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;
55 static unsigned long page_size;
56 static unsigned long mmap_window = 32;
58 static char default_parent_pattern[] = "^sys_|^do_page_fault";
59 static char *parent_pattern = default_parent_pattern;
60 static regex_t parent_regex;
62 static int exclude_other = 1;
64 static char callchain_default_opt[] = "fractal,0.5";
69 struct callchain_param callchain_param = {
70 .mode = CHAIN_GRAPH_ABS,
74 static u64 sample_type;
77 struct perf_event_header header;
80 unsigned char __more_data[];
84 struct perf_event_header header;
89 char filename[PATH_MAX];
93 struct perf_event_header header;
99 struct perf_event_header header;
103 struct period_event {
104 struct perf_event_header header;
111 struct perf_event_header header;
117 struct perf_event_header header;
123 typedef union event_union {
124 struct perf_event_header header;
126 struct mmap_event mmap;
127 struct comm_event comm;
128 struct fork_event fork;
129 struct period_event period;
130 struct lost_event lost;
131 struct read_event read;
134 static int repsep_fprintf(FILE *fp, const char *fmt, ...)
141 n = vfprintf(fp, fmt, ap);
144 n = vasprintf(&bf, fmt, ap);
148 sep = strchr(sep, *field_sep);
161 static LIST_HEAD(dsos);
162 static struct dso *kernel_dso;
163 static struct dso *vdso;
164 static struct dso *hypervisor_dso;
166 static void dsos__add(struct dso *dso)
168 list_add_tail(&dso->node, &dsos);
171 static struct dso *dsos__find(const char *name)
175 list_for_each_entry(pos, &dsos, node)
176 if (strcmp(pos->name, name) == 0)
181 static struct dso *dsos__findnew(const char *name)
183 struct dso *dso = dsos__find(name);
189 dso = dso__new(name, 0);
193 nr = dso__load(dso, NULL, verbose);
195 eprintf("Failed to open: %s\n", name);
199 eprintf("No symbols found in: %s, maybe install a debug package?\n", name);
210 static void dsos__fprintf(FILE *fp)
214 list_for_each_entry(pos, &dsos, node)
215 dso__fprintf(pos, fp);
218 static struct symbol *vdso__find_symbol(struct dso *dso, u64 ip)
220 return dso__find_symbol(dso, ip);
223 static int load_kernel(void)
227 kernel_dso = dso__new("[kernel]", 0);
231 err = dso__load_kernel(kernel_dso, vmlinux, NULL, verbose, modules);
233 dso__delete(kernel_dso);
236 dsos__add(kernel_dso);
238 vdso = dso__new("[vdso]", 0);
242 vdso->find_symbol = vdso__find_symbol;
246 hypervisor_dso = dso__new("[hypervisor]", 0);
249 dsos__add(hypervisor_dso);
254 static char __cwd[PATH_MAX];
255 static char *cwd = __cwd;
258 static int strcommon(const char *pathname)
262 while (pathname[n] == cwd[n] && n < cwdlen)
269 struct list_head node;
273 u64 (*map_ip)(struct map *, u64);
277 static u64 map__map_ip(struct map *map, u64 ip)
279 return ip - map->start + map->pgoff;
282 static u64 vdso__map_ip(struct map *map __used, u64 ip)
287 static inline int is_anon_memory(const char *filename)
289 return strcmp(filename, "//anon") == 0;
292 static struct map *map__new(struct mmap_event *event)
294 struct map *self = malloc(sizeof(*self));
297 const char *filename = event->filename;
298 char newfilename[PATH_MAX];
302 int n = strcommon(filename);
305 snprintf(newfilename, sizeof(newfilename),
306 ".%s", filename + n);
307 filename = newfilename;
311 anon = is_anon_memory(filename);
314 snprintf(newfilename, sizeof(newfilename), "/tmp/perf-%d.map", event->pid);
315 filename = newfilename;
318 self->start = event->start;
319 self->end = event->start + event->len;
320 self->pgoff = event->pgoff;
322 self->dso = dsos__findnew(filename);
323 if (self->dso == NULL)
326 if (self->dso == vdso || anon)
327 self->map_ip = vdso__map_ip;
329 self->map_ip = map__map_ip;
337 static struct map *map__clone(struct map *self)
339 struct map *map = malloc(sizeof(*self));
344 memcpy(map, self, sizeof(*self));
349 static int map__overlap(struct map *l, struct map *r)
351 if (l->start > r->start) {
357 if (l->end > r->start)
363 static size_t map__fprintf(struct map *self, FILE *fp)
365 return fprintf(fp, " %Lx-%Lx %Lx %s\n",
366 self->start, self->end, self->pgoff, self->dso->name);
