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,symbol";
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
;
106 struct perf_event_header header
;
112 struct perf_event_header header
;
118 typedef union event_union
{
119 struct perf_event_header header
;
121 struct mmap_event mmap
;
122 struct comm_event comm
;
123 struct fork_event fork
;
124 struct lost_event lost
;
125 struct read_event read
;
128 static int repsep_fprintf(FILE *fp
, const char *fmt
, ...)
135 n
= vfprintf(fp
, fmt
, ap
);
138 n
= vasprintf(&bf
, fmt
, ap
);
142 sep
= strchr(sep
, *field_sep
);
155 static LIST_HEAD(dsos
);
156 static struct dso
*kernel_dso
;
157 static struct dso
*vdso
;
158 static struct dso
*hypervisor_dso
;
160 static void dsos__add(struct dso
*dso
)
162 list_add_tail(&dso
->node
, &dsos
);
165 static struct dso
*dsos__find(const char *name
)
169 list_for_each_entry(pos
, &dsos
, node
)
170 if (strcmp(pos
->name
, name
) == 0)
175 static struct dso
*dsos__findnew(const char *name
)
177 struct dso
*dso
= dsos__find(name
);
183 dso
= dso__new(name
, 0);
187 nr
= dso__load(dso
, NULL
, verbose
);
189 eprintf("Failed to open: %s\n", name
);
193 eprintf("No symbols found in: %s, maybe install a debug package?\n", name
);
204 static void dsos__fprintf(FILE *fp
)
208 list_for_each_entry(pos
, &dsos
, node
)
209 dso__fprintf(pos
, fp
);
212 static struct symbol
*vdso__find_symbol(struct dso
*dso
, u64 ip
)
214 return dso__find_symbol(dso
, ip
);
217 static int load_kernel(void)
221 kernel_dso
= dso__new("[kernel]", 0);
225 err
= dso__load_kernel(kernel_dso
, vmlinux
, NULL
, verbose
, modules
);
227 dso__delete(kernel_dso
);
230 dsos__add(kernel_dso
);
232 vdso
= dso__new("[vdso]", 0);
236 vdso
->find_symbol
= vdso__find_symbol
;
240 hypervisor_dso
= dso__new("[hypervisor]", 0);
243 dsos__add(hypervisor_dso
);
248 static char __cwd
[PATH_MAX
];
249 static char *cwd
= __cwd
;
252 static int strcommon(const char *pathname
)
256 while (n
< cwdlen
&& pathname
[n
] == cwd
[n
])
263 struct list_head node
;
267 u64 (*map_ip
)(struct map
*, u64
);
271 static u64
map__map_ip(struct map
*map
, u64 ip
)
273 return ip
- map
->start
+ map
->pgoff
;
276 static u64
vdso__map_ip(struct map
*map __used
, u64 ip
)
281 static inline int is_anon_memory(const char *filename
)
283 return strcmp(filename
, "//anon") == 0;
286 static struct map
*map__new(struct mmap_event
*event
)
288 struct map
*self
= malloc(sizeof(*self
));
291 const char *filename
= event
->filename
;
292 char newfilename
[PATH_MAX
];
296 int n
= strcommon(filename
);
299 snprintf(newfilename
, sizeof(newfilename
),
300 ".%s", filename
+ n
);
301 filename
= newfilename
;
305 anon
= is_anon_memory(filename
);
308 snprintf(newfilename
, sizeof(newfilename
), "/tmp/perf-%d.map", event
->pid
);
309 filename
= newfilename
;
312 self
->start
= event
->start
;
313 self
->end
= event
->start
+ event
->len
;
314 self
->pgoff
= event
->pgoff
;
316 self
->dso
= dsos__findnew(filename
);
317 if (self
->dso
== NULL
)
320 if (self
->dso
== vdso
|| anon
)
321 self
->map_ip
= vdso__map_ip
;
323 self
->map_ip
= map__map_ip
;
331 static struct map
*map__clone(struct map
*self
)
333 struct map
*map
= malloc(sizeof(*self
));
338 memcpy(map
, self
, sizeof(*self
));
343 static int map__overlap(struct map
*l
, struct map
*r
)
345 if (l
->start
> r
->start
) {
351 if (l
->end
> r
->start
)
357 static size_t map__fprintf(struct map
*self
, FILE *fp
)
359 return fprintf(fp
, " %Lx-%Lx %Lx %s\n",
360 self
->start
, self
->end
, self
->pgoff
, self
->dso
->name
);
365 struct rb_node rb_node
;
366 struct list_head maps
;
371 static struct thread
*thread__new(pid_t pid
)
373 struct thread
*self
= malloc(sizeof(*self
));
377 self
->comm
= malloc(32);
379 snprintf(self
->comm
, 32, ":%d", self
->pid
);
380 INIT_LIST_HEAD(&self
->maps
);
386 static unsigned int dsos__col_width
,
390 static int thread__set_comm(struct thread
*self
, const char *comm
)
394 self
->comm
= strdup(comm
);
398 if (!col_width_list_str
&& !field_sep
&&
399 (!comm_list
|| strlist__has_entry(comm_list
, comm
))) {
