perf report: Add --symbols parameter
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / tools / perf / builtin-report.c
blob135b7837e6bf3de20575d8c28bf0ee63037aab9e
1 /*
2 * builtin-report.c
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.
7 */
8 #include "builtin.h"
10 #include "util/util.h"
12 #include "util/color.h"
13 #include "util/list.h"
14 #include "util/cache.h"
15 #include "util/rbtree.h"
16 #include "util/symbol.h"
17 #include "util/string.h"
18 #include "util/callchain.h"
19 #include "util/strlist.h"
21 #include "perf.h"
22 #include "util/header.h"
24 #include "util/parse-options.h"
25 #include "util/parse-events.h"
27 #define SHOW_KERNEL 1
28 #define SHOW_USER 2
29 #define SHOW_HV 4
31 static char const *input_name = "perf.data";
32 static char *vmlinux = NULL;
34 static char default_sort_order[] = "comm,dso";
35 static char *sort_order = default_sort_order;
36 static char *dso_list_str, *comm_list_str, *sym_list_str;
37 static struct strlist *dso_list, *comm_list, *sym_list;
39 static int input;
40 static int show_mask = SHOW_KERNEL | SHOW_USER | SHOW_HV;
42 static int dump_trace = 0;
43 #define dprintf(x...) do { if (dump_trace) printf(x); } while (0)
44 #define cdprintf(x...) do { if (dump_trace) color_fprintf(stdout, color, x); } while (0)
46 static int verbose;
47 #define eprintf(x...) do { if (verbose) fprintf(stderr, x); } while (0)
49 static int full_paths;
51 static unsigned long page_size;
52 static unsigned long mmap_window = 32;
54 static char default_parent_pattern[] = "^sys_|^do_page_fault";
55 static char *parent_pattern = default_parent_pattern;
56 static regex_t parent_regex;
58 static int exclude_other = 1;
59 static int callchain;
61 static u64 sample_type;
63 struct ip_event {
64 struct perf_event_header header;
65 u64 ip;
66 u32 pid, tid;
67 unsigned char __more_data[];
70 struct mmap_event {
71 struct perf_event_header header;
72 u32 pid, tid;
73 u64 start;
74 u64 len;
75 u64 pgoff;
76 char filename[PATH_MAX];
79 struct comm_event {
80 struct perf_event_header header;
81 u32 pid, tid;
82 char comm[16];
85 struct fork_event {
86 struct perf_event_header header;
87 u32 pid, ppid;
90 struct period_event {
91 struct perf_event_header header;
92 u64 time;
93 u64 id;
94 u64 sample_period;
97 struct lost_event {
98 struct perf_event_header header;
99 u64 id;
100 u64 lost;
103 struct read_event {
104 struct perf_event_header header;
105 u32 pid,tid;
106 u64 value;
107 u64 format[3];
110 typedef union event_union {
111 struct perf_event_header header;
112 struct ip_event ip;
113 struct mmap_event mmap;
114 struct comm_event comm;
115 struct fork_event fork;
116 struct period_event period;
117 struct lost_event lost;
118 struct read_event read;
119 } event_t;
121 static LIST_HEAD(dsos);
122 static struct dso *kernel_dso;
123 static struct dso *vdso;
125 static void dsos__add(struct dso *dso)
127 list_add_tail(&dso->node, &dsos);
130 static struct dso *dsos__find(const char *name)
132 struct dso *pos;
134 list_for_each_entry(pos, &dsos, node)
135 if (strcmp(pos->name, name) == 0)
136 return pos;
137 return NULL;
140 static struct dso *dsos__findnew(const char *name)
142 struct dso *dso = dsos__find(name);
143 int nr;
145 if (dso)
146 return dso;
148 dso = dso__new(name, 0);
149 if (!dso)
150 goto out_delete_dso;
152 nr = dso__load(dso, NULL, verbose);
153 if (nr < 0) {
154 eprintf("Failed to open: %s\n", name);
155 goto out_delete_dso;
157 if (!nr)
158 eprintf("No symbols found in: %s, maybe install a debug package?\n", name);
160 dsos__add(dso);
162 return dso;
164 out_delete_dso:
165 dso__delete(dso);
166 return NULL;
169 static void dsos__fprintf(FILE *fp)
171 struct dso *pos;
173 list_for_each_entry(pos, &dsos, node)
174 dso__fprintf(pos, fp);
177 static struct symbol *vdso__find_symbol(struct dso *dso, u64 ip)
179 return dso__find_symbol(kernel_dso, ip);
182 static int load_kernel(void)
184 int err;
186 kernel_dso = dso__new("[kernel]", 0);
187 if (!kernel_dso)
188 return -1;
190 err = dso__load_kernel(kernel_dso, vmlinux, NULL, verbose);
191 if (err) {
192 dso__delete(kernel_dso);
193 kernel_dso = NULL;
194 } else
195 dsos__add(kernel_dso);
197 vdso = dso__new("[vdso]", 0);
198 if (!vdso)
199 return -1;
201 vdso->find_symbol = vdso__find_symbol;
203 dsos__add(vdso);
205 return err;
208 static char __cwd[PATH_MAX];
209 static char *cwd = __cwd;
210 static int cwdlen;
212 static int strcommon(const char *pathname)
214 int n = 0;
216 while (pathname[n] == cwd[n] && n < cwdlen)
217 ++n;
219 return n;
222 struct map {
223 struct list_head node;
224 u64 start;
225 u64 end;
226 u64 pgoff;
227 u64 (*map_ip)(struct map *, u64);
228 struct dso *dso;
231 static u64 map__map_ip(struct map *map, u64 ip)
233 return ip - map->start + map->pgoff;
236 static u64 vdso__map_ip(struct map *map, u64 ip)
238 return ip;
