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
;
104 struct period_event
{
105 struct perf_event_header header
;
112 struct perf_event_header header
;
118 struct perf_event_header header
;
124 typedef union event_union
{
125 struct perf_event_header header
;
127 struct mmap_event mmap
;
128 struct comm_event comm
;
129 struct fork_event fork
;
130 struct period_event period
;
131 struct lost_event lost
;
132 struct read_event read
;
135 static int repsep_fprintf(FILE *fp
, const char *fmt
, ...)
142 n
= vfprintf(fp
, fmt
, ap
);
145 n
= vasprintf(&bf
, fmt
, ap
);
149 sep
= strchr(sep
, *field_sep
);
162 static LIST_HEAD(dsos
);
163 static struct dso
*kernel_dso
;
164 static struct dso
*vdso
;
165 static struct dso
*hypervisor_dso
;
167 static void dsos__add(struct dso
*dso
)
169 list_add_tail(&dso
->node
, &dsos
);
172 static struct dso
*dsos__find(const char *name
)
176 list_for_each_entry(pos
, &dsos
, node
)
177 if (strcmp(pos
->name
, name
) == 0)
182 static struct dso
*dsos__findnew(const char *name
)
184 struct dso
*dso
= dsos__find(name
);
190 dso
= dso__new(name
, 0);
194 nr
= dso__load(dso
, NULL
, verbose
);
196 eprintf("Failed to open: %s\n", name
);
200 eprintf("No symbols found in: %s, maybe install a debug package?\n", name
);
211 static void dsos__fprintf(FILE *fp
)
215 list_for_each_entry(pos
, &dsos
, node
)
216 dso__fprintf(pos
, fp
);
219 static struct symbol
*vdso__find_symbol(struct dso
*dso
, u64 ip
)
221 return dso__find_symbol(dso
, ip
);
224 static int load_kernel(void)
228 kernel_dso
= dso__new("[kernel]", 0);
232 err
= dso__load_kernel(kernel_dso
, vmlinux
, NULL
, verbose
, modules
);
234 dso__delete(kernel_dso
);
237 dsos__add(kernel_dso
);
239 vdso
= dso__new("[vdso]", 0);
243 vdso
->find_symbol
= vdso__find_symbol
;
247 hypervisor_dso
= dso__new("[hypervisor]", 0);
250 dsos__add(hypervisor_dso
);
255 static char __cwd
[PATH_MAX
];
256 static char *cwd
= __cwd
;
259 static int strcommon(const char *pathname
)
263 while (pathname
[n
] == cwd
[n
] && n
< cwdlen
)
270 struct list_head node
;
274 u64 (*map_ip
)(struct map
*, u64
);
278 static u64
map__map_ip(struct map
*map
, u64 ip
)
280 return ip
- map
->start
+ map
->pgoff
;
283 static u64
vdso__map_ip(struct map
*map __used
, u64 ip
)
288 static inline int is_anon_memory(const char *filename
)
290 return strcmp(filename
, "//anon") == 0;
293 static struct map
*map__new(struct mmap_event
*event
)
295 struct map
*self
= malloc(sizeof(*self
));
298 const char *filename
= event
->filename
;
299 char newfilename
[PATH_MAX
];
303 int n
= strcommon(filename
);
306 snprintf(newfilename
, sizeof(newfilename
),
307 ".%s", filename
+ n
);
308 filename
= newfilename
;
312 anon
= is_anon_memory(filename
);
315 snprintf(newfilename
, sizeof(newfilename
), "/tmp/perf-%d.map", event
->pid
);
316 filename
= newfilename
;
319 self
->start
= event
->start
;
320 self
->end
= event
->start
+ event
->len
;
321 self
->pgoff
= event
->pgoff
;
323 self
->dso
= dsos__findnew(filename
);
324 if (self
->dso
== NULL
)
327 if (self
->dso
== vdso
|| anon
)
328 self
->map_ip
= vdso__map_ip
;
330 self
->map_ip
= map__map_ip
;
338 static struct map
*map__clone(struct map
*self
)
340 struct map
*map
= malloc(sizeof(*self
));
345 memcpy(map
, self
, sizeof(*self
));
350 static int map__overlap(struct map
*l
, struct map
*r
)
352 if (l
->start
> r
->start
) {
358 if (l
->end
> r
->start
)
364 static size_t map__fprintf(struct map
*self
, FILE *fp
)
366 return fprintf(fp
, " %Lx-%Lx %Lx %s\n",
367 self
->start
, self
->end
, self
->pgoff
, self
->dso
->name
);
372 struct rb_node rb_node
;
373 struct list_head maps
;
378 static struct thread
*thread__new(pid_t pid
)
380 struct thread
*self
= malloc(sizeof(*self
));
384 self
->comm
= malloc(32);
386 snprintf(self
->comm
, 32, ":%d", self
->pid
);
387 INIT_LIST_HEAD(&self
->maps
);
393 static unsigned int dsos__col_width
,
397 static int thread__set_comm(struct thread
*self
, const char *comm
)
401 self
->comm
= strdup(comm
);
405 if (!col_width_list_str
&& !field_sep
&&
406 (!comm_list
|| strlist__has_entry(comm_list