371 struct rb_node rb_node;
372 struct list_head maps;
377 static struct thread *thread__new(pid_t pid)
379 struct thread *self = malloc(sizeof(*self));
383 self->comm = malloc(32);
385 snprintf(self->comm, 32, ":%d", self->pid);
386 INIT_LIST_HEAD(&self->maps);
392 static unsigned int dsos__col_width,
396 static int thread__set_comm(struct thread *self, const char *comm)
400 self->comm = strdup(comm);
404 if (!col_width_list_str && !field_sep &&
405 (!comm_list || strlist__has_entry(comm_list, comm))) {
406 unsigned int slen = strlen(comm);
407 if (slen > comms__col_width) {
408 comms__col_width = slen;
409 threads__col_width = slen + 6;
416 static size_t thread__fprintf(struct thread *self, FILE *fp)
419 size_t ret = fprintf(fp, "Thread %d %s\n", self->pid, self->comm);
421 list_for_each_entry(pos, &self->maps, node)
422 ret += map__fprintf(pos, fp);
428 static struct rb_root threads;
429 static struct thread *last_match;
431 static struct thread *threads__findnew(pid_t pid)
433 struct rb_node **p = &threads.rb_node;
434 struct rb_node *parent = NULL;
438 * Font-end cache - PID lookups come in blocks,
439 * so most of the time we dont have to look up
442 if (last_match && last_match->pid == pid)
447 th = rb_entry(parent, struct thread, rb_node);
449 if (th->pid == pid) {
460 th = thread__new(pid);
462 rb_link_node(&th->rb_node, parent, p);
463 rb_insert_color(&th->rb_node, &threads);
470 static void thread__insert_map(struct thread *self, struct map *map)
472 struct map *pos, *tmp;
474 list_for_each_entry_safe(pos, tmp, &self->maps, node) {
475 if (map__overlap(pos, map)) {
477 printf("overlapping maps:\n");
478 map__fprintf(map, stdout);
479 map__fprintf(pos, stdout);
482 if (map->start <= pos->start && map->end > pos->start)
483 pos->start = map->end;
485 if (map->end >= pos->end && map->start < pos->end)
486 pos->end = map->start;
489 printf("after collision:\n");
490 map__fprintf(pos, stdout);
493 if (pos->start >= pos->end) {
494 list_del_init(&pos->node);
500 list_add_tail(&map->node, &self->maps);
503 static int thread__fork(struct thread *self, struct thread *parent)
509 self->comm = strdup(parent->comm);
513 list_for_each_entry(map, &parent->maps, node) {
514 struct map *new = map__clone(map);
517 thread__insert_map(self, new);
523 static struct map *thread__find_map(struct thread *self, u64 ip)
530 list_for_each_entry(pos, &self->maps, node)
531 if (ip >= pos->start && ip <= pos->end)
537 static size_t threads__fprintf(FILE *fp)
542 for (nd = rb_first(&threads); nd; nd = rb_next(nd)) {
543 struct thread *pos = rb_entry(nd, struct thread, rb_node);
545 ret += thread__fprintf(pos, fp);
552 * histogram, sorted on item, collects counts
555 static struct rb_root hist;
558 struct rb_node rb_node;
560 struct thread *thread;
564 struct symbol *parent;
567 struct callchain_node callchain;
568 struct rb_root sorted_chain;
574 * configurable sorting bits
578 struct list_head list;
582 int64_t (*cmp)(struct hist_entry *, struct hist_entry *);
583 int64_t (*collapse)(struct hist_entry *, struct hist_entry *);
584 size_t (*print)(FILE *fp, struct hist_entry *, unsigned int width);
588 static int64_t cmp_null(void *l, void *r)
601 sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
603 return right->thread->pid - left->thread->pid;
607 sort__thread_print(FILE *fp, struct hist_entry *self, unsigned int width)
609 return repsep_fprintf(fp, "%*s:%5d", width - 6,
610 self->thread->comm ?: "", self->thread->pid);
613 static struct sort_entry sort_thread = {
614 .header = "Command: Pid",
615 .cmp = sort__thread_cmp,
616 .print = sort__thread_print,
617 .width = &threads__col_width,
623 sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
625 return right->thread->pid - left->thread->pid;
629 sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
631 char *comm_l = left->thread->comm;
632 char *comm_r = right->thread->comm;
634 if (!comm_l || !comm_r)
635 return cmp_null(comm_l, comm_r);
637 return strcmp(comm_l, comm_r);
641 sort__comm_print(FILE *fp, struct hist_entry *self, unsigned int width)
