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/header.h"
25 #include "util/parse-options.h"
26 #include "util/parse-events.h"
28 #include "util/thread.h"
30 static char const *input_name = "perf.data";
32 static char default_sort_order[] = "comm,dso,symbol";
33 static char *sort_order = default_sort_order;
34 static char *dso_list_str, *comm_list_str, *sym_list_str,
36 static struct strlist *dso_list, *comm_list, *sym_list;
37 static char *field_sep;
40 static int show_mask = SHOW_KERNEL | SHOW_USER | SHOW_HV;
42 #define cdprintf(x...) do { if (dump_trace) color_fprintf(stdout, color, x); } while (0)
44 static int full_paths;
45 static int show_nr_samples;
47 static int show_threads;
48 static struct perf_read_values show_threads_values;
50 static char default_pretty_printing_style[] = "normal";
51 static char *pretty_printing_style = default_pretty_printing_style;
53 static unsigned long page_size;
54 static unsigned long mmap_window = 32;
56 static char default_parent_pattern[] = "^sys_|^do_page_fault";
57 static char *parent_pattern = default_parent_pattern;
58 static regex_t parent_regex;
60 static int exclude_other = 1;
62 static char callchain_default_opt[] = "fractal,0.5";
66 static char __cwd[PATH_MAX];
67 static char *cwd = __cwd;
70 static struct rb_root threads;
71 static struct thread *last_match;
74 struct callchain_param callchain_param = {
75 .mode = CHAIN_GRAPH_REL,
79 static u64 sample_type;
81 static int repsep_fprintf(FILE *fp, const char *fmt, ...)
88 n = vfprintf(fp, fmt, ap);
91 n = vasprintf(&bf, fmt, ap);
96 sep = strchr(sep, *field_sep);
109 static unsigned int dsos__col_width,
114 * histogram, sorted on item, collects counts
117 static struct rb_root hist;
120 struct rb_node rb_node;
122 struct thread *thread;
126 struct symbol *parent;
129 struct callchain_node callchain;
130 struct rb_root sorted_chain;
136 * configurable sorting bits
140 struct list_head list;
144 int64_t (*cmp)(struct hist_entry *, struct hist_entry *);
145 int64_t (*collapse)(struct hist_entry *, struct hist_entry *);
146 size_t (*print)(FILE *fp, struct hist_entry *, unsigned int width);
151 static int64_t cmp_null(void *l, void *r)
164 sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
166 return right->thread->pid - left->thread->pid;
170 sort__thread_print(FILE *fp, struct hist_entry *self, unsigned int width)
172 return repsep_fprintf(fp, "%*s:%5d", width - 6,
173 self->thread->comm ?: "", self->thread->pid);
176 static struct sort_entry sort_thread = {
177 .header = "Command: Pid",
178 .cmp = sort__thread_cmp,
179 .print = sort__thread_print,
180 .width = &threads__col_width,
186 sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
188 return right->thread->pid - left->thread->pid;
192 sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
194 char *comm_l = left->thread->comm;
195 char *comm_r = right->thread->comm;
197 if (!comm_l || !comm_r)
198 return cmp_null(comm_l, comm_r);
200 return strcmp(comm_l, comm_r);
204 sort__comm_print(FILE *fp, struct hist_entry *self, unsigned int width)
206 return repsep_fprintf(fp, "%*s", width, self->thread->comm);
209 static struct sort_entry sort_comm = {
211 .cmp = sort__comm_cmp,
212 .collapse = sort__comm_collapse,
213 .print = sort__comm_print,
214 .width = &comms__col_width,
220 sort__dso_cmp(struct hist_entry *left, struct hist_entry *right)
222 struct dso *dso_l = left->dso;
223 struct dso *dso_r = right->dso;
225 if (!dso_l || !dso_r)
226 return cmp_null(dso_l, dso_r);
228 return strcmp(dso_l->name, dso_r->name);
232 sort__dso_print(FILE *fp, struct hist_entry *self, unsigned int width)
235 return repsep_fprintf(fp, "%-*s", width, self->dso->name);
237 return repsep_fprintf(fp, "%*llx", width, (u64)self->ip);
240 static struct sort_entry sort_dso = {
241 .header = "Shared Object",
242 .cmp = sort__dso_cmp,
243 .print = sort__dso_print,
244 .width = &dsos__col_width,
250 sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
254 if (left->sym == right->sym)
257 ip_l = left->sym ? left->sym->start : left->ip;
258 ip_r = right->sym ? right->sym->start : right->ip;
260 return (int64_t)(ip_r - ip_l);
264 sort__sym_print(FILE *fp, struct hist_entry *self, unsigned int width __used)
269 ret += repsep_fprintf(fp, "%#018llx %c ", (u64)self->ip,
270 dso__symtab_origin(self->dso));
272 ret += repsep_fprintf(fp, "[%c] ", self->level);