, comm
))) {
407 unsigned int slen
= strlen(comm
);
408 if (slen
> comms__col_width
) {
409 comms__col_width
= slen
;
410 threads__col_width
= slen
+ 6;
417 static size_t thread__fprintf(struct thread
*self
, FILE *fp
)
420 size_t ret
= fprintf(fp
, "Thread %d %s\n", self
->pid
, self
->comm
);
422 list_for_each_entry(pos
, &self
->maps
, node
)
423 ret
+= map__fprintf(pos
, fp
);
429 static struct rb_root threads
;
430 static struct thread
*last_match
;
432 static struct thread
*threads__findnew(pid_t pid
)
434 struct rb_node
**p
= &threads
.rb_node
;
435 struct rb_node
*parent
= NULL
;
439 * Font-end cache - PID lookups come in blocks,
440 * so most of the time we dont have to look up
443 if (last_match
&& last_match
->pid
== pid
)
448 th
= rb_entry(parent
, struct thread
, rb_node
);
450 if (th
->pid
== pid
) {
461 th
= thread__new(pid
);
463 rb_link_node(&th
->rb_node
, parent
, p
);
464 rb_insert_color(&th
->rb_node
, &threads
);
471 static void thread__insert_map(struct thread
*self
, struct map
*map
)
473 struct map
*pos
, *tmp
;
475 list_for_each_entry_safe(pos
, tmp
, &self
->maps
, node
) {
476 if (map__overlap(pos
, map
)) {
478 printf("overlapping maps:\n");
479 map__fprintf(map
, stdout
);
480 map__fprintf(pos
, stdout
);
483 if (map
->start
<= pos
->start
&& map
->end
> pos
->start
)
484 pos
->start
= map
->end
;
486 if (map
->end
>= pos
->end
&& map
->start
< pos
->end
)
487 pos
->end
= map
->start
;
490 printf("after collision:\n");
491 map__fprintf(pos
, stdout
);
494 if (pos
->start
>= pos
->end
) {
495 list_del_init(&pos
->node
);
501 list_add_tail(&map
->node
, &self
->maps
);
504 static int thread__fork(struct thread
*self
, struct thread
*parent
)
510 self
->comm
= strdup(parent
->comm
);
514 list_for_each_entry(map
, &parent
->maps
, node
) {
515 struct map
*new = map__clone(map
);
518 thread__insert_map(self
, new);
524 static struct map
*thread__find_map(struct thread
*self
, u64 ip
)
531 list_for_each_entry(pos
, &self
->maps
, node
)
532 if (ip
>= pos
->start
&& ip
<= pos
->end
)
538 static size_t threads__fprintf(FILE *fp
)
543 for (nd
= rb_first(&threads
); nd
; nd
= rb_next(nd
)) {
544 struct thread
*pos
= rb_entry(nd
, struct thread
, rb_node
);
546 ret
+= thread__fprintf(pos
, fp
);
553 * histogram, sorted on item, collects counts
556 static struct rb_root hist
;
559 struct rb_node rb_node
;
561 struct thread
*thread
;
565 struct symbol
*parent
;
568 struct callchain_node callchain
;
569 struct rb_root sorted_chain
;
575 * configurable sorting bits
579 struct list_head list
;
583 int64_t (*cmp
)(struct hist_entry
*, struct hist_entry
*);
584 int64_t (*collapse
)(struct hist_entry
*, struct hist_entry
*);
585 size_t (*print
)(FILE *fp
, struct hist_entry
*, unsigned int width
);
590 static int64_t cmp_null(void *l
, void *r
)
603 sort__thread_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
605 return right
->thread
->pid
- left
->thread
->pid
;
609 sort__thread_print(FILE *fp
, struct hist_entry
*self
, unsigned int width
)
611 return repsep_fprintf(fp
, "%*s:%5d", width
- 6,
612 self
->thread
->comm
?: "", self
->thread
->pid
);
615 static struct sort_entry sort_thread
= {
616 .header
= "Command: Pid",
617 .cmp
= sort__thread_cmp
,
618 .print
= sort__thread_print
,
619 .width
= &threads__col_width
,
625 sort__comm_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
627 return right
->thread
->pid
- left
->thread
->pid
;
631 sort__comm_collapse(struct hist_entry
*left
, struct hist_entry
*right
)
633 char *comm_l
= left
->thread
->comm
;
634 char *comm_r
= right
->thread
->comm
;
636 if (!comm_l
|| !comm_r
)
637 return cmp_null(comm_l
, comm_r
);
639 return strcmp(comm_l
, comm_r
);
643 sort__comm_print(FILE *fp
, struct hist_entry
*self
, unsigned int width
)
645 return repsep_fprintf(fp
, "%*s", width
, self
->thread
->comm
);
648 static struct sort_entry sort_comm
= {
650 .cmp
= sort__comm_cmp
,
651 .collapse
= sort__comm_collapse
,
652 .print
= sort__comm_print
,
653 .width
= &comms__col_width
,
659 sort__dso_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