643 return repsep_fprintf(fp, "%*s", width, self->thread->comm);
646 static struct sort_entry sort_comm = {
648 .cmp = sort__comm_cmp,
649 .collapse = sort__comm_collapse,
650 .print = sort__comm_print,
651 .width = &comms__col_width,
657 sort__dso_cmp(struct hist_entry *left, struct hist_entry *right)
659 struct dso *dso_l = left->dso;
660 struct dso *dso_r = right->dso;
662 if (!dso_l || !dso_r)
663 return cmp_null(dso_l, dso_r);
665 return strcmp(dso_l->name, dso_r->name);
669 sort__dso_print(FILE *fp, struct hist_entry *self, unsigned int width)
672 return repsep_fprintf(fp, "%-*s", width, self->dso->name);
674 return repsep_fprintf(fp, "%*llx", width, (u64)self->ip);
677 static struct sort_entry sort_dso = {
678 .header = "Shared Object",
679 .cmp = sort__dso_cmp,
680 .print = sort__dso_print,
681 .width = &dsos__col_width,
687 sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
691 if (left->sym == right->sym)
694 ip_l = left->sym ? left->sym->start : left->ip;
695 ip_r = right->sym ? right->sym->start : right->ip;
697 return (int64_t)(ip_r - ip_l);
701 sort__sym_print(FILE *fp, struct hist_entry *self, unsigned int width __used)
706 ret += repsep_fprintf(fp, "%#018llx ", (u64)self->ip);
709 ret += repsep_fprintf(fp, "[%c] %s",
710 self->dso == kernel_dso ? 'k' :
711 self->dso == hypervisor_dso ? 'h' : '.', self->sym->name);
713 if (self->sym->module)
714 ret += repsep_fprintf(fp, "\t[%s]",
715 self->sym->module->name);
717 ret += repsep_fprintf(fp, "%#016llx", (u64)self->ip);
723 static struct sort_entry sort_sym = {
725 .cmp = sort__sym_cmp,
726 .print = sort__sym_print,
732 sort__parent_cmp(struct hist_entry *left, struct hist_entry *right)
734 struct symbol *sym_l = left->parent;
735 struct symbol *sym_r = right->parent;
737 if (!sym_l || !sym_r)
738 return cmp_null(sym_l, sym_r);
740 return strcmp(sym_l->name, sym_r->name);
744 sort__parent_print(FILE *fp, struct hist_entry *self, unsigned int width)
746 return repsep_fprintf(fp, "%-*s", width,
747 self->parent ? self->parent->name : "[other]");
750 static unsigned int parent_symbol__col_width;
752 static struct sort_entry sort_parent = {
753 .header = "Parent symbol",
754 .cmp = sort__parent_cmp,
755 .print = sort__parent_print,
756 .width = &parent_symbol__col_width,
759 static int sort__need_collapse = 0;
760 static int sort__has_parent = 0;
762 struct sort_dimension {
764 struct sort_entry *entry;
768 static struct sort_dimension sort_dimensions[] = {
769 { .name = "pid", .entry = &sort_thread, },
770 { .name = "comm", .entry = &sort_comm, },
771 { .name = "dso", .entry = &sort_dso, },
772 { .name = "symbol", .entry = &sort_sym, },
773 { .name = "parent", .entry = &sort_parent, },
776 static LIST_HEAD(hist_entry__sort_list);
778 static int sort_dimension__add(char *tok)
782 for (i = 0; i < ARRAY_SIZE(sort_dimensions); i++) {
783 struct sort_dimension *sd = &sort_dimensions[i];
788 if (strncasecmp(tok, sd->name, strlen(tok)))
791 if (sd->entry->collapse)
792 sort__need_collapse = 1;
794 if (sd->entry == &sort_parent) {
795 int ret = regcomp(&parent_regex, parent_pattern, REG_EXTENDED);
799 regerror(ret, &parent_regex, err, sizeof(err));
800 fprintf(stderr, "Invalid regex: %s\n%s",
801 parent_pattern, err);
804 sort__has_parent = 1;
807 list_add_tail(&sd->entry->list, &hist_entry__sort_list);
817 hist_entry__cmp(struct hist_entry *left, struct hist_entry *right)
819 struct sort_entry *se;
822 list_for_each_entry(se, &hist_entry__sort_list, list) {
823 cmp = se->cmp(left, right);
832 hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
834 struct sort_entry *se;
837 list_for_each_entry(se, &hist_entry__sort_list, list) {
838 int64_t (*f)(struct hist_entry *, struct hist_entry *);
840 f = se->collapse ?: se->cmp;
842 cmp = f(left, right);
850 static size_t ipchain__fprintf_graph_line(FILE *fp, int depth, int depth_mask)
855 ret += fprintf(fp, "%s", " ");
857 for (i = 0; i < depth; i++)
858 if (depth_mask & (1 << i))
859 ret += fprintf(fp, "| ");
861 ret += fprintf(fp, " ");
863 ret += fprintf(fp, "\n");
