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"
20 #include "util/values.h"
23 #include "util/debug.h"
24 #include "util/header.h"
26 #include "util/parse-options.h"
27 #include "util/parse-events.h"
29 #include "util/thread.h"
31 static char const *input_name = "perf.data";
33 static char default_sort_order[] = "comm,dso,symbol";
34 static char *sort_order = default_sort_order;
35 static char *dso_list_str, *comm_list_str, *sym_list_str,
37 static struct strlist *dso_list, *comm_list, *sym_list;
38 static char *field_sep;
41 static int show_mask = SHOW_KERNEL | SHOW_USER | SHOW_HV;
43 static int full_paths;
44 static int show_nr_samples;
46 static int show_threads;
47 static struct perf_read_values show_threads_values;
49 static char default_pretty_printing_style[] = "normal";
50 static char *pretty_printing_style = default_pretty_printing_style;
52 static unsigned long page_size;
53 static unsigned long mmap_window = 32;
55 static char default_parent_pattern[] = "^sys_|^do_page_fault";
56 static char *parent_pattern = default_parent_pattern;
57 static regex_t parent_regex;
59 static int exclude_other = 1;
61 static char callchain_default_opt[] = "fractal,0.5";
65 static char __cwd[PATH_MAX];
66 static char *cwd = __cwd;
69 static struct rb_root threads;
70 static struct thread *last_match;
72 static struct perf_header *header;
75 struct callchain_param callchain_param = {
76 .mode = CHAIN_GRAPH_REL,
80 static u64 sample_type;
82 static int repsep_fprintf(FILE *fp, const char *fmt, ...)
89 n = vfprintf(fp, fmt, ap);
92 n = vasprintf(&bf, fmt, ap);
97 sep = strchr(sep, *field_sep);
110 static unsigned int dsos__col_width,
115 * histogram, sorted on item, collects counts
118 static struct rb_root hist;
121 struct rb_node rb_node;
123 struct thread *thread;
127 struct symbol *parent;
130 struct callchain_node callchain;
131 struct rb_root sorted_chain;
137 * configurable sorting bits
141 struct list_head list;
145 int64_t (*cmp)(struct hist_entry *, struct hist_entry *);
146 int64_t (*collapse)(struct hist_entry *, struct hist_entry *);
147 size_t (*print)(FILE *fp, struct hist_entry *, unsigned int width);
152 static int64_t cmp_null(void *l, void *r)
165 sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
167 return right->thread->pid - left->thread->pid;
171 sort__thread_print(FILE *fp, struct hist_entry *self, unsigned int width)
173 return repsep_fprintf(fp, "%*s:%5d", width - 6,
174 self->thread->comm ?: "", self->thread->pid);
177 static struct sort_entry sort_thread = {
178 .header = "Command: Pid",
179 .cmp = sort__thread_cmp,
180 .print = sort__thread_print,
181 .width = &threads__col_width,
187 sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
189 return right->thread->pid - left->thread->pid;
193 sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
195 char *comm_l = left->thread->comm;
196 char *comm_r = right->thread->comm;
198 if (!comm_l || !comm_r)
199 return cmp_null(comm_l, comm_r);
201 return strcmp(comm_l, comm_r);
205 sort__comm_print(FILE *fp, struct hist_entry *self, unsigned int width)
207 return repsep_fprintf(fp, "%*s", width, self->thread->comm);
210 static struct sort_entry sort_comm = {
212 .cmp = sort__comm_cmp,
213 .collapse = sort__comm_collapse,
214 .print = sort__comm_print,
215 .width = &comms__col_width,
221 sort__dso_cmp(struct hist_entry *left, struct hist_entry *right)
223 struct dso *dso_l = left->dso;
224 struct dso *dso_r = right->dso;
226 if (!dso_l || !dso_r)
227 return cmp_null(dso_l, dso_r);
229 return strcmp(dso_l->name, dso_r->name);
233 sort__dso_print(FILE *fp, struct hist_entry *self, unsigned int width)
236 return repsep_fprintf(fp, "%-*s", width, self->dso->name);
238 return repsep_fprintf(fp, "%*llx", width, (u64)self->ip);
241 static struct sort_entry sort_dso = {
242 .header = "Shared Object",
243 .cmp = sort__dso_cmp,
244 .print = sort__dso_print,
245 .width = &dsos__col_width,
251 sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
255 if (left->sym == right->sym)
258 ip_l = left->sym ? left->sym->start : left->ip;
259 ip_r = right->sym ? right->sym->start : right->ip;
261 return (int64_t)(ip_r - ip_l);
265 sort__sym_print(FILE *fp, struct hist_entry *self, unsigned int width __used)
270 ret += repsep_fprintf(fp, "%#018llx %c ", (u64)self->ip,
271 dso__symtab_origin(self->dso));
273 ret += repsep_fprintf(fp, "[%c] ", self->level);