274 ret += repsep_fprintf(fp, "%s", self->sym->name);
276 if (self->sym->module)
277 ret += repsep_fprintf(fp, "\t[%s]",
278 self->sym->module->name);
280 ret += repsep_fprintf(fp, "%#016llx", (u64)self->ip);
286 static struct sort_entry sort_sym = {
288 .cmp = sort__sym_cmp,
289 .print = sort__sym_print,
295 sort__parent_cmp(struct hist_entry *left, struct hist_entry *right)
297 struct symbol *sym_l = left->parent;
298 struct symbol *sym_r = right->parent;
300 if (!sym_l || !sym_r)
301 return cmp_null(sym_l, sym_r);
303 return strcmp(sym_l->name, sym_r->name);
307 sort__parent_print(FILE *fp, struct hist_entry *self, unsigned int width)
309 return repsep_fprintf(fp, "%-*s", width,
310 self->parent ? self->parent->name : "[other]");
313 static unsigned int parent_symbol__col_width;
315 static struct sort_entry sort_parent = {
316 .header = "Parent symbol",
317 .cmp = sort__parent_cmp,
318 .print = sort__parent_print,
319 .width = &parent_symbol__col_width,
322 static int sort__need_collapse = 0;
323 static int sort__has_parent = 0;
325 struct sort_dimension {
327 struct sort_entry *entry;
331 static struct sort_dimension sort_dimensions[] = {
332 { .name = "pid", .entry = &sort_thread, },
333 { .name = "comm", .entry = &sort_comm, },
334 { .name = "dso", .entry = &sort_dso, },
335 { .name = "symbol", .entry = &sort_sym, },
336 { .name = "parent", .entry = &sort_parent, },
339 static LIST_HEAD(hist_entry__sort_list);
341 static int sort_dimension__add(const char *tok)
345 for (i = 0; i < ARRAY_SIZE(sort_dimensions); i++) {
346 struct sort_dimension *sd = &sort_dimensions[i];
351 if (strncasecmp(tok, sd->name, strlen(tok)))
354 if (sd->entry->collapse)
355 sort__need_collapse = 1;
357 if (sd->entry == &sort_parent) {
358 int ret = regcomp(&parent_regex, parent_pattern, REG_EXTENDED);
362 regerror(ret, &parent_regex, err, sizeof(err));
363 fprintf(stderr, "Invalid regex: %s\n%s",
364 parent_pattern, err);
367 sort__has_parent = 1;
370 list_add_tail(&sd->entry->list, &hist_entry__sort_list);
380 hist_entry__cmp(struct hist_entry *left, struct hist_entry *right)
382 struct sort_entry *se;
385 list_for_each_entry(se, &hist_entry__sort_list, list) {
386 cmp = se->cmp(left, right);
395 hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
397 struct sort_entry *se;
400 list_for_each_entry(se, &hist_entry__sort_list, list) {
401 int64_t (*f)(struct hist_entry *, struct hist_entry *);
403 f = se->collapse ?: se->cmp;
405 cmp = f(left, right);
413 static size_t ipchain__fprintf_graph_line(FILE *fp, int depth, int depth_mask)
418 ret += fprintf(fp, "%s", " ");
420 for (i = 0; i < depth; i++)
421 if (depth_mask & (1 << i))
422 ret += fprintf(fp, "| ");
424 ret += fprintf(fp, " ");
426 ret += fprintf(fp, "\n");
431 ipchain__fprintf_graph(FILE *fp, struct callchain_list *chain, int depth,
432 int depth_mask, int count, u64 total_samples,
438 ret += fprintf(fp, "%s", " ");
439 for (i = 0; i < depth; i++) {
440 if (depth_mask & (1 << i))
441 ret += fprintf(fp, "|");
443 ret += fprintf(fp, " ");
444 if (!count && i == depth - 1) {
447 percent = hits * 100.0 / total_samples;
448 ret += percent_color_fprintf(fp, "--%2.2f%%-- ", percent);
450 ret += fprintf(fp, "%s", " ");
453 ret += fprintf(fp, "%s\n", chain->sym->name);
455 ret += fprintf(fp, "%p\n", (void *)(long)chain->ip);
460 static struct symbol *rem_sq_bracket;
461 static struct callchain_list rem_hits;
463 static void init_rem_hits(void)
465 rem_sq_bracket = malloc(sizeof(*rem_sq_bracket) + 6);
466 if (!rem_sq_bracket) {
467 fprintf(stderr, "Not enough memory to display remaining hits\n");
471 strcpy(rem_sq_bracket->name, "[...]");
472 rem_hits.sym = rem_sq_bracket;
476 callchain__fprintf_graph(FILE *fp, struct callchain_node *self,
477 u64 total_samples, int depth, int depth_mask)
479 struct rb_node *node, *next;
480 struct callchain_node *child;
481 struct callchain_list *chain;
482 int new_depth_mask = depth_mask;
488 if (callchain_param.mode == CHAIN_GRAPH_REL)
489 new_total = self->children_hit;
491 new_total = total_samples;
493 remaining = new_total;
495 node = rb_first(&self->rb_root);
499 child = rb_entry(node, struct callchain_node, rb_node);
500 cumul = cumul_hits(child);
504 * The depth mask manages the output of pipes that show
505 * the depth. We don't want to keep the pipes of the current
506 * level for the last child of this depth.