868 ipchain__fprintf_graph(FILE *fp, struct callchain_list *chain, int depth,
869 int depth_mask, int count, u64 total_samples,
875 ret += fprintf(fp, "%s", " ");
876 for (i = 0; i < depth; i++) {
877 if (depth_mask & (1 << i))
878 ret += fprintf(fp, "|");
880 ret += fprintf(fp, " ");
881 if (!count && i == depth - 1) {
884 percent = hits * 100.0 / total_samples;
885 ret += percent_color_fprintf(fp, "--%2.2f%%-- ", percent);
887 ret += fprintf(fp, "%s", " ");
890 ret += fprintf(fp, "%s\n", chain->sym->name);
892 ret += fprintf(fp, "%p\n", (void *)(long)chain->ip);
898 callchain__fprintf_graph(FILE *fp, struct callchain_node *self,
899 u64 total_samples, int depth, int depth_mask)
901 struct rb_node *node, *next;
902 struct callchain_node *child;
903 struct callchain_list *chain;
904 int new_depth_mask = depth_mask;
909 if (callchain_param.mode == CHAIN_GRAPH_REL)
910 new_total = self->cumul_hit;
912 new_total = total_samples;
914 node = rb_first(&self->rb_root);
916 child = rb_entry(node, struct callchain_node, rb_node);
919 * The depth mask manages the output of pipes that show
920 * the depth. We don't want to keep the pipes of the current
921 * level for the last child of this depth
923 next = rb_next(node);
925 new_depth_mask &= ~(1 << (depth - 1));
928 * But we keep the older depth mask for the line seperator
929 * to keep the level link until we reach the last child
931 ret += ipchain__fprintf_graph_line(fp, depth, depth_mask);
933 list_for_each_entry(chain, &child->val, list) {
934 if (chain->ip >= PERF_CONTEXT_MAX)
936 ret += ipchain__fprintf_graph(fp, chain, depth,
941 ret += callchain__fprintf_graph(fp, child, new_total,
943 new_depth_mask | (1 << depth));
951 callchain__fprintf_flat(FILE *fp, struct callchain_node *self,
954 struct callchain_list *chain;
960 ret += callchain__fprintf_flat(fp, self->parent, total_samples);
963 list_for_each_entry(chain, &self->val, list) {
964 if (chain->ip >= PERF_CONTEXT_MAX)
967 ret += fprintf(fp, " %s\n", chain->sym->name);
969 ret += fprintf(fp, " %p\n",
970 (void *)(long)chain->ip);
977 hist_entry_callchain__fprintf(FILE *fp, struct hist_entry *self,
980 struct rb_node *rb_node;
981 struct callchain_node *chain;
984 rb_node = rb_first(&self->sorted_chain);
988 chain = rb_entry(rb_node, struct callchain_node, rb_node);
989 percent = chain->hit * 100.0 / total_samples;
990 switch (callchain_param.mode) {
992 ret += percent_color_fprintf(fp, " %6.2f%%\n",
994 ret += callchain__fprintf_flat(fp, chain, total_samples);
996 case CHAIN_GRAPH_ABS: /* Falldown */
997 case CHAIN_GRAPH_REL:
998 ret += callchain__fprintf_graph(fp, chain,
999 total_samples, 1, 1);
1003 ret += fprintf(fp, "\n");
1004 rb_node = rb_next(rb_node);
1012 hist_entry__fprintf(FILE *fp, struct hist_entry *self, u64 total_samples)
1014 struct sort_entry *se;
1017 if (exclude_other && !self->parent)
1021 ret = percent_color_fprintf(fp,
1022 field_sep ? "%.2f" : " %6.2f%%",
1023 (self->count * 100.0) / total_samples);
1025 ret = fprintf(fp, field_sep ? "%lld" : "%12lld ", self->count);
1027 list_for_each_entry(se, &hist_entry__sort_list, list) {
1028 if (exclude_other && (se == &sort_parent))
1031 fprintf(fp, "%s", field_sep ?: " ");
1032 ret += se->print(fp, self, se->width ? *se->width : 0);
1035 ret += fprintf(fp, "\n");
1038 hist_entry_callchain__fprintf(fp, self, total_samples);
1047 static void dso__calc_col_width(struct dso *self)
1049 if (!col_width_list_str && !field_sep &&
1050 (!dso_list || strlist__has_entry(dso_list, self->name))) {
1051 unsigned int slen = strlen(self->name);
1052 if (slen > dsos__col_width)
1053 dsos__col_width = slen;
1056 self->slen_calculated = 1;
1059 static struct symbol *
1060 resolve_symbol(struct thread *thread, struct map **mapp,
1061 struct dso **dsop, u64 *ipp)
1063 struct dso *dso = dsop ? *dsop : NULL;
1064 struct map *map = mapp ? *mapp : NULL;
1076 map = thread__find_map(thread, ip);
1079 * We have to do this here as we may have a dso
1080 * with no symbol hit that has a name longer than
1081 * the ones with symbols sampled.
