4 * Builtin report command: Analyze the perf.data input file,
5 * look up and read DSOs and symbol information and display
6 * a histogram of results, along various sorting keys.
10 #include "util/util.h"
12 #include "util/color.h"
13 #include <linux/list.h>
14 #include "util/cache.h"
15 #include <linux/rbtree.h>
16 #include "util/symbol.h"
17 #include "util/string.h"
18 #include "util/callchain.h"
19 #include "util/strlist.h"
22 #include "util/header.h"
24 #include "util/parse-options.h"
25 #include "util/parse-events.h"
31 static char const *input_name = "perf.data";
32 static char *vmlinux = NULL;
34 static char default_sort_order[] = "comm,dso,symbol";
35 static char *sort_order = default_sort_order;
36 static char *dso_list_str, *comm_list_str, *sym_list_str,
38 static struct strlist *dso_list, *comm_list, *sym_list;
39 static char *field_sep;
43 static int show_mask = SHOW_KERNEL | SHOW_USER | SHOW_HV;
45 static int dump_trace = 0;
46 #define dprintf(x...) do { if (dump_trace) printf(x); } while (0)
47 #define cdprintf(x...) do { if (dump_trace) color_fprintf(stdout, color, x); } while (0)
50 #define eprintf(x...) do { if (verbose) fprintf(stderr, x); } while (0)
54 static int full_paths;
55 static int show_nr_samples;
57 static unsigned long page_size;
58 static unsigned long mmap_window = 32;
60 static char default_parent_pattern[] = "^sys_|^do_page_fault";
61 static char *parent_pattern = default_parent_pattern;
62 static regex_t parent_regex;
64 static int exclude_other = 1;
66 static char callchain_default_opt[] = "fractal,0.5";
71 struct callchain_param callchain_param = {
72 .mode = CHAIN_GRAPH_REL,
76 static u64 sample_type;
79 struct perf_event_header header;
82 unsigned char __more_data[];
86 struct perf_event_header header;
91 char filename[PATH_MAX];
95 struct perf_event_header header;
101 struct perf_event_header header;
107 struct perf_event_header header;
113 struct perf_event_header header;
121 typedef union event_union {
122 struct perf_event_header header;
124 struct mmap_event mmap;
125 struct comm_event comm;
126 struct fork_event fork;
127 struct lost_event lost;
128 struct read_event read;
131 static int repsep_fprintf(FILE *fp, const char *fmt, ...)
138 n = vfprintf(fp, fmt, ap);
141 n = vasprintf(&bf, fmt, ap);
145 sep = strchr(sep, *field_sep);
158 static LIST_HEAD(dsos);
159 static struct dso *kernel_dso;
160 static struct dso *vdso;
161 static struct dso *hypervisor_dso;
163 static void dsos__add(struct dso *dso)
165 list_add_tail(&dso->node, &dsos);
168 static struct dso *dsos__find(const char *name)
172 list_for_each_entry(pos, &dsos, node)
173 if (strcmp(pos->name, name) == 0)
178 static struct dso *dsos__findnew(const char *name)
180 struct dso *dso = dsos__find(name);
186 dso = dso__new(name, 0);
190 nr = dso__load(dso, NULL, verbose);
192 eprintf("Failed to open: %s\n", name);
196 eprintf("No symbols found in: %s, maybe install a debug package?\n", name);
207 static void dsos__fprintf(FILE *fp)
211 list_for_each_entry(pos, &dsos, node)
212 dso__fprintf(pos, fp);
215 static struct symbol *vdso__find_symbol(struct dso *dso, u64 ip)
217 return dso__find_symbol(dso, ip);
220 static int load_kernel(void)
224 kernel_dso = dso__new("[kernel]", 0);