400 unsigned int slen
= strlen(comm
);
401 if (slen
> comms__col_width
) {
402 comms__col_width
= slen
;
403 threads__col_width
= slen
+ 6;
410 static size_t thread__fprintf(struct thread
*self
, FILE *fp
)
413 size_t ret
= fprintf(fp
, "Thread %d %s\n", self
->pid
, self
->comm
);
415 list_for_each_entry(pos
, &self
->maps
, node
)
416 ret
+= map__fprintf(pos
, fp
);
422 static struct rb_root threads
;
423 static struct thread
*last_match
;
425 static struct thread
*threads__findnew(pid_t pid
)
427 struct rb_node
**p
= &threads
.rb_node
;
428 struct rb_node
*parent
= NULL
;
432 * Font-end cache - PID lookups come in blocks,
433 * so most of the time we dont have to look up
436 if (last_match
&& last_match
->pid
== pid
)
441 th
= rb_entry(parent
, struct thread
, rb_node
);
443 if (th
->pid
== pid
) {
454 th
= thread__new(pid
);
456 rb_link_node(&th
->rb_node
, parent
, p
);
457 rb_insert_color(&th
->rb_node
, &threads
);
464 static void thread__insert_map(struct thread
*self
, struct map
*map
)
466 struct map
*pos
, *tmp
;
468 list_for_each_entry_safe(pos
, tmp
, &self
->maps
, node
) {
469 if (map__overlap(pos
, map
)) {
471 printf("overlapping maps:\n");
472 map__fprintf(map
, stdout
);
473 map__fprintf(pos
, stdout
);
476 if (map
->start
<= pos
->start
&& map
->end
> pos
->start
)
477 pos
->start
= map
->end
;
479 if (map
->end
>= pos
->end
&& map
->start
< pos
->end
)
480 pos
->end
= map
->start
;
483 printf("after collision:\n");
484 map__fprintf(pos
, stdout
);
487 if (pos
->start
>= pos
->end
) {
488 list_del_init(&pos
->node
);
494 list_add_tail(&map
->node
, &self
->maps
);
497 static int thread__fork(struct thread
*self
, struct thread
*parent
)
503 self
->comm
= strdup(parent
->comm
);
507 list_for_each_entry(map
, &parent
->maps
, node
) {
508 struct map
*new = map__clone(map
);
511 thread__insert_map(self
, new);
517 static struct map
*thread__find_map(struct thread
*self
, u64 ip
)
524 list_for_each_entry(pos
, &self
->maps
, node
)
525 if (ip
>= pos
->start
&& ip
<= pos
->end
)
531 static size_t threads__fprintf(FILE *fp
)
536 for (nd
= rb_first(&threads
); nd
; nd
= rb_next(nd
)) {
537 struct thread
*pos
= rb_entry(nd
, struct thread
, rb_node
);
539 ret
+= thread__fprintf(pos
, fp
);
546 * histogram, sorted on item, collects counts
549 static struct rb_root hist
;
552 struct rb_node rb_node
;
554 struct thread
*thread
;
558 struct symbol
*parent
;
561 struct callchain_node callchain
;
562 struct rb_root sorted_chain
;
568 * configurable sorting bits
572 struct list_head list
;
576 int64_t (*cmp
)(struct hist_entry
*, struct hist_entry
*);
577 int64_t (*collapse
)(struct hist_entry
*, struct hist_entry
*);
578 size_t (*print
)(FILE *fp
, struct hist_entry
*, unsigned int width
);
583 static int64_t cmp_null(void *l
, void *r
)
596 sort__thread_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
598 return right
->thread
->pid
- left
->thread
->pid
;
602 sort__thread_print(FILE *fp
, struct hist_entry
*self
, unsigned int width
)
604 return repsep_fprintf(fp
, "%*s:%5d", width
- 6,
605 self
->thread
->comm
?: "", self
->thread
->pid
);
608 static struct sort_entry sort_thread
= {
609 .header
= "Command: Pid",
610 .cmp
= sort__thread_cmp
,
611 .print
= sort__thread_print
,
612 .width
= &threads__col_width
,
618 sort__comm_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
620 return right
->thread
->pid
- left
->thread
->pid
;
624 sort__comm_collapse(struct hist_entry
*left
, struct hist_entry
*right
)
626 char *comm_l
= left
->thread
->comm
;
627 char *comm_r
= right
->thread
->comm
;
629 if (!comm_l
|| !comm_r
)
630 return cmp_null(comm_l
, comm_r
);
632 return strcmp(comm_l
, comm_r
);
636 sort__comm_print(FILE *fp
, struct hist_entry
*self
, unsigned int width
)
638 return repsep_fprintf(fp
, "%*s", width
, self
->thread
->comm
);
641 static struct sort_entry sort_comm
= {
643 .cmp
= sort__comm_cmp
,
644 .collapse
= sort__comm_collapse
,
645 .print
= sort__comm_print
,
646 .width
= &comms__col_width
,
652 sort__dso_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
654 struct dso