507 * Except if we have remaining filtered hits. They will
508 * supersede the last child
510 next = rb_next(node);
511 if (!next && (callchain_param.mode != CHAIN_GRAPH_REL || !remaining))
512 new_depth_mask &= ~(1 << (depth - 1));
515 * But we keep the older depth mask for the line seperator
516 * to keep the level link until we reach the last child
518 ret += ipchain__fprintf_graph_line(fp, depth, depth_mask);
520 list_for_each_entry(chain, &child->val, list) {
521 if (chain->ip >= PERF_CONTEXT_MAX)
523 ret += ipchain__fprintf_graph(fp, chain, depth,
528 ret += callchain__fprintf_graph(fp, child, new_total,
530 new_depth_mask | (1 << depth));
534 if (callchain_param.mode == CHAIN_GRAPH_REL &&
535 remaining && remaining != new_total) {
540 new_depth_mask &= ~(1 << (depth - 1));
542 ret += ipchain__fprintf_graph(fp, &rem_hits, depth,
543 new_depth_mask, 0, new_total,
551 callchain__fprintf_flat(FILE *fp, struct callchain_node *self,
554 struct callchain_list *chain;
560 ret += callchain__fprintf_flat(fp, self->parent, total_samples);
563 list_for_each_entry(chain, &self->val, list) {
564 if (chain->ip >= PERF_CONTEXT_MAX)
567 ret += fprintf(fp, " %s\n", chain->sym->name);
569 ret += fprintf(fp, " %p\n",
570 (void *)(long)chain->ip);
577 hist_entry_callchain__fprintf(FILE *fp, struct hist_entry *self,
580 struct rb_node *rb_node;
581 struct callchain_node *chain;
584 rb_node = rb_first(&self->sorted_chain);
588 chain = rb_entry(rb_node, struct callchain_node, rb_node);
589 percent = chain->hit * 100.0 / total_samples;
590 switch (callchain_param.mode) {
592 ret += percent_color_fprintf(fp, " %6.2f%%\n",
594 ret += callchain__fprintf_flat(fp, chain, total_samples);
596 case CHAIN_GRAPH_ABS: /* Falldown */
597 case CHAIN_GRAPH_REL:
598 ret += callchain__fprintf_graph(fp, chain,
599 total_samples, 1, 1);
604 ret += fprintf(fp, "\n");
605 rb_node = rb_next(rb_node);
613 hist_entry__fprintf(FILE *fp, struct hist_entry *self, u64 total_samples)
615 struct sort_entry *se;
618 if (exclude_other && !self->parent)
622 ret = percent_color_fprintf(fp,
623 field_sep ? "%.2f" : " %6.2f%%",
624 (self->count * 100.0) / total_samples);
626 ret = fprintf(fp, field_sep ? "%lld" : "%12lld ", self->count);
628 if (show_nr_samples) {
630 fprintf(fp, "%c%lld", *field_sep, self->count);
632 fprintf(fp, "%11lld", self->count);
635 list_for_each_entry(se, &hist_entry__sort_list, list) {
639 fprintf(fp, "%s", field_sep ?: " ");
640 ret += se->print(fp, self, se->width ? *se->width : 0);
643 ret += fprintf(fp, "\n");
646 hist_entry_callchain__fprintf(fp, self, total_samples);
655 static void dso__calc_col_width(struct dso *self)
657 if (!col_width_list_str && !field_sep &&
658 (!dso_list || strlist__has_entry(dso_list, self->name))) {
659 unsigned int slen = strlen(self->name);
660 if (slen > dsos__col_width)
661 dsos__col_width = slen;
664 self->slen_calculated = 1;
667 static struct symbol *
668 resolve_symbol(struct thread *thread, struct map **mapp,
669 struct dso **dsop, u64 *ipp)
671 struct dso *dso = dsop ? *dsop : NULL;
672 struct map *map = mapp ? *mapp : NULL;
684 map = thread__find_map(thread, ip);
687 * We have to do this here as we may have a dso
688 * with no symbol hit that has a name longer than
689 * the ones with symbols sampled.
