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 struct symbol *
669 resolve_symbol(struct thread *thread, struct map **mapp,
670 struct dso **dsop, u64 *ipp)
672 struct dso *dso = dsop ? *dsop : NULL;
673 struct map *map = mapp ? *mapp : NULL;
685 map = thread__find_map(thread, ip);
688 * We have to do this here as we may have a dso
689 * with no symbol hit that has a name longer than
690 * the ones with symbols sampled.
692 if (!sort_dso.elide && !map->dso->slen_calculated)
693 dso__calc_col_width(map->dso);
698 ip = map->map_ip(map, ip);
703 * If this is outside of all known maps,
704 * and is a negative address, try to look it
705 * up in the kernel dso, as it might be a
706 * vsyscall (which executes in user-mode):
708 if ((long long)ip < 0)
711 dump_printf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
712 dump_printf(" ...... map: %Lx -> %Lx\n", *ipp, ip);
721 return dso->find_symbol(dso, ip);
724 static int call__match(struct symbol *sym)
726 if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
732 static struct symbol **
733 resolve_callchain(struct thread *thread, struct map *map __used,
734 struct ip_callchain *chain, struct hist_entry *entry)
736 u64 context = PERF_CONTEXT_MAX;
737 struct symbol **syms = NULL;
741 syms = calloc(chain->nr, sizeof(*syms));
743 fprintf(stderr, "Can't allocate memory for symbols\n");
748 for (i = 0; i < chain->nr; i++) {
749 u64 ip = chain->ips[i];
750 struct dso *dso = NULL;
753 if (ip >= PERF_CONTEXT_MAX) {
759 case PERF_CONTEXT_HV:
760 dso = hypervisor_dso;
762 case PERF_CONTEXT_KERNEL:
769 sym = resolve_symbol(thread, NULL, &dso, &ip);
772 if (sort__has_parent && call__match(sym) &&
785 * collect histogram counts
789 hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
790 struct symbol *sym, u64 ip, struct ip_callchain *chain,
791 char level, u64 count)
793 struct rb_node **p = &hist.rb_node;
794 struct rb_node *parent = NULL;
795 struct hist_entry *he;
796 struct symbol **syms = NULL;
797 struct hist_entry entry = {
806 .sorted_chain = RB_ROOT
810 if ((sort__has_parent || callchain) && chain)
811 syms = resolve_callchain(thread, map, chain, &entry);
815 he = rb_entry(parent, struct hist_entry, rb_node);
817 cmp = hist_entry__cmp(&entry, he);
822 append_chain(&he->callchain, chain, syms);
834 he = malloc(sizeof(*he));
839 callchain_init(&he->callchain);
840 append_chain(&he->callchain, chain, syms);
843 rb_link_node(&he->rb_node, parent, p);
844 rb_insert_color(&he->rb_node, &hist);
849 static void hist_entry__free(struct hist_entry *he)
855 * collapse the histogram
858 static struct rb_root collapse_hists;
860 static void collapse__insert_entry(struct hist_entry *he)
862 struct rb_node **p = &collapse_hists.rb_node;
863 struct rb_node *parent = NULL;
864 struct hist_entry *iter;
869 iter = rb_entry(parent, struct hist_entry, rb_node);
871 cmp = hist_entry__collapse(iter, he);
874 iter->count += he->count;
875 hist_entry__free(he);
885 rb_link_node(&he->rb_node, parent, p);
886 rb_insert_color(&he->rb_node, &collapse_hists);
889 static void collapse__resort(void)
891 struct rb_node *next;
892 struct hist_entry *n;
894 if (!sort__need_collapse)
897 next = rb_first(&hist);
899 n = rb_entry(next, struct hist_entry, rb_node);
900 next = rb_next(&n->rb_node);
902 rb_erase(&n->rb_node, &hist);
903 collapse__insert_entry(n);
908 * reverse the map, sort on count.