*dso_l
= left
->dso
;
655 struct dso
*dso_r
= right
->dso
;
657 if (!dso_l
|| !dso_r
)
658 return cmp_null(dso_l
, dso_r
);
660 return strcmp(dso_l
->name
, dso_r
->name
);
664 sort__dso_print(FILE *fp
, struct hist_entry
*self
, unsigned int width
)
667 return repsep_fprintf(fp
, "%-*s", width
, self
->dso
->name
);
669 return repsep_fprintf(fp
, "%*llx", width
, (u64
)self
->ip
);
672 static struct sort_entry sort_dso
= {
673 .header
= "Shared Object",
674 .cmp
= sort__dso_cmp
,
675 .print
= sort__dso_print
,
676 .width
= &dsos__col_width
,
682 sort__sym_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
686 if (left
->sym
== right
->sym
)
689 ip_l
= left
->sym
? left
->sym
->start
: left
->ip
;
690 ip_r
= right
->sym
? right
->sym
->start
: right
->ip
;
692 return (int64_t)(ip_r
- ip_l
);
696 sort__sym_print(FILE *fp
, struct hist_entry
*self
, unsigned int width __used
)
701 ret
+= repsep_fprintf(fp
, "%#018llx ", (u64
)self
->ip
);
703 ret
+= repsep_fprintf(fp
, "[%c] ", self
->level
);
705 ret
+= repsep_fprintf(fp
, "%s", self
->sym
->name
);
707 if (self
->sym
->module
)
708 ret
+= repsep_fprintf(fp
, "\t[%s]",
709 self
->sym
->module
->name
);
711 ret
+= repsep_fprintf(fp
, "%#016llx", (u64
)self
->ip
);
717 static struct sort_entry sort_sym
= {
719 .cmp
= sort__sym_cmp
,
720 .print
= sort__sym_print
,
726 sort__parent_cmp(struct hist_entry
*left
, struct hist_entry
*right
)
728 struct symbol
*sym_l
= left
->parent
;
729 struct symbol
*sym_r
= right
->parent
;
731 if (!sym_l
|| !sym_r
)
732 return cmp_null(sym_l
, sym_r
);
734 return strcmp(sym_l
->name
, sym_r
->name
);
738 sort__parent_print(FILE *fp
, struct hist_entry
*self
, unsigned int width
)
740 return repsep_fprintf(fp
, "%-*s", width
,
741 self
->parent
? self
->parent
->name
: "[other]");
744 static unsigned int parent_symbol__col_width
;
746 static struct sort_entry sort_parent
= {
747 .header
= "Parent symbol",
748 .cmp
= sort__parent_cmp
,
749 .print
= sort__parent_print
,
750 .width
= &parent_symbol__col_width
,
753 static int sort__need_collapse
= 0;
754 static int sort__has_parent
= 0;
756 struct sort_dimension
{
758 struct sort_entry
*entry
;
762 static struct sort_dimension sort_dimensions
[] = {
763 { .name
= "pid", .entry
= &sort_thread
, },
764 { .name
= "comm", .entry
= &sort_comm
, },
765 { .name
= "dso", .entry
= &sort_dso
, },
766 { .name
= "symbol", .entry
= &sort_sym
, },
767 { .name
= "parent", .entry
= &sort_parent
, },
770 static LIST_HEAD(hist_entry__sort_list
);
772 static int sort_dimension__add(char *tok
)
776 for (i
= 0; i
< ARRAY_SIZE(sort_dimensions
); i
++) {
777 struct sort_dimension
*sd
= &sort_dimensions
[i
];
782 if (strncasecmp(tok
, sd
->name
, strlen(tok
)))
785 if (sd
->entry
->collapse
)
786 sort__need_collapse
= 1;
788 if (sd
->entry
== &sort_parent
) {
789 int ret
= regcomp(&parent_regex
, parent_pattern
, REG_EXTENDED
);
793 regerror(ret
, &parent_regex
, err
, sizeof(err
));
794 fprintf(stderr
, "Invalid regex: %s\n%s",
795 parent_pattern
, err
);
798 sort__has_parent
= 1;
801 list_add_tail(&sd
->entry
->list
, &hist_entry__sort_list
);
811 hist_entry__cmp(struct hist_entry
*left
, struct hist_entry
*right
)
813 struct sort_entry
*se
;
816 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
817 cmp
= se
->cmp(left
, right
);
826 hist_entry__collapse(struct hist_entry
*left
, struct hist_entry
*right
)
828 struct sort_entry
*se
;
831 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
832 int64_t (*f
)(struct hist_entry
*, struct hist_entry
*);
834 f
= se
->collapse
?: se
->cmp
;
836 cmp
= f(left
, right
);
844 static size_t ipchain__fprintf_graph_line(FILE *fp
, int depth
, int depth_mask
)
849 ret
+= fprintf(fp
, "%s", " ");
851 for (i
= 0; i
< depth
; i
++)
852 if (depth_mask
& (1 << i
))
853 ret
+= fprintf(fp
, "| ");
855 ret
+= fprintf(fp
, " ");
857 ret
+= fprintf(fp
, "\n");
862 ipchain__fprintf_graph(FILE *fp
, struct callchain_list
*chain
, int depth
,
863 int depth_mask
, int count
, u64 total_samples
,
869 ret
+= fprintf(fp
, "%s", " ");
870 for (i
= 0; i
< depth
; i