691 if (!sort_dso.elide && !map->dso->slen_calculated)
692 dso__calc_col_width(map->dso);
697 ip = map->map_ip(map, ip);
702 * If this is outside of all known maps,
703 * and is a negative address, try to look it
704 * up in the kernel dso, as it might be a
705 * vsyscall (which executes in user-mode):
707 if ((long long)ip < 0)
710 dump_printf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
711 dump_printf(" ...... map: %Lx -> %Lx\n", *ipp, ip);
720 return dso->find_symbol(dso, ip);
723 static int call__match(struct symbol *sym)
725 if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
731 static struct symbol **
732 resolve_callchain(struct thread *thread, struct map *map __used,
733 struct ip_callchain *chain, struct hist_entry *entry)
735 u64 context = PERF_CONTEXT_MAX;
736 struct symbol **syms = NULL;
740 syms = calloc(chain->nr, sizeof(*syms));
742 fprintf(stderr, "Can't allocate memory for symbols\n");
747 for (i = 0; i < chain->nr; i++) {
748 u64 ip = chain->ips[i];
749 struct dso *dso = NULL;
752 if (ip >= PERF_CONTEXT_MAX) {
758 case PERF_CONTEXT_HV:
759 dso = hypervisor_dso;
761 case PERF_CONTEXT_KERNEL:
768 sym = resolve_symbol(thread, NULL, &dso, &ip);
771 if (sort__has_parent && call__match(sym) &&
784 * collect histogram counts
788 hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
789 struct symbol *sym, u64 ip, struct ip_callchain *chain,
790 char level, u64 count)
792 struct rb_node **p = &hist.rb_node;
793 struct rb_node *parent = NULL;
794 struct hist_entry *he;
795 struct symbol **syms = NULL;
796 struct hist_entry entry = {
805 .sorted_chain = RB_ROOT
809 if ((sort__has_parent || callchain) && chain)
810 syms = resolve_callchain(thread, map, chain, &entry);
814 he = rb_entry(parent, struct hist_entry, rb_node);
816 cmp = hist_entry__cmp(&entry, he);
821 append_chain(&he->callchain, chain, syms);
833 he = malloc(sizeof(*he));
838 callchain_init(&he->callchain);
839 append_chain(&he->callchain, chain, syms);
842 rb_link_node(&he->rb_node, parent, p);
843 rb_insert_color(&he->rb_node, &hist);
848 static void hist_entry__free(struct hist_entry *he)
854 * collapse the histogram
857 static struct rb_root collapse_hists;
859 static void collapse__insert_entry(struct hist_entry *he)
861 struct rb_node **p = &collapse_hists.rb_node;
862 struct rb_node *parent = NULL;
863 struct hist_entry *iter;
868 iter = rb_entry(parent, struct hist_entry, rb_node);
870 cmp = hist_entry__collapse(iter, he);
873 iter->count += he->count;
874 hist_entry__free(he);
884 rb_link_node(&he->rb_node, parent, p);
885 rb_insert_color(&he->rb_node, &collapse_hists);
888 static void collapse__resort(void)
890 struct rb_node *next;
891 struct hist_entry *n;
893 if (!sort__need_collapse)
896 next = rb_first(&hist);
898 n = rb_entry(next, struct hist_entry, rb_node);
899 next = rb_next(&n->rb_node);
901 rb_erase(&n->rb_node, &hist);
902 collapse__insert_entry(n);
907 * reverse the map, sort on count.