911 static struct rb_root output_hists;
913 static void output__insert_entry(struct hist_entry *he, u64 min_callchain_hits)
915 struct rb_node **p = &output_hists.rb_node;
916 struct rb_node *parent = NULL;
917 struct hist_entry *iter;
920 callchain_param.sort(&he->sorted_chain, &he->callchain,
921 min_callchain_hits, &callchain_param);
925 iter = rb_entry(parent, struct hist_entry, rb_node);
927 if (he->count > iter->count)
933 rb_link_node(&he->rb_node, parent, p);
934 rb_insert_color(&he->rb_node, &output_hists);
937 static void output__resort(u64 total_samples)
939 struct rb_node *next;
940 struct hist_entry *n;
941 struct rb_root *tree = &hist;
942 u64 min_callchain_hits;
944 min_callchain_hits = total_samples * (callchain_param.min_percent / 100);
946 if (sort__need_collapse)
947 tree = &collapse_hists;
949 next = rb_first(tree);
952 n = rb_entry(next, struct hist_entry, rb_node);
953 next = rb_next(&n->rb_node);
955 rb_erase(&n->rb_node, tree);
956 output__insert_entry(n, min_callchain_hits);
960 static size_t output__fprintf(FILE *fp, u64 total_samples)
962 struct hist_entry *pos;
963 struct sort_entry *se;
967 char *col_width = col_width_list_str;
968 int raw_printing_style;
970 raw_printing_style = !strcmp(pretty_printing_style, "raw");
974 fprintf(fp, "# Samples: %Ld\n", (u64)total_samples);
977 fprintf(fp, "# Overhead");
978 if (show_nr_samples) {
980 fprintf(fp, "%cSamples", *field_sep);
982 fputs(" Samples ", fp);
984 list_for_each_entry(se, &hist_entry__sort_list, list) {
988 fprintf(fp, "%c%s", *field_sep, se->header);
991 width = strlen(se->header);
993 if (col_width_list_str) {
995 *se->width = atoi(col_width);
996 col_width = strchr(col_width, ',');
1001 width = *se->width = max(*se->width, width);
1003 fprintf(fp, " %*s", width, se->header);
1010 fprintf(fp, "# ........");
1011 if (show_nr_samples)
1012 fprintf(fp, " ..........");
1013 list_for_each_entry(se, &hist_entry__sort_list, list) {
1023 width = strlen(se->header);
1024 for (i = 0; i < width; i++)
1032 for (nd = rb_first(&output_hists); nd; nd = rb_next(nd)) {
1033 pos = rb_entry(nd, struct hist_entry, rb_node);
1034 ret += hist_entry__fprintf(fp, pos, total_samples);
1037 if (sort_order == default_sort_order &&
1038 parent_pattern == default_parent_pattern) {
1040 fprintf(fp, "# (For a higher level overview, try: perf report --sort comm,dso)\n");
1045 free(rem_sq_bracket);
1048 perf_read_values_display(fp, &show_threads_values,
1049 raw_printing_style);
1054 static void register_idle_thread(void)
1056 struct thread *thread = threads__findnew(0, &threads, &last_match);
1058 if (thread == NULL ||
1059 thread__set_comm(thread, "[idle]")) {
1060 fprintf(stderr, "problem inserting idle task.\n");
1065 static unsigned long total = 0,
1072 static int validate_chain(struct ip_callchain *chain, event_t *event)
1074 unsigned int chain_size;
1076 chain_size = event->header.size;
1077 chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;
1079 if (chain->nr*sizeof(u64) > chain_size)
1086 process_sample_event(event_t *event, unsigned long offset, unsigned long head)
1090 struct dso *dso = NULL;
1091 struct thread *thread;
1092 u64 ip = event->ip.ip;
1094 struct map *map = NULL;
1095 void *more_data = event->ip.__more_data;
1096 struct ip_callchain *chain = NULL;
1099 thread = threads__findnew(event->ip.pid, &threads, &last_match);
1101 if (sample_type & PERF_SAMPLE_PERIOD) {
1102 period = *(u64 *)more_data;
1103 more_data += sizeof(u64);
1106 dump_printf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n",
1107 (void *)(offset + head),
1108 (void *)(long)(event->header.size),
1110 event->ip.pid, event->ip.tid,