++) {
871 if (depth_mask
& (1 << i
))
872 ret
+= fprintf(fp
, "|");
874 ret
+= fprintf(fp
, " ");
875 if (!count
&& i
== depth
- 1) {
878 percent
= hits
* 100.0 / total_samples
;
879 ret
+= percent_color_fprintf(fp
, "--%2.2f%%-- ", percent
);
881 ret
+= fprintf(fp
, "%s", " ");
884 ret
+= fprintf(fp
, "%s\n", chain
->sym
->name
);
886 ret
+= fprintf(fp
, "%p\n", (void *)(long)chain
->ip
);
892 callchain__fprintf_graph(FILE *fp
, struct callchain_node
*self
,
893 u64 total_samples
, int depth
, int depth_mask
)
895 struct rb_node
*node
, *next
;
896 struct callchain_node
*child
;
897 struct callchain_list
*chain
;
898 int new_depth_mask
= depth_mask
;
903 if (callchain_param
.mode
== CHAIN_GRAPH_REL
)
904 new_total
= self
->cumul_hit
;
906 new_total
= total_samples
;
908 node
= rb_first(&self
->rb_root
);
910 child
= rb_entry(node
, struct callchain_node
, rb_node
);
913 * The depth mask manages the output of pipes that show
914 * the depth. We don't want to keep the pipes of the current
915 * level for the last child of this depth
917 next
= rb_next(node
);
919 new_depth_mask
&= ~(1 << (depth
- 1));
922 * But we keep the older depth mask for the line seperator
923 * to keep the level link until we reach the last child
925 ret
+= ipchain__fprintf_graph_line(fp
, depth
, depth_mask
);
927 list_for_each_entry(chain
, &child
->val
, list
) {
928 if (chain
->ip
>= PERF_CONTEXT_MAX
)
930 ret
+= ipchain__fprintf_graph(fp
, chain
, depth
,
935 ret
+= callchain__fprintf_graph(fp
, child
, new_total
,
937 new_depth_mask
| (1 << depth
));
945 callchain__fprintf_flat(FILE *fp
, struct callchain_node
*self
,
948 struct callchain_list
*chain
;
954 ret
+= callchain__fprintf_flat(fp
, self
->parent
, total_samples
);
957 list_for_each_entry(chain
, &self
->val
, list
) {
958 if (chain
->ip
>= PERF_CONTEXT_MAX
)
961 ret
+= fprintf(fp
, " %s\n", chain
->sym
->name
);
963 ret
+= fprintf(fp
, " %p\n",
964 (void *)(long)chain
->ip
);
971 hist_entry_callchain__fprintf(FILE *fp
, struct hist_entry
*self
,
974 struct rb_node
*rb_node
;
975 struct callchain_node
*chain
;
978 rb_node
= rb_first(&self
->sorted_chain
);
982 chain
= rb_entry(rb_node
, struct callchain_node
, rb_node
);
983 percent
= chain
->hit
* 100.0 / total_samples
;
984 switch (callchain_param
.mode
) {
986 ret
+= percent_color_fprintf(fp
, " %6.2f%%\n",
988 ret
+= callchain__fprintf_flat(fp
, chain
, total_samples
);
990 case CHAIN_GRAPH_ABS
: /* Falldown */
991 case CHAIN_GRAPH_REL
:
992 ret
+= callchain__fprintf_graph(fp
, chain
,
993 total_samples
, 1, 1);
997 ret
+= fprintf(fp
, "\n");
998 rb_node
= rb_next(rb_node
);
1006 hist_entry__fprintf(FILE *fp
, struct hist_entry
*self
, u64 total_samples
)
1008 struct sort_entry
*se
;
1011 if (exclude_other
&& !self
->parent
)
1015 ret
= percent_color_fprintf(fp
,
1016 field_sep
? "%.2f" : " %6.2f%%",
1017 (self
->count
* 100.0) / total_samples
);
1019 ret
= fprintf(fp
, field_sep
? "%lld" : "%12lld ", self
->count
);
1021 if (show_nr_samples
) {
1023 fprintf(fp
, "%c%lld", *field_sep
, self
->count
);
1025 fprintf(fp
, "%11lld", self
->count
);
1028 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
1032 fprintf(fp
, "%s", field_sep
?: " ");
1033 ret
+= se
->print(fp
, self
, se
->width
? *se
->width
: 0);
1036 ret
+= fprintf(fp
, "\n");
1039 hist_entry_callchain__fprintf(fp
, self
, total_samples
);
1048 static void dso__calc_col_width(struct dso
*self
)
1050 if (!col_width_list_str
&& !field_sep
&&
1051 (!dso_list
|| strlist__has_entry(dso_list
, self
->name
))) {
1052 unsigned int slen
= strlen(self
->name
);
1053 if (slen
> dsos__col_width
)
1054 dsos__col_width
= slen
;
1057 self
->slen_calculated
= 1;
1060 static struct symbol
*
1061 resolve_symbol(struct thread
*thread
, struct map
**mapp
,
1062 struct dso
**dsop
, u64
*ipp
)
1064 struct dso
*dso
= dsop
? *dsop
: NULL
;
1065 struct map
*map
= mapp
? *mapp
: NULL
;
1077 map
= thread__find_map(thread
, ip
);
1080 * We have to do this here as we may have a dso
1081 * with no symbol hit that has a name longer than
1082 * the ones with symbols sampled.