275 ret += repsep_fprintf(fp, "%s", self->sym->name);
277 if (self->sym->module)
278 ret += repsep_fprintf(fp, "\t[%s]",
279 self->sym->module->name);
281 ret += repsep_fprintf(fp, "%#016llx", (u64)self->ip);
287 static struct sort_entry sort_sym = {
289 .cmp = sort__sym_cmp,
290 .print = sort__sym_print,
296 sort__parent_cmp(struct hist_entry *left, struct hist_entry *right)
298 struct symbol *sym_l = left->parent;
299 struct symbol *sym_r = right->parent;
301 if (!sym_l || !sym_r)
302 return cmp_null(sym_l, sym_r);
304 return strcmp(sym_l->name, sym_r->name);
308 sort__parent_print(FILE *fp, struct hist_entry *self, unsigned int width)
310 return repsep_fprintf(fp, "%-*s", width,
311 self->parent ? self->parent->name : "[other]");
314 static unsigned int parent_symbol__col_width;
316 static struct sort_entry sort_parent = {
317 .header = "Parent symbol",
318 .cmp = sort__parent_cmp,
319 .print = sort__parent_print,
320 .width = &parent_symbol__col_width,
323 static int sort__need_collapse = 0;
324 static int sort__has_parent = 0;
326 struct sort_dimension {
328 struct sort_entry *entry;
332 static struct sort_dimension sort_dimensions[] = {
333 { .name = "pid", .entry = &sort_thread, },
334 { .name = "comm", .entry = &sort_comm, },
335 { .name = "dso", .entry = &sort_dso, },
336 { .name = "symbol", .entry = &sort_sym, },
337 { .name = "parent", .entry = &sort_parent, },
340 static LIST_HEAD(hist_entry__sort_list);
342 static int sort_dimension__add(const char *tok)
346 for (i = 0; i < ARRAY_SIZE(sort_dimensions); i++) {
347 struct sort_dimension *sd = &sort_dimensions[i];
352 if (strncasecmp(tok, sd->name, strlen(tok)))
355 if (sd->entry->collapse)
356 sort__need_collapse = 1;
358 if (sd->entry == &sort_parent) {
359 int ret = regcomp(&parent_regex, parent_pattern, REG_EXTENDED);
363 regerror(ret, &parent_regex, err, sizeof(err));
364 fprintf(stderr, "Invalid regex: %s\n%s",
365 parent_pattern, err);
368 sort__has_parent = 1;
371 list_add_tail(&sd->entry->list, &hist_entry__sort_list);
381 hist_entry__cmp(struct hist_entry *left, struct hist_entry *right)
383 struct sort_entry *se;
386 list_for_each_entry(se, &hist_entry__sort_list, list) {
387 cmp = se->cmp(left, right);
396 hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
398 struct sort_entry *se;
401 list_for_each_entry(se, &hist_entry__sort_list, list) {
402 int64_t (*f)(struct hist_entry *, struct hist_entry *);
404 f = se->collapse ?: se->cmp;
406 cmp = f(left, right);
414 static size_t ipchain__fprintf_graph_line(FILE *fp, int depth, int depth_mask)
419 ret += fprintf(fp, "%s", " ");
421 for (i = 0; i < depth; i++)
422 if (depth_mask & (1 << i))
423 ret += fprintf(fp, "| ");
425 ret += fprintf(fp, " ");
427 ret += fprintf(fp, "\n");
432 ipchain__fprintf_graph(FILE *fp, struct callchain_list *chain, int depth,
433 int depth_mask, int count, u64 total_samples,
439 ret += fprintf(fp, "%s", " ");
440 for (i = 0; i < depth; i++) {
441 if (depth_mask & (1 << i))
442 ret += fprintf(fp, "|");
444 ret += fprintf(fp, " ");
445 if (!count && i == depth - 1) {
448 percent = hits * 100.0 / total_samples;
449 ret += percent_color_fprintf(fp, "--%2.2f%%-- ", percent);
451 ret += fprintf(fp, "%s", " ");
454 ret += fprintf(fp, "%s\n", chain->sym->name);
456 ret += fprintf(fp, "%p\n", (void *)(long)chain->ip);
461 static struct symbol *rem_sq_bracket;
462 static struct callchain_list rem_hits;
464 static void init_rem_hits(void)
466 rem_sq_bracket = malloc(sizeof(*rem_sq_bracket) + 6);
467 if (!rem_sq_bracket) {
468 fprintf(stderr, "Not enough memory to display remaining hits\n");
472 strcpy(rem_sq_bracket->name, "[...]");
473 rem_hits.sym = rem_sq_bracket;
477 callchain__fprintf_graph(FILE *fp, struct callchain_node *self,
478 u64 total_samples, int depth, int depth_mask)
480 struct rb_node *node, *next;
481 struct callchain_node *child;
482 struct callchain_list *chain;
483 int new_depth_mask = depth_mask;
489 if (callchain_param.mode == CHAIN_GRAPH_REL)
490 new_total = self->children_hit;
492 new_total = total_samples;
494 remaining = new_total;
496 node = rb_first(&self->rb_root);
500 child = rb_entry(node, struct callchain_node, rb_node);
501 cumul = cumul_hits(child);
505 * The depth mask manages the output of pipes that show
506 * the depth. We don't want to keep the pipes of the current
507 * level for the last child of this depth.
