf(stderr
, "Invalid regex: %s\n%s",
794 parent_pattern
, err
);
797 sort__has_parent
= 1;
800 list_add_tail(&sd
->entry
->list
, &hist_entry__sort_list
);
810 hist_entry__cmp(struct hist_entry
*left
, struct hist_entry
*right
)
812 struct sort_entry
*se
;
815 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
816 cmp
= se
->cmp(left
, right
);
825 hist_entry__collapse(struct hist_entry
*left
, struct hist_entry
*right
)
827 struct sort_entry
*se
;
830 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
831 int64_t (*f
)(struct hist_entry
*, struct hist_entry
*);
833 f
= se
->collapse
?: se
->cmp
;
835 cmp
= f(left
, right
);
843 static size_t ipchain__fprintf_graph_line(FILE *fp
, int depth
, int depth_mask
)
848 ret
+= fprintf(fp
, "%s", " ");
850 for (i
= 0; i
< depth
; i
++)
851 if (depth_mask
& (1 << i
))
852 ret
+= fprintf(fp
, "| ");
854 ret
+= fprintf(fp
, " ");
856 ret
+= fprintf(fp
, "\n");
861 ipchain__fprintf_graph(FILE *fp
, struct callchain_list
*chain
, int depth
,
862 int depth_mask
, int count
, u64 total_samples
,
868 ret
+= fprintf(fp
, "%s", " ");
869 for (i
= 0; i
< depth
; i
++) {
870 if (depth_mask
& (1 << i
))
871 ret
+= fprintf(fp
, "|");
873 ret
+= fprintf(fp
, " ");
874 if (!count
&& i
== depth
- 1) {
877 percent
= hits
* 100.0 / total_samples
;
878 ret
+= percent_color_fprintf(fp
, "--%2.2f%%-- ", percent
);
880 ret
+= fprintf(fp
, "%s", " ");
883 ret
+= fprintf(fp
, "%s\n", chain
->sym
->name
);
885 ret
+= fprintf(fp
, "%p\n", (void *)(long)chain
->ip
);
891 callchain__fprintf_graph(FILE *fp
, struct callchain_node
*self
,
892 u64 total_samples
, int depth
, int depth_mask
)
894 struct rb_node
*node
, *next
;
895 struct callchain_node
*child
;
896 struct callchain_list
*chain
;
897 int new_depth_mask
= depth_mask
;
902 if (callchain_param
.mode
== CHAIN_GRAPH_REL
)
903 new_total
= self
->cumul_hit
;
905 new_total
= total_samples
;
907 node
= rb_first(&self
->rb_root
);
909 child
= rb_entry(node
, struct callchain_node
, rb_node
);
912 * The depth mask manages the output of pipes that show
913 * the depth. We don't want to keep the pipes of the current
914 * level for the last child of this depth
916 next
= rb_next(node
);
918 new_depth_mask
&= ~(1 << (depth
- 1));
921 * But we keep the older depth mask for the line seperator
922 * to keep the level link until we reach the last child
924 ret
+= ipchain__fprintf_graph_line(fp
, depth
, depth_mask
);
926 list_for_each_entry(chain
, &child
->val
, list
) {
927 if (chain
->ip
>= PERF_CONTEXT_MAX
)
929 ret
+= ipchain__fprintf_graph(fp
, chain
, depth
,
934 ret
+= callchain__fprintf_graph(fp
, child
, new_total
,
936 new_depth_mask
| (1 << depth
));
944 callchain__fprintf_flat(FILE *fp
, struct callchain_node
*self
,
947 struct callchain_list
*chain
;
953 ret
+= callchain__fprintf_flat(fp
, self
->parent
, total_samples
);
956 list_for_each_entry(chain
, &self
->val
, list
) {
957 if (chain
->ip
>= PERF_CONTEXT_MAX
)
960 ret
+= fprintf(fp
, " %s\n", chain
->sym
->name
);
962 ret
+= fprintf(fp
, " %p\n",
963 (void *)(long)chain
->ip
);
970 hist_entry_callchain__fprintf(FILE *fp
, struct hist_entry
*self
,
973 struct rb_node
*rb_node
;
974 struct callchain_node
*chain
;
977 rb_node
= rb_first(&self
->sorted_chain
);
981 chain
= rb_entry(rb_node
, struct callchain_node
, rb_node
);
982 percent
= chain
->hit
* 100.0 / total_samples
;
983 switch (callchain_param
.mode
) {
985 ret
+= percent_color_fprintf(fp