241 static inline int is_anon_memory(const char *filename)
243 return strcmp(filename, "//anon") == 0;
246 static struct map *map__new(struct mmap_event *event)
248 struct map *self = malloc(sizeof(*self));
250 if (self != NULL) {
251 const char *filename = event->filename;
252 char newfilename[PATH_MAX];
253 int anon;
255 if (cwd) {
256 int n = strcommon(filename);
258 if (n == cwdlen) {
259 snprintf(newfilename, sizeof(newfilename),
260 ".%s", filename + n);
261 filename = newfilename;
265 anon = is_anon_memory(filename);
267 if (anon) {
268 snprintf(newfilename, sizeof(newfilename), "/tmp/perf-%d.map", event->pid);
269 filename = newfilename;
272 self->start = event->start;
273 self->end = event->start + event->len;
274 self->pgoff = event->pgoff;
276 self->dso = dsos__findnew(filename);
277 if (self->dso == NULL)
278 goto out_delete;
280 if (self->dso == vdso || anon)
281 self->map_ip = vdso__map_ip;
282 else
283 self->map_ip = map__map_ip;
285 return self;
286 out_delete:
287 free(self);
288 return NULL;
291 static struct map *map__clone(struct map *self)
293 struct map *map = malloc(sizeof(*self));
295 if (!map)
296 return NULL;
298 memcpy(map, self, sizeof(*self));
300 return map;
303 static int map__overlap(struct map *l, struct map *r)
305 if (l->start > r->start) {
306 struct map *t = l;
307 l = r;
308 r = t;
311 if (l->end > r->start)
312 return 1;
314 return 0;
317 static size_t map__fprintf(struct map *self, FILE *fp)
319 return fprintf(fp, " %Lx-%Lx %Lx %s\n",
320 self->start, self->end, self->pgoff, self->dso->name);
324 struct thread {
325 struct rb_node rb_node;
326 struct list_head maps;
327 pid_t pid;
328 char *comm;
331 static struct thread *thread__new(pid_t pid)
333 struct thread *self = malloc(sizeof(*self));
335 if (self != NULL) {
336 self->pid = pid;
337 self->comm = malloc(32);
338 if (self->comm)
339 snprintf(self->comm, 32, ":%d", self->pid);
340 INIT_LIST_HEAD(&self->maps);
343 return self;
346 static int thread__set_comm(struct thread *self, const char *comm)
348 if (self->comm)
349 free(self->comm);
350 self->comm = strdup(comm);
351 return self->comm ? 0 : -ENOMEM;
354 static size_t thread__fprintf(struct thread *self, FILE *fp)
356 struct map *pos;
357 size_t ret = fprintf(fp, "Thread %d %s\n", self->pid, self->comm);
359 list_for_each_entry(pos, &self->maps, node)
360 ret += map__fprintf(pos, fp);
362 return ret;
366 static struct rb_root threads;
367 static struct thread *last_match;
369 static struct thread *threads__findnew(pid_t pid)
371 struct rb_node **p = &threads.rb_node;
372 struct rb_node *parent = NULL;
373 struct thread *th;
376 * Font-end cache - PID lookups come in blocks,
377 * so most of the time we dont have to look up
378 * the full rbtree:
380 if (last_match && last_match->pid == pid)
381 return last_match;
383 while (*p != NULL) {
384 parent = *p;
385 th = rb_entry(parent, struct thread, rb_node);
387 if (th->pid == pid) {
388 last_match = th;
389 return th;
392 if (pid < th->pid)
393 p = &(*p)->rb_left;
394 else
395 p = &(*p)->rb_right;
398 th = thread__new(pid);
399 if (th != NULL) {
400 rb_link_node(&th->rb_node, parent, p);
401 rb_insert_color(&th->rb_node, &threads);
402 last_match = th;
405 return th;
408 static void thread__insert_map(struct thread *self, struct map *map)
410 struct map *pos, *tmp;
412 list_for_each_entry_safe(pos, tmp, &self->maps, node) {
413 if (map__overlap(pos, map)) {
414 if (verbose >= 2) {
415 printf("overlapping maps:\n");
416 map__fprintf(map, stdout);
417 map__fprintf(pos, stdout);
420 if (map->start <= pos->start && map->end > pos->start)
421 pos->start = map->end;
423 if (map->end >= pos->end && map->start < pos->end)
424 pos->end = map->start;
426 if (verbose >= 2) {
427 printf("after collision:\n");
428 map__fprintf(pos, stdout);
431 if (pos->start >= pos->end) {
432 list_del_init(&pos->node);
433 free(pos);
438 list_add_tail(&map->node, &self->maps);
441 static int thread__fork(struct thread *self, struct thread *parent)
443 struct map *map;
445 if (self->comm)
446 free(self->comm);
447 self->comm = strdup(parent->comm);
448 if (!self->comm)
449 return -ENOMEM;
451 list_for_each_entry(map, &parent->maps, node) {
452 struct map *new = map__clone(map);
453 if (!new)
454 return -ENOMEM;
455 thread__insert_map(self, new);
458 return 0;
461 static struct map *thread__find_map(struct thread *self, u64 ip)
463 struct map *pos;
465 if (self == NULL)
466 return NULL;
468 list_for_each_entry(pos, &self->maps, node)
469 if (ip >= pos->start && ip <= pos->end)
470 return pos;
472 return NULL;
475 static size_t threads__fprintf(FILE *fp)
477 size_t ret = 0;
478 struct rb_node *nd;
480 for (nd = rb_first(&threads); nd; nd = rb_next(nd)) {
481 struct thread *pos = rb_entry(nd, struct thread, rb_node);
483 ret += thread__fprintf(pos, fp);
486 return ret;