508 * Except if we have remaining filtered hits. They will
509 * supersede the last child
511 next = rb_next(node);
512 if (!next && (callchain_param.mode != CHAIN_GRAPH_REL || !remaining))
513 new_depth_mask &= ~(1 << (depth - 1));
516 * But we keep the older depth mask for the line seperator
517 * to keep the level link until we reach the last child
519 ret += ipchain__fprintf_graph_line(fp, depth, depth_mask);
521 list_for_each_entry(chain, &child->val, list) {
522 if (chain->ip >= PERF_CONTEXT_MAX)
524 ret += ipchain__fprintf_graph(fp, chain, depth,
529 ret += callchain__fprintf_graph(fp, child, new_total,
531 new_depth_mask | (1 << depth));
535 if (callchain_param.mode == CHAIN_GRAPH_REL &&
536 remaining && remaining != new_total) {
541 new_depth_mask &= ~(1 << (depth - 1));
543 ret += ipchain__fprintf_graph(fp, &rem_hits, depth,
544 new_depth_mask, 0, new_total,
552 callchain__fprintf_flat(FILE *fp, struct callchain_node *self,
555 struct callchain_list *chain;
561 ret += callchain__fprintf_flat(fp, self->parent, total_samples);
564 list_for_each_entry(chain, &self->val, list) {
565 if (chain->ip >= PERF_CONTEXT_MAX)
568 ret += fprintf(fp, " %s\n", chain->sym->name);
570 ret += fprintf(fp, " %p\n",
571 (void *)(long)chain->ip);
578 hist_entry_callchain__fprintf(FILE *fp, struct hist_entry *self,
581 struct rb_node *rb_node;
582 struct callchain_node *chain;
585 rb_node = rb_first(&self->sorted_chain);
589 chain = rb_entry(rb_node, struct callchain_node, rb_node);
590 percent = chain->hit * 100.0 / total_samples;
591 switch (callchain_param.mode) {
593 ret += percent_color_fprintf(fp, " %6.2f%%\n",
595 ret += callchain__fprintf_flat(fp, chain, total_samples);
597 case CHAIN_GRAPH_ABS: /* Falldown */
598 case CHAIN_GRAPH_REL:
599 ret += callchain__fprintf_graph(fp, chain,
600 total_samples, 1, 1);
605 ret += fprintf(fp, "\n");
606 rb_node = rb_next(rb_node);
614 hist_entry__fprintf(FILE *fp, struct hist_entry *self, u64 total_samples)
616 struct sort_entry *se;
619 if (exclude_other && !self->parent)
623 ret = percent_color_fprintf(fp,
624 field_sep ? "%.2f" : " %6.2f%%",
625 (self->count * 100.0) / total_samples);
627 ret = fprintf(fp, field_sep ? "%lld" : "%12lld ", self->count);
629 if (show_nr_samples) {
631 fprintf(fp, "%c%lld", *field_sep, self->count);
633 fprintf(fp, "%11lld", self->count);
636 list_for_each_entry(se, &hist_entry__sort_list, list) {
640 fprintf(fp, "%s", field_sep ?: " ");
641 ret += se->print(fp, self, se->width ? *se->width : 0);
644 ret += fprintf(fp, "\n");
647 hist_entry_callchain__fprintf(fp, self, total_samples);
656 static void dso__calc_col_width(struct dso *self)
658 if (!col_width_list_str && !field_sep &&
659 (!dso_list || strlist__has_entry(dso_list, self->name))) {
660 unsigned int slen = strlen(self->name);
661 if (slen > dsos__col_width)
662 dsos__col_width = slen;
665 self->slen_calculated = 1;
668 static int thread__set_comm_adjust(struct thread *self, const char *comm)
670 int ret = thread__set_comm(self, comm);
675 if (!col_width_list_str && !field_sep &&
676 (!comm_list || strlist__has_entry(comm_list, comm))) {
677 unsigned int slen = strlen(comm);
679 if (slen > comms__col_width) {
680 comms__col_width = slen;
681 threads__col_width = slen + 6;
689 static struct symbol *
690 resolve_symbol(struct thread *thread, struct map **mapp,
691 struct dso **dsop, u64 *ipp)
693 struct dso *dso = dsop ? *dsop : NULL;
694 struct map *map = mapp ? *mapp : NULL;
706 map = thread__find_map(thread, ip);
709 * We have to do this here as we may have a dso
710 * with no symbol hit that has a name longer than
711 * the ones with symbols sampled.