1083 if (!map->dso->slen_calculated)
1084 dso__calc_col_width(map->dso);
1089 ip = map->map_ip(map, ip);
1094 * If this is outside of all known maps,
1095 * and is a negative address, try to look it
1096 * up in the kernel dso, as it might be a
1097 * vsyscall (which executes in user-mode):
1099 if ((long long)ip < 0)
1102 dprintf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
1103 dprintf(" ...... map: %Lx -> %Lx\n", *ipp, ip);
1112 return dso->find_symbol(dso, ip);
1115 static int call__match(struct symbol *sym)
1117 if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
1123 static struct symbol **
1124 resolve_callchain(struct thread *thread, struct map *map __used,
1125 struct ip_callchain *chain, struct hist_entry *entry)
1127 u64 context = PERF_CONTEXT_MAX;
1128 struct symbol **syms = NULL;
1132 syms = calloc(chain->nr, sizeof(*syms));
1134 fprintf(stderr, "Can't allocate memory for symbols\n");
1139 for (i = 0; i < chain->nr; i++) {
1140 u64 ip = chain->ips[i];
1141 struct dso *dso = NULL;
1144 if (ip >= PERF_CONTEXT_MAX) {
1150 case PERF_CONTEXT_HV:
1151 dso = hypervisor_dso;
1153 case PERF_CONTEXT_KERNEL:
1160 sym = resolve_symbol(thread, NULL, &dso, &ip);
1163 if (sort__has_parent && call__match(sym) &&
1165 entry->parent = sym;
1176 * collect histogram counts
1180 hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
1181 struct symbol *sym, u64 ip, struct ip_callchain *chain,
1182 char level, u64 count)
1184 struct rb_node **p = &hist.rb_node;
1185 struct rb_node *parent = NULL;
1186 struct hist_entry *he;
1187 struct symbol **syms = NULL;
1188 struct hist_entry entry = {
1197 .sorted_chain = RB_ROOT
1201 if ((sort__has_parent || callchain) && chain)
1202 syms = resolve_callchain(thread, map, chain, &entry);
1204 while (*p != NULL) {
1206 he = rb_entry(parent, struct hist_entry, rb_node);
1208 cmp = hist_entry__cmp(&entry, he);
1213 append_chain(&he->callchain, chain, syms);
1222 p = &(*p)->rb_right;
1225 he = malloc(sizeof(*he));
1230 callchain_init(&he->callchain);
1231 append_chain(&he->callchain, chain, syms);
1234 rb_link_node(&he->rb_node, parent, p);
1235 rb_insert_color(&he->rb_node, &hist);
1240 static void hist_entry__free(struct hist_entry *he)
1246 * collapse the histogram
1249 static struct rb_root collapse_hists;
1251 static void collapse__insert_entry(struct hist_entry *he)
1253 struct rb_node **p = &collapse_hists.rb_node;
1254 struct rb_node *parent = NULL;
1255 struct hist_entry *iter;
1258 while (*p != NULL) {
1260 iter = rb_entry(parent, struct hist_entry, rb_node);
1262 cmp = hist_entry__collapse(iter, he);
1265 iter->count += he->count;
1266 hist_entry__free(he);
1273 p = &(*p)->rb_right;
1276 rb_link_node(&he->rb_node, parent, p);
1277 rb_insert_color(&he->rb_node, &collapse_hists);
1280 static void collapse__resort(void)
1282 struct rb_node *next;
1283 struct hist_entry *n;
1285 if (!sort__need_collapse)
1288 next = rb_first(&hist);
1290 n = rb_entry(next, struct hist_entry, rb_node);
1291 next = rb_next(&n->rb_node);
1293 rb_erase(&n->rb_node, &hist);
1294 collapse__insert_entry(n);
1299 * reverse the map, sort on count.
1302 static struct rb_root output_hists;
1304 static void output__insert_entry(struct hist_entry *he, u64 min_callchain_hits)
1306 struct rb_node **p = &output_hists.rb_node;
1307 struct rb_node *parent = NULL;
1308 struct hist_entry *iter;
1311 callchain_param.sort(&he->sorted_chain, &he->callchain,
1312 min_callchain_hits, &callchain_param);
1314 while (*p != NULL) {
1316 iter = rb_entry(parent, struct hist_entry, rb_node);
1318 if (he->count > iter->count)
1321 p = &(*p)->rb_right;
1324 rb_link_node(&he->rb_node, parent, p);
1325 rb_insert_color(&he->rb_node, &output_hists);
1328 static void output__resort(u64 total_samples)
1330 struct rb_node *next;
1331 struct hist_entry *n;
1332 struct rb_root *tree = &hist;
1333 u64 min_callchain_hits;
1335 min_callchain_hits = total_samples * (callchain_param.min_percent / 100);
1337 if (sort__need_collapse)
1338 tree = &collapse_hists;
1340 next = rb_first(tree);
1343 n = rb_entry(next, struct hist_entry, rb_node);
1344 next = rb_next(&n->rb_node);
1346 rb_erase(&n->rb_node, tree);
1347 output__insert_entry(n, min_callchain_hits);
1351 static size_t output__fprintf(FILE *fp, u64 total_samples)
1353 struct hist_entry *pos;
1354 struct sort_entry *se;
1358 char *col_width = col_width_list_str;
1362 fprintf(fp, "# (%Ld samples)\n", (u64)total_samples);
1365 fprintf(fp, "# Overhead");
1366 list_for_each_entry(se, &hist_entry__sort_list, list) {
1367 if (exclude_other && (se == &sort_parent))
1370 fprintf(fp, "%c%s", *field_sep, se->header);
1373 width = strlen(se->header);