490 * histogram, sorted on item, collects counts
493 static struct rb_root hist;
495 struct hist_entry {
496 struct rb_node rb_node;
498 struct thread *thread;
499 struct map *map;
500 struct dso *dso;
501 struct symbol *sym;
502 struct symbol *parent;
503 u64 ip;
504 char level;
505 struct callchain_node callchain;
506 struct rb_root sorted_chain;
508 u64 count;
512 * configurable sorting bits
515 struct sort_entry {
516 struct list_head list;
518 char *header;
520 int64_t (*cmp)(struct hist_entry *, struct hist_entry *);
521 int64_t (*collapse)(struct hist_entry *, struct hist_entry *);
522 size_t (*print)(FILE *fp, struct hist_entry *);
525 static int64_t cmp_null(void *l, void *r)
527 if (!l && !r)
528 return 0;
529 else if (!l)
530 return -1;
531 else
532 return 1;
535 /* --sort pid */
537 static int64_t
538 sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
540 return right->thread->pid - left->thread->pid;
543 static size_t
544 sort__thread_print(FILE *fp, struct hist_entry *self)
546 return fprintf(fp, "%16s:%5d", self->thread->comm ?: "", self->thread->pid);
549 static struct sort_entry sort_thread = {
550 .header = " Command: Pid",
551 .cmp = sort__thread_cmp,
552 .print = sort__thread_print,
555 /* --sort comm */
557 static int64_t
558 sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
560 return right->thread->pid - left->thread->pid;
563 static int64_t
564 sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
566 char *comm_l = left->thread->comm;
567 char *comm_r = right->thread->comm;
569 if (!comm_l || !comm_r)
570 return cmp_null(comm_l, comm_r);
572 return strcmp(comm_l, comm_r);
575 static size_t
576 sort__comm_print(FILE *fp, struct hist_entry *self)
578 return fprintf(fp, "%16s", self->thread->comm);
581 static struct sort_entry sort_comm = {
582 .header = " Command",
583 .cmp = sort__comm_cmp,
584 .collapse = sort__comm_collapse,
585 .print = sort__comm_print,
588 /* --sort dso */
590 static int64_t
591 sort__dso_cmp(struct hist_entry *left, struct hist_entry *right)
593 struct dso *dso_l = left->dso;
594 struct dso *dso_r = right->dso;
596 if (!dso_l || !dso_r)
597 return cmp_null(dso_l, dso_r);
599 return strcmp(dso_l->name, dso_r->name);
602 static size_t
603 sort__dso_print(FILE *fp, struct hist_entry *self)
605 if (self->dso)
606 return fprintf(fp, "%-25s", self->dso->name);
608 return fprintf(fp, "%016llx ", (u64)self->ip);
611 static struct sort_entry sort_dso = {
612 .header = "Shared Object ",
613 .cmp = sort__dso_cmp,
614 .print = sort__dso_print,
617 /* --sort symbol */
619 static int64_t
620 sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
622 u64 ip_l, ip_r;
624 if (left->sym == right->sym)
625 return 0;
627 ip_l = left->sym ? left->sym->start : left->ip;
628 ip_r = right->sym ? right->sym->start : right->ip;
630 return (int64_t)(ip_r - ip_l);
633 static size_t
634 sort__sym_print(FILE *fp, struct hist_entry *self)
636 size_t ret = 0;
638 if (verbose)
639 ret += fprintf(fp, "%#018llx ", (u64)self->ip);
641 if (self->sym) {
642 ret += fprintf(fp, "[%c] %s",
643 self->dso == kernel_dso ? 'k' : '.', self->sym->name);
644 } else {
645 ret += fprintf(fp, "%#016llx", (u64)self->ip);
648 return ret;
651 static struct sort_entry sort_sym = {
652 .header = "Symbol",
653 .cmp = sort__sym_cmp,
654 .print = sort__sym_print,
657 /* --sort parent */
659 static int64_t
660 sort__parent_cmp(struct hist_entry *left, struct hist_entry *right)
662 struct symbol *sym_l = left->parent;
663 struct symbol *sym_r = right->parent;
665 if (!sym_l || !sym_r)
666 return cmp_null(sym_l, sym_r);
668 return strcmp(sym_l->name, sym_r->name);
671 static size_t
672 sort__parent_print(FILE *fp, struct hist_entry *self)
674 size_t ret = 0;
676 ret += fprintf(fp, "%-20s", self->parent ? self->parent->name : "[other]");
678 return ret;
681 static struct sort_entry sort_parent = {
682 .header = "Parent symbol ",
683 .cmp = sort__parent_cmp,
684 .print = sort__parent_print,
687 static int sort__need_collapse = 0;
688 static int sort__has_parent = 0;
690 struct sort_dimension {
691 char *name;
692 struct sort_entry *entry;
693 int taken;
696 static struct sort_dimension sort_dimensions[] = {
697 { .name = "pid", .entry = &sort_thread, },
698 { .name = "comm", .entry = &sort_comm, },
699 { .name = "dso", .entry = &sort_dso, },
700 { .name = "symbol", .entry = &sort_sym, },
701 { .name = "parent", .entry = &sort_parent, },
704 static LIST_HEAD(hist_entry__sort_list);
706 static int sort_dimension__add(char *tok)
708 int i;
710 for (i = 0; i < ARRAY_SIZE(sort_dimensions); i++) {
711 struct sort_dimension *sd = &sort_dimensions[i];
713 if (sd->taken)
714 continue;
716 if (strncasecmp(tok, sd->name, strlen(tok)))
717 continue;
719 if (sd->entry->collapse)
720 sort__need_collapse = 1;
722 if (sd->entry == &sort_parent) {