1084 if (!sort_dso
.elide
&& !map
->dso
->slen_calculated
)
1085 dso__calc_col_width(map
->dso
);
1090 ip
= map
->map_ip(map
, ip
);
1095 * If this is outside of all known maps,
1096 * and is a negative address, try to look it
1097 * up in the kernel dso, as it might be a
1098 * vsyscall (which executes in user-mode):
1100 if ((long long)ip
< 0)
1103 dprintf(" ...... dso: %s\n", dso
? dso
->name
: "<not found>");
1104 dprintf(" ...... map: %Lx -> %Lx\n", *ipp
, ip
);
1113 return dso
->find_symbol(dso
, ip
);
1116 static int call__match(struct symbol
*sym
)
1118 if (sym
->name
&& !regexec(&parent_regex
, sym
->name
, 0, NULL
, 0))
1124 static struct symbol
**
1125 resolve_callchain(struct thread
*thread
, struct map
*map __used
,
1126 struct ip_callchain
*chain
, struct hist_entry
*entry
)
1128 u64 context
= PERF_CONTEXT_MAX
;
1129 struct symbol
**syms
= NULL
;
1133 syms
= calloc(chain
->nr
, sizeof(*syms
));
1135 fprintf(stderr
, "Can't allocate memory for symbols\n");
1140 for (i
= 0; i
< chain
->nr
; i
++) {
1141 u64 ip
= chain
->ips
[i
];
1142 struct dso
*dso
= NULL
;
1145 if (ip
>= PERF_CONTEXT_MAX
) {
1151 case PERF_CONTEXT_HV
:
1152 dso
= hypervisor_dso
;
1154 case PERF_CONTEXT_KERNEL
:
1161 sym
= resolve_symbol(thread
, NULL
, &dso
, &ip
);
1164 if (sort__has_parent
&& call__match(sym
) &&
1166 entry
->parent
= sym
;
1177 * collect histogram counts
1181 hist_entry__add(struct thread
*thread
, struct map
*map
, struct dso
*dso
,
1182 struct symbol
*sym
, u64 ip
, struct ip_callchain
*chain
,
1183 char level
, u64 count
)
1185 struct rb_node
**p
= &hist
.rb_node
;
1186 struct rb_node
*parent
= NULL
;
1187 struct hist_entry
*he
;
1188 struct symbol
**syms
= NULL
;
1189 struct hist_entry entry
= {
1198 .sorted_chain
= RB_ROOT
1202 if ((sort__has_parent
|| callchain
) && chain
)
1203 syms
= resolve_callchain(thread
, map
, chain
, &entry
);
1205 while (*p
!= NULL
) {
1207 he
= rb_entry(parent
, struct hist_entry
, rb_node
);
1209 cmp
= hist_entry__cmp(&entry
, he
);
1214 append_chain(&he
->callchain
, chain
, syms
);
1223 p
= &(*p
)->rb_right
;
1226 he
= malloc(sizeof(*he
));
1231 callchain_init(&he
->callchain
);
1232 append_chain(&he
->callchain
, chain
, syms
);
1235 rb_link_node(&he
->rb_node
, parent
, p
);
1236 rb_insert_color(&he
->rb_node
, &hist
);
1241 static void hist_entry__free(struct hist_entry
*he
)
1247 * collapse the histogram
1250 static struct rb_root collapse_hists
;
1252 static void collapse__insert_entry(struct hist_entry
*he
)
1254 struct rb_node
**p
= &collapse_hists
.rb_node
;
1255 struct rb_node
*parent
= NULL
;
1256 struct hist_entry
*iter
;
1259 while (*p
!= NULL
) {
1261 iter
= rb_entry(parent
, struct hist_entry
, rb_node
);
1263 cmp
= hist_entry__collapse(iter
, he
);
1266 iter
->count
+= he
->count
;
1267 hist_entry__free(he
);
1274 p
= &(*p
)->rb_right
;
1277 rb_link_node(&he
->rb_node
, parent
, p
);
1278 rb_insert_color(&he
->rb_node
, &collapse_hists
);
1281 static void collapse__resort(void)
1283 struct rb_node
*next
;
1284 struct hist_entry
*n
;
1286 if (!sort__need_collapse
)
1289 next
= rb_first(&hist
);
1291 n
= rb_entry(next
, struct hist_entry
, rb_node
);
1292 next
= rb_next(&n
->rb_node
);
1294 rb_erase(&n
->rb_node
, &hist
);
1295 collapse__insert_entry(n
);
1300 * reverse the map, sort on count.
1303 static struct rb_root output_hists
;
1305 static void output__insert_entry(struct hist_entry
*he
, u64 min_callchain_hits
)
1307 struct rb_node
**p
= &output_hists
.rb_node
;
1308 struct rb_node
*parent
= NULL
;
1309 struct hist_entry
*iter
;
1312 callchain_param
.sort(&he
->sorted_chain
, &he
->callchain
,
1313 min_callchain_hits
, &callchain_param
);
1315 while (*p
!= NULL
) {
1317 iter
= rb_entry(parent
, struct hist_entry
, rb_node
);
1319 if (he
->count
> iter
->count
)
1322 p
= &(*p
)->rb_right
;
1325 rb_link_node(&he
->rb_node
, parent
, p
);
1326 rb_insert_color(&he
->rb_node
, &output_hists
);
1329 static void output__resort(u64 total_samples
)
1331 struct rb_node
*next
;
1332 struct hist_entry
*n
;
1333 struct rb_root
*tree
= &hist
;
1334 u64 min_callchain_hits
;
1336 min_callchain_hits
= total_samples
* (callchain_param
.min_percent
/ 100);
1338 if (sort__need_collapse
)
1339 tree
= &collapse_hists
;
1341 next
= rb_first(tree
);
1344 n
= rb_entry(next
, struct hist_entry
, rb_node
);
1345 next
= rb_next(&n
->rb_node
);
1347 rb_erase(&n
->rb_node
, tree
);
1348 output__insert_entry(n
, min_callchain_hits
);
1352 static size_t output__fprintf(FILE *fp
, u64 total_samples
)
1354 struct hist_entry
*pos
;
1355 struct sort_entry
*se
;
1359 char *col_width
= col_width_list_str
;
1361 fprintf(fp
, "# Samples: %Ld\n", (u64
)total_samples
);
1364 fprintf(fp
, "# Overhead");
1365 if (show_nr_samples
) {
1367 fprintf(fp
, "%cSamples", *field_sep
);
1369 fputs(" Samples ", fp
);