, " %6.2f%%\n",
987 ret
+= callchain__fprintf_flat(fp
, chain
, total_samples
);
989 case CHAIN_GRAPH_ABS
: /* Falldown */
990 case CHAIN_GRAPH_REL
:
991 ret
+= callchain__fprintf_graph(fp
, chain
,
992 total_samples
, 1, 1);
996 ret
+= fprintf(fp
, "\n");
997 rb_node
= rb_next(rb_node
);
1005 hist_entry__fprintf(FILE *fp
, struct hist_entry
*self
, u64 total_samples
)
1007 struct sort_entry
*se
;
1010 if (exclude_other
&& !self
->parent
)
1014 ret
= percent_color_fprintf(fp
,
1015 field_sep
? "%.2f" : " %6.2f%%",
1016 (self
->count
* 100.0) / total_samples
);
1018 ret
= fprintf(fp
, field_sep
? "%lld" : "%12lld ", self
->count
);
1020 if (show_nr_samples
) {
1022 fprintf(fp
, "%c%lld", *field_sep
, self
->count
);
1024 fprintf(fp
, "%11lld", self
->count
);
1027 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
1031 fprintf(fp
, "%s", field_sep
?: " ");
1032 ret
+= se
->print(fp
, self
, se
->width
? *se
->width
: 0);
1035 ret
+= fprintf(fp
, "\n");
1038 hist_entry_callchain__fprintf(fp
, self
, total_samples
);
1047 static void dso__calc_col_width(struct dso
*self
)
1049 if (!col_width_list_str
&& !field_sep
&&
1050 (!dso_list
|| strlist__has_entry(dso_list
, self
->name
))) {
1051 unsigned int slen
= strlen(self
->name
);
1052 if (slen
> dsos__col_width
)
1053 dsos__col_width
= slen
;
1056 self
->slen_calculated
= 1;
1059 static struct symbol
*
1060 resolve_symbol(struct thread
*thread
, struct map
**mapp
,
1061 struct dso
**dsop
, u64
*ipp
)
1063 struct dso
*dso
= dsop
? *dsop
: NULL
;
1064 struct map
*map
= mapp
? *mapp
: NULL
;
1076 map
= thread__find_map(thread
, ip
);
1079 * We have to do this here as we may have a dso
1080 * with no symbol hit that has a name longer than
1081 * the ones with symbols sampled.
1083 if (!sort_dso
.elide
&& !map
->dso
->slen_calculated
)
1084 dso__calc_col_width(map
->dso
);
1089 ip
= map
->map_ip(map
, ip
);
1094 * If this is outside of all known maps,
1095 * and is a negative address, try to look it
1096 * up in the kernel dso, as it might be a
1097 * vsyscall (which executes in user-mode):
1099 if ((long long)ip
< 0)
1102 dprintf(" ...... dso: %s\n", dso
? dso
->name
: "<not found>");
1103 dprintf(" ...... map: %Lx -> %Lx\n", *ipp
, ip
);
1112 return dso
->find_symbol(dso
, ip
);
1115 static int call__match(struct symbol
*sym
)
1117 if (sym
->name
&& !regexec(&parent_regex
, sym
->name
, 0, NULL
, 0))
1123 static struct symbol
**
1124 resolve_callchain(struct thread
*thread
, struct map
*map __used
,
1125 struct ip_callchain
*chain
, struct hist_entry
*entry
)
1127 u64 context
= PERF_CONTEXT_MAX
;
1128 struct symbol
**syms
= NULL
;
1132 syms
= calloc(chain
->nr
, sizeof(*syms
));
1134 fprintf(stderr
, "Can't allocate memory for symbols\n");
1139 for (i
= 0; i
< chain
->nr
; i
++) {
1140 u64 ip
= chain
->ips
[i
];
1141 struct dso
*dso
= NULL
;
1144 if (ip
>= PERF_CONTEXT_MAX
) {
1150 case PERF_CONTEXT_HV
:
1151 dso
= hypervisor_dso
;
1153 case PERF_CONTEXT_KERNEL
:
1160 sym
= resolve_symbol(thread
, NULL
, &dso
, &ip
);
1163 if (sort__has_parent
&& call__match(sym
) &&
1165 entry
->parent
= sym
;
1176 * collect histogram counts
1180 hist_entry__add(struct thread
*thread
, struct map
*map
, struct dso
*dso
,
1181 struct symbol
*sym
, u64 ip
, struct ip_callchain
*chain
,
1182 char level
, u64 count
)
1184 struct rb_node
**p
= &hist
.rb_node
;
1185 struct rb_node
*parent
= NULL
;
1186 struct hist_entry
*he
;
1187 struct symbol