228 err = dso__load_kernel(kernel_dso, vmlinux, NULL, verbose, modules);
230 dso__delete(kernel_dso);
233 dsos__add(kernel_dso);
235 vdso = dso__new("[vdso]", 0);
239 vdso->find_symbol = vdso__find_symbol;
243 hypervisor_dso = dso__new("[hypervisor]", 0);
246 dsos__add(hypervisor_dso);
251 static char __cwd[PATH_MAX];
252 static char *cwd = __cwd;
255 static int strcommon(const char *pathname)
259 while (n < cwdlen && pathname[n] == cwd[n])
266 struct list_head node;
270 u64 (*map_ip)(struct map *, u64);
274 static u64 map__map_ip(struct map *map, u64 ip)
276 return ip - map->start + map->pgoff;
279 static u64 vdso__map_ip(struct map *map __used, u64 ip)
284 static inline int is_anon_memory(const char *filename)
286 return strcmp(filename, "//anon") == 0;
289 static struct map *map__new(struct mmap_event *event)
291 struct map *self = malloc(sizeof(*self));
294 const char *filename = event->filename;
295 char newfilename[PATH_MAX];
299 int n = strcommon(filename);
302 snprintf(newfilename, sizeof(newfilename),
303 ".%s", filename + n);
304 filename = newfilename;
308 anon = is_anon_memory(filename);
311 snprintf(newfilename, sizeof(newfilename), "/tmp/perf-%d.map", event->pid);
312 filename = newfilename;
315 self->start = event->start;
316 self->end = event->start + event->len;
317 self->pgoff = event->pgoff;
319 self->dso = dsos__findnew(filename);
320 if (self->dso == NULL)
323 if (self->dso == vdso || anon)
324 self->map_ip = vdso__map_ip;
326 self->map_ip = map__map_ip;
334 static struct map *map__clone(struct map *self)
336 struct map *map = malloc(sizeof(*self));
341 memcpy(map, self, sizeof(*self));
346 static int map__overlap(struct map *l, struct map *r)
348 if (l->start > r->start) {
354 if (l->end > r->start)
360 static size_t map__fprintf(struct map *self, FILE *fp)
362 return fprintf(fp, " %Lx-%Lx %Lx %s\n",
363 self->start, self->end, self->pgoff, self->dso->name);
368 struct rb_node rb_node;
369 struct list_head maps;
374 static struct thread *thread__new(pid_t pid)
376 struct thread *self = malloc(sizeof(*self));
380 self->comm = malloc(32);
382 snprintf(self->comm, 32, ":%d", self->pid);
383 INIT_LIST_HEAD(&self->maps);
389 static unsigned int dsos__col_width,
393 static int thread__set_comm(struct thread *self, const char *comm)
397 self->comm = strdup(comm);
401 if (!col_width_list_str && !field_sep &&
402 (!comm_list || strlist__has_entry(comm_list, comm))) {
403 unsigned int slen = strlen(comm);
404 if (slen > comms__col_width) {
405 comms__col_width = slen;
406 threads__col_width = slen + 6;
413 static size_t thread__fprintf(struct thread *self, FILE *fp)
416 size_t ret = fprintf(fp, "Thread %d %s\n", self->pid, self->comm);
418 list_for_each_entry(pos, &self->maps, node)
419 ret += map__fprintf(pos, fp);
425 static struct rb_root threads;
426 static struct thread *last_match;
428 static struct thread *threads__findnew(pid_t pid)
430 struct rb_node **p = &threads.rb_node;
431 struct rb_node *parent = NULL;
435 * Font-end cache - PID lookups come in blocks,
436 * so most of the time we dont have to look up
439 if (last_match && last_match->pid == pid)
444 th = rb_entry(parent, struct thread, rb_node);
446 if (th->pid == pid) {
457 th = thread__new(pid);
459 rb_link_node(&th->rb_node, parent, p);