661 struct dso
*dso_l
= left
->dso
;
662 struct dso
*dso_r
= right
->dso
;
664 if (!dso_l
|| !dso_r
)
665 return cmp_null(dso_l
, dso_r
);
667 return strcmp(dso_l
->name
, dso_r
->name
);
671 sort__dso_print(FILE *fp
, struct hist_entry
*self
, unsigned int width
)
674 return repsep_fprintf(fp
, "%-*s", width
, self
->dso
->name
);
676 return repsep_fprintf(fp
, "%*llx", width
, (u64
)self
->ip
);
679 static struct sort_entry sort_dso
= {
680 .header
= "Shared Object",
681 .cmp
= sort__dso_cmp
,
682 .print
= sort__dso_print
,
683 .width
= &dsos__col_width
,
689 sort__sym_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
693 if (left
->sym
== right
->sym
)
696 ip_l
= left
->sym
? left
->sym
->start
: left
->ip
;
697 ip_r
= right
->sym
? right
->sym
->start
: right
->ip
;
699 return (int64_t)(ip_r
- ip_l
);
703 sort__sym_print(FILE *fp
, struct hist_entry
*self
, unsigned int width __used
)
708 ret
+= repsep_fprintf(fp
, "%#018llx ", (u64
)self
->ip
);
710 ret
+= repsep_fprintf(fp
, "[%c] ", self
->level
);
712 ret
+= repsep_fprintf(fp
, "%s", self
->sym
->name
);
714 if (self
->sym
->module
)
715 ret
+= repsep_fprintf(fp
, "\t[%s]",
716 self
->sym
->module
->name
);
718 ret
+= repsep_fprintf(fp
, "%#016llx", (u64
)self
->ip
);
724 static struct sort_entry sort_sym
= {
726 .cmp
= sort__sym_cmp
,
727 .print
= sort__sym_print
,
733 sort__parent_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
735 struct symbol
*sym_l
= left
->parent
;
736 struct symbol
*sym_r
= right
->parent
;
738 if (!sym_l
|| !sym_r
)
739 return cmp_null(sym_l
, sym_r
);
741 return strcmp(sym_l
->name
, sym_r
->name
);
745 sort__parent_print(FILE *fp
, struct hist_entry
*self
, unsigned int width
)
747 return repsep_fprintf(fp
, "%-*s", width
,
748 self
->parent
? self
->parent
->name
: "[other]");
751 static unsigned int parent_symbol__col_width
;
753 static struct sort_entry sort_parent
= {
754 .header
= "Parent symbol",
755 .cmp
= sort__parent_cmp
,
756 .print
= sort__parent_print
,
757 .width
= &parent_symbol__col_width
,
760 static int sort__need_collapse
= 0;
761 static int sort__has_parent
= 0;
763 struct sort_dimension
{
765 struct sort_entry
*entry
;
769 static struct sort_dimension sort_dimensions
[] = {
770 { .name
= "pid", .entry
= &sort_thread
, },
771 { .name
= "comm", .entry
= &sort_comm
, },
772 { .name
= "dso", .entry
= &sort_dso
, },
773 { .name
= "symbol", .entry
= &sort_sym
, },
774 { .name
= "parent", .entry
= &sort_parent
, },
777 static LIST_HEAD(hist_entry__sort_list
);
779 static int sort_dimension__add(char *tok
)
783 for (i
= 0; i
< ARRAY_SIZE(sort_dimensions
); i
++) {
784 struct sort_dimension
*sd
= &sort_dimensions
[i
];
789 if (strncasecmp(tok
, sd
->name
, strlen(tok
)))
792 if (sd
->entry
->collapse
)
793 sort__need_collapse
= 1;
795 if (sd
->entry
== &sort_parent
) {
796 int ret
= regcomp(&parent_regex
, parent_pattern
, REG_EXTENDED
);
800 regerror(ret
, &parent_regex
, err
, sizeof(err
));
801 fprintf(stderr
, "Invalid regex: %s\n%s",
802 parent_pattern
, err
);
805 sort__has_parent
= 1;
808 list_add_tail(&sd
->entry
->list
, &hist_entry__sort_list
);
818 hist_entry__cmp(struct hist_entry
*left
, struct hist_entry
*right
)
820 struct sort_entry
*se
;
823 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
824 cmp
= se
->cmp(left
, right
);
833 hist_entry__collapse(struct hist_entry
*left
, struct hist_entry
*right
)
835 struct sort_entry
*se
;
838 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
839 int64_t (*f
)(struct hist_entry
*, struct hist_entry
*);
841 f
= se
->collapse
?: se
->cmp
;
843 cmp
= f(left
, right
);
851 static size_t ipchain__fprintf_graph_line(FILE *fp
, int depth
, int depth_mask
)
856 ret
+= fprintf(fp
, "%s", " ");
858 for (i
= 0; i
< depth
; i
++)
859 if (depth_mask
& (1 << i
))
860 ret
+= fprintf(fp
, "| ");
862 ret
+= fprintf(fp
, " ");
864 ret
+= fprintf(fp
, "\n");
869 ipchain__fprintf_graph(FILE *fp
, struct callchain_list
*chain
, int depth
,
870 int depth_mask
, int count
, u64 total_samples
,
876 ret
+= fprintf(fp
, "%s", " ");
877 for (i
= 0; i
< depth