**syms
= NULL
;
1188 struct hist_entry entry
= {
1197 .sorted_chain
= RB_ROOT
1201 if ((sort__has_parent
|| callchain
) && chain
)
1202 syms
= resolve_callchain(thread
, map
, chain
, &entry
);
1204 while (*p
!= NULL
) {
1206 he
= rb_entry(parent
, struct hist_entry
, rb_node
);
1208 cmp
= hist_entry__cmp(&entry
, he
);
1213 append_chain(&he
->callchain
, chain
, syms
);
1222 p
= &(*p
)->rb_right
;
1225 he
= malloc(sizeof(*he
));
1230 callchain_init(&he
->callchain
);
1231 append_chain(&he
->callchain
, chain
, syms
);
1234 rb_link_node(&he
->rb_node
, parent
, p
);
1235 rb_insert_color(&he
->rb_node
, &hist
);
1240 static void hist_entry__free(struct hist_entry
*he
)
1246 * collapse the histogram
1249 static struct rb_root collapse_hists
;
1251 static void collapse__insert_entry(struct hist_entry
*he
)
1253 struct rb_node
**p
= &collapse_hists
.rb_node
;
1254 struct rb_node
*parent
= NULL
;
1255 struct hist_entry
*iter
;
1258 while (*p
!= NULL
) {
1260 iter
= rb_entry(parent
, struct hist_entry
, rb_node
);
1262 cmp
= hist_entry__collapse(iter
, he
);
1265 iter
->count
+= he
->count
;
1266 hist_entry__free(he
);
1273 p
= &(*p
)->rb_right
;
1276 rb_link_node(&he
->rb_node
, parent
, p
);
1277 rb_insert_color(&he
->rb_node
, &collapse_hists
);
1280 static void collapse__resort(void)
1282 struct rb_node
*next
;
1283 struct hist_entry
*n
;
1285 if (!sort__need_collapse
)
1288 next
= rb_first(&hist
);
1290 n
= rb_entry(next
, struct hist_entry
, rb_node
);
1291 next
= rb_next(&n
->rb_node
);
1293 rb_erase(&n
->rb_node
, &hist
);
1294 collapse__insert_entry(n
);
1299 * reverse the map, sort on count.
1302 static struct rb_root output_hists
;
1304 static void output__insert_entry(struct hist_entry
*he
, u64 min_callchain_hits
)
1306 struct rb_node
**p
= &output_hists
.rb_node
;
1307 struct rb_node
*parent
= NULL
;
1308 struct hist_entry
*iter
;
1311 callchain_param
.sort(&he
->sorted_chain
, &he
->callchain
,
1312 min_callchain_hits
, &callchain_param
);
1314 while (*p
!= NULL
) {
1316 iter
= rb_entry(parent
, struct hist_entry
, rb_node
);
1318 if (he
->count
> iter
->count
)
1321 p
= &(*p
)->rb_right
;
1324 rb_link_node(&he
->rb_node
, parent
, p
);
1325 rb_insert_color(&he
->rb_node
, &output_hists
);
1328 static void output__resort(u64 total_samples
)
1330 struct rb_node
*next
;
1331 struct hist_entry
*n
;
1332 struct rb_root
*tree
= &hist
;
1333 u64 min_callchain_hits
;
1335 min_callchain_hits
= total_samples
* (callchain_param
.min_percent
/ 100);
1337 if (sort__need_collapse
)
1338 tree
= &collapse_hists
;
1340 next
= rb_first(tree
);
1343 n
= rb_entry(next
, struct hist_entry
, rb_node
);
1344 next
= rb_next(&n
->rb_node
);
1346 rb_erase(&n
->rb_node
, tree
);
1347 output__insert_entry(n
, min_callchain_hits
);
1351 static size_t output__fprintf(FILE *fp
, u64 total_samples
)
1353 struct hist_entry
*pos
;
1354 struct sort_entry
*se
;
1358 char *col_width
= col_width_list_str
;
1360 fprintf(fp
, "# Samples: %Ld\n", (u64
)total_samples
);
1363 fprintf(fp
, "# Overhead");
1364 if (show_nr_samples
) {
1366 fprintf(fp
, "%cSamples", *field_sep
);
1368 fputs(" Samples ", fp
);
1370 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
1374 fprintf(fp
, "%c%s", *field_sep
, se
->header
);
1377 width
= strlen(se
->header
);
1379 if (col_width_list_str
) {
1381 *se
->width
= atoi(col_width
);
1382 col_width
= strchr(col_width
, ',');
1387 width
= *se
->width
= max(*se