460 rb_insert_color(&th->rb_node, &threads);
467 static void thread__insert_map(struct thread *self, struct map *map)
469 struct map *pos, *tmp;
471 list_for_each_entry_safe(pos, tmp, &self->maps, node) {
472 if (map__overlap(pos, map)) {
474 printf("overlapping maps:\n");
475 map__fprintf(map, stdout);
476 map__fprintf(pos, stdout);
479 if (map->start <= pos->start && map->end > pos->start)
480 pos->start = map->end;
482 if (map->end >= pos->end && map->start < pos->end)
483 pos->end = map->start;
486 printf("after collision:\n");
487 map__fprintf(pos, stdout);
490 if (pos->start >= pos->end) {
491 list_del_init(&pos->node);
497 list_add_tail(&map->node, &self->maps);
500 static int thread__fork(struct thread *self, struct thread *parent)
506 self->comm = strdup(parent->comm);
510 list_for_each_entry(map, &parent->maps, node) {
511 struct map *new = map__clone(map);
514 thread__insert_map(self, new);
520 static struct map *thread__find_map(struct thread *self, u64 ip)
527 list_for_each_entry(pos, &self->maps, node)
528 if (ip >= pos->start && ip <= pos->end)
534 static size_t threads__fprintf(FILE *fp)
539 for (nd = rb_first(&threads); nd; nd = rb_next(nd)) {
540 struct thread *pos = rb_entry(nd, struct thread, rb_node);
542 ret += thread__fprintf(pos, fp);
549 * histogram, sorted on item, collects counts
552 static struct rb_root hist;
555 struct rb_node rb_node;
557 struct thread *thread;
561 struct symbol *parent;
564 struct callchain_node callchain;
565 struct rb_root sorted_chain;
571 * configurable sorting bits
575 struct list_head list;
579 int64_t (*cmp)(struct hist_entry *, struct hist_entry *);
580 int64_t (*collapse)(struct hist_entry *, struct hist_entry *);
581 size_t (*print)(FILE *fp, struct hist_entry *, unsigned int width);
586 static int64_t cmp_null(void *l, void *r)
599 sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
601 return right->thread->pid - left->thread->pid;
605 sort__thread_print(FILE *fp, struct hist_entry *self, unsigned int width)
607 return repsep_fprintf(fp, "%*s:%5d", width - 6,
608 self->thread->comm ?: "", self->thread->pid);
611 static struct sort_entry sort_thread = {
612 .header = "Command: Pid",
613 .cmp = sort__thread_cmp,
614 .print = sort__thread_print,
615 .width = &threads__col_width,
621 sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
623 return right->thread->pid - left->thread->pid;
627 sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
629 char *comm_l = left->thread->comm;
630 char *comm_r = right->thread->comm;
632 if (!comm_l || !comm_r)
633 return cmp_null(comm_l, comm_r);
635 return strcmp(comm_l, comm_r);
639 sort__comm_print(FILE *fp, struct hist_entry *self, unsigned int width)
641 return repsep_fprintf(fp, "%*s", width, self->thread->comm);
644 static struct sort_entry sort_comm = {
646 .cmp = sort__comm_cmp,
647 .collapse = sort__comm_collapse,
648 .print = sort__comm_print,
649 .width = &comms__col_width,
655 sort__dso_cmp(struct hist_entry *left, struct hist_entry *right)
657 struct dso *dso_l = left->dso;
658 struct dso *dso_r = right->dso;
660 if (!dso_l || !dso_r)
661 return cmp_null(dso_l, dso_r);
663 return strcmp(dso_l->name, dso_r->name);
667 sort__dso_print(FILE *fp, struct hist_entry *self, unsigned int width)