910 static struct rb_root output_hists;
912 static void output__insert_entry(struct hist_entry *he, u64 min_callchain_hits)
914 struct rb_node **p = &output_hists.rb_node;
915 struct rb_node *parent = NULL;
916 struct hist_entry *iter;
919 callchain_param.sort(&he->sorted_chain, &he->callchain,
920 min_callchain_hits, &callchain_param);
924 iter = rb_entry(parent, struct hist_entry, rb_node);
926 if (he->count > iter->count)
932 rb_link_node(&he->rb_node, parent, p);
933 rb_insert_color(&he->rb_node, &output_hists);
936 static void output__resort(u64 total_samples)
938 struct rb_node *next;
939 struct hist_entry *n;
940 struct rb_root *tree = &hist;
941 u64 min_callchain_hits;
943 min_callchain_hits = total_samples * (callchain_param.min_percent / 100);
945 if (sort__need_collapse)
946 tree = &collapse_hists;
948 next = rb_first(tree);
951 n = rb_entry(next, struct hist_entry, rb_node);
952 next = rb_next(&n->rb_node);
954 rb_erase(&n->rb_node, tree);
955 output__insert_entry(n, min_callchain_hits);
959 static size_t output__fprintf(FILE *fp, u64 total_samples)
961 struct hist_entry *pos;
962 struct sort_entry *se;
966 char *col_width = col_width_list_str;
967 int raw_printing_style;
969 raw_printing_style = !strcmp(pretty_printing_style, "raw");
973 fprintf(fp, "# Samples: %Ld\n", (u64)total_samples);
976 fprintf(fp, "# Overhead");
977 if (show_nr_samples) {
979 fprintf(fp, "%cSamples", *field_sep);
981 fputs(" Samples ", fp);
983 list_for_each_entry(se, &hist_entry__sort_list, list) {
987 fprintf(fp, "%c%s", *field_sep, se->header);
990 width = strlen(se->header);
992 if (col_width_list_str) {
994 *se->width = atoi(col_width);
995 col_width = strchr(col_width, ',');
1000 width = *se->width = max(*se->width, width);
1002 fprintf(fp, " %*s", width, se->header);
1009 fprintf(fp, "# ........");
1010 if (show_nr_samples)
1011 fprintf(fp, " ..........");
1012 list_for_each_entry(se, &hist_entry__sort_list, list) {
1022 width = strlen(se->header);
1023 for (i = 0; i < width; i++)
1031 for (nd = rb_first(&output_hists); nd; nd = rb_next(nd)) {
1032 pos = rb_entry(nd, struct hist_entry, rb_node);
1033 ret += hist_entry__fprintf(fp, pos, total_samples);
1036 if (sort_order == default_sort_order &&
1037 parent_pattern == default_parent_pattern) {
1039 fprintf(fp, "# (For a higher level overview, try: perf report --sort comm,dso)\n");
1044 free(rem_sq_bracket);
1047 perf_read_values_display(fp, &show_threads_values,
1048 raw_printing_style);
1053 static void register_idle_thread(void)
1055 struct thread *thread = threads__findnew(0, &threads, &last_match);
1057 if (thread == NULL ||
1058 thread__set_comm(thread, "[idle]")) {
1059 fprintf(stderr, "problem inserting idle task.\n");
1064 static unsigned long total = 0,
1071 static int validate_chain(struct ip_callchain *chain, event_t *event)
1073 unsigned int chain_size;
1075 chain_size = event->header.size;
1076 chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;
1078 if (chain->nr*sizeof(u64) > chain_size)
1085 process_sample_event(event_t *event, unsigned long offset, unsigned long head)
1089 struct dso *dso = NULL;
1090 struct thread *thread;
1091 u64 ip = event->ip.ip;
1093 struct map *map = NULL;
1094 void *more_data = event->ip.__more_data;
1095 struct ip_callchain *chain = NULL;
1098 thread = threads__findnew(event->ip.pid, &threads, &last_match);
1100 if (sample_type & PERF_SAMPLE_PERIOD) {
1101 period = *(u64 *)more_data;
1102 more_data += sizeof(u64);
1105 dump_printf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n",
1106 (void *)(offset + head),
1107 (void *)(long)(event->header.size),
1109 event->ip.pid, event->ip.tid,
1113 if (sample_type & PERF_SAMPLE_CALLCHAIN) {
1116 chain = (void *)more_data;
1118 dump_printf("... chain: nr:%Lu\n", chain->nr);
1120 if (validate_chain(chain, event) < 0) {
1121 eprintf("call-chain problem with event, skipping it.\n");
1126 for (i = 0; i < chain->nr; i++)
1127 dump_printf("..... %2d: %016Lx\n", i, chain->ips[i]);
1131 dump_printf(" ... thread: %s:%d\n", thread->comm, thread->pid);