1114 if (sample_type & PERF_SAMPLE_CALLCHAIN) {
1117 chain = (void *)more_data;
1119 dump_printf("... chain: nr:%Lu\n", chain->nr);
1121 if (validate_chain(chain, event) < 0) {
1122 eprintf("call-chain problem with event, skipping it.\n");
1127 for (i = 0; i < chain->nr; i++)
1128 dump_printf("..... %2d: %016Lx\n", i, chain->ips[i]);
1132 dump_printf(" ... thread: %s:%d\n", thread->comm, thread->pid);
1134 if (thread == NULL) {
1135 eprintf("problem processing %d event, skipping it.\n",
1136 event->header.type);
1140 if (comm_list && !strlist__has_entry(comm_list, thread->comm))
1143 cpumode = event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK;
1145 if (cpumode == PERF_EVENT_MISC_KERNEL) {
1151 dump_printf(" ...... dso: %s\n", dso->name);
1153 } else if (cpumode == PERF_EVENT_MISC_USER) {
1162 dso = hypervisor_dso;
1164 dump_printf(" ...... dso: [hypervisor]\n");
1167 if (show & show_mask) {
1168 struct symbol *sym = resolve_symbol(thread, &map, &dso, &ip);
1170 if (dso_list && (!dso || !dso->name ||
1171 !strlist__has_entry(dso_list, dso->name)))
1174 if (sym_list && (!sym || !strlist__has_entry(sym_list, sym->name)))
1177 if (hist_entry__add(thread, map, dso, sym, ip, chain, level, period)) {
1178 eprintf("problem incrementing symbol count, skipping event\n");
1188 process_mmap_event(event_t *event, unsigned long offset, unsigned long head)
1190 struct thread *thread;
1191 struct map *map = map__new(&event->mmap, cwd, cwdlen);
1193 thread = threads__findnew(event->mmap.pid, &threads, &last_match);
1195 dump_printf("%p [%p]: PERF_EVENT_MMAP %d/%d: [%p(%p) @ %p]: %s\n",
1196 (void *)(offset + head),
1197 (void *)(long)(event->header.size),
1200 (void *)(long)event->mmap.start,
1201 (void *)(long)event->mmap.len,
1202 (void *)(long)event->mmap.pgoff,
1203 event->mmap.filename);
1205 if (thread == NULL || map == NULL) {
1206 dump_printf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1210 thread__insert_map(thread, map);
1217 process_comm_event(event_t *event, unsigned long offset, unsigned long head)
1219 struct thread *thread;
1221 thread = threads__findnew(event->comm.pid, &threads, &last_match);
1223 dump_printf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
1224 (void *)(offset + head),
1225 (void *)(long)(event->header.size),
1226 event->comm.comm, event->comm.pid);
1228 if (thread == NULL ||
1229 thread__set_comm(thread, event->comm.comm)) {
1230 dump_printf("problem processing PERF_EVENT_COMM, skipping event.\n");
1239 process_task_event(event_t *event, unsigned long offset, unsigned long head)
1241 struct thread *thread;
1242 struct thread *parent;
1244 thread = threads__findnew(event->fork.pid, &threads, &last_match);
1245 parent = threads__findnew(event->fork.ppid, &threads, &last_match);
1247 dump_printf("%p [%p]: PERF_EVENT_%s: (%d:%d):(%d:%d)\n",
1248 (void *)(offset + head),
1249 (void *)(long)(event->header.size),
1250 event->header.type == PERF_EVENT_FORK ? "FORK" : "EXIT",
1251 event->fork.pid, event->fork.tid,
1252 event->fork.ppid, event->fork.ptid);
1255 * A thread clone will have the same PID for both
1258 if (thread == parent)
1261 if (event->header.type == PERF_EVENT_EXIT)
1264 if (!thread || !parent || thread__fork(thread, parent)) {
1265 dump_printf("problem processing PERF_EVENT_FORK, skipping event.\n");
1274 process_lost_event(event_t *event, unsigned long offset, unsigned long head)
1276 dump_printf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n",
1277 (void *)(offset + head),
1278 (void *)(long)(event->header.size),
1282 total_lost += event->lost.lost;
1288 process_read_event(event_t *event, unsigned long offset, unsigned long head)
1290 struct perf_counter_attr *attr;