723 int ret = regcomp(&parent_regex, parent_pattern, REG_EXTENDED);
724 if (ret) {
725 char err[BUFSIZ];
727 regerror(ret, &parent_regex, err, sizeof(err));
728 fprintf(stderr, "Invalid regex: %s\n%s",
729 parent_pattern, err);
730 exit(-1);
732 sort__has_parent = 1;
735 list_add_tail(&sd->entry->list, &hist_entry__sort_list);
736 sd->taken = 1;
738 return 0;
741 return -ESRCH;
744 static int64_t
745 hist_entry__cmp(struct hist_entry *left, struct hist_entry *right)
747 struct sort_entry *se;
748 int64_t cmp = 0;
750 list_for_each_entry(se, &hist_entry__sort_list, list) {
751 cmp = se->cmp(left, right);
752 if (cmp)
753 break;
756 return cmp;
759 static int64_t
760 hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
762 struct sort_entry *se;
763 int64_t cmp = 0;
765 list_for_each_entry(se, &hist_entry__sort_list, list) {
766 int64_t (*f)(struct hist_entry *, struct hist_entry *);
768 f = se->collapse ?: se->cmp;
770 cmp = f(left, right);
771 if (cmp)
772 break;
775 return cmp;
778 static size_t
779 callchain__fprintf(FILE *fp, struct callchain_node *self, u64 total_samples)
781 struct callchain_list *chain;
782 size_t ret = 0;
784 if (!self)
785 return 0;
787 ret += callchain__fprintf(fp, self->parent, total_samples);
790 list_for_each_entry(chain, &self->val, list)
791 ret += fprintf(fp, " %p\n", (void *)chain->ip);
793 return ret;
796 static size_t
797 hist_entry_callchain__fprintf(FILE *fp, struct hist_entry *self,
798 u64 total_samples)
800 struct rb_node *rb_node;
801 struct callchain_node *chain;
802 size_t ret = 0;
804 rb_node = rb_first(&self->sorted_chain);
805 while (rb_node) {
806 double percent;
808 chain = rb_entry(rb_node, struct callchain_node, rb_node);
809 percent = chain->hit * 100.0 / total_samples;
810 ret += fprintf(fp, " %6.2f%%\n", percent);
811 ret += callchain__fprintf(fp, chain, total_samples);
812 ret += fprintf(fp, "\n");
813 rb_node = rb_next(rb_node);
816 return ret;
820 static size_t
821 hist_entry__fprintf(FILE *fp, struct hist_entry *self, u64 total_samples)
823 struct sort_entry *se;
824 size_t ret;
826 if (exclude_other && !self->parent)
827 return 0;
829 if (total_samples) {
830 double percent = self->count * 100.0 / total_samples;
831 char *color = PERF_COLOR_NORMAL;
834 * We color high-overhead entries in red, mid-overhead
835 * entries in green - and keep the low overhead places
836 * normal:
838 if (percent >= 5.0) {
839 color = PERF_COLOR_RED;
840 } else {
841 if (percent >= 0.5)
842 color = PERF_COLOR_GREEN;
845 ret = color_fprintf(fp, color, " %6.2f%%",
846 (self->count * 100.0) / total_samples);
847 } else
848 ret = fprintf(fp, "%12Ld ", self->count);
850 list_for_each_entry(se, &hist_entry__sort_list, list) {
851 if (exclude_other && (se == &sort_parent))
852 continue;
854 fprintf(fp, " ");
855 ret += se->print(fp, self);
858 ret += fprintf(fp, "\n");
860 if (callchain)
861 hist_entry_callchain__fprintf(fp, self, total_samples);
863 return ret;
870 static struct symbol *
871 resolve_symbol(struct thread *thread, struct map **mapp,
872 struct dso **dsop, u64 *ipp)
874 struct dso *dso = dsop ? *dsop : NULL;
875 struct map *map = mapp ? *mapp : NULL;
876 u64 ip = *ipp;
878 if (!thread)
879 return NULL;
881 if (dso)
882 goto got_dso;
884 if (map)
885 goto got_map;
887 map = thread__find_map(thread, ip);
888 if (map != NULL) {
889 if (mapp)
890 *mapp = map;
891 got_map:
892 ip = map->map_ip(map, ip);
894 dso = map->dso;
895 } else {
897 * If this is outside of all known maps,
898 * and is a negative address, try to look it
899 * up in the kernel dso, as it might be a
900 * vsyscall (which executes in user-mode):
902 if ((long long)ip < 0)
903 dso = kernel_dso;
905 dprintf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
906 dprintf(" ...... map: %Lx -> %Lx\n", *ipp, ip);
907 *ipp = ip;
909 if (dsop)
910 *dsop = dso;
912 if (!dso)
913 return NULL;
914 got_dso:
915 return dso->find_symbol(dso, ip);
918 static int call__match(struct symbol *sym)
920 if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
921 return 1;
923 return 0;
927 * collect histogram counts
930 static int
931 hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
932 struct symbol *sym, u64 ip, struct ip_callchain *chain,
933 char level, u64 count)
935 struct rb_node **p = &hist.rb_node;
936 struct rb_node *parent = NULL;
937 struct hist_entry *he;
938 struct hist_entry entry = {
939 .thread = thread,
940 .map = map,
941 .dso = dso,
942 .sym = sym,
943 .ip = ip,
944 .level = level,
945 .count = count,
946 .parent = NULL,
947 .sorted_chain = RB_ROOT
949 int cmp;
951 if (sort__has_parent && chain) {
952 u64 context = PERF_CONTEXT_MAX;
953 int i;
955 for (i = 0; i < chain->nr; i++) {
956 u64 ip = chain->ips[i];
957 struct dso *dso = NULL;
958 struct symbol *sym;
960 if (ip >= PERF_CONTEXT_MAX) {
961 context = ip;
962 continue;