1375 if (col_width_list_str) {
1377 *se->width = atoi(col_width);
1378 col_width = strchr(col_width, ',');
1383 width = *se->width = max(*se->width, width);
1385 fprintf(fp, " %*s", width, se->header);
1392 fprintf(fp, "# ........");
1393 list_for_each_entry(se, &hist_entry__sort_list, list) {
1396 if (exclude_other && (se == &sort_parent))
1403 width = strlen(se->header);
1404 for (i = 0; i < width; i++)
1412 for (nd = rb_first(&output_hists); nd; nd = rb_next(nd)) {
1413 pos = rb_entry(nd, struct hist_entry, rb_node);
1414 ret += hist_entry__fprintf(fp, pos, total_samples);
1417 if (sort_order == default_sort_order &&
1418 parent_pattern == default_parent_pattern) {
1420 fprintf(fp, "# (For more details, try: perf report --sort comm,dso,symbol)\n");
1428 static void register_idle_thread(void)
1430 struct thread *thread = threads__findnew(0);
1432 if (thread == NULL ||
1433 thread__set_comm(thread, "[idle]")) {
1434 fprintf(stderr, "problem inserting idle task.\n");
1439 static unsigned long total = 0,
1446 static int validate_chain(struct ip_callchain *chain, event_t *event)
1448 unsigned int chain_size;
1450 chain_size = event->header.size;
1451 chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;
1453 if (chain->nr*sizeof(u64) > chain_size)
1460 process_sample_event(event_t *event, unsigned long offset, unsigned long head)
1464 struct dso *dso = NULL;
1465 struct thread *thread = threads__findnew(event->ip.pid);
1466 u64 ip = event->ip.ip;
1468 struct map *map = NULL;
1469 void *more_data = event->ip.__more_data;
1470 struct ip_callchain *chain = NULL;
1473 if (sample_type & PERF_SAMPLE_PERIOD) {
1474 period = *(u64 *)more_data;
1475 more_data += sizeof(u64);
1478 dprintf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d: %p period: %Ld\n",
1479 (void *)(offset + head),
1480 (void *)(long)(event->header.size),
1486 if (sample_type & PERF_SAMPLE_CALLCHAIN) {
1489 chain = (void *)more_data;
1491 dprintf("... chain: nr:%Lu\n", chain->nr);
1493 if (validate_chain(chain, event) < 0) {
1494 eprintf("call-chain problem with event, skipping it.\n");
1499 for (i = 0; i < chain->nr; i++)
1500 dprintf("..... %2d: %016Lx\n", i, chain->ips[i]);
1504 dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid);
1506 if (thread == NULL) {
1507 eprintf("problem processing %d event, skipping it.\n",
1508 event->header.type);
1512 if (comm_list && !strlist__has_entry(comm_list, thread->comm))
1515 cpumode = event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK;
1517 if (cpumode == PERF_EVENT_MISC_KERNEL) {
1523 dprintf(" ...... dso: %s\n", dso->name);
1525 } else if (cpumode == PERF_EVENT_MISC_USER) {
1534 dso = hypervisor_dso;
1536 dprintf(" ...... dso: [hypervisor]\n");
1539 if (show & show_mask) {
1540 struct symbol *sym = resolve_symbol(thread, &map, &dso, &ip);
1542 if (dso_list && dso && dso->name && !strlist__has_entry(dso_list, dso->name))
1545 if (sym_list && sym && !strlist__has_entry(sym_list, sym->name))
1548 if (hist_entry__add(thread, map, dso, sym, ip, chain, level, period)) {
1549 eprintf("problem incrementing symbol count, skipping event\n");
1559 process_mmap_event(event_t *event, unsigned long offset, unsigned long head)
1561 struct thread *thread = threads__findnew(event->mmap.pid);
1562 struct map *map = map__new(&event->mmap);
1564 dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n",
1565 (void *)(offset + head),
1566 (void *)(long)(event->header.size),
1568 (void *)(long)event->mmap.start,
1569 (void *)(long)event->mmap.len,
1570 (void *)(long)event->mmap.pgoff,
1571 event->mmap.filename);
1573 if (thread == NULL || map == NULL) {
1574 dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1578 thread__insert_map(thread, map);
1585 process_comm_event(event_t *event, unsigned long offset, unsigned long head)
1587 struct thread *thread = threads__findnew(event->comm.pid);
1589 dprintf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
1590 (void *)(offset + head),
1591 (void *)(long)(event->header.size),
1592 event->comm.comm, event->comm.pid);
1594 if (thread == NULL ||
1595 thread__set_comm(thread, event->comm.comm)) {
1596 dprintf("problem processing PERF_EVENT_COMM, skipping event.\n");
1605 process_fork_event(event_t *event, unsigned long offset, unsigned long head)
1607 struct thread *thread = threads__findnew(event->fork.pid);
1608 struct thread *parent = threads__findnew(event->fork.ppid);
1610 dprintf("%p [%p]: PERF_EVENT_FORK: %d:%d\n",
1611 (void *)(offset + head),
1612 (void *)(long)(event->header.size),
1613 event->fork.pid, event->fork.ppid);
1615 if (!thread || !parent || thread__fork(thread, parent)) {