1133 if (thread == NULL) {
1134 eprintf("problem processing %d event, skipping it.\n",
1135 event->header.type);
1139 if (comm_list && !strlist__has_entry(comm_list, thread->comm))
1142 cpumode = event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK;
1144 if (cpumode == PERF_EVENT_MISC_KERNEL) {
1150 dump_printf(" ...... dso: %s\n", dso->name);
1152 } else if (cpumode == PERF_EVENT_MISC_USER) {
1161 dso = hypervisor_dso;
1163 dump_printf(" ...... dso: [hypervisor]\n");
1166 if (show & show_mask) {
1167 struct symbol *sym = resolve_symbol(thread, &map, &dso, &ip);
1169 if (dso_list && (!dso || !dso->name ||
1170 !strlist__has_entry(dso_list, dso->name)))
1173 if (sym_list && (!sym || !strlist__has_entry(sym_list, sym->name)))
1176 if (hist_entry__add(thread, map, dso, sym, ip, chain, level, period)) {
1177 eprintf("problem incrementing symbol count, skipping event\n");
1187 process_mmap_event(event_t *event, unsigned long offset, unsigned long head)
1189 struct thread *thread;
1190 struct map *map = map__new(&event->mmap, cwd, cwdlen);
1192 thread = threads__findnew(event->mmap.pid, &threads, &last_match);
1194 dump_printf("%p [%p]: PERF_EVENT_MMAP %d/%d: [%p(%p) @ %p]: %s\n",
1195 (void *)(offset + head),
1196 (void *)(long)(event->header.size),
1199 (void *)(long)event->mmap.start,
1200 (void *)(long)event->mmap.len,
1201 (void *)(long)event->mmap.pgoff,
1202 event->mmap.filename);
1204 if (thread == NULL || map == NULL) {
1205 dump_printf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1209 thread__insert_map(thread, map);
1216 process_comm_event(event_t *event, unsigned long offset, unsigned long head)
1218 struct thread *thread;
1220 thread = threads__findnew(event->comm.pid, &threads, &last_match);
1222 dump_printf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
1223 (void *)(offset + head),
1224 (void *)(long)(event->header.size),
1225 event->comm.comm, event->comm.pid);
1227 if (thread == NULL ||
1228 thread__set_comm(thread, event->comm.comm)) {
1229 dump_printf("problem processing PERF_EVENT_COMM, skipping event.\n");
1238 process_task_event(event_t *event, unsigned long offset, unsigned long head)
1240 struct thread *thread;
1241 struct thread *parent;
1243 thread = threads__findnew(event->fork.pid, &threads, &last_match);
1244 parent = threads__findnew(event->fork.ppid, &threads, &last_match);
1246 dump_printf("%p [%p]: PERF_EVENT_%s: (%d:%d):(%d:%d)\n",
1247 (void *)(offset + head),
1248 (void *)(long)(event->header.size),
1249 event->header.type == PERF_EVENT_FORK ? "FORK" : "EXIT",
1250 event->fork.pid, event->fork.tid,
1251 event->fork.ppid, event->fork.ptid);
1254 * A thread clone will have the same PID for both
1257 if (thread == parent)
1260 if (event->header.type == PERF_EVENT_EXIT)
1263 if (!thread || !parent || thread__fork(thread, parent)) {
1264 dump_printf("problem processing PERF_EVENT_FORK, skipping event.\n");
1273 process_lost_event(event_t *event, unsigned long offset, unsigned long head)
1275 dump_printf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n",
1276 (void *)(offset + head),
1277 (void *)(long)(event->header.size),
1281 total_lost += event->lost.lost;
1286 static void trace_event(event_t *event)
1288 unsigned char *raw_event = (void *)event;
1289 const char *color = PERF_COLOR_BLUE;
1296 cdprintf("\n. ... raw event: size %d bytes\n", event->header.size);
1298 for (i = 0; i < event->header.size; i++) {
1299 if ((i & 15) == 0) {
1301 cdprintf(" %04x: ", i);
1304 cdprintf(" %02x", raw_event[i]);
1306 if (((i & 15) == 15) || i == event->header.size-1) {
1308 for (j = 0; j < 15-(i & 15); j++)
1310 for (j = 0; j < (i & 15); j++) {
1311 if (isprint(raw_event[i-15+j]))
1312 cdprintf("%c", raw_event[i-15+j]);
1322 static struct perf_header *header;
1324 static struct perf_counter_attr *perf_header__find_attr(u64 id)
1328 for (i = 0; i < header->attrs; i++) {
1329 struct perf_header_attr *attr = header->attr[i];
1332 for (j = 0; j < attr->ids; j++) {