965 switch (context) {
966 case PERF_CONTEXT_KERNEL:
967 dso = kernel_dso;
968 break;
969 default:
970 break;
973 sym = resolve_symbol(thread, NULL, &dso, &ip);
975 if (sym && call__match(sym)) {
976 entry.parent = sym;
977 break;
982 while (*p != NULL) {
983 parent = *p;
984 he = rb_entry(parent, struct hist_entry, rb_node);
986 cmp = hist_entry__cmp(&entry, he);
988 if (!cmp) {
989 he->count += count;
990 if (callchain)
991 append_chain(&he->callchain, chain);
992 return 0;
995 if (cmp < 0)
996 p = &(*p)->rb_left;
997 else
998 p = &(*p)->rb_right;
1001 he = malloc(sizeof(*he));
1002 if (!he)
1003 return -ENOMEM;
1004 *he = entry;
1005 if (callchain) {
1006 callchain_init(&he->callchain);
1007 append_chain(&he->callchain, chain);
1009 rb_link_node(&he->rb_node, parent, p);
1010 rb_insert_color(&he->rb_node, &hist);
1012 return 0;
1015 static void hist_entry__free(struct hist_entry *he)
1017 free(he);
1021 * collapse the histogram
1024 static struct rb_root collapse_hists;
1026 static void collapse__insert_entry(struct hist_entry *he)
1028 struct rb_node **p = &collapse_hists.rb_node;
1029 struct rb_node *parent = NULL;
1030 struct hist_entry *iter;
1031 int64_t cmp;
1033 while (*p != NULL) {
1034 parent = *p;
1035 iter = rb_entry(parent, struct hist_entry, rb_node);
1037 cmp = hist_entry__collapse(iter, he);
1039 if (!cmp) {
1040 iter->count += he->count;
1041 hist_entry__free(he);
1042 return;
1045 if (cmp < 0)
1046 p = &(*p)->rb_left;
1047 else
1048 p = &(*p)->rb_right;
1051 rb_link_node(&he->rb_node, parent, p);
1052 rb_insert_color(&he->rb_node, &collapse_hists);
1055 static void collapse__resort(void)
1057 struct rb_node *next;
1058 struct hist_entry *n;
1060 if (!sort__need_collapse)
1061 return;
1063 next = rb_first(&hist);
1064 while (next) {
1065 n = rb_entry(next, struct hist_entry, rb_node);
1066 next = rb_next(&n->rb_node);
1068 rb_erase(&n->rb_node, &hist);
1069 collapse__insert_entry(n);
1074 * reverse the map, sort on count.
1077 static struct rb_root output_hists;
1079 static void output__insert_entry(struct hist_entry *he)
1081 struct rb_node **p = &output_hists.rb_node;
1082 struct rb_node *parent = NULL;
1083 struct hist_entry *iter;
1085 if (callchain)
1086 sort_chain_to_rbtree(&he->sorted_chain, &he->callchain);
1088 while (*p != NULL) {
1089 parent = *p;
1090 iter = rb_entry(parent, struct hist_entry, rb_node);
1092 if (he->count > iter->count)
1093 p = &(*p)->rb_left;
1094 else
1095 p = &(*p)->rb_right;
1098 rb_link_node(&he->rb_node, parent, p);
1099 rb_insert_color(&he->rb_node, &output_hists);
1102 static void output__resort(void)
1104 struct rb_node *next;
1105 struct hist_entry *n;
1106 struct rb_root *tree = &hist;
1108 if (sort__need_collapse)
1109 tree = &collapse_hists;
1111 next = rb_first(tree);
1113 while (next) {
1114 n = rb_entry(next, struct hist_entry, rb_node);
1115 next = rb_next(&n->rb_node);
1117 rb_erase(&n->rb_node, tree);
1118 output__insert_entry(n);
1122 static size_t output__fprintf(FILE *fp, u64 total_samples)
1124 struct hist_entry *pos;
1125 struct sort_entry *se;
1126 struct rb_node *nd;
1127 size_t ret = 0;
1129 fprintf(fp, "\n");
1130 fprintf(fp, "#\n");
1131 fprintf(fp, "# (%Ld samples)\n", (u64)total_samples);
1132 fprintf(fp, "#\n");
1134 fprintf(fp, "# Overhead");
1135 list_for_each_entry(se, &hist_entry__sort_list, list) {
1136 if (exclude_other && (se == &sort_parent))
1137 continue;
1138 fprintf(fp, " %s", se->header);
1140 fprintf(fp, "\n");
1142 fprintf(fp, "# ........");
1143 list_for_each_entry(se, &hist_entry__sort_list, list) {
1144 int i;
1146 if (exclude_other && (se == &sort_parent))
1147 continue;
1149 fprintf(fp, " ");
1150 for (i = 0; i < strlen(se->header); i++)
1151 fprintf(fp, ".");
1153 fprintf(fp, "\n");
1155 fprintf(fp, "#\n");
1157 for (nd = rb_first(&output_hists); nd; nd = rb_next(nd)) {
1158 pos = rb_entry(nd, struct hist_entry, rb_node);
1159 ret += hist_entry__fprintf(fp, pos, total_samples);
1162 if (sort_order == default_sort_order &&
1163 parent_pattern == default_parent_pattern) {
1164 fprintf(fp, "#\n");
1165 fprintf(fp, "# (For more details, try: perf report --sort comm,dso,symbol)\n");
1166 fprintf(fp, "#\n");
1168 fprintf(fp, "\n");
1170 return ret;
1173 static void register_idle_thread(void)
1175 struct thread *thread = threads__findnew(0);
1177 if (thread == NULL ||
1178 thread__set_comm(thread, "[idle]")) {
1179 fprintf(stderr, "problem inserting idle task.\n");
1180 exit(-1);
1184 static unsigned long total = 0,
1185 total_mmap = 0,
1186 total_comm = 0,
1187 total_fork = 0,
1188 total_unknown = 0,
1189 total_lost = 0;
1191 static int validate_chain(struct ip_callchain *chain, event_t *event)
1193 unsigned int chain_size;
1195 chain_size = event->header.size;
1196 chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;
1198 if (chain->nr*sizeof(u64) > chain_size)