713 if (!sort_dso.elide && !map->dso->slen_calculated)
714 dso__calc_col_width(map->dso);
719 ip = map->map_ip(map, ip);
724 * If this is outside of all known maps,
725 * and is a negative address, try to look it
726 * up in the kernel dso, as it might be a
727 * vsyscall (which executes in user-mode):
729 if ((long long)ip < 0)
732 dump_printf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
733 dump_printf(" ...... map: %Lx -> %Lx\n", *ipp, ip);
742 return dso->find_symbol(dso, ip);
745 static int call__match(struct symbol *sym)
747 if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
753 static struct symbol **
754 resolve_callchain(struct thread *thread, struct map *map __used,
755 struct ip_callchain *chain, struct hist_entry *entry)
757 u64 context = PERF_CONTEXT_MAX;
758 struct symbol **syms = NULL;
762 syms = calloc(chain->nr, sizeof(*syms));
764 fprintf(stderr, "Can't allocate memory for symbols\n");
769 for (i = 0; i < chain->nr; i++) {
770 u64 ip = chain->ips[i];
771 struct dso *dso = NULL;
774 if (ip >= PERF_CONTEXT_MAX) {
780 case PERF_CONTEXT_HV:
781 dso = hypervisor_dso;
783 case PERF_CONTEXT_KERNEL:
790 sym = resolve_symbol(thread, NULL, &dso, &ip);
793 if (sort__has_parent && call__match(sym) &&
806 * collect histogram counts
810 hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
811 struct symbol *sym, u64 ip, struct ip_callchain *chain,
812 char level, u64 count)
814 struct rb_node **p = &hist.rb_node;
815 struct rb_node *parent = NULL;
816 struct hist_entry *he;
817 struct symbol **syms = NULL;
818 struct hist_entry entry = {
827 .sorted_chain = RB_ROOT
831 if ((sort__has_parent || callchain) && chain)
832 syms = resolve_callchain(thread, map, chain, &entry);
836 he = rb_entry(parent, struct hist_entry, rb_node);
838 cmp = hist_entry__cmp(&entry, he);
843 append_chain(&he->callchain, chain, syms);
855 he = malloc(sizeof(*he));
860 callchain_init(&he->callchain);
861 append_chain(&he->callchain, chain, syms);
864 rb_link_node(&he->rb_node, parent, p);
865 rb_insert_color(&he->rb_node, &hist);
870 static void hist_entry__free(struct hist_entry *he)
876 * collapse the histogram
879 static struct rb_root collapse_hists;
881 static void collapse__insert_entry(struct hist_entry *he)
883 struct rb_node **p = &collapse_hists.rb_node;
884 struct rb_node *parent = NULL;
885 struct hist_entry *iter;
890 iter = rb_entry(parent, struct hist_entry, rb_node);
892 cmp = hist_entry__collapse(iter, he);
895 iter->count += he->count;
896 hist_entry__free(he);
906 rb_link_node(&he->rb_node, parent, p);
907 rb_insert_color(&he->rb_node, &collapse_hists);
910 static void collapse__resort(void)
912 struct rb_node *next;
913 struct hist_entry *n;
915 if (!sort__need_collapse)
918 next = rb_first(&hist);
920 n = rb_entry(next, struct hist_entry, rb_node);
921 next = rb_next(&n->rb_node);
923 rb_erase(&n->rb_node, &hist);
924 collapse__insert_entry(n);
929 * reverse the map, sort on count.