670 return repsep_fprintf(fp, "%-*s", width, self->dso->name);
672 return repsep_fprintf(fp, "%*llx", width, (u64)self->ip);
675 static struct sort_entry sort_dso = {
676 .header = "Shared Object",
677 .cmp = sort__dso_cmp,
678 .print = sort__dso_print,
679 .width = &dsos__col_width,
685 sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
689 if (left->sym == right->sym)
692 ip_l = left->sym ? left->sym->start : left->ip;
693 ip_r = right->sym ? right->sym->start : right->ip;
695 return (int64_t)(ip_r - ip_l);
699 sort__sym_print(FILE *fp, struct hist_entry *self, unsigned int width __used)
704 ret += repsep_fprintf(fp, "%#018llx %c ", (u64)self->ip,
705 dso__symtab_origin(self->dso));
707 ret += repsep_fprintf(fp, "[%c] ", self->level);
709 ret += repsep_fprintf(fp, "%s", self->sym->name);
711 if (self->sym->module)
712 ret += repsep_fprintf(fp, "\t[%s]",
713 self->sym->module->name);
715 ret += repsep_fprintf(fp, "%#016llx", (u64)self->ip);
721 static struct sort_entry sort_sym = {
723 .cmp = sort__sym_cmp,
724 .print = sort__sym_print,
730 sort__parent_cmp(struct hist_entry *left, struct hist_entry *right)
732 struct symbol *sym_l = left->parent;
733 struct symbol *sym_r = right->parent;
735 if (!sym_l || !sym_r)
736 return cmp_null(sym_l, sym_r);
738 return strcmp(sym_l->name, sym_r->name);
742 sort__parent_print(FILE *fp, struct hist_entry *self, unsigned int width)
744 return repsep_fprintf(fp, "%-*s", width,
745 self->parent ? self->parent->name : "[other]");
748 static unsigned int parent_symbol__col_width;
750 static struct sort_entry sort_parent = {
751 .header = "Parent symbol",
752 .cmp = sort__parent_cmp,
753 .print = sort__parent_print,
754 .width = &parent_symbol__col_width,
757 static int sort__need_collapse = 0;
758 static int sort__has_parent = 0;
760 struct sort_dimension {
762 struct sort_entry *entry;
766 static struct sort_dimension sort_dimensions[] = {
767 { .name = "pid", .entry = &sort_thread, },
768 { .name = "comm", .entry = &sort_comm, },
769 { .name = "dso", .entry = &sort_dso, },
770 { .name = "symbol", .entry = &sort_sym, },
771 { .name = "parent", .entry = &sort_parent, },
774 static LIST_HEAD(hist_entry__sort_list);
776 static int sort_dimension__add(char *tok)
780 for (i = 0; i < ARRAY_SIZE(sort_dimensions); i++) {
781 struct sort_dimension *sd = &sort_dimensions[i];
786 if (strncasecmp(tok, sd->name, strlen(tok)))
789 if (sd->entry->collapse)
790 sort__need_collapse = 1;
792 if (sd->entry == &sort_parent) {
793 int ret = regcomp(&parent_regex, parent_pattern, REG_EXTENDED);
797 regerror(ret, &parent_regex, err, sizeof(err));
798 fprintf(stderr, "Invalid regex: %s\n%s",
799 parent_pattern, err);
802 sort__has_parent = 1;
805 list_add_tail(&sd->entry->list, &hist_entry__sort_list);
815 hist_entry__cmp(struct hist_entry *left, struct hist_entry *right)
817 struct sort_entry *se;
820 list_for_each_entry(se, &hist_entry__sort_list, list) {
821 cmp = se->cmp(left, right);
830 hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
832 struct sort_entry *se;
835 list_for_each_entry(se, &hist_entry__sort_list, list) {
836 int64_t (*f)(struct hist_entry *, struct hist_entry *);
838 f = se->collapse ?: se->cmp;
840 cmp = f(left, right);