1292 attr = perf_header__find_attr(event->read.id, header);
1295 const char *name = attr ? __event_name(attr->type, attr->config)
1297 perf_read_values_add_value(&show_threads_values,
1298 event->read.pid, event->read.tid,
1304 dump_printf("%p [%p]: PERF_EVENT_READ: %d %d %s %Lu\n",
1305 (void *)(offset + head),
1306 (void *)(long)(event->header.size),
1309 attr ? __event_name(attr->type, attr->config)
1317 process_event(event_t *event, unsigned long offset, unsigned long head)
1321 switch (event->header.type) {
1322 case PERF_EVENT_SAMPLE:
1323 return process_sample_event(event, offset, head);
1325 case PERF_EVENT_MMAP:
1326 return process_mmap_event(event, offset, head);
1328 case PERF_EVENT_COMM:
1329 return process_comm_event(event, offset, head);
1331 case PERF_EVENT_FORK:
1332 case PERF_EVENT_EXIT:
1333 return process_task_event(event, offset, head);
1335 case PERF_EVENT_LOST:
1336 return process_lost_event(event, offset, head);
1338 case PERF_EVENT_READ:
1339 return process_read_event(event, offset, head);
1342 * We dont process them right now but they are fine:
1345 case PERF_EVENT_THROTTLE:
1346 case PERF_EVENT_UNTHROTTLE:
1356 static int __cmd_report(void)
1358 int ret, rc = EXIT_FAILURE;
1359 unsigned long offset = 0;
1360 unsigned long head, shift;
1361 struct stat input_stat;
1366 register_idle_thread();
1369 perf_read_values_init(&show_threads_values);
1371 input = open(input_name, O_RDONLY);
1373 fprintf(stderr, " failed to open file: %s", input_name);
1374 if (!strcmp(input_name, "perf.data"))
1375 fprintf(stderr, " (try 'perf record' first)");
1376 fprintf(stderr, "\n");
1380 ret = fstat(input, &input_stat);
1382 perror("failed to stat file");
1386 if (!input_stat.st_size) {
1387 fprintf(stderr, "zero-sized file, nothing to do!\n");
1391 header = perf_header__read(input);
1392 head = header->data_offset;
1394 sample_type = perf_header__sample_type(header);
1396 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
1397 if (sort__has_parent) {
1398 fprintf(stderr, "selected --sort parent, but no"
1399 " callchain data. Did you call"
1400 " perf record without -g?\n");
1404 fprintf(stderr, "selected -c but no callchain data."
1405 " Did you call perf record without"
1409 } else if (callchain_param.mode != CHAIN_NONE && !callchain) {
1411 if (register_callchain_param(&callchain_param) < 0) {
1412 fprintf(stderr, "Can't register callchain"
1418 if (load_kernel() < 0) {
1419 perror("failed to load kernel symbols");
1420 return EXIT_FAILURE;
1424 if (getcwd(__cwd, sizeof(__cwd)) == NULL) {
1425 perror("failed to get the current directory");
1426 return EXIT_FAILURE;
1428 cwdlen = strlen(cwd);
1434 shift = page_size * (head / page_size);
1439 buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
1440 MAP_SHARED, input, offset);
1441 if (buf == MAP_FAILED) {
1442 perror("failed to mmap file");
1447 event = (event_t *)(buf + head);
1449 size = event->header.size;
1453 if (head + event->header.size >= page_size * mmap_window) {
1456 shift = page_size * (head / page_size);
1458 munmap_ret = munmap(buf, page_size * mmap_window);
1459 assert(munmap_ret == 0);
1466 size = event->header.size;
1468 dump_printf("\n%p [%p]: event: %d\n",
1469 (void *)(offset + head),
1470 (void *)(long)event->header.size,
1471 event->header.type);
1473 if (!size || process_event(event, offset, head) < 0) {
1475 dump_printf("%p [%p]: skipping unknown header type: %d\n",
1476 (void *)(offset + head),
1477 (void *)(long)(event->header.size),
1478 event->header.type);
1483 * assume we lost track of the stream, check alignment, and
1484 * increment a single u64 in the hope to catch on again 'soon'.