1333 if (attr->id[j] == id)
1342 process_read_event(event_t *event, unsigned long offset, unsigned long head)
1344 struct perf_counter_attr *attr = perf_header__find_attr(event->read.id);
1347 const char *name = attr ? __event_name(attr->type, attr->config)
1349 perf_read_values_add_value(&show_threads_values,
1350 event->read.pid, event->read.tid,
1356 dump_printf("%p [%p]: PERF_EVENT_READ: %d %d %s %Lu\n",
1357 (void *)(offset + head),
1358 (void *)(long)(event->header.size),
1361 attr ? __event_name(attr->type, attr->config)
1369 process_event(event_t *event, unsigned long offset, unsigned long head)
1373 switch (event->header.type) {
1374 case PERF_EVENT_SAMPLE:
1375 return process_sample_event(event, offset, head);
1377 case PERF_EVENT_MMAP:
1378 return process_mmap_event(event, offset, head);
1380 case PERF_EVENT_COMM:
1381 return process_comm_event(event, offset, head);
1383 case PERF_EVENT_FORK:
1384 case PERF_EVENT_EXIT:
1385 return process_task_event(event, offset, head);
1387 case PERF_EVENT_LOST:
1388 return process_lost_event(event, offset, head);
1390 case PERF_EVENT_READ:
1391 return process_read_event(event, offset, head);
1394 * We dont process them right now but they are fine:
1397 case PERF_EVENT_THROTTLE:
1398 case PERF_EVENT_UNTHROTTLE:
1408 static u64 perf_header__sample_type(void)
1413 for (i = 0; i < header->attrs; i++) {
1414 struct perf_header_attr *attr = header->attr[i];
1417 type = attr->attr.sample_type;
1418 else if (type != attr->attr.sample_type)
1419 die("non matching sample_type");
1425 static int __cmd_report(void)
1427 int ret, rc = EXIT_FAILURE;
1428 unsigned long offset = 0;
1429 unsigned long head, shift;
1430 struct stat input_stat;
1435 register_idle_thread();
1438 perf_read_values_init(&show_threads_values);
1440 input = open(input_name, O_RDONLY);
1442 fprintf(stderr, " failed to open file: %s", input_name);
1443 if (!strcmp(input_name, "perf.data"))
1444 fprintf(stderr, " (try 'perf record' first)");
1445 fprintf(stderr, "\n");
1449 ret = fstat(input, &input_stat);
1451 perror("failed to stat file");
1455 if (!input_stat.st_size) {
1456 fprintf(stderr, "zero-sized file, nothing to do!\n");
1460 header = perf_header__read(input);
1461 head = header->data_offset;
1463 sample_type = perf_header__sample_type();
1465 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
1466 if (sort__has_parent) {
1467 fprintf(stderr, "selected --sort parent, but no"
1468 " callchain data. Did you call"
1469 " perf record without -g?\n");
1473 fprintf(stderr, "selected -c but no callchain data."
1474 " Did you call perf record without"
1478 } else if (callchain_param.mode != CHAIN_NONE && !callchain) {
1480 if (register_callchain_param(&callchain_param) < 0) {
1481 fprintf(stderr, "Can't register callchain"
1487 if (load_kernel() < 0) {
1488 perror("failed to load kernel symbols");
1489 return EXIT_FAILURE;
1493 if (getcwd(__cwd, sizeof(__cwd)) == NULL) {
1494 perror("failed to get the current directory");
1495 return EXIT_FAILURE;
1497 cwdlen = strlen(cwd);
1503 shift = page_size * (head / page_size);
1508 buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
1509 MAP_SHARED, input, offset);
1510 if (buf == MAP_FAILED) {
1511 perror("failed to mmap file");
1516 event = (event_t *)(buf + head);
1518 size = event->header.size;
1522 if (head + event->header.size >= page_size * mmap_window) {
1525 shift = page_size * (head / page_size);
1527 munmap_ret = munmap(buf, page_size * mmap_window);
1528 assert(munmap_ret == 0);
1535 size = event->header.size;
1537 dump_printf("\n%p [%p]: event: %d\n",
1538 (void *)(offset + head),
1539 (void *)(long)event->header.size,
1540 event->header.type);
1542 if (!size || process_event(event, offset, head) < 0) {
1544 dump_printf("%p [%p]: skipping unknown header type: %d\n",
1545 (void *)(offset + head),
1546 (void *)(long)(event->header.size),
1547 event->header.type);
1552 * assume we lost track of the stream, check alignment, and
1553 * increment a single u64 in the hope to catch on again 'soon'.