932 static struct rb_root output_hists;
934 static void output__insert_entry(struct hist_entry *he, u64 min_callchain_hits)
936 struct rb_node **p = &output_hists.rb_node;
937 struct rb_node *parent = NULL;
938 struct hist_entry *iter;
941 callchain_param.sort(&he->sorted_chain, &he->callchain,
942 min_callchain_hits, &callchain_param);
946 iter = rb_entry(parent, struct hist_entry, rb_node);
948 if (he->count > iter->count)
954 rb_link_node(&he->rb_node, parent, p);
955 rb_insert_color(&he->rb_node, &output_hists);
958 static void output__resort(u64 total_samples)
960 struct rb_node *next;
961 struct hist_entry *n;
962 struct rb_root *tree = &hist;
963 u64 min_callchain_hits;
965 min_callchain_hits = total_samples * (callchain_param.min_percent / 100);
967 if (sort__need_collapse)
968 tree = &collapse_hists;
970 next = rb_first(tree);
973 n = rb_entry(next, struct hist_entry, rb_node);
974 next = rb_next(&n->rb_node);
976 rb_erase(&n->rb_node, tree);
977 output__insert_entry(n, min_callchain_hits);
981 static size_t output__fprintf(FILE *fp, u64 total_samples)
983 struct hist_entry *pos;
984 struct sort_entry *se;
988 char *col_width = col_width_list_str;
989 int raw_printing_style;
991 raw_printing_style = !strcmp(pretty_printing_style, "raw");
995 fprintf(fp, "# Samples: %Ld\n", (u64)total_samples);
998 fprintf(fp, "# Overhead");
999 if (show_nr_samples) {
1001 fprintf(fp, "%cSamples", *field_sep);
1003 fputs(" Samples ", fp);
1005 list_for_each_entry(se, &hist_entry__sort_list, list) {
1009 fprintf(fp, "%c%s", *field_sep, se->header);
1012 width = strlen(se->header);
1014 if (col_width_list_str) {
1016 *se->width = atoi(col_width);
1017 col_width = strchr(col_width, ',');
1022 width = *se->width = max(*se->width, width);
1024 fprintf(fp, " %*s", width, se->header);
1031 fprintf(fp, "# ........");
1032 if (show_nr_samples)
1033 fprintf(fp, " ..........");
1034 list_for_each_entry(se, &hist_entry__sort_list, list) {
1044 width = strlen(se->header);
1045 for (i = 0; i < width; i++)
1053 for (nd = rb_first(&output_hists); nd; nd = rb_next(nd)) {
1054 pos = rb_entry(nd, struct hist_entry, rb_node);
1055 ret += hist_entry__fprintf(fp, pos, total_samples);
1058 if (sort_order == default_sort_order &&
1059 parent_pattern == default_parent_pattern) {
1061 fprintf(fp, "# (For a higher level overview, try: perf report --sort comm,dso)\n");
1066 free(rem_sq_bracket);
1069 perf_read_values_display(fp, &show_threads_values,
1070 raw_printing_style);
1075 static void register_idle_thread(void)
1077 struct thread *thread = threads__findnew(0, &threads, &last_match);
1079 if (thread == NULL ||
1080 thread__set_comm_adjust(thread, "[idle]")) {
1081 fprintf(stderr, "problem inserting idle task.\n");
1086 static unsigned long total = 0,
1093 static int validate_chain(struct ip_callchain *chain, event_t *event)
1095 unsigned int chain_size;
1097 chain_size = event->header.size;
1098 chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;
1100 if (chain->nr*sizeof(u64) > chain_size)
1107 process_sample_event(event_t *event, unsigned long offset, unsigned long head)
1111 struct dso *dso = NULL;
1112 struct thread *thread;
1113 u64 ip = event->ip.ip;
1115 struct map *map = NULL;
1116 void *more_data = event->ip.__more_data;
1117 struct ip_callchain *chain = NULL;
1120 thread = threads__findnew(event->ip.pid, &threads, &last_match);
1122 if (sample_type & PERF_SAMPLE_PERIOD) {
1123 period = *(u64 *)more_data;
1124 more_data += sizeof(u64);
1127 dump_printf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n",
1128 (void *)(offset + head),
1129 (void *)(long)(event->header.size),
1131 event->ip.pid, event->ip.tid,