1487 if (unlikely(head & 7))
1495 if (offset + head >= header->data_offset + header->data_size)
1498 if (offset + head < (unsigned long)input_stat.st_size)
1505 dump_printf(" IP events: %10ld\n", total);
1506 dump_printf(" mmap events: %10ld\n", total_mmap);
1507 dump_printf(" comm events: %10ld\n", total_comm);
1508 dump_printf(" fork events: %10ld\n", total_fork);
1509 dump_printf(" lost events: %10ld\n", total_lost);
1510 dump_printf(" unknown events: %10ld\n", total_unknown);
1516 threads__fprintf(stdout, &threads);
1519 dsos__fprintf(stdout);
1522 output__resort(total);
1523 output__fprintf(stdout, total);
1526 perf_read_values_destroy(&show_threads_values);
1532 parse_callchain_opt(const struct option *opt __used, const char *arg,
1543 tok = strtok((char *)arg, ",");
1547 /* get the output mode */
1548 if (!strncmp(tok, "graph", strlen(arg)))
1549 callchain_param.mode = CHAIN_GRAPH_ABS;
1551 else if (!strncmp(tok, "flat", strlen(arg)))
1552 callchain_param.mode = CHAIN_FLAT;
1554 else if (!strncmp(tok, "fractal", strlen(arg)))
1555 callchain_param.mode = CHAIN_GRAPH_REL;
1557 else if (!strncmp(tok, "none", strlen(arg))) {
1558 callchain_param.mode = CHAIN_NONE;
1567 /* get the min percentage */
1568 tok = strtok(NULL, ",");
1572 callchain_param.min_percent = strtod(tok, &endptr);
1577 if (register_callchain_param(&callchain_param) < 0) {
1578 fprintf(stderr, "Can't register callchain params\n");
1584 static const char * const report_usage[] = {
1585 "perf report [<options>] <command>",
1589 static const struct option options[] = {
1590 OPT_STRING('i', "input", &input_name, "file",
1592 OPT_BOOLEAN('v', "verbose", &verbose,
1593 "be more verbose (show symbol address, etc)"),
1594 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1595 "dump raw trace in ASCII"),
1596 OPT_STRING('k', "vmlinux", &vmlinux_name, "file", "vmlinux pathname"),
1597 OPT_BOOLEAN('m', "modules", &modules,
1598 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1599 OPT_BOOLEAN('n', "show-nr-samples", &show_nr_samples,
1600 "Show a column with the number of samples"),
1601 OPT_BOOLEAN('T', "threads", &show_threads,
1602 "Show per-thread event counters"),
1603 OPT_STRING(0, "pretty", &pretty_printing_style, "key",
1604 "pretty printing style key: normal raw"),
1605 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1606 "sort by key(s): pid, comm, dso, symbol, parent"),
1607 OPT_BOOLEAN('P', "full-paths", &full_paths,
1608 "Don't shorten the pathnames taking into account the cwd"),
1609 OPT_STRING('p', "parent", &parent_pattern, "regex",
1610 "regex filter to identify parent, see: '--sort parent'"),
1611 OPT_BOOLEAN('x', "exclude-other", &exclude_other,
1612 "Only display entries with parent-match"),
1613 OPT_CALLBACK_DEFAULT('g', "call-graph", NULL, "output_type,min_percent",
1614 "Display callchains using output_type and min percent threshold. "
1615 "Default: fractal,0.5", &parse_callchain_opt, callchain_default_opt),
1616 OPT_STRING('d', "dsos", &dso_list_str, "dso[,dso...]",
1617 "only consider symbols in these dsos"),
1618 OPT_STRING('C', "comms", &comm_list_str, "comm[,comm...]",
1619 "only consider symbols in these comms"),
1620 OPT_STRING('S', "symbols", &sym_list_str, "symbol[,symbol...]",
1621 "only consider these symbols"),
1622 OPT_STRING('w', "column-widths", &col_width_list_str,
1624 "don't try to adjust column width, use these fixed values"),
1625 OPT_STRING('t', "field-separator", &field_sep, "separator",
1626 "separator for columns, no spaces will be added between "
1627 "columns '.' is reserved."),
1631 static void setup_sorting(void)
1633 char *tmp, *tok, *str = strdup(sort_order);
1635 for (tok = strtok_r(str, ", ", &tmp);
1636 tok; tok = strtok_r(NULL, ", ", &tmp)) {
1637 if (sort_dimension__add(tok) < 0) {
1638 error("Unknown --sort key: `%s'", tok);
1639 usage_with_options(report_usage, options);
1646 static void setup_list(struct strlist **list, const char *list_str,
1647 struct sort_entry *se, const char *list_name,
1651 *list = strlist__new(true, list_str);
1653 fprintf(stderr, "problems parsing %s list\n",
1657 if (strlist__nr_entries(*list) == 1) {
1658 fprintf(fp, "# %s: %s\n", list_name,
1659 strlist__entry(*list, 0)->s);
1665 int cmd_report(int argc, const char **argv, const char *prefix __used)
1669 page_size = getpagesize();
1671 argc = parse_options(argc, argv, options, report_usage, 0);
1675 if (parent_pattern != default_parent_pattern) {
1676 sort_dimension__add("parent");
1677 sort_parent.elide = 1;
1682 * Any (unrecognized) arguments left?
1685 usage_with_options(report_usage, options);
1689 setup_list(&dso_list, dso_list_str, &sort_dso, "dso", stdout);
1690 setup_list(&comm_list, comm_list_str, &sort_comm, "comm", stdout);
1691 setup_list(&sym_list, sym_list_str, &sort_sym, "symbol", stdout);
1693 if (field_sep && *field_sep == '.') {
1694 fputs("'.' is the only non valid --field-separator argument\n",
1699 return __cmd_report();