848 static size_t ipchain__fprintf_graph_line(FILE *fp, int depth, int depth_mask)
853 ret += fprintf(fp, "%s", " ");
855 for (i = 0; i < depth; i++)
856 if (depth_mask & (1 << i))
857 ret += fprintf(fp, "| ");
859 ret += fprintf(fp, " ");
861 ret += fprintf(fp, "\n");
866 ipchain__fprintf_graph(FILE *fp, struct callchain_list *chain, int depth,
867 int depth_mask, int count, u64 total_samples,
873 ret += fprintf(fp, "%s", " ");
874 for (i = 0; i < depth; i++) {
875 if (depth_mask & (1 << i))
876 ret += fprintf(fp, "|");
878 ret += fprintf(fp, " ");
879 if (!count && i == depth - 1) {
882 percent = hits * 100.0 / total_samples;
883 ret += percent_color_fprintf(fp, "--%2.2f%%-- ", percent);
885 ret += fprintf(fp, "%s", " ");
888 ret += fprintf(fp, "%s\n", chain->sym->name);
890 ret += fprintf(fp, "%p\n", (void *)(long)chain->ip);
895 static struct symbol *rem_sq_bracket;
896 static struct callchain_list rem_hits;
898 static void init_rem_hits(void)
900 rem_sq_bracket = malloc(sizeof(*rem_sq_bracket) + 6);
901 if (!rem_sq_bracket) {
902 fprintf(stderr, "Not enough memory to display remaining hits\n");
906 strcpy(rem_sq_bracket->name, "[...]");
907 rem_hits.sym = rem_sq_bracket;
911 callchain__fprintf_graph(FILE *fp, struct callchain_node *self,
912 u64 total_samples, int depth, int depth_mask)
914 struct rb_node *node, *next;
915 struct callchain_node *child;
916 struct callchain_list *chain;
917 int new_depth_mask = depth_mask;
923 if (callchain_param.mode == CHAIN_GRAPH_REL)
924 new_total = self->children_hit;
926 new_total = total_samples;
928 remaining = new_total;
930 node = rb_first(&self->rb_root);
934 child = rb_entry(node, struct callchain_node, rb_node);
935 cumul = cumul_hits(child);
939 * The depth mask manages the output of pipes that show
940 * the depth. We don't want to keep the pipes of the current
941 * level for the last child of this depth.
942 * Except if we have remaining filtered hits. They will
943 * supersede the last child
945 next = rb_next(node);
946 if (!next && (callchain_param.mode != CHAIN_GRAPH_REL || !remaining))
947 new_depth_mask &= ~(1 << (depth - 1));
950 * But we keep the older depth mask for the line seperator
951 * to keep the level link until we reach the last child
953 ret += ipchain__fprintf_graph_line(fp, depth, depth_mask);
955 list_for_each_entry(chain, &child->val, list) {
956 if (chain->ip >= PERF_CONTEXT_MAX)
958 ret += ipchain__fprintf_graph(fp, chain, depth,
963 ret += callchain__fprintf_graph(fp, child, new_total,
965 new_depth_mask | (1 << depth));
969 if (callchain_param.mode == CHAIN_GRAPH_REL &&
970 remaining && remaining != new_total) {
975 new_depth_mask &= ~(1 << (depth - 1));
977 ret += ipchain__fprintf_graph(fp, &rem_hits, depth,
978 new_depth_mask, 0, new_total,
986 callchain__fprintf_flat(FILE *fp, struct callchain_node *self,
989 struct callchain_list *chain;
995 ret += callchain__fprintf_flat(fp, self->parent, total_samples);
998 list_for_each_entry(chain, &self->val, list) {
999 if (chain->ip >= PERF_CONTEXT_MAX)
1002 ret += fprintf(fp, " %s\n", chain->sym->name);
1004 ret += fprintf(fp, " %p\n",
1005 (void *)(long)chain->ip);
1012 hist_entry_callchain__fprintf(FILE *fp, struct hist_entry *self,
1015 struct rb_node *rb_node;
1016 struct callchain_node *chain;