1616 dprintf("problem processing PERF_EVENT_FORK, skipping event.\n");
1625 process_period_event(event_t *event, unsigned long offset, unsigned long head)
1627 dprintf("%p [%p]: PERF_EVENT_PERIOD: time:%Ld, id:%Ld: period:%Ld\n",
1628 (void *)(offset + head),
1629 (void *)(long)(event->header.size),
1632 event->period.sample_period);
1638 process_lost_event(event_t *event, unsigned long offset, unsigned long head)
1640 dprintf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n",
1641 (void *)(offset + head),
1642 (void *)(long)(event->header.size),
1646 total_lost += event->lost.lost;
1651 static void trace_event(event_t *event)
1653 unsigned char *raw_event = (void *)event;
1654 char *color = PERF_COLOR_BLUE;
1661 cdprintf("\n. ... raw event: size %d bytes\n", event->header.size);
1663 for (i = 0; i < event->header.size; i++) {
1664 if ((i & 15) == 0) {
1666 cdprintf(" %04x: ", i);
1669 cdprintf(" %02x", raw_event[i]);
1671 if (((i & 15) == 15) || i == event->header.size-1) {
1673 for (j = 0; j < 15-(i & 15); j++)
1675 for (j = 0; j < (i & 15); j++) {
1676 if (isprint(raw_event[i-15+j]))
1677 cdprintf("%c", raw_event[i-15+j]);
1688 process_read_event(event_t *event, unsigned long offset, unsigned long head)
1690 dprintf("%p [%p]: PERF_EVENT_READ: %d %d %Lu\n",
1691 (void *)(offset + head),
1692 (void *)(long)(event->header.size),
1701 process_event(event_t *event, unsigned long offset, unsigned long head)
1705 switch (event->header.type) {
1706 case PERF_EVENT_SAMPLE:
1707 return process_sample_event(event, offset, head);
1709 case PERF_EVENT_MMAP:
1710 return process_mmap_event(event, offset, head);
1712 case PERF_EVENT_COMM:
1713 return process_comm_event(event, offset, head);
1715 case PERF_EVENT_FORK:
1716 return process_fork_event(event, offset, head);
1718 case PERF_EVENT_PERIOD:
1719 return process_period_event(event, offset, head);
1721 case PERF_EVENT_LOST:
1722 return process_lost_event(event, offset, head);
1724 case PERF_EVENT_READ:
1725 return process_read_event(event, offset, head);
1728 * We dont process them right now but they are fine:
1731 case PERF_EVENT_THROTTLE:
1732 case PERF_EVENT_UNTHROTTLE:
1742 static struct perf_header *header;
1744 static u64 perf_header__sample_type(void)
1746 u64 sample_type = 0;
1749 for (i = 0; i < header->attrs; i++) {
1750 struct perf_header_attr *attr = header->attr[i];
1753 sample_type = attr->attr.sample_type;
1754 else if (sample_type != attr->attr.sample_type)
1755 die("non matching sample_type");
1761 static int __cmd_report(void)
1763 int ret, rc = EXIT_FAILURE;
1764 unsigned long offset = 0;
1765 unsigned long head, shift;
1771 register_idle_thread();
1773 input = open(input_name, O_RDONLY);
1775 fprintf(stderr, " failed to open file: %s", input_name);
1776 if (!strcmp(input_name, "perf.data"))
1777 fprintf(stderr, " (try 'perf record' first)");
1778 fprintf(stderr, "\n");
1782 ret = fstat(input, &stat);
1784 perror("failed to stat file");
1788 if (!stat.st_size) {
1789 fprintf(stderr, "zero-sized file, nothing to do!\n");
1793 header = perf_header__read(input);
1794 head = header->data_offset;
1796 sample_type = perf_header__sample_type();
1798 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
1799 if (sort__has_parent) {
1800 fprintf(stderr, "selected --sort parent, but no"
1801 " callchain data. Did you call"
1802 " perf record without -g?\n");
1806 fprintf(stderr, "selected -c but no callchain data."
1807 " Did you call perf record without"
1813 if (load_kernel() < 0) {
1814 perror("failed to load kernel symbols");
1815 return EXIT_FAILURE;
1819 if (getcwd(__cwd, sizeof(__cwd)) == NULL) {
1820 perror("failed to get the current directory");
1821 return EXIT_FAILURE;
1823 cwdlen = strlen(cwd);
1829 shift = page_size * (head / page_size);
1834 buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
1835 MAP_SHARED, input, offset);
1836 if (buf == MAP_FAILED) {
1837 perror("failed to mmap file");
1842 event = (event_t *)(buf + head);
1844 size = event->header.size;
1848 if (head + event->header.size >= page_size * mmap_window) {
1851 shift = page_size * (head / page_size);
1853 ret = munmap(buf, page_size * mmap_window);
1861 size = event->header.size;
1863 dprintf("\n%p [%p]: event: %d\n",
1864 (void *)(offset + head),
1865 (void *)(long)event->header.size,
1866 event->header.type);
1868 if (!size || process_event(event, offset, head) < 0) {
1870 dprintf("%p [%p]: skipping unknown header type: %d\n",
1871 (void *)(offset + head),
1872 (void *)(long)(event->header.size),
1873 event->header.type);
1878 * assume we lost track of the stream, check alignment, and
1879 * increment a single u64 in the hope to catch on again 'soon'.