; i
++) {
878 if (depth_mask
& (1 << i
))
879 ret
+= fprintf(fp
, "|");
881 ret
+= fprintf(fp
, " ");
882 if (!count
&& i
== depth
- 1) {
885 percent
= hits
* 100.0 / total_samples
;
886 ret
+= percent_color_fprintf(fp
, "--%2.2f%%-- ", percent
);
888 ret
+= fprintf(fp
, "%s", " ");
891 ret
+= fprintf(fp
, "%s\n", chain
->sym
->name
);
893 ret
+= fprintf(fp
, "%p\n", (void *)(long)chain
->ip
);
899 callchain__fprintf_graph(FILE *fp
, struct callchain_node
*self
,
900 u64 total_samples
, int depth
, int depth_mask
)
902 struct rb_node
*node
, *next
;
903 struct callchain_node
*child
;
904 struct callchain_list
*chain
;
905 int new_depth_mask
= depth_mask
;
910 if (callchain_param
.mode
== CHAIN_GRAPH_REL
)
911 new_total
= self
->cumul_hit
;
913 new_total
= total_samples
;
915 node
= rb_first(&self
->rb_root
);
917 child
= rb_entry(node
, struct callchain_node
, rb_node
);
920 * The depth mask manages the output of pipes that show
921 * the depth. We don't want to keep the pipes of the current
922 * level for the last child of this depth
924 next
= rb_next(node
);
926 new_depth_mask
&= ~(1 << (depth
- 1));
929 * But we keep the older depth mask for the line seperator
930 * to keep the level link until we reach the last child
932 ret
+= ipchain__fprintf_graph_line(fp
, depth
, depth_mask
);
934 list_for_each_entry(chain
, &child
->val
, list
) {
935 if (chain
->ip
>= PERF_CONTEXT_MAX
)
937 ret
+= ipchain__fprintf_graph(fp
, chain
, depth
,
942 ret
+= callchain__fprintf_graph(fp
, child
, new_total
,
944 new_depth_mask
| (1 << depth
));
952 callchain__fprintf_flat(FILE *fp
, struct callchain_node
*self
,
955 struct callchain_list
*chain
;
961 ret
+= callchain__fprintf_flat(fp
, self
->parent
, total_samples
);
964 list_for_each_entry(chain
, &self
->val
, list
) {
965 if (chain
->ip
>= PERF_CONTEXT_MAX
)
968 ret
+= fprintf(fp
, " %s\n", chain
->sym
->name
);
970 ret
+= fprintf(fp
, " %p\n",
971 (void *)(long)chain
->ip
);
978 hist_entry_callchain__fprintf(FILE *fp
, struct hist_entry
*self
,
981 struct rb_node
*rb_node
;
982 struct callchain_node
*chain
;
985 rb_node
= rb_first(&self
->sorted_chain
);
989 chain
= rb_entry(rb_node
, struct callchain_node
, rb_node
);
990 percent
= chain
->hit
* 100.0 / total_samples
;
991 switch (callchain_param
.mode
) {
993 ret
+= percent_color_fprintf(fp
, " %6.2f%%\n",
995 ret
+= callchain__fprintf_flat(fp
, chain
, total_samples
);
997 case CHAIN_GRAPH_ABS
: /* Falldown */
998 case CHAIN_GRAPH_REL
:
999 ret
+= callchain__fprintf_graph(fp
, chain
,
1000 total_samples
, 1, 1);
1004 ret
+= fprintf(fp
, "\n");
1005 rb_node
= rb_next(rb_node
);
1013 hist_entry__fprintf(FILE *fp
, struct hist_entry
*self
, u64 total_samples
)
1015 struct sort_entry
*se
;
1018 if (exclude_other
&& !self
->parent
)
1022 ret
= percent_color_fprintf(fp
,
1023 field_sep
? "%.2f" : " %6.2f%%",
1024 (self
->count
* 100.0) / total_samples
);
1026 ret
= fprintf(fp
, field_sep
? "%lld" : "%12lld ", self
->count
);
1028 if (show_nr_samples
) {
1030 fprintf(fp
, "%c%lld", *field_sep
, self
->count
);
1032 fprintf(fp
, "%11lld", self
->count
);
1035 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
1039 fprintf(fp
, "%s", field_sep
?: " ");
1040 ret
+= se
->print(fp
, self
, se
->width
? *se
->width
: 0);
1043 ret
+= fprintf(fp
, "\n");
1046 hist_entry_callchain__fprintf(fp
, self
, total_samples
);
1055 static void dso__calc_col_width(struct dso
*self
)
1057 if (!col_width_list_str
&& !field_sep
&&
1058 (!dso_list
|| strlist__has_entry(dso_list
, self
->name
))) {
1059 unsigned int slen
= strlen(self
->name
);
1060 if (slen
> dsos__col_width
)
1061 dsos__col_width
= slen
;
1064 self
->slen_calculated
= 1;
1067 static struct symbol
*
1068 resolve_symbol(struct thread
*thread
, struct map
**mapp
,
1069 struct dso
**dsop
, u64
*ipp
)
1071 struct dso
*dso
= dsop
? *dsop
: NULL
;
1072 struct map
*map
= mapp
? *mapp
: NULL
;
1084 map
= thread__find_map(thread
, ip
);
1087 * We have to do this here as we may have a dso
1088 * with no symbol hit that has a name longer than
1089 * the ones with symbols sampled.