1019 rb_node = rb_first(&self->sorted_chain);
1023 chain = rb_entry(rb_node, struct callchain_node, rb_node);
1024 percent = chain->hit * 100.0 / total_samples;
1025 switch (callchain_param.mode) {
1027 ret += percent_color_fprintf(fp, " %6.2f%%\n",
1029 ret += callchain__fprintf_flat(fp, chain, total_samples);
1031 case CHAIN_GRAPH_ABS: /* Falldown */
1032 case CHAIN_GRAPH_REL:
1033 ret += callchain__fprintf_graph(fp, chain,
1034 total_samples, 1, 1);
1038 ret += fprintf(fp, "\n");
1039 rb_node = rb_next(rb_node);
1047 hist_entry__fprintf(FILE *fp, struct hist_entry *self, u64 total_samples)
1049 struct sort_entry *se;
1052 if (exclude_other && !self->parent)
1056 ret = percent_color_fprintf(fp,
1057 field_sep ? "%.2f" : " %6.2f%%",
1058 (self->count * 100.0) / total_samples);
1060 ret = fprintf(fp, field_sep ? "%lld" : "%12lld ", self->count);
1062 if (show_nr_samples) {
1064 fprintf(fp, "%c%lld", *field_sep, self->count);
1066 fprintf(fp, "%11lld", self->count);
1069 list_for_each_entry(se, &hist_entry__sort_list, list) {
1073 fprintf(fp, "%s", field_sep ?: " ");
1074 ret += se->print(fp, self, se->width ? *se->width : 0);
1077 ret += fprintf(fp, "\n");
1080 hist_entry_callchain__fprintf(fp, self, total_samples);
1089 static void dso__calc_col_width(struct dso *self)
1091 if (!col_width_list_str && !field_sep &&
1092 (!dso_list || strlist__has_entry(dso_list, self->name))) {
1093 unsigned int slen = strlen(self->name);
1094 if (slen > dsos__col_width)
1095 dsos__col_width = slen;
1098 self->slen_calculated = 1;
1101 static struct symbol *
1102 resolve_symbol(struct thread *thread, struct map **mapp,
1103 struct dso **dsop, u64 *ipp)
1105 struct dso *dso = dsop ? *dsop : NULL;
1106 struct map *map = mapp ? *mapp : NULL;
1118 map = thread__find_map(thread, ip);
1121 * We have to do this here as we may have a dso
1122 * with no symbol hit that has a name longer than
1123 * the ones with symbols sampled.
1125 if (!sort_dso.elide && !map->dso->slen_calculated)
1126 dso__calc_col_width(map->dso);
1131 ip = map->map_ip(map, ip);
1136 * If this is outside of all known maps,
1137 * and is a negative address, try to look it
1138 * up in the kernel dso, as it might be a
1139 * vsyscall (which executes in user-mode):
1141 if ((long long)ip < 0)
1144 dprintf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
1145 dprintf(" ...... map: %Lx -> %Lx\n", *ipp, ip);
1154 return dso->find_symbol(dso, ip);
1157 static int call__match(struct symbol *sym)
1159 if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
1165 static struct symbol **
1166 resolve_callchain(struct thread *thread, struct map *map __used,
1167 struct ip_callchain *chain, struct hist_entry *entry)
1169 u64 context = PERF_CONTEXT_MAX;
1170 struct symbol **syms = NULL;
1174 syms = calloc(chain->nr, sizeof(*syms));
1176 fprintf(stderr, "Can't allocate memory for symbols\n");
1181 for (i = 0; i < chain->nr; i++) {
1182 u64 ip = chain->ips[i];
1183 struct dso *dso = NULL;
1186 if (ip >= PERF_CONTEXT_MAX) {
1192 case PERF_CONTEXT_HV:
1193 dso = hypervisor_dso;
1195 case PERF_CONTEXT_KERNEL:
1202 sym = resolve_symbol(thread, NULL, &dso, &ip);
1205 if (sort__has_parent && call__match(sym) &&
1207 entry->parent = sym;
1218 * collect histogram counts