->width
, width
);
1389 fprintf(fp
, " %*s", width
, se
->header
);
1396 fprintf(fp
, "# ........");
1397 if (show_nr_samples
)
1398 fprintf(fp
, " ..........");
1399 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
1409 width
= strlen(se
->header
);
1410 for (i
= 0; i
< width
; i
++)
1418 for (nd
= rb_first(&output_hists
); nd
; nd
= rb_next(nd
)) {
1419 pos
= rb_entry(nd
, struct hist_entry
, rb_node
);
1420 ret
+= hist_entry__fprintf(fp
, pos
, total_samples
);
1423 if (sort_order
== default_sort_order
&&
1424 parent_pattern
== default_parent_pattern
) {
1426 fprintf(fp
, "# (For more details, try: perf report --sort comm,dso,symbol)\n");
1434 static void register_idle_thread(void)
1436 struct thread
*thread
= threads__findnew(0);
1438 if (thread
== NULL
||
1439 thread__set_comm(thread
, "[idle]")) {
1440 fprintf(stderr
, "problem inserting idle task.\n");
1445 static unsigned long total
= 0,
1452 static int validate_chain(struct ip_callchain
*chain
, event_t
*event
)
1454 unsigned int chain_size
;
1456 chain_size
= event
->header
.size
;
1457 chain_size
-= (unsigned long)&event
->ip
.__more_data
- (unsigned long)event
;
1459 if (chain
->nr
*sizeof(u64
) > chain_size
)
1466 process_sample_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1470 struct dso
*dso
= NULL
;
1471 struct thread
*thread
= threads__findnew(event
->ip
.pid
);
1472 u64 ip
= event
->ip
.ip
;
1474 struct map
*map
= NULL
;
1475 void *more_data
= event
->ip
.__more_data
;
1476 struct ip_callchain
*chain
= NULL
;
1479 if (sample_type
& PERF_SAMPLE_PERIOD
) {
1480 period
= *(u64
*)more_data
;
1481 more_data
+= sizeof(u64
);
1484 dprintf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d: %p period: %Ld\n",
1485 (void *)(offset
+ head
),
1486 (void *)(long)(event
->header
.size
),
1492 if (sample_type
& PERF_SAMPLE_CALLCHAIN
) {
1495 chain
= (void *)more_data
;
1497 dprintf("... chain: nr:%Lu\n", chain
->nr
);
1499 if (validate_chain(chain
, event
) < 0) {
1500 eprintf("call-chain problem with event, skipping it.\n");
1505 for (i
= 0; i
< chain
->nr
; i
++)
1506 dprintf("..... %2d: %016Lx\n", i
, chain
->ips
[i
]);
1510 dprintf(" ... thread: %s:%d\n", thread
->comm
, thread
->pid
);
1512 if (thread
== NULL
) {
1513 eprintf("problem processing %d event, skipping it.\n",
1514 event
->header
.type
);
1518 if (comm_list
&& !strlist__has_entry(comm_list
, thread
->comm
))
1521 cpumode
= event
->header
.misc
& PERF_EVENT_MISC_CPUMODE_MASK
;
1523 if (cpumode
== PERF_EVENT_MISC_KERNEL
) {
1529 dprintf(" ...... dso: %s\n", dso
->name
);
1531 } else if (cpumode
== PERF_EVENT_MISC_USER
) {
1540 dso
= hypervisor_dso
;
1542 dprintf(" ...... dso: [hypervisor]\n");
1545 if (show
& show_mask
) {
1546 struct symbol
*sym
= resolve_symbol(thread
, &map
, &dso
, &ip
);
1548 if (dso_list
&& dso
&& dso
->name
&& !strlist__has_entry(dso_list
, dso
->name
))
1551 if (sym_list
&& sym
&& !strlist__has_entry(sym_list
, sym
->name
))
1554 if (hist_entry__add(thread
, map
, dso
, sym
, ip
, chain
, level
, period
)) {
1555 eprintf("problem incrementing symbol count, skipping event\n");
1565 process_mmap_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1567 struct thread
*thread
= threads__findnew(event
->mmap
.pid
);
1568 struct map
*map
= map__new(&event
->mmap
);
1570 dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n",
1571 (void *)(offset
+ head
),
1572 (void *)(long)(event
->header
.size
),
1574 (void *)(long)event