1199 return -1;
1201 return 0;
1204 static int
1205 process_sample_event(event_t *event, unsigned long offset, unsigned long head)
1207 char level;
1208 int show = 0;
1209 struct dso *dso = NULL;
1210 struct thread *thread = threads__findnew(event->ip.pid);
1211 u64 ip = event->ip.ip;
1212 u64 period = 1;
1213 struct map *map = NULL;
1214 void *more_data = event->ip.__more_data;
1215 struct ip_callchain *chain = NULL;
1217 if (sample_type & PERF_SAMPLE_PERIOD) {
1218 period = *(u64 *)more_data;
1219 more_data += sizeof(u64);
1222 dprintf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d: %p period: %Ld\n",
1223 (void *)(offset + head),
1224 (void *)(long)(event->header.size),
1225 event->header.misc,
1226 event->ip.pid,
1227 (void *)(long)ip,
1228 (long long)period);
1230 if (sample_type & PERF_SAMPLE_CALLCHAIN) {
1231 int i;
1233 chain = (void *)more_data;
1235 dprintf("... chain: nr:%Lu\n", chain->nr);
1237 if (validate_chain(chain, event) < 0) {
1238 eprintf("call-chain problem with event, skipping it.\n");
1239 return 0;
1242 if (dump_trace) {
1243 for (i = 0; i < chain->nr; i++)
1244 dprintf("..... %2d: %016Lx\n", i, chain->ips[i]);
1248 dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid);
1250 if (thread == NULL) {
1251 eprintf("problem processing %d event, skipping it.\n",
1252 event->header.type);
1253 return -1;
1256 if (comm_list && !strlist__has_entry(comm_list, thread->comm))
1257 return 0;
1259 if (event->header.misc & PERF_EVENT_MISC_KERNEL) {
1260 show = SHOW_KERNEL;
1261 level = 'k';
1263 dso = kernel_dso;
1265 dprintf(" ...... dso: %s\n", dso->name);
1267 } else if (event->header.misc & PERF_EVENT_MISC_USER) {
1269 show = SHOW_USER;
1270 level = '.';
1272 } else {
1273 show = SHOW_HV;
1274 level = 'H';
1275 dprintf(" ...... dso: [hypervisor]\n");
1278 if (show & show_mask) {
1279 struct symbol *sym = resolve_symbol(thread, &map, &dso, &ip);
1281 if (dso_list && dso && dso->name && !strlist__has_entry(dso_list, dso->name))
1282 return 0;
1284 if (sym_list && sym && !strlist__has_entry(sym_list, sym->name))
1285 return 0;
1287 if (hist_entry__add(thread, map, dso, sym, ip, chain, level, period)) {
1288 eprintf("problem incrementing symbol count, skipping event\n");
1289 return -1;
1292 total += period;
1294 return 0;
1297 static int
1298 process_mmap_event(event_t *event, unsigned long offset, unsigned long head)
1300 struct thread *thread = threads__findnew(event->mmap.pid);
1301 struct map *map = map__new(&event->mmap);
1303 dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n",
1304 (void *)(offset + head),
1305 (void *)(long)(event->header.size),
1306 event->mmap.pid,
1307 (void *)(long)event->mmap.start,
1308 (void *)(long)event->mmap.len,
1309 (void *)(long)event->mmap.pgoff,
1310 event->mmap.filename);
1312 if (thread == NULL || map == NULL) {
1313 dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1314 return 0;
1317 thread__insert_map(thread, map);
1318 total_mmap++;
1320 return 0;
1323 static int
1324 process_comm_event(event_t *event, unsigned long offset, unsigned long head)
1326 struct thread *thread = threads__findnew(event->comm.pid);
1328 dprintf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
1329 (void *)(offset + head),
1330 (void *)(long)(event->header.size),
1331 event->comm.comm, event->comm.pid);
1333 if (thread == NULL ||
1334 thread__set_comm(thread, event->comm.comm)) {
1335 dprintf("problem processing PERF_EVENT_COMM, skipping event.\n");
1336 return -1;
1338 total_comm++;
1340 return 0;
1343 static int
1344 process_fork_event(event_t *event, unsigned long offset, unsigned long head)
1346 struct thread *thread = threads__findnew(event->fork.pid);
1347 struct thread *parent = threads__findnew(event->fork.ppid);
1349 dprintf("%p [%p]: PERF_EVENT_FORK: %d:%d\n",
1350 (void *)(offset + head),
1351 (void *)(long)(event->header.size),
1352 event->fork.pid, event->fork.ppid);
1354 if (!thread || !parent || thread__fork(thread, parent)) {
1355 dprintf("problem processing PERF_EVENT_FORK, skipping event.\n");
1356 return -1;
1358 total_fork++;
1360 return 0;
1363 static int
1364 process_period_event(event_t *event, unsigned long offset, unsigned long head)
1366 dprintf("%p [%p]: PERF_EVENT_PERIOD: time:%Ld, id:%Ld: period:%Ld\n",
1367 (void *)(offset + head),
1368 (void *)(long)(event->header.size),
1369 event->period.time,
1370 event->period.id,
1371 event->period.sample_period);
1373 return 0;
1376 static int
1377 process_lost_event(event_t *event, unsigned long offset, unsigned long head)
1379 dprintf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n",
1380 (void *)(offset + head),
1381 (void *)(long)(event->header.size),
1382 event->lost.id,
1383 event->lost.lost);
1385 total_lost += event->lost.lost;
1387 return 0;
1390 static void trace_event(event_t *event)