1091 if (!sort_dso
.elide
&& !map
->dso
->slen_calculated
)
1092 dso__calc_col_width(map
->dso
);
1097 ip
= map
->map_ip(map
, ip
);
1102 * If this is outside of all known maps,
1103 * and is a negative address, try to look it
1104 * up in the kernel dso, as it might be a
1105 * vsyscall (which executes in user-mode):
1107 if ((long long)ip
< 0)
1110 dprintf(" ...... dso: %s\n", dso
? dso
->name
: "<not found>");
1111 dprintf(" ...... map: %Lx -> %Lx\n", *ipp
, ip
);
1120 return dso
->find_symbol(dso
, ip
);
1123 static int call__match(struct symbol
*sym
)
1125 if (sym
->name
&& !regexec(&parent_regex
, sym
->name
, 0, NULL
, 0))
1131 static struct symbol
**
1132 resolve_callchain(struct thread
*thread
, struct map
*map __used
,
1133 struct ip_callchain
*chain
, struct hist_entry
*entry
)
1135 u64 context
= PERF_CONTEXT_MAX
;
1136 struct symbol
**syms
= NULL
;
1140 syms
= calloc(chain
->nr
, sizeof(*syms
));
1142 fprintf(stderr
, "Can't allocate memory for symbols\n");
1147 for (i
= 0; i
< chain
->nr
; i
++) {
1148 u64 ip
= chain
->ips
[i
];
1149 struct dso
*dso
= NULL
;
1152 if (ip
>= PERF_CONTEXT_MAX
) {
1158 case PERF_CONTEXT_HV
:
1159 dso
= hypervisor_dso
;
1161 case PERF_CONTEXT_KERNEL
:
1168 sym
= resolve_symbol(thread
, NULL
, &dso
, &ip
);
1171 if (sort__has_parent
&& call__match(sym
) &&
1173 entry
->parent
= sym
;
1184 * collect histogram counts
1188 hist_entry__add(struct thread
*thread
, struct map
*map
, struct dso
*dso
,
1189 struct symbol
*sym
, u64 ip
, struct ip_callchain
*chain
,
1190 char level
, u64 count
)
1192 struct rb_node
**p
= &hist
.rb_node
;
1193 struct rb_node
*parent
= NULL
;
1194 struct hist_entry
*he
;
1195 struct symbol
**syms
= NULL
;
1196 struct hist_entry entry
= {
1205 .sorted_chain
= RB_ROOT
1209 if ((sort__has_parent
|| callchain
) && chain
)
1210 syms
= resolve_callchain(thread
, map
, chain
, &entry
);
1212 while (*p
!= NULL
) {
1214 he
= rb_entry(parent
, struct hist_entry
, rb_node
);
1216 cmp
= hist_entry__cmp(&entry
, he
);
1221 append_chain(&he
->callchain
, chain
, syms
);
1230 p
= &(*p
)->rb_right
;
1233 he
= malloc(sizeof(*he
));
1238 callchain_init(&he
->callchain
);
1239 append_chain(&he
->callchain
, chain
, syms
);
1242 rb_link_node(&he
->rb_node
, parent
, p
);
1243 rb_insert_color(&he
->rb_node
, &hist
);
1248 static void hist_entry__free(struct hist_entry
*he
)
1254 * collapse the histogram
1257 static struct rb_root collapse_hists
;
1259 static void collapse__insert_entry(struct hist_entry
*he
)
1261 struct rb_node
**p
= &collapse_hists
.rb_node
;
1262 struct rb_node
*parent
= NULL
;
1263 struct hist_entry
*iter
;
1266 while (*p
!= NULL
) {
1268 iter
= rb_entry(parent
, struct hist_entry
, rb_node
);
1270 cmp
= hist_entry__collapse(iter
, he
);
1273 iter
->count
+= he
->count
;
1274 hist_entry__free(he
);
1281 p
= &(*p
)->rb_right
;
1284 rb_link_node(&he
->rb_node
, parent
, p
);
1285 rb_insert_color(&he
->rb_node
, &collapse_hists
);
1288 static void collapse__resort(void)
1290 struct rb_node
*next
;
1291 struct hist_entry
*n
;
1293 if (!sort__need_collapse
)
1296 next
= rb_first(&hist
);
1298 n
= rb_entry(next
, struct hist_entry
, rb_node
);
1299 next
= rb_next(&n
->rb_node
);
1301 rb_erase(&n
->rb_node
, &hist
);
1302 collapse__insert_entry(n
);
1307 * reverse the map, sort on count.
1310 static struct rb_root output_hists
;
1312 static void output__insert_entry(struct hist_entry
*he
, u64 min_callchain_hits
)
1314 struct rb_node
**p
= &output_hists
.rb_node
;
1315 struct rb_node
*parent
= NULL
;
1316 struct hist_entry
*iter
;
1319 callchain_param
.sort(&he
->sorted_chain
, &he
->callchain
,
1320 min_callchain_hits
, &callchain_param
);
1322 while (*p
!= NULL
) {
1324 iter
= rb_entry(parent
, struct hist_entry
, rb_node
);
1326 if (he
->count
> iter
->count
)
1329 p
= &(*p
)->rb_right
;
1332 rb_link_node(&he
->rb_node
, parent
, p
);
1333 rb_insert_color(&he
->rb_node
, &output_hists
);
1336 static void output__resort(u64 total_samples
)
1338 struct rb_node
*next
;
1339 struct hist_entry
*n
;
1340 struct rb_root
*tree
= &hist
;
1341 u64 min_callchain_hits
;
1343 min_callchain_hits
= total_samples
* (callchain_param
.min_percent
/ 100);
1345 if (sort__need_collapse
)
1346 tree
= &collapse_hists
;
1348 next
= rb_first(tree
);
1351 n
= rb_entry(next
, struct hist_entry
, rb_node
);
1352 next
= rb_next(&n
->rb_node
);
1354 rb_erase(&n
->rb_node
, tree
);
1355 output__insert_entry(n
, min_callchain_hits
);
1359 static size_t output__fprintf(FILE *fp
, u64 total_samples
)
1361 struct hist_entry
*pos
;
1362 struct sort_entry
*se
;
1366 char *col_width
= col_width_list_str
;
1368 fprintf(fp
, "# Samples: %Ld\n", (u64
)total_samples
);
1371 fprintf(fp
, "# Overhead");
1372 if (show_nr_samples
) {
1374 fprintf(fp
, "%cSamples", *field_sep
);
1376 fputs(" Samples ", fp
);
1378 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