->mmap
.start
,
1575 (void *)(long)event
->mmap
.len
,
1576 (void *)(long)event
->mmap
.pgoff
,
1577 event
->mmap
.filename
);
1579 if (thread
== NULL
|| map
== NULL
) {
1580 dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1584 thread__insert_map(thread
, map
);
1591 process_comm_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1593 struct thread
*thread
= threads__findnew(event
->comm
.pid
);
1595 dprintf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
1596 (void *)(offset
+ head
),
1597 (void *)(long)(event
->header
.size
),
1598 event
->comm
.comm
, event
->comm
.pid
);
1600 if (thread
== NULL
||
1601 thread__set_comm(thread
, event
->comm
.comm
)) {
1602 dprintf("problem processing PERF_EVENT_COMM, skipping event.\n");
1611 process_fork_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1613 struct thread
*thread
= threads__findnew(event
->fork
.pid
);
1614 struct thread
*parent
= threads__findnew(event
->fork
.ppid
);
1616 dprintf("%p [%p]: PERF_EVENT_FORK: %d:%d\n",
1617 (void *)(offset
+ head
),
1618 (void *)(long)(event
->header
.size
),
1619 event
->fork
.pid
, event
->fork
.ppid
);
1621 if (!thread
|| !parent
|| thread__fork(thread
, parent
)) {
1622 dprintf("problem processing PERF_EVENT_FORK, skipping event.\n");
1631 process_lost_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1633 dprintf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n",
1634 (void *)(offset
+ head
),
1635 (void *)(long)(event
->header
.size
),
1639 total_lost
+= event
->lost
.lost
;
1644 static void trace_event(event_t
*event
)
1646 unsigned char *raw_event
= (void *)event
;
1647 char *color
= PERF_COLOR_BLUE
;
1654 cdprintf("\n. ... raw event: size %d bytes\n", event
->header
.size
);
1656 for (i
= 0; i
< event
->header
.size
; i
++) {
1657 if ((i
& 15) == 0) {
1659 cdprintf(" %04x: ", i
);
1662 cdprintf(" %02x", raw_event
[i
]);
1664 if (((i
& 15) == 15) || i
== event
->header
.size
-1) {
1666 for (j
= 0; j
< 15-(i
& 15); j
++)
1668 for (j
= 0; j
< (i
& 15); j
++) {
1669 if (isprint(raw_event
[i
-15+j
]))
1670 cdprintf("%c", raw_event
[i
-15+j
]);
1681 process_read_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1683 dprintf("%p [%p]: PERF_EVENT_READ: %d %d %Lu\n",
1684 (void *)(offset
+ head
),
1685 (void *)(long)(event
->header
.size
),
1694 process_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1698 switch (event
->header
.type
) {
1699 case PERF_EVENT_SAMPLE
:
1700 return process_sample_event(event
, offset
, head
);
1702 case PERF_EVENT_MMAP
:
1703 return process_mmap_event(event
, offset
, head
);
1705 case PERF_EVENT_COMM
:
1706 return process_comm_event(event
, offset
, head
);
1708 case PERF_EVENT_FORK
:
1709 return process_fork_event(event
, offset
, head
);
1711 case PERF_EVENT_LOST
:
1712 return process_lost_event(event
, offset
, head
);
1714 case PERF_EVENT_READ
:
1715 return process_read_event(event
, offset
, head
);
1718 * We dont process them right now but they are fine:
1721 case PERF_EVENT_THROTTLE
:
1722 case PERF_EVENT_UNTHROTTLE
:
1732 static struct perf_header
*header
;
1734 static u64
perf_header__sample_type(void)
1736 u64 sample_type
= 0;
1739 for (i
= 0; i
< header
->attrs
; i
++) {
1740 struct perf_header_attr
*attr
= header
->attr
[i
];
1743 sample_type
= attr
->attr
.sample_type
;
1744 else if (sample_type
!= attr
->attr
.sample_type
)
1745 die("non matching sample_type");
1751 static int __cmd_report(void)
1753 int ret