1392 unsigned char *raw_event = (void *)event;
1393 char *color = PERF_COLOR_BLUE;
1394 int i, j;
1396 if (!dump_trace)
1397 return;
1399 dprintf(".");
1400 cdprintf("\n. ... raw event: size %d bytes\n", event->header.size);
1402 for (i = 0; i < event->header.size; i++) {
1403 if ((i & 15) == 0) {
1404 dprintf(".");
1405 cdprintf(" %04x: ", i);
1408 cdprintf(" %02x", raw_event[i]);
1410 if (((i & 15) == 15) || i == event->header.size-1) {
1411 cdprintf(" ");
1412 for (j = 0; j < 15-(i & 15); j++)
1413 cdprintf(" ");
1414 for (j = 0; j < (i & 15); j++) {
1415 if (isprint(raw_event[i-15+j]))
1416 cdprintf("%c", raw_event[i-15+j]);
1417 else
1418 cdprintf(".");
1420 cdprintf("\n");
1423 dprintf(".\n");
1426 static int
1427 process_read_event(event_t *event, unsigned long offset, unsigned long head)
1429 dprintf("%p [%p]: PERF_EVENT_READ: %d %d %Lu\n",
1430 (void *)(offset + head),
1431 (void *)(long)(event->header.size),
1432 event->read.pid,
1433 event->read.tid,
1434 event->read.value);
1436 return 0;
1439 static int
1440 process_event(event_t *event, unsigned long offset, unsigned long head)
1442 trace_event(event);
1444 switch (event->header.type) {
1445 case PERF_EVENT_SAMPLE:
1446 return process_sample_event(event, offset, head);
1448 case PERF_EVENT_MMAP:
1449 return process_mmap_event(event, offset, head);
1451 case PERF_EVENT_COMM:
1452 return process_comm_event(event, offset, head);
1454 case PERF_EVENT_FORK:
1455 return process_fork_event(event, offset, head);
1457 case PERF_EVENT_PERIOD:
1458 return process_period_event(event, offset, head);
1460 case PERF_EVENT_LOST:
1461 return process_lost_event(event, offset, head);
1463 case PERF_EVENT_READ:
1464 return process_read_event(event, offset, head);
1467 * We dont process them right now but they are fine:
1470 case PERF_EVENT_THROTTLE:
1471 case PERF_EVENT_UNTHROTTLE:
1472 return 0;
1474 default:
1475 return -1;
1478 return 0;
1481 static struct perf_header *header;
1483 static u64 perf_header__sample_type(void)
1485 u64 sample_type = 0;
1486 int i;
1488 for (i = 0; i < header->attrs; i++) {
1489 struct perf_header_attr *attr = header->attr[i];
1491 if (!sample_type)
1492 sample_type = attr->attr.sample_type;
1493 else if (sample_type != attr->attr.sample_type)
1494 die("non matching sample_type");
1497 return sample_type;
1500 static int __cmd_report(void)
1502 int ret, rc = EXIT_FAILURE;
1503 unsigned long offset = 0;
1504 unsigned long head, shift;
1505 struct stat stat;
1506 event_t *event;
1507 uint32_t size;
1508 char *buf;
1510 register_idle_thread();
1512 input = open(input_name, O_RDONLY);
1513 if (input < 0) {
1514 fprintf(stderr, " failed to open file: %s", input_name);
1515 if (!strcmp(input_name, "perf.data"))
1516 fprintf(stderr, " (try 'perf record' first)");
1517 fprintf(stderr, "\n");
1518 exit(-1);
1521 ret = fstat(input, &stat);
1522 if (ret < 0) {
1523 perror("failed to stat file");
1524 exit(-1);
1527 if (!stat.st_size) {
1528 fprintf(stderr, "zero-sized file, nothing to do!\n");
1529 exit(0);
1532 header = perf_header__read(input);
1533 head = header->data_offset;
1535 sample_type = perf_header__sample_type();
1537 if (sort__has_parent && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
1538 fprintf(stderr, "selected --sort parent, but no callchain data\n");
1539 exit(-1);
1542 if (load_kernel() < 0) {
1543 perror("failed to load kernel symbols");
1544 return EXIT_FAILURE;
1547 if (!full_paths) {
1548 if (getcwd(__cwd, sizeof(__cwd)) == NULL) {
1549 perror("failed to get the current directory");
1550 return EXIT_FAILURE;
1552 cwdlen = strlen(cwd);
1553 } else {
1554 cwd = NULL;
1555 cwdlen = 0;
1558 shift = page_size * (head / page_size);
1559 offset += shift;
1560 head -= shift;
1562 remap:
1563 buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
1564 MAP_SHARED, input, offset);
1565 if (buf == MAP_FAILED) {
1566 perror("failed to mmap file");
1567 exit(-1);
1570 more:
1571 event = (event_t *)(buf + head);
1573 size = event->header.size;
1574 if (!size)
1575 size = 8;
1577 if (head + event->header.size >= page_size * mmap_window) {
1578 int ret;
1580 shift = page_size * (head / page_size);
1582 ret = munmap(buf, page_size * mmap_window);
1583 assert(ret == 0);
1585 offset += shift;
1586 head -= shift;
1587 goto remap;
1590 size = event->header.size;
1592 dprintf("\n%p [%p]: event: %d\n",
1593 (void *)(offset + head),
1594 (void *)(long)event->header.size,
1595 event->header.type);
1597 if (!size || process_event(event, offset, head) < 0) {
1599 dprintf("%p [%p]: skipping unknown header type: %d\n",
1600 (void *)(offset + head),
1601 (void *)(long)(event->header.size),
1602 event->header.type);
1604 total_unknown++;
1607 * assume we lost track of the stream, check alignment, and
1608 * increment a single u64 in the hope to catch on again 'soon'.