1135 if (sample_type & PERF_SAMPLE_CALLCHAIN) {
1138 chain = (void *)more_data;
1140 dump_printf("... chain: nr:%Lu\n", chain->nr);
1142 if (validate_chain(chain, event) < 0) {
1143 eprintf("call-chain problem with event, skipping it.\n");
1148 for (i = 0; i < chain->nr; i++)
1149 dump_printf("..... %2d: %016Lx\n", i, chain->ips[i]);
1153 dump_printf(" ... thread: %s:%d\n", thread->comm, thread->pid);
1155 if (thread == NULL) {
1156 eprintf("problem processing %d event, skipping it.\n",
1157 event->header.type);
1161 if (comm_list && !strlist__has_entry(comm_list, thread->comm))
1164 cpumode = event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK;
1166 if (cpumode == PERF_EVENT_MISC_KERNEL) {
1172 dump_printf(" ...... dso: %s\n", dso->name);
1174 } else if (cpumode == PERF_EVENT_MISC_USER) {
1183 dso = hypervisor_dso;
1185 dump_printf(" ...... dso: [hypervisor]\n");
1188 if (show & show_mask) {
1189 struct symbol *sym = resolve_symbol(thread, &map, &dso, &ip);
1191 if (dso_list && (!dso || !dso->name ||
1192 !strlist__has_entry(dso_list, dso->name)))
1195 if (sym_list && (!sym || !strlist__has_entry(sym_list, sym->name)))
1198 if (hist_entry__add(thread, map, dso, sym, ip, chain, level, period)) {
1199 eprintf("problem incrementing symbol count, skipping event\n");
1209 process_mmap_event(event_t *event, unsigned long offset, unsigned long head)
1211 struct thread *thread;
1212 struct map *map = map__new(&event->mmap, cwd, cwdlen);
1214 thread = threads__findnew(event->mmap.pid, &threads, &last_match);
1216 dump_printf("%p [%p]: PERF_EVENT_MMAP %d/%d: [%p(%p) @ %p]: %s\n",
1217 (void *)(offset + head),
1218 (void *)(long)(event->header.size),
1221 (void *)(long)event->mmap.start,
1222 (void *)(long)event->mmap.len,
1223 (void *)(long)event->mmap.pgoff,
1224 event->mmap.filename);
1226 if (thread == NULL || map == NULL) {
1227 dump_printf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1231 thread__insert_map(thread, map);
1238 process_comm_event(event_t *event, unsigned long offset, unsigned long head)
1240 struct thread *thread;
1242 thread = threads__findnew(event->comm.pid, &threads, &last_match);
1244 dump_printf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
1245 (void *)(offset + head),
1246 (void *)(long)(event->header.size),
1247 event->comm.comm, event->comm.pid);
1249 if (thread == NULL ||
1250 thread__set_comm_adjust(thread, event->comm.comm)) {
1251 dump_printf("problem processing PERF_EVENT_COMM, skipping event.\n");
1260 process_task_event(event_t *event, unsigned long offset, unsigned long head)
1262 struct thread *thread;
1263 struct thread *parent;
1265 thread = threads__findnew(event->fork.pid, &threads, &last_match);
1266 parent = threads__findnew(event->fork.ppid, &threads, &last_match);
1268 dump_printf("%p [%p]: PERF_EVENT_%s: (%d:%d):(%d:%d)\n",
1269 (void *)(offset + head),
1270 (void *)(long)(event->header.size),
1271 event->header.type == PERF_EVENT_FORK ? "FORK" : "EXIT",
1272 event->fork.pid, event->fork.tid,
1273 event->fork.ppid, event->fork.ptid);
1276 * A thread clone will have the same PID for both
1279 if (thread == parent)
1282 if (event->header.type == PERF_EVENT_EXIT)
1285 if (!thread || !parent || thread__fork(thread, parent)) {
1286 dump_printf("problem processing PERF_EVENT_FORK, skipping event.\n");
1295 process_lost_event(event_t *event, unsigned long offset, unsigned long head)
1297 dump_printf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n",
1298 (void *)(offset + head),
1299 (void *)(long)(event->header.size),
1303 total_lost += event->lost.lost;
1309 process_read_event(event_t *event, unsigned long offset, unsigned long head)