1556 if (unlikely(head & 7))
1564 if (offset + head >= header->data_offset + header->data_size)
1567 if (offset + head < (unsigned long)input_stat.st_size)
1574 dump_printf(" IP events: %10ld\n", total);
1575 dump_printf(" mmap events: %10ld\n", total_mmap);
1576 dump_printf(" comm events: %10ld\n", total_comm);
1577 dump_printf(" fork events: %10ld\n", total_fork);
1578 dump_printf(" lost events: %10ld\n", total_lost);
1579 dump_printf(" unknown events: %10ld\n", total_unknown);
1585 threads__fprintf(stdout, &threads);
1588 dsos__fprintf(stdout);
1591 output__resort(total);
1592 output__fprintf(stdout, total);
1595 perf_read_values_destroy(&show_threads_values);
1601 parse_callchain_opt(const struct option *opt __used, const char *arg,
1612 tok = strtok((char *)arg, ",");
1616 /* get the output mode */
1617 if (!strncmp(tok, "graph", strlen(arg)))
1618 callchain_param.mode = CHAIN_GRAPH_ABS;
1620 else if (!strncmp(tok, "flat", strlen(arg)))
1621 callchain_param.mode = CHAIN_FLAT;
1623 else if (!strncmp(tok, "fractal", strlen(arg)))
1624 callchain_param.mode = CHAIN_GRAPH_REL;
1626 else if (!strncmp(tok, "none", strlen(arg))) {
1627 callchain_param.mode = CHAIN_NONE;
1636 /* get the min percentage */
1637 tok = strtok(NULL, ",");
1641 callchain_param.min_percent = strtod(tok, &endptr);
1646 if (register_callchain_param(&callchain_param) < 0) {
1647 fprintf(stderr, "Can't register callchain params\n");
1653 static const char * const report_usage[] = {
1654 "perf report [<options>] <command>",
1658 static const struct option options[] = {
1659 OPT_STRING('i', "input", &input_name, "file",
1661 OPT_BOOLEAN('v', "verbose", &verbose,
1662 "be more verbose (show symbol address, etc)"),
1663 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1664 "dump raw trace in ASCII"),
1665 OPT_STRING('k', "vmlinux", &vmlinux_name, "file", "vmlinux pathname"),
1666 OPT_BOOLEAN('m', "modules", &modules,
1667 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1668 OPT_BOOLEAN('n', "show-nr-samples", &show_nr_samples,
1669 "Show a column with the number of samples"),
1670 OPT_BOOLEAN('T', "threads", &show_threads,
1671 "Show per-thread event counters"),
1672 OPT_STRING(0, "pretty", &pretty_printing_style, "key",
1673 "pretty printing style key: normal raw"),
1674 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1675 "sort by key(s): pid, comm, dso, symbol, parent"),
1676 OPT_BOOLEAN('P', "full-paths", &full_paths,
1677 "Don't shorten the pathnames taking into account the cwd"),
1678 OPT_STRING('p', "parent", &parent_pattern, "regex",
1679 "regex filter to identify parent, see: '--sort parent'"),
1680 OPT_BOOLEAN('x', "exclude-other", &exclude_other,
1681 "Only display entries with parent-match"),
1682 OPT_CALLBACK_DEFAULT('g', "call-graph", NULL, "output_type,min_percent",
1683 "Display callchains using output_type and min percent threshold. "
1684 "Default: fractal,0.5", &parse_callchain_opt, callchain_default_opt),
1685 OPT_STRING('d', "dsos", &dso_list_str, "dso[,dso...]",
1686 "only consider symbols in these dsos"),
1687 OPT_STRING('C', "comms", &comm_list_str, "comm[,comm...]",
1688 "only consider symbols in these comms"),
1689 OPT_STRING('S', "symbols", &sym_list_str, "symbol[,symbol...]",
1690 "only consider these symbols"),
1691 OPT_STRING('w', "column-widths", &col_width_list_str,
1693 "don't try to adjust column width, use these fixed values"),
1694 OPT_STRING('t', "field-separator", &field_sep, "separator",
1695 "separator for columns, no spaces will be added between "
1696 "columns '.' is reserved."),
1700 static void setup_sorting(void)
1702 char *tmp, *tok, *str = strdup(sort_order);
1704 for (tok = strtok_r(str, ", ", &tmp);
1705 tok; tok = strtok_r(NULL, ", ", &tmp)) {
1706 if (sort_dimension__add(tok) < 0) {
1707 error("Unknown --sort key: `%s'", tok);
1708 usage_with_options(report_usage, options);
1715 static void setup_list(struct strlist **list, const char *list_str,
1716 struct sort_entry *se, const char *list_name,
1720 *list = strlist__new(true, list_str);
1722 fprintf(stderr, "problems parsing %s list\n",
1726 if (strlist__nr_entries(*list) == 1) {
1727 fprintf(fp, "# %s: %s\n", list_name,
1728 strlist__entry(*list, 0)->s);
1734 int cmd_report(int argc, const char **argv, const char *prefix __used)
1738 page_size = getpagesize();
1740 argc = parse_options(argc, argv, options, report_usage, 0);
1744 if (parent_pattern != default_parent_pattern) {
1745 sort_dimension__add("parent");
1746 sort_parent.elide = 1;
1751 * Any (unrecognized) arguments left?
1754 usage_with_options(report_usage, options);
1758 setup_list(&dso_list, dso_list_str, &sort_dso, "dso", stdout);
1759 setup_list(&comm_list, comm_list_str, &sort_comm, "comm", stdout);
1760 setup_list(&sym_list, sym_list_str, &sort_sym, "symbol", stdout);
1762 if (field_sep && *field_sep == '.') {
1763 fputs("'.' is the only non valid --field-separator argument\n",
1768 return __cmd_report();