1371 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
1375 fprintf(fp
, "%c%s", *field_sep
, se
->header
);
1378 width
= strlen(se
->header
);
1380 if (col_width_list_str
) {
1382 *se
->width
= atoi(col_width
);
1383 col_width
= strchr(col_width
, ',');
1388 width
= *se
->width
= max(*se
->width
, width
);
1390 fprintf(fp
, " %*s", width
, se
->header
);
1397 fprintf(fp
, "# ........");
1398 if (show_nr_samples
)
1399 fprintf(fp
, " ..........");
1400 list_for_each_entry(se
, &hist_entry__sort_list
, list
) {
1410 width
= strlen(se
->header
);
1411 for (i
= 0; i
< width
; i
++)
1419 for (nd
= rb_first(&output_hists
); nd
; nd
= rb_next(nd
)) {
1420 pos
= rb_entry(nd
, struct hist_entry
, rb_node
);
1421 ret
+= hist_entry__fprintf(fp
, pos
, total_samples
);
1424 if (sort_order
== default_sort_order
&&
1425 parent_pattern
== default_parent_pattern
) {
1427 fprintf(fp
, "# (For a higher level overview, try: perf report --sort comm,dso)\n");
1435 static void register_idle_thread(void)
1437 struct thread
*thread
= threads__findnew(0);
1439 if (thread
== NULL
||
1440 thread__set_comm(thread
, "[idle]")) {
1441 fprintf(stderr
, "problem inserting idle task.\n");
1446 static unsigned long total
= 0,
1453 static int validate_chain(struct ip_callchain
*chain
, event_t
*event
)
1455 unsigned int chain_size
;
1457 chain_size
= event
->header
.size
;
1458 chain_size
-= (unsigned long)&event
->ip
.__more_data
- (unsigned long)event
;
1460 if (chain
->nr
*sizeof(u64
) > chain_size
)
1467 process_sample_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1471 struct dso
*dso
= NULL
;
1472 struct thread
*thread
= threads__findnew(event
->ip
.pid
);
1473 u64 ip
= event
->ip
.ip
;
1475 struct map
*map
= NULL
;
1476 void *more_data
= event
->ip
.__more_data
;
1477 struct ip_callchain
*chain
= NULL
;
1480 if (sample_type
& PERF_SAMPLE_PERIOD
) {
1481 period
= *(u64
*)more_data
;
1482 more_data
+= sizeof(u64
);
1485 dprintf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d: %p period: %Ld\n",
1486 (void *)(offset
+ head
),
1487 (void *)(long)(event
->header
.size
),
1493 if (sample_type
& PERF_SAMPLE_CALLCHAIN
) {
1496 chain
= (void *)more_data
;
1498 dprintf("... chain: nr:%Lu\n", chain
->nr
);
1500 if (validate_chain(chain
, event
) < 0) {
1501 eprintf("call-chain problem with event, skipping it.\n");
1506 for (i
= 0; i
< chain
->nr
; i
++)
1507 dprintf("..... %2d: %016Lx\n", i
, chain
->ips
[i
]);
1511 dprintf(" ... thread: %s:%d\n", thread
->comm
, thread
->pid
);
1513 if (thread
== NULL
) {
1514 eprintf("problem processing %d event, skipping it.\n",
1515 event
->header
.type
);
1519 if (comm_list
&& !strlist__has_entry(comm_list
, thread
->comm
))
1522 cpumode
= event
->header
.misc
& PERF_EVENT_MISC_CPUMODE_MASK
;
1524 if (cpumode
== PERF_EVENT_MISC_KERNEL
) {
1530 dprintf(" ...... dso: %s\n", dso
->name
);
1532 } else if (cpumode
== PERF_EVENT_MISC_USER
) {
1541 dso
= hypervisor_dso
;
1543 dprintf(" ...... dso: [hypervisor]\n");
1546 if (show
& show_mask
) {
1547 struct symbol
*sym
= resolve_symbol(thread
, &map
, &dso
, &ip
);
1549 if (dso_list
&& dso
&& dso
->name
&& !strlist__has_entry(dso_list
, dso
->name
))
1552 if (sym_list
&& sym
&& !strlist__has_entry(sym_list
, sym
->name
))
1555 if (hist_entry__add(thread
, map
, dso
, sym
, ip
, chain
, level
, period
)) {
1556 eprintf("problem incrementing symbol count, skipping event\n");
1566 process_mmap_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1568 struct thread
*thread
= threads__findnew(event
->mmap
.pid
);
1569 struct map
*map
= map__new(&event
->mmap
);
1571 dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n",
1572 (void *)(offset
+ head
),
1573 (void *)(long)(event
->header
.size
),
1575 (void *)(long)event
->mmap
.start
,
1576 (void *)(long)event
->mmap
.len
,
1577 (void *)(long)event
->mmap
.pgoff
,
1578 event
->mmap
.filename
);
1580 if (thread
== NULL
|| map
== NULL
) {
1581 dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1585 thread__insert_map(thread
, map
);
1592 process_comm_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1594 struct thread
*thread
= threads__findnew(event
->comm
.pid
);
1596 dprintf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
1597 (void *)(offset
+ head
),
1598 (void *)(long)(event
->header
.size
),
1599 event
->comm
.comm
, event
->comm
.pid
);
1601 if (thread
== NULL
||
1602 thread__set_comm(thread
, event
->comm
.comm
)) {
1603 dprintf("problem processing PERF_EVENT_COMM, skipping event.\n");
1612 process_task_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1614 struct thread
*thread
= threads__findnew(event
->fork
.pid
);
1615 struct thread
*parent
= threads__findnew(event
->fork
.ppid
);
1617 dprintf("%p [%p]: PERF_EVENT_%s: (%d:%d):(%d:%d)\n",
1618 (void *)(offset
+ head
),
1619 (void *)(long)(event
->header
.size
),
1620 event