1382 fprintf(fp
, "%c%s", *field_sep
, se
->header
);
1385 width
= strlen(se
->header
);
1387 if (col_width_list_str
) {
1389 *se
->width
= atoi(col_width
);
1390 col_width
= strchr(col_width
, ',');
1395 width
= *se
->width
= max(*se
->width
, width
);
1397 fprintf(fp
, " %*s", width
, se
->header
);
1404 fprintf(fp
, "# ........");
1405 if (show_nr_samples
)
1406 fprintf(fp
, " ..........");
1407 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
1417 width
= strlen(se
->header
);
1418 for (i
= 0; i
< width
; i
++)
1426 for (nd
= rb_first(&output_hists
); nd
; nd
= rb_next(nd
)) {
1427 pos
= rb_entry(nd
, struct hist_entry
, rb_node
);
1428 ret
+= hist_entry__fprintf(fp
, pos
, total_samples
);
1431 if (sort_order
== default_sort_order
&&
1432 parent_pattern
== default_parent_pattern
) {
1434 fprintf(fp
, "# (For more details, try: perf report --sort comm,dso,symbol)\n");
1442 static void register_idle_thread(void)
1444 struct thread
*thread
= threads__findnew(0);
1446 if (thread
== NULL
||
1447 thread__set_comm(thread
, "[idle]")) {
1448 fprintf(stderr
, "problem inserting idle task.\n");
1453 static unsigned long total
= 0,
1460 static int validate_chain(struct ip_callchain
*chain
, event_t
*event
)
1462 unsigned int chain_size
;
1464 chain_size
= event
->header
.size
;
1465 chain_size
-= (unsigned long)&event
->ip
.__more_data
- (unsigned long)event
;
1467 if (chain
->nr
*sizeof(u64
) > chain_size
)
1474 process_sample_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1478 struct dso
*dso
= NULL
;
1479 struct thread
*thread
= threads__findnew(event
->ip
.pid
);
1480 u64 ip
= event
->ip
.ip
;
1482 struct map
*map
= NULL
;
1483 void *more_data
= event
->ip
.__more_data
;
1484 struct ip_callchain
*chain
= NULL
;
1487 if (sample_type
& PERF_SAMPLE_PERIOD
) {
1488 period
= *(u64
*)more_data
;
1489 more_data
+= sizeof(u64
);
1492 dprintf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d: %p period: %Ld\n",
1493 (void *)(offset
+ head
),
1494 (void *)(long)(event
->header
.size
),
1500 if (sample_type
& PERF_SAMPLE_CALLCHAIN
) {
1503 chain
= (void *)more_data
;
1505 dprintf("... chain: nr:%Lu\n", chain
->nr
);
1507 if (validate_chain(chain
, event
) < 0) {
1508 eprintf("call-chain problem with event, skipping it.\n");
1513 for (i
= 0; i
< chain
->nr
; i
++)
1514 dprintf("..... %2d: %016Lx\n", i
, chain
->ips
[i
]);
1518 dprintf(" ... thread: %s:%d\n", thread
->comm
, thread
->pid
);
1520 if (thread
== NULL
) {
1521 eprintf("problem processing %d event, skipping it.\n",
1522 event
->header
.type
);
1526 if (comm_list
&& !strlist__has_entry(comm_list
, thread
->comm
))
1529 cpumode
= event
->header
.misc
& PERF_EVENT_MISC_CPUMODE_MASK
;
1531 if (cpumode
== PERF_EVENT_MISC_KERNEL
) {
1537 dprintf(" ...... dso: %s\n", dso
->name
);
1539 } else if (cpumode
== PERF_EVENT_MISC_USER
) {
1548 dso
= hypervisor_dso
;
1550 dprintf(" ...... dso: [hypervisor]\n");
1553 if (show
& show_mask
) {
1554 struct symbol
*sym
= resolve_symbol(thread
, &map
, &dso
, &ip
);
1556 if (dso_list
&& dso
&& dso
->name
&& !strlist__has_entry(dso_list
, dso
->name
))
1559 if (sym_list
&& sym
&& !strlist__has_entry(sym_list
, sym
->name
))
1562 if (hist_entry__add(thread
, map
, dso
, sym
, ip
, chain
, level
, period
)) {
1563 eprintf("problem incrementing symbol count, skipping event\n");
1573 process_mmap_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1575 struct thread
*thread
= threads__findnew(event
->mmap
.pid
);
1576 struct map
*map
= map__new(&event
->mmap
);
1578 dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n",
1579 (void *)(offset
+ head
),
1580 (void *)(long)(event
->header
.size
),
1582 (void *)(long)event
->mmap
.start
,
1583 (void *)(long)event
->mmap
.len
,
1584 (void *)(long)event
->mmap
.pgoff
,
1585 event
->mmap
.filename
);
1587 if (thread
== NULL
|| map
== NULL
) {
1588 dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1592 thread__insert_map(thread
, map
);
1599 process_comm_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1601 struct thread
*thread
= threads__findnew(event
->comm
.pid
);
1603 dprintf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
1604 (void *)(offset
+ head
),
1605 (void *)(long)(event
->header
.size
),
1606 event
->comm
.comm
, event
->comm
.pid
);
1608 if (thread
== NULL
||
1609 thread__set_comm(thread
, event
->comm
.comm
)) {
1610 dprintf("problem processing PERF_EVENT_COMM, skipping event.\n");
1619 process_fork_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1621 struct thread
*thread
= threads__findnew(event
->fork
.pid
);
1622 struct thread
*parent
= threads__findnew(event
->fork
.ppid
);
1624 dprintf("%p [%p]: PERF_EVENT_FORK: %d:%d\n",
1625 (void *)(offset
+ head
),
1626 (void *)(long)(event
->header
.size
),
1627 event
->fork
.pid
, event
->fork
.ppid
);
1629 if (!thread