1311 struct perf_counter_attr *attr;
1313 attr = perf_header__find_attr(event->read.id, header);
1316 const char *name = attr ? __event_name(attr->type, attr->config)
1318 perf_read_values_add_value(&show_threads_values,
1319 event->read.pid, event->read.tid,
1325 dump_printf("%p [%p]: PERF_EVENT_READ: %d %d %s %Lu\n",
1326 (void *)(offset + head),
1327 (void *)(long)(event->header.size),
1330 attr ? __event_name(attr->type, attr->config)
1338 process_event(event_t *event, unsigned long offset, unsigned long head)
1342 switch (event->header.type) {
1343 case PERF_EVENT_SAMPLE:
1344 return process_sample_event(event, offset, head);
1346 case PERF_EVENT_MMAP:
1347 return process_mmap_event(event, offset, head);
1349 case PERF_EVENT_COMM:
1350 return process_comm_event(event, offset, head);
1352 case PERF_EVENT_FORK:
1353 case PERF_EVENT_EXIT:
1354 return process_task_event(event, offset, head);
1356 case PERF_EVENT_LOST:
1357 return process_lost_event(event, offset, head);
1359 case PERF_EVENT_READ:
1360 return process_read_event(event, offset, head);
1363 * We dont process them right now but they are fine:
1366 case PERF_EVENT_THROTTLE:
1367 case PERF_EVENT_UNTHROTTLE:
1377 static int __cmd_report(void)
1379 int ret, rc = EXIT_FAILURE;
1380 unsigned long offset = 0;
1381 unsigned long head, shift;
1382 struct stat input_stat;
1387 register_idle_thread();
1390 perf_read_values_init(&show_threads_values);
1392 input = open(input_name, O_RDONLY);
1394 fprintf(stderr, " failed to open file: %s", input_name);
1395 if (!strcmp(input_name, "perf.data"))
1396 fprintf(stderr, " (try 'perf record' first)");
1397 fprintf(stderr, "\n");
1401 ret = fstat(input, &input_stat);
1403 perror("failed to stat file");
1407 if (!input_stat.st_size) {
1408 fprintf(stderr, "zero-sized file, nothing to do!\n");
1412 header = perf_header__read(input);
1413 head = header->data_offset;
1415 sample_type = perf_header__sample_type(header);
1417 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
1418 if (sort__has_parent) {
1419 fprintf(stderr, "selected --sort parent, but no"
1420 " callchain data. Did you call"
1421 " perf record without -g?\n");
1425 fprintf(stderr, "selected -c but no callchain data."
1426 " Did you call perf record without"
1430 } else if (callchain_param.mode != CHAIN_NONE && !callchain) {
1432 if (register_callchain_param(&callchain_param) < 0) {
1433 fprintf(stderr, "Can't register callchain"
1439 if (load_kernel() < 0) {
1440 perror("failed to load kernel symbols");
1441 return EXIT_FAILURE;
1445 if (getcwd(__cwd, sizeof(__cwd)) == NULL) {
1446 perror("failed to get the current directory");
1447 return EXIT_FAILURE;
1449 cwdlen = strlen(cwd);
1455 shift = page_size * (head / page_size);
1460 buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
1461 MAP_SHARED, input, offset);
1462 if (buf == MAP_FAILED) {
1463 perror("failed to mmap file");
1468 event = (event_t *)(buf + head);
1470 size = event->header.size;
1474 if (head + event->header.size >= page_size * mmap_window) {
1477 shift = page_size * (head / page_size);
1479 munmap_ret = munmap(buf, page_size * mmap_window);
1480 assert(munmap_ret == 0);
1487 size = event->header.size;
1489 dump_printf("\n%p [%p]: event: %d\n",
1490 (void *)(offset + head),
1491 (void *)(long)event->header.size,
1492 event->header.type);
1494 if (!size || process_event(event, offset, head) < 0) {
1496 dump_printf("%p [%p]: skipping unknown header type: %d\n",
1497 (void *)(offset + head),
1498 (void *)(long)(event->header.size),
1499 event->header.type);
1504 * assume we lost track of the stream, check alignment, and
1505 * increment a single u64 in the hope to catch on again 'soon'.