1882 if (unlikely(head & 7))
1890 if (offset + head >= header->data_offset + header->data_size)
1893 if (offset + head < (unsigned long)stat.st_size)
1900 dprintf(" IP events: %10ld\n", total);
1901 dprintf(" mmap events: %10ld\n", total_mmap);
1902 dprintf(" comm events: %10ld\n", total_comm);
1903 dprintf(" fork events: %10ld\n", total_fork);
1904 dprintf(" lost events: %10ld\n", total_lost);
1905 dprintf(" unknown events: %10ld\n", total_unknown);
1911 threads__fprintf(stdout);
1914 dsos__fprintf(stdout);
1917 output__resort(total);
1918 output__fprintf(stdout, total);
1924 parse_callchain_opt(const struct option *opt __used, const char *arg,
1935 tok = strtok((char *)arg, ",");
1939 /* get the output mode */
1940 if (!strncmp(tok, "graph", strlen(arg)))
1941 callchain_param.mode = CHAIN_GRAPH_ABS;
1943 else if (!strncmp(tok, "flat", strlen(arg)))
1944 callchain_param.mode = CHAIN_FLAT;
1946 else if (!strncmp(tok, "fractal", strlen(arg)))
1947 callchain_param.mode = CHAIN_GRAPH_REL;
1952 /* get the min percentage */
1953 tok = strtok(NULL, ",");
1957 callchain_param.min_percent = strtod(tok, &endptr);
1962 if (register_callchain_param(&callchain_param) < 0) {
1963 fprintf(stderr, "Can't register callchain params\n");
1969 static const char * const report_usage[] = {
1970 "perf report [<options>] <command>",
1974 static const struct option options[] = {
1975 OPT_STRING('i', "input", &input_name, "file",
1977 OPT_BOOLEAN('v', "verbose", &verbose,
1978 "be more verbose (show symbol address, etc)"),
1979 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1980 "dump raw trace in ASCII"),
1981 OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"),
1982 OPT_BOOLEAN('m', "modules", &modules,
1983 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1984 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1985 "sort by key(s): pid, comm, dso, symbol, parent"),
1986 OPT_BOOLEAN('P', "full-paths", &full_paths,
1987 "Don't shorten the pathnames taking into account the cwd"),
1988 OPT_STRING('p', "parent", &parent_pattern, "regex",
1989 "regex filter to identify parent, see: '--sort parent'"),
1990 OPT_BOOLEAN('x', "exclude-other", &exclude_other,
1991 "Only display entries with parent-match"),
1992 OPT_CALLBACK_DEFAULT('c', "callchain", NULL, "output_type,min_percent",
1993 "Display callchains using output_type and min percent threshold. "
1994 "Default: flat,0", &parse_callchain_opt, callchain_default_opt),
1995 OPT_STRING('d', "dsos", &dso_list_str, "dso[,dso...]",
1996 "only consider symbols in these dsos"),
1997 OPT_STRING('C', "comms", &comm_list_str, "comm[,comm...]",
1998 "only consider symbols in these comms"),
1999 OPT_STRING('S', "symbols", &sym_list_str, "symbol[,symbol...]",
2000 "only consider these symbols"),
2001 OPT_STRING('w', "column-widths", &col_width_list_str,
2003 "don't try to adjust column width, use these fixed values"),
2004 OPT_STRING('t', "field-separator", &field_sep, "separator",
2005 "separator for columns, no spaces will be added between "
2006 "columns '.' is reserved."),
2010 static void setup_sorting(void)
2012 char *tmp, *tok, *str = strdup(sort_order);
2014 for (tok = strtok_r(str, ", ", &tmp);
2015 tok; tok = strtok_r(NULL, ", ", &tmp)) {
2016 if (sort_dimension__add(tok) < 0) {
2017 error("Unknown --sort key: `%s'", tok);
2018 usage_with_options(report_usage, options);
2025 static void setup_list(struct strlist **list, const char *list_str,
2026 const char *list_name)
2029 *list = strlist__new(true, list_str);
2031 fprintf(stderr, "problems parsing %s list\n",
2038 int cmd_report(int argc, const char **argv, const char *prefix __used)
2042 page_size = getpagesize();
2044 argc = parse_options(argc, argv, options, report_usage, 0);
2048 if (parent_pattern != default_parent_pattern)
2049 sort_dimension__add("parent");
2054 * Any (unrecognized) arguments left?
2057 usage_with_options(report_usage, options);
2059 setup_list(&dso_list, dso_list_str, "dso");
2060 setup_list(&comm_list, comm_list_str, "comm");
2061 setup_list(&sym_list, sym_list_str, "symbol");
2063 if (field_sep && *field_sep == '.') {
2064 fputs("'.' is the only non valid --field-separator argument\n",
2071 return __cmd_report();