|| !parent
|| thread__fork(thread
, parent
)) {
1630 dprintf("problem processing PERF_EVENT_FORK, skipping event.\n");
1639 process_period_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1641 dprintf("%p [%p]: PERF_EVENT_PERIOD: time:%Ld, id:%Ld: period:%Ld\n",
1642 (void *)(offset
+ head
),
1643 (void *)(long)(event
->header
.size
),
1646 event
->period
.sample_period
);
1652 process_lost_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1654 dprintf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n",
1655 (void *)(offset
+ head
),
1656 (void *)(long)(event
->header
.size
),
1660 total_lost
+= event
->lost
.lost
;
1665 static void trace_event(event_t
*event
)
1667 unsigned char *raw_event
= (void *)event
;
1668 char *color
= PERF_COLOR_BLUE
;
1675 cdprintf("\n. ... raw event: size %d bytes\n", event
->header
.size
);
1677 for (i
= 0; i
< event
->header
.size
; i
++) {
1678 if ((i
& 15) == 0) {
1680 cdprintf(" %04x: ", i
);
1683 cdprintf(" %02x", raw_event
[i
]);
1685 if (((i
& 15) == 15) || i
== event
->header
.size
-1) {
1687 for (j
= 0; j
< 15-(i
& 15); j
++)
1689 for (j
= 0; j
< (i
& 15); j
++) {
1690 if (isprint(raw_event
[i
-15+j
]))
1691 cdprintf("%c", raw_event
[i
-15+j
]);
1702 process_read_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1704 dprintf("%p [%p]: PERF_EVENT_READ: %d %d %Lu\n",
1705 (void *)(offset
+ head
),
1706 (void *)(long)(event
->header
.size
),
1715 process_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1719 switch (event
->header
.type
) {
1720 case PERF_EVENT_SAMPLE
:
1721 return process_sample_event(event
, offset
, head
);
1723 case PERF_EVENT_MMAP
:
1724 return process_mmap_event(event
, offset
, head
);
1726 case PERF_EVENT_COMM
:
1727 return process_comm_event(event
, offset
, head
);
1729 case PERF_EVENT_FORK
:
1730 return process_fork_event(event
, offset
, head
);
1732 case PERF_EVENT_PERIOD
:
1733 return process_period_event(event
, offset
, head
);
1735 case PERF_EVENT_LOST
:
1736 return process_lost_event(event
, offset
, head
);
1738 case PERF_EVENT_READ
:
1739 return process_read_event(event
, offset
, head
);
1742 * We dont process them right now but they are fine:
1745 case PERF_EVENT_THROTTLE
:
1746 case PERF_EVENT_UNTHROTTLE
:
1756 static struct perf_header
*header
;
1758 static u64
perf_header__sample_type(void)
1760 u64 sample_type
= 0;
1763 for (i
= 0; i
< header
->attrs
; i
++) {
1764 struct perf_header_attr
*attr
= header
->attr
[i
];
1767 sample_type
= attr
->attr
.sample_type
;
1768 else if (sample_type
!= attr
->attr
.sample_type
)
1769 die("non matching sample_type");
1775 static int __cmd_report(void)
1777 int ret
, rc
= EXIT_FAILURE
;
1778 unsigned long offset
= 0;
1779 unsigned long head
, shift
;
1785 register_idle_thread();
1787 input
= open(input_name
, O_RDONLY
);
1789 fprintf(stderr
, " failed to open file: %s", input_name
);
1790 if (!strcmp(input_name
, "perf.data"))
1791 fprintf(stderr
, " (try 'perf record' first)");
1792 fprintf(stderr
, "\n");
1796 ret
= fstat(input
, &stat
);
1798 perror("failed to stat file");
1802 if (!stat
.st_size
) {
1803 fprintf(stderr
, "zero-sized file, nothing to do!\n");
1807 header
= perf_header__read(input
);
1808 head
= header
->data_offset
;
1810 sample_type
= perf_header__sample_type();
1812 if (!(sample_type
& PERF_SAMPLE_CALLCHAIN
)) {
1813 if (sort__has_parent
) {
1814 fprintf(stderr
, "selected --sort parent, but no"
1815 " callchain data. Did you call"
1816 " perf record without -g?\n");
1820 fprintf(stderr
, "selected -c but no callchain data."
1821 " Did you call perf record without"
1827 if (load_kernel() < 0) {
1828 perror("failed to load kernel symbols");
1829 return EXIT_FAILURE
;
1833 if (getcwd(__cwd
, sizeof(__cwd
)) == NULL
) {
1834 perror("failed to get the current directory");
1835 return EXIT_FAILURE
;
1837 cwdlen
= strlen(cwd
);
1843 shift
= page_size
* (head
/ page_size
);
1848 buf
= (char *)mmap(NULL
, page_size
* mmap_window
, PROT_READ
,
1849 MAP_SHARED
, input
, offset
);
1850 if (buf
== MAP_FAILED
) {
1851 perror("failed to mmap file");
1856 event
= (event_t
*)(buf
+ head
);
1858 size
= event
->header
.size
;
1862 if (head
+ event
->header
.size
>= page_size
* mmap_window
) {
1865 shift
= page_size
* (head
/ page_size
);
1867 ret
= munmap(buf
, page_size
* mmap_window
);
1875 size
= event
->header
.size
;
1877 dprintf("\n%p [%p]: event: %d\n",
1878 (void *)(offset
+ head
),
1879 (void *)(long)event
->header
.size
,
1880 event
->header
.type
);
1882 if (!size
|| process_event(event
, offset
, head
) < 0) {
1884 dprintf("%p [%p]: skipping unknown header type: %d\n",
1885 (void *)(offset
+ head
),
1886 (void *)(long)(event
->header
.size
),
1887 event
->header
.type
);
1892 * assume we lost track of the stream, check alignment, and
1893 * increment a single u64 in the hope to catch on again 'soon'.