, rc
= EXIT_FAILURE
;
1754 unsigned long offset
= 0;
1755 unsigned long head
, shift
;
1761 register_idle_thread();
1763 input
= open(input_name
, O_RDONLY
);
1765 fprintf(stderr
, " failed to open file: %s", input_name
);
1766 if (!strcmp(input_name
, "perf.data"))
1767 fprintf(stderr
, " (try 'perf record' first)");
1768 fprintf(stderr
, "\n");
1772 ret
= fstat(input
, &stat
);
1774 perror("failed to stat file");
1778 if (!stat
.st_size
) {
1779 fprintf(stderr
, "zero-sized file, nothing to do!\n");
1783 header
= perf_header__read(input
);
1784 head
= header
->data_offset
;
1786 sample_type
= perf_header__sample_type();
1788 if (!(sample_type
& PERF_SAMPLE_CALLCHAIN
)) {
1789 if (sort__has_parent
) {
1790 fprintf(stderr
, "selected --sort parent, but no"
1791 " callchain data. Did you call"
1792 " perf record without -g?\n");
1796 fprintf(stderr
, "selected -c but no callchain data."
1797 " Did you call perf record without"
1803 if (load_kernel() < 0) {
1804 perror("failed to load kernel symbols");
1805 return EXIT_FAILURE
;
1809 if (getcwd(__cwd
, sizeof(__cwd
)) == NULL
) {
1810 perror("failed to get the current directory");
1811 return EXIT_FAILURE
;
1813 cwdlen
= strlen(cwd
);
1819 shift
= page_size
* (head
/ page_size
);
1824 buf
= (char *)mmap(NULL
, page_size
* mmap_window
, PROT_READ
,
1825 MAP_SHARED
, input
, offset
);
1826 if (buf
== MAP_FAILED
) {
1827 perror("failed to mmap file");
1832 event
= (event_t
*)(buf
+ head
);
1834 size
= event
->header
.size
;
1838 if (head
+ event
->header
.size
>= page_size
* mmap_window
) {
1841 shift
= page_size
* (head
/ page_size
);
1843 ret
= munmap(buf
, page_size
* mmap_window
);
1851 size
= event
->header
.size
;
1853 dprintf("\n%p [%p]: event: %d\n",
1854 (void *)(offset
+ head
),
1855 (void *)(long)event
->header
.size
,
1856 event
->header
.type
);
1858 if (!size
|| process_event(event
, offset
, head
) < 0) {
1860 dprintf("%p [%p]: skipping unknown header type: %d\n",
1861 (void *)(offset
+ head
),
1862 (void *)(long)(event
->header
.size
),
1863 event
->header
.type
);
1868 * assume we lost track of the stream, check alignment, and
1869 * increment a single u64 in the hope to catch on again 'soon'.
1872 if (unlikely(head
& 7))
1880 if (offset
+ head
>= header
->data_offset
+ header
->data_size
)
1883 if (offset
+ head
< (unsigned long)stat
.st_size
)
1890 dprintf(" IP events: %10ld\n", total
);
1891 dprintf(" mmap events: %10ld\n", total_mmap
);
1892 dprintf(" comm events: %10ld\n", total_comm
);
1893 dprintf(" fork events: %10ld\n", total_fork
);
1894 dprintf(" lost events: %10ld\n", total_lost
);
1895 dprintf(" unknown events: %10ld\n", total_unknown
);
1901 threads__fprintf(stdout
);
1904 dsos__fprintf(stdout
);
1907 output__resort(total
);
1908 output__fprintf(stdout
, total
);
1914 parse_callchain_opt(const struct option
*opt __used
, const char *arg
,
1925 tok
= strtok((char *)arg
, ",");
1929 /* get the output mode */
1930 if (!strncmp(tok
, "graph", strlen(arg
)))
1931 callchain_param
.mode
= CHAIN_GRAPH_ABS
;
1933 else if (!strncmp(tok
, "flat", strlen(arg
)))
1934 callchain_param
.mode
= CHAIN_FLAT
;
1936 else if (!strncmp(tok
, "fractal", strlen(arg
)))
1937 callchain_param
.mode
= CHAIN_GRAPH_REL
;
1942 /* get the min percentage */
1943 tok
= strtok(NULL
, ",");
1947 callchain_param
.min_percent
= strtod(tok
, &endptr
);
1952 if (register_callchain_param(&callchain_param
) < 0) {
1953 fprintf(stderr