1611 if (unlikely(head & 7))
1612 head &= ~7ULL;
1614 size = 8;
1617 head += size;
1619 if (offset + head >= header->data_offset + header->data_size)
1620 goto done;
1622 if (offset + head < stat.st_size)
1623 goto more;
1625 done:
1626 rc = EXIT_SUCCESS;
1627 close(input);
1629 dprintf(" IP events: %10ld\n", total);
1630 dprintf(" mmap events: %10ld\n", total_mmap);
1631 dprintf(" comm events: %10ld\n", total_comm);
1632 dprintf(" fork events: %10ld\n", total_fork);
1633 dprintf(" lost events: %10ld\n", total_lost);
1634 dprintf(" unknown events: %10ld\n", total_unknown);
1636 if (dump_trace)
1637 return 0;
1639 if (verbose >= 3)
1640 threads__fprintf(stdout);
1642 if (verbose >= 2)
1643 dsos__fprintf(stdout);
1645 collapse__resort();
1646 output__resort();
1647 output__fprintf(stdout, total);
1649 return rc;
1652 static const char * const report_usage[] = {
1653 "perf report [<options>] <command>",
1654 NULL
1657 static const struct option options[] = {
1658 OPT_STRING('i', "input", &input_name, "file",
1659 "input file name"),
1660 OPT_BOOLEAN('v', "verbose", &verbose,
1661 "be more verbose (show symbol address, etc)"),
1662 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1663 "dump raw trace in ASCII"),
1664 OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"),
1665 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1666 "sort by key(s): pid, comm, dso, symbol, parent"),
1667 OPT_BOOLEAN('P', "full-paths", &full_paths,
1668 "Don't shorten the pathnames taking into account the cwd"),
1669 OPT_STRING('p', "parent", &parent_pattern, "regex",
1670 "regex filter to identify parent, see: '--sort parent'"),
1671 OPT_BOOLEAN('x', "exclude-other", &exclude_other,
1672 "Only display entries with parent-match"),
1673 OPT_BOOLEAN('c', "callchain", &callchain, "Display callchains"),
1674 OPT_STRING('d', "dsos", &dso_list_str, "dso[,dso...]",
1675 "only consider symbols in these dsos"),
1676 OPT_STRING('C', "comms", &comm_list_str, "comm[,comm...]",
1677 "only consider symbols in these comms"),
1678 OPT_STRING('S', "symbols", &sym_list_str, "symbol[,symbol...]",
1679 "only consider these symbols"),
1680 OPT_END()
1683 static void setup_sorting(void)
1685 char *tmp, *tok, *str = strdup(sort_order);
1687 for (tok = strtok_r(str, ", ", &tmp);
1688 tok; tok = strtok_r(NULL, ", ", &tmp)) {
1689 if (sort_dimension__add(tok) < 0) {
1690 error("Unknown --sort key: `%s'", tok);
1691 usage_with_options(report_usage, options);
1695 free(str);
1698 static void setup_list(struct strlist **list, const char *list_str,
1699 const char *list_name)
1701 if (list_str) {
1702 *list = strlist__new(true, list_str);
1703 if (!*list) {
1704 fprintf(stderr, "problems parsing %s list\n",
1705 list_name);
1706 exit(129);
1711 int cmd_report(int argc, const char **argv, const char *prefix)
1713 symbol__init();
1715 page_size = getpagesize();
1717 argc = parse_options(argc, argv, options, report_usage, 0);
1719 setup_sorting();
1721 if (parent_pattern != default_parent_pattern)
1722 sort_dimension__add("parent");
1723 else
1724 exclude_other = 0;
1727 * Any (unrecognized) arguments left?
1729 if (argc)
1730 usage_with_options(report_usage, options);
1732 setup_list(&dso_list, dso_list_str, "dso");
1733 setup_list(&comm_list, comm_list_str, "comm");
1734 setup_list(&sym_list, sym_list_str, "symbol");
1736 setup_pager();
1738 return __cmd_report();