1896 if (unlikely(head
& 7))
1904 if (offset
+ head
>= header
->data_offset
+ header
->data_size
)
1907 if (offset
+ head
< (unsigned long)stat
.st_size
)
1914 dprintf(" IP events: %10ld\n", total
);
1915 dprintf(" mmap events: %10ld\n", total_mmap
);
1916 dprintf(" comm events: %10ld\n", total_comm
);
1917 dprintf(" fork events: %10ld\n", total_fork
);
1918 dprintf(" lost events: %10ld\n", total_lost
);
1919 dprintf(" unknown events: %10ld\n", total_unknown
);
1925 threads__fprintf(stdout
);
1928 dsos__fprintf(stdout
);
1931 output__resort(total
);
1932 output__fprintf(stdout
, total
);
1938 parse_callchain_opt(const struct option
*opt __used
, const char *arg
,
1949 tok
= strtok((char *)arg
, ",");
1953 /* get the output mode */
1954 if (!strncmp(tok
, "graph", strlen(arg
)))
1955 callchain_param
.mode
= CHAIN_GRAPH_ABS
;
1957 else if (!strncmp(tok
, "flat", strlen(arg
)))
1958 callchain_param
.mode
= CHAIN_FLAT
;
1960 else if (!strncmp(tok
, "fractal", strlen(arg
)))
1961 callchain_param
.mode
= CHAIN_GRAPH_REL
;
1966 /* get the min percentage */
1967 tok
= strtok(NULL
, ",");
1971 callchain_param
.min_percent
= strtod(tok
, &endptr
);
1976 if (register_callchain_param(&callchain_param
) < 0) {
1977 fprintf(stderr
, "Can't register callchain params\n");
1983 static const char * const report_usage
[] = {
1984 "perf report [<options>] <command>",
1988 static const struct option options
[] = {
1989 OPT_STRING('i', "input", &input_name
, "file",
1991 OPT_BOOLEAN('v', "verbose", &verbose
,
1992 "be more verbose (show symbol address, etc)"),
1993 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace
,
1994 "dump raw trace in ASCII"),
1995 OPT_STRING('k', "vmlinux", &vmlinux
, "file", "vmlinux pathname"),
1996 OPT_BOOLEAN('m', "modules", &modules
,
1997 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1998 OPT_BOOLEAN('n', "show-nr-samples", &show_nr_samples
,
1999 "Show a column with the number of samples"),
2000 OPT_STRING('s', "sort", &sort_order
, "key[,key2...]",
2001 "sort by key(s): pid, comm, dso, symbol, parent"),
2002 OPT_BOOLEAN('P', "full-paths", &full_paths
,
2003 "Don't shorten the pathnames taking into account the cwd"),
2004 OPT_STRING('p', "parent", &parent_pattern
, "regex",
2005 "regex filter to identify parent, see: '--sort parent'"),
2006 OPT_BOOLEAN('x', "exclude-other", &exclude_other
,
2007 "Only display entries with parent-match"),
2008 OPT_CALLBACK_DEFAULT('g', "call-graph", NULL
, "output_type,min_percent",
2009 "Display callchains using output_type and min percent threshold. "
2010 "Default: fractal,0.5", &parse_callchain_opt
, callchain_default_opt
),
2011 OPT_STRING('d', "dsos", &dso_list_str
, "dso[,dso...]",
2012 "only consider symbols in these dsos"),
2013 OPT_STRING('C', "comms", &comm_list_str
, "comm[,comm...]",
2014 "only consider symbols in these comms"),
2015 OPT_STRING('S', "symbols", &sym_list_str
, "symbol[,symbol...]",
2016 "only consider these symbols"),
2017 OPT_STRING('w', "column-widths", &col_width_list_str
,
2019 "don't try to adjust column width, use these fixed values"),
2020 OPT_STRING('t', "field-separator", &field_sep
, "separator",
2021 "separator for columns, no spaces will be added between "
2022 "columns '.' is reserved."),
2026 static void setup_sorting(void)
2028 char *tmp
, *tok
, *str
= strdup(sort_order
);
2030 for (tok
= strtok_r(str
, ", ", &tmp
);
2031 tok
; tok
= strtok_r(NULL
, ", ", &tmp
)) {
2032 if (sort_dimension__add(tok
) < 0) {
2033 error("Unknown --sort key: `%s'", tok
);
2034 usage_with_options(report_usage
, options
);
2041 static void setup_list(struct strlist
**list
, const char *list_str
,
2042 struct sort_entry
*se
, const char *list_name
,
2046 *list
= strlist__new(true, list_str
);
2048 fprintf(stderr
, "problems parsing %s list\n",
2052 if (strlist__nr_entries(*list
) == 1) {
2053 fprintf(fp
, "# %s: %s\n", list_name
,
2054 strlist__entry(*list
, 0)->s
);
2060 int cmd_report(int argc
, const char **argv
, const char *prefix __used
)
2064 page_size
= getpagesize();
2066 argc
= parse_options(argc
, argv
, options
, report_usage
, 0);
2070 if (parent_pattern
!= default_parent_pattern
) {
2071 sort_dimension__add("parent");
2072 sort_parent
.elide
= 1;
2077 * Any (unrecognized) arguments left?
2080 usage_with_options(report_usage
, options
);
2084 setup_list(&dso_list
, dso_list_str
, &sort_dso
, "dso", stdout
);
2085 setup_list(&comm_list
, comm_list_str
, &sort_comm
, "comm", stdout
);
2086 setup_list(&sym_list
, sym_list_str
, &sort_sym
, "symbol", stdout
);
2088 if (field_sep
&& *field_sep
== '.') {
2089 fputs("'.' is the only non valid --field-separator argument\n",
2094 return __cmd_report();