->header
.type
== PERF_EVENT_FORK
? "FORK" : "EXIT",
1621 event
->fork
.pid
, event
->fork
.tid
,
1622 event
->fork
.ppid
, event
->fork
.ptid
);
1625 * A thread clone will have the same PID for both
1628 if (thread
== parent
)
1631 if (event
->header
.type
== PERF_EVENT_EXIT
)
1634 if (!thread
|| !parent
|| thread__fork(thread
, parent
)) {
1635 dprintf("problem processing PERF_EVENT_FORK, skipping event.\n");
1644 process_lost_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1646 dprintf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n",
1647 (void *)(offset
+ head
),
1648 (void *)(long)(event
->header
.size
),
1652 total_lost
+= event
->lost
.lost
;
1657 static void trace_event(event_t
*event
)
1659 unsigned char *raw_event
= (void *)event
;
1660 char *color
= PERF_COLOR_BLUE
;
1667 cdprintf("\n. ... raw event: size %d bytes\n", event
->header
.size
);
1669 for (i
= 0; i
< event
->header
.size
; i
++) {
1670 if ((i
& 15) == 0) {
1672 cdprintf(" %04x: ", i
);
1675 cdprintf(" %02x", raw_event
[i
]);
1677 if (((i
& 15) == 15) || i
== event
->header
.size
-1) {
1679 for (j
= 0; j
< 15-(i
& 15); j
++)
1681 for (j
= 0; j
< (i
& 15); j
++) {
1682 if (isprint(raw_event
[i
-15+j
]))
1683 cdprintf("%c", raw_event
[i
-15+j
]);
1694 process_read_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1696 dprintf("%p [%p]: PERF_EVENT_READ: %d %d %Lu\n",
1697 (void *)(offset
+ head
),
1698 (void *)(long)(event
->header
.size
),
1707 process_event(event_t
*event
, unsigned long offset
, unsigned long head
)
1711 switch (event
->header
.type
) {
1712 case PERF_EVENT_SAMPLE
:
1713 return process_sample_event(event
, offset
, head
);
1715 case PERF_EVENT_MMAP
:
1716 return process_mmap_event(event
, offset
, head
);
1718 case PERF_EVENT_COMM
:
1719 return process_comm_event(event
, offset
, head
);
1721 case PERF_EVENT_FORK
:
1722 case PERF_EVENT_EXIT
:
1723 return process_task_event(event
, offset
, head
);
1725 case PERF_EVENT_LOST
:
1726 return process_lost_event(event
, offset
, head
);
1728 case PERF_EVENT_READ
:
1729 return process_read_event(event
, offset
, head
);
1732 * We dont process them right now but they are fine:
1735 case PERF_EVENT_THROTTLE
:
1736 case PERF_EVENT_UNTHROTTLE
:
1746 static struct perf_header
*header
;
1748 static u64
perf_header__sample_type(void)
1750 u64 sample_type
= 0;
1753 for (i
= 0; i
< header
->attrs
; i
++) {
1754 struct perf_header_attr
*attr
= header
->attr
[i
];
1757 sample_type
= attr
->attr
.sample_type
;
1758 else if (sample_type
!= attr
->attr
.sample_type
)
1759 die("non matching sample_type");
1765 static int __cmd_report(void)
1767 int ret
, rc
= EXIT_FAILURE
;
1768 unsigned long offset
= 0;
1769 unsigned long head
, shift
;
1775 register_idle_thread();
1777 input
= open(input_name
, O_RDONLY
);
1779 fprintf(stderr
, " failed to open file: %s", input_name
);
1780 if (!strcmp(input_name
, "perf.data"))
1781 fprintf(stderr
, " (try 'perf record' first)");
1782 fprintf(stderr
, "\n");
1786 ret
= fstat(input
, &stat
);
1788 perror("failed to stat file");
1792 if (!stat
.st_size
) {
1793 fprintf(stderr
, "zero-sized file, nothing to do!\n");
1797 header
= perf_header__read(input
);
1798 head
= header
->data_offset
;
1800 sample_type
= perf_header__sample_type();
1802 if (!(sample_type
& PERF_SAMPLE_CALLCHAIN
)) {
1803 if (sort__has_parent
) {
1804 fprintf(stderr
, "selected --sort parent, but no"
1805 " callchain data. Did you call"
1806 " perf record without -g?\n");
1810 fprintf(stderr
, "selected -c but no callchain data."
1811 " Did you call perf record without"
1817 if (load_kernel() < 0) {
1818 perror("failed to load kernel symbols");
1819 return EXIT_FAILURE
;
1823 if (getcwd(__cwd
, sizeof(__cwd
)) == NULL
) {
1824 perror("failed to get the current directory");
1825 return EXIT_FAILURE
;
1827 cwdlen
= strlen(cwd
);
1833 shift
= page_size
* (head
/ page_size
);
1838 buf
= (char *)mmap(NULL
, page_size
* mmap_window
, PROT_READ
,
1839 MAP_SHARED
, input
, offset
);
1840 if (buf
== MAP_FAILED
) {
1841 perror("failed to mmap file");
1846 event
= (event_t
*)(buf
+ head
);
1848 size
= event
->header
.size
;
1852 if (head
+ event
->header
.size
>= page_size
* mmap_window
) {
1855 shift
= page_size
* (head
/ page_size
);
1857 ret
= munmap(buf
, page_size
* mmap_window
);
1865 size
= event
->header
.size
;
1867 dprintf("\n%p [%p]: event: %d\n",
1868 (void *)(offset
+ head
),
1869 (void *)(long)event
->header
.size
,
1870 event
->header
.type
);
1872 if (!size
|| process_event(event
, offset
, head
) < 0) {
1874 dprintf("%p [%p]: skipping unknown header type: %d\n",
1875 (void *)(offset
+ head
),
1876 (void *)(long)(event
->header
.size
),
1877 event
->header
.type
);
1882 * assume we lost track of the stream, check alignment, and
1883 * increment a single u64 in the hope to catch on again 'soon'.