, "Can't register callchain params\n");
1959 static const char * const report_usage
[] = {
1960 "perf report [<options>] <command>",
1964 static const struct option options
[] = {
1965 OPT_STRING('i', "input", &input_name
, "file",
1967 OPT_BOOLEAN('v', "verbose", &verbose
,
1968 "be more verbose (show symbol address, etc)"),
1969 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace
,
1970 "dump raw trace in ASCII"),
1971 OPT_STRING('k', "vmlinux", &vmlinux
, "file", "vmlinux pathname"),
1972 OPT_BOOLEAN('m', "modules", &modules
,
1973 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1974 OPT_BOOLEAN('n', "show-nr-samples", &show_nr_samples
,
1975 "Show a column with the number of samples"),
1976 OPT_STRING('s', "sort", &sort_order
, "key[,key2...]",
1977 "sort by key(s): pid, comm, dso, symbol, parent"),
1978 OPT_BOOLEAN('P', "full-paths", &full_paths
,
1979 "Don't shorten the pathnames taking into account the cwd"),
1980 OPT_STRING('p', "parent", &parent_pattern
, "regex",
1981 "regex filter to identify parent, see: '--sort parent'"),
1982 OPT_BOOLEAN('x', "exclude-other", &exclude_other
,
1983 "Only display entries with parent-match"),
1984 OPT_CALLBACK_DEFAULT('g', "call-graph", NULL
, "output_type,min_percent",
1985 "Display callchains using output_type and min percent threshold. "
1986 "Default: fractal,0.5", &parse_callchain_opt
, callchain_default_opt
),
1987 OPT_STRING('d', "dsos", &dso_list_str
, "dso[,dso...]",
1988 "only consider symbols in these dsos"),
1989 OPT_STRING('C', "comms", &comm_list_str
, "comm[,comm...]",
1990 "only consider symbols in these comms"),
1991 OPT_STRING('S', "symbols", &sym_list_str
, "symbol[,symbol...]",
1992 "only consider these symbols"),
1993 OPT_STRING('w', "column-widths", &col_width_list_str
,
1995 "don't try to adjust column width, use these fixed values"),
1996 OPT_STRING('t', "field-separator", &field_sep
, "separator",
1997 "separator for columns, no spaces will be added between "
1998 "columns '.' is reserved."),
2002 static void setup_sorting(void)
2004 char *tmp
, *tok
, *str
= strdup(sort_order
);
2006 for (tok
= strtok_r(str
, ", ", &tmp
);
2007 tok
; tok
= strtok_r(NULL
, ", ", &tmp
)) {
2008 if (sort_dimension__add(tok
) < 0) {
2009 error("Unknown --sort key: `%s'", tok
);
2010 usage_with_options(report_usage
, options
);
2017 static void setup_list(struct strlist
**list
, const char *list_str
,
2018 struct sort_entry
*se
, const char *list_name
,
2022 *list
= strlist__new(true, list_str
);
2024 fprintf(stderr
, "problems parsing %s list\n",
2028 if (strlist__nr_entries(*list
) == 1) {
2029 fprintf(fp
, "# %s: %s\n", list_name
,
2030 strlist__entry(*list
, 0)->s
);
2036 int cmd_report(int argc
, const char **argv
, const char *prefix __used
)
2040 page_size
= getpagesize();
2042 argc
= parse_options(argc
, argv
, options
, report_usage
, 0);
2046 if (parent_pattern
!= default_parent_pattern
) {
2047 sort_dimension__add("parent");
2048 sort_parent
.elide
= 1;
2053 * Any (unrecognized) arguments left?
2056 usage_with_options(report_usage
, options
);
2060 setup_list(&dso_list
, dso_list_str
, &sort_dso
, "dso", stdout
);
2061 setup_list(&comm_list
, comm_list_str
, &sort_comm
, "comm", stdout
);
2062 setup_list(&sym_list
, sym_list_str
, &sort_sym
, "symbol", stdout
);
2064 if (field_sep
&& *field_sep
== '.') {
2065 fputs("'.' is the only non valid --field-separator argument\n",
2070 return __cmd_report();