1508 if (unlikely(head & 7))
1516 if (offset + head >= header->data_offset + header->data_size)
1519 if (offset + head < (unsigned long)input_stat.st_size)
1526 dump_printf(" IP events: %10ld\n", total);
1527 dump_printf(" mmap events: %10ld\n", total_mmap);
1528 dump_printf(" comm events: %10ld\n", total_comm);
1529 dump_printf(" fork events: %10ld\n", total_fork);
1530 dump_printf(" lost events: %10ld\n", total_lost);
1531 dump_printf(" unknown events: %10ld\n", total_unknown);
1537 threads__fprintf(stdout, &threads);
1540 dsos__fprintf(stdout);
1543 output__resort(total);
1544 output__fprintf(stdout, total);
1547 perf_read_values_destroy(&show_threads_values);
1553 parse_callchain_opt(const struct option *opt __used, const char *arg,
1564 tok = strtok((char *)arg, ",");
1568 /* get the output mode */
1569 if (!strncmp(tok, "graph", strlen(arg)))
1570 callchain_param.mode = CHAIN_GRAPH_ABS;
1572 else if (!strncmp(tok, "flat", strlen(arg)))
1573 callchain_param.mode = CHAIN_FLAT;
1575 else if (!strncmp(tok, "fractal", strlen(arg)))
1576 callchain_param.mode = CHAIN_GRAPH_REL;
1578 else if (!strncmp(tok, "none", strlen(arg))) {
1579 callchain_param.mode = CHAIN_NONE;
1588 /* get the min percentage */
1589 tok = strtok(NULL, ",");
1593 callchain_param.min_percent = strtod(tok, &endptr);
1598 if (register_callchain_param(&callchain_param) < 0) {
1599 fprintf(stderr, "Can't register callchain params\n");
1605 static const char * const report_usage[] = {
1606 "perf report [<options>] <command>",
1610 static const struct option options[] = {
1611 OPT_STRING('i', "input", &input_name, "file",
1613 OPT_BOOLEAN('v', "verbose", &verbose,
1614 "be more verbose (show symbol address, etc)"),
1615 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1616 "dump raw trace in ASCII"),
1617 OPT_STRING('k', "vmlinux", &vmlinux_name, "file", "vmlinux pathname"),
1618 OPT_BOOLEAN('m', "modules", &modules,
1619 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1620 OPT_BOOLEAN('n', "show-nr-samples", &show_nr_samples,
1621 "Show a column with the number of samples"),
1622 OPT_BOOLEAN('T', "threads", &show_threads,
1623 "Show per-thread event counters"),
1624 OPT_STRING(0, "pretty", &pretty_printing_style, "key",
1625 "pretty printing style key: normal raw"),
1626 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1627 "sort by key(s): pid, comm, dso, symbol, parent"),
1628 OPT_BOOLEAN('P', "full-paths", &full_paths,
1629 "Don't shorten the pathnames taking into account the cwd"),
1630 OPT_STRING('p', "parent", &parent_pattern, "regex",
1631 "regex filter to identify parent, see: '--sort parent'"),
1632 OPT_BOOLEAN('x', "exclude-other", &exclude_other,
1633 "Only display entries with parent-match"),
1634 OPT_CALLBACK_DEFAULT('g', "call-graph", NULL, "output_type,min_percent",
1635 "Display callchains using output_type and min percent threshold. "
1636 "Default: fractal,0.5", &parse_callchain_opt, callchain_default_opt),
1637 OPT_STRING('d', "dsos", &dso_list_str, "dso[,dso...]",
1638 "only consider symbols in these dsos"),
1639 OPT_STRING('C', "comms", &comm_list_str, "comm[,comm...]",
1640 "only consider symbols in these comms"),
1641 OPT_STRING('S', "symbols", &sym_list_str, "symbol[,symbol...]",
1642 "only consider these symbols"),
1643 OPT_STRING('w', "column-widths", &col_width_list_str,
1645 "don't try to adjust column width, use these fixed values"),
1646 OPT_STRING('t', "field-separator", &field_sep, "separator",
1647 "separator for columns, no spaces will be added between "
1648 "columns '.' is reserved."),
1652 static void setup_sorting(void)
1654 char *tmp, *tok, *str = strdup(sort_order);
1656 for (tok = strtok_r(str, ", ", &tmp);
1657 tok; tok = strtok_r(NULL, ", ", &tmp)) {
1658 if (sort_dimension__add(tok) < 0) {
1659 error("Unknown --sort key: `%s'", tok);
1660 usage_with_options(report_usage, options);
1667 static void setup_list(struct strlist **list, const char *list_str,
1668 struct sort_entry *se, const char *list_name,
1672 *list = strlist__new(true, list_str);
1674 fprintf(stderr, "problems parsing %s list\n",
1678 if (strlist__nr_entries(*list) == 1) {
1679 fprintf(fp, "# %s: %s\n", list_name,
1680 strlist__entry(*list, 0)->s);
1686 int cmd_report(int argc, const char **argv, const char *prefix __used)
1690 page_size = getpagesize();
1692 argc = parse_options(argc, argv, options, report_usage, 0);
1696 if (parent_pattern != default_parent_pattern) {
1697 sort_dimension__add("parent");
1698 sort_parent.elide = 1;
1703 * Any (unrecognized) arguments left?
1706 usage_with_options(report_usage, options);
1710 setup_list(&dso_list, dso_list_str, &sort_dso, "dso", stdout);
1711 setup_list(&comm_list, comm_list_str, &sort_comm, "comm", stdout);
1712 setup_list(&sym_list, sym_list_str, &sort_sym, "symbol", stdout);
1714 if (field_sep && *field_sep == '.') {
1715 fputs("'.' is the only non valid --field-separator argument\n",
1720 return __cmd_report();