1886 if (unlikely(head
& 7))
1894 if (offset
+ head
>= header
->data_offset
+ header
->data_size
)
1897 if (offset
+ head
< (unsigned long)stat
.st_size
)
1904 dprintf(" IP events: %10ld\n", total
);
1905 dprintf(" mmap events: %10ld\n", total_mmap
);
1906 dprintf(" comm events: %10ld\n", total_comm
);
1907 dprintf(" fork events: %10ld\n", total_fork
);
1908 dprintf(" lost events: %10ld\n", total_lost
);
1909 dprintf(" unknown events: %10ld\n", total_unknown
);
1915 threads__fprintf(stdout
);
1918 dsos__fprintf(stdout
);
1921 output__resort(total
);
1922 output__fprintf(stdout
, total
);
1928 parse_callchain_opt(const struct option
*opt __used
, const char *arg
,
1939 tok
= strtok((char *)arg
, ",");
1943 /* get the output mode */
1944 if (!strncmp(tok
, "graph", strlen(arg
)))
1945 callchain_param
.mode
= CHAIN_GRAPH_ABS
;
1947 else if (!strncmp(tok
, "flat", strlen(arg
)))
1948 callchain_param
.mode
= CHAIN_FLAT
;
1950 else if (!strncmp(tok
, "fractal", strlen(arg
)))
1951 callchain_param
.mode
= CHAIN_GRAPH_REL
;
1956 /* get the min percentage */
1957 tok
= strtok(NULL
, ",");
1961 callchain_param
.min_percent
= strtod(tok
, &endptr
);
1966 if (register_callchain_param(&callchain_param
) < 0) {
1967 fprintf(stderr
, "Can't register callchain params\n");
1973 static const char * const report_usage
[] = {
1974 "perf report [<options>] <command>",
1978 static const struct option options
[] = {
1979 OPT_STRING('i', "input", &input_name
, "file",
1981 OPT_BOOLEAN('v', "verbose", &verbose
,
1982 "be more verbose (show symbol address, etc)"),
1983 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace
,
1984 "dump raw trace in ASCII"),
1985 OPT_STRING('k', "vmlinux", &vmlinux
, "file", "vmlinux pathname"),
1986 OPT_BOOLEAN('m', "modules", &modules
,
1987 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1988 OPT_BOOLEAN('n', "show-nr-samples", &show_nr_samples
,
1989 "Show a column with the number of samples"),
1990 OPT_STRING('s', "sort", &sort_order
, "key[,key2...]",
1991 "sort by key(s): pid, comm, dso, symbol, parent"),
1992 OPT_BOOLEAN('P', "full-paths", &full_paths
,
1993 "Don't shorten the pathnames taking into account the cwd"),
1994 OPT_STRING('p', "parent", &parent_pattern
, "regex",
1995 "regex filter to identify parent, see: '--sort parent'"),
1996 OPT_BOOLEAN('x', "exclude-other", &exclude_other
,
1997 "Only display entries with parent-match"),
1998 OPT_CALLBACK_DEFAULT('g', "call-graph", NULL
, "output_type,min_percent",
1999 "Display callchains using output_type and min percent threshold. "
2000 "Default: fractal,0.5", &parse_callchain_opt
, callchain_default_opt
),
2001 OPT_STRING('d', "dsos", &dso_list_str
, "dso[,dso...]",
2002 "only consider symbols in these dsos"),
2003 OPT_STRING('C', "comms", &comm_list_str
, "comm[,comm...]",
2004 "only consider symbols in these comms"),
2005 OPT_STRING('S', "symbols", &sym_list_str
, "symbol[,symbol...]",
2006 "only consider these symbols"),
2007 OPT_STRING('w', "column-widths", &col_width_list_str
,
2009 "don't try to adjust column width, use these fixed values"),
2010 OPT_STRING('t', "field-separator", &field_sep
, "separator",
2011 "separator for columns, no spaces will be added between "
2012 "columns '.' is reserved."),
2016 static void setup_sorting(void)
2018 char *tmp
, *tok
, *str
= strdup(sort_order
);
2020 for (tok
= strtok_r(str
, ", ", &tmp
);
2021 tok
; tok
= strtok_r(NULL
, ", ", &tmp
)) {
2022 if (sort_dimension__add(tok
) < 0) {
2023 error("Unknown --sort key: `%s'", tok
);
2024 usage_with_options(report_usage
, options
);
2031 static void setup_list(struct strlist
**list
, const char *list_str
,
2032 struct sort_entry
*se
, const char *list_name
,
2036 *list
= strlist__new(true, list_str
);
2038 fprintf(stderr
, "problems parsing %s list\n",
2042 if (strlist__nr_entries(*list
) == 1) {
2043 fprintf(fp
, "# %s: %s\n", list_name
,
2044 strlist__entry(*list
, 0)->s
);
2050 int cmd_report(int argc
, const char **argv
, const char *prefix __used
)
2054 page_size
= getpagesize();
2056 argc
= parse_options(argc
, argv
, options
, report_usage
, 0);
2060 if (parent_pattern
!= default_parent_pattern
) {
2061 sort_dimension__add("parent");
2062 sort_parent
.elide
= 1;
2067 * Any (unrecognized) arguments left?
2070 usage_with_options(report_usage
, options
);
2074 setup_list(&dso_list
, dso_list_str
, &sort_dso
, "dso", stdout
);
2075 setup_list(&comm_list
, comm_list_str
, &sort_comm
, "comm", stdout
);
2076 setup_list(&sym_list
, sym_list_str
, &sort_sym
, "symbol", stdout
);
2078 if (field_sep
&& *field_sep
== '.') {
2079 fputs("'.' is the only non valid --field-separator argument\n",
2084 return __cmd_report();