1222 hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
1223 struct symbol *sym, u64 ip, struct ip_callchain *chain,
1224 char level, u64 count)
1226 struct rb_node **p = &hist.rb_node;
1227 struct rb_node *parent = NULL;
1228 struct hist_entry *he;
1229 struct symbol **syms = NULL;
1230 struct hist_entry entry = {
1239 .sorted_chain = RB_ROOT
1243 if ((sort__has_parent || callchain) && chain)
1244 syms = resolve_callchain(thread, map, chain, &entry);
1246 while (*p != NULL) {
1248 he = rb_entry(parent, struct hist_entry, rb_node);
1250 cmp = hist_entry__cmp(&entry, he);
1255 append_chain(&he->callchain, chain, syms);
1264 p = &(*p)->rb_right;
1267 he = malloc(sizeof(*he));
1272 callchain_init(&he->callchain);
1273 append_chain(&he->callchain, chain, syms);
1276 rb_link_node(&he->rb_node, parent, p);
1277 rb_insert_color(&he->rb_node, &hist);
1282 static void hist_entry__free(struct hist_entry *he)
1288 * collapse the histogram
1291 static struct rb_root collapse_hists;
1293 static void collapse__insert_entry(struct hist_entry *he)
1295 struct rb_node **p = &collapse_hists.rb_node;
1296 struct rb_node *parent = NULL;
1297 struct hist_entry *iter;
1300 while (*p != NULL) {
1302 iter = rb_entry(parent, struct hist_entry, rb_node);
1304 cmp = hist_entry__collapse(iter, he);
1307 iter->count += he->count;
1308 hist_entry__free(he);
1315 p = &(*p)->rb_right;
1318 rb_link_node(&he->rb_node, parent, p);
1319 rb_insert_color(&he->rb_node, &collapse_hists);
1322 static void collapse__resort(void)
1324 struct rb_node *next;
1325 struct hist_entry *n;
1327 if (!sort__need_collapse)
1330 next = rb_first(&hist);
1332 n = rb_entry(next, struct hist_entry, rb_node);
1333 next = rb_next(&n->rb_node);
1335 rb_erase(&n->rb_node, &hist);
1336 collapse__insert_entry(n);
1341 * reverse the map, sort on count.
1344 static struct rb_root output_hists;
1346 static void output__insert_entry(struct hist_entry *he, u64 min_callchain_hits)
1348 struct rb_node **p = &output_hists.rb_node;
1349 struct rb_node *parent = NULL;
1350 struct hist_entry *iter;
1353 callchain_param.sort(&he->sorted_chain, &he->callchain,
1354 min_callchain_hits, &callchain_param);
1356 while (*p != NULL) {
1358 iter = rb_entry(parent, struct hist_entry, rb_node);
1360 if (he->count > iter->count)
1363 p = &(*p)->rb_right;
1366 rb_link_node(&he->rb_node, parent, p);
1367 rb_insert_color(&he->rb_node, &output_hists);
1370 static void output__resort(u64 total_samples)
1372 struct rb_node *next;
1373 struct hist_entry *n;
1374 struct rb_root *tree = &hist;
1375 u64 min_callchain_hits;
1377 min_callchain_hits = total_samples * (callchain_param.min_percent / 100);
1379 if (sort__need_collapse)
1380 tree = &collapse_hists;
1382 next = rb_first(tree);
1385 n = rb_entry(next, struct hist_entry, rb_node);
1386 next = rb_next(&n->rb_node);
1388 rb_erase(&n->rb_node, tree);
1389 output__insert_entry(n, min_callchain_hits);
1393 static size_t output__fprintf(FILE *fp, u64 total_samples)
1395 struct hist_entry *pos;
1396 struct sort_entry *se;
1400 char *col_width = col_width_list_str;
1404 fprintf(fp, "# Samples: %Ld\n", (u64)total_samples);
1407 fprintf(fp, "# Overhead");
1408 if (show_nr_samples) {
1410 fprintf(fp, "%cSamples", *field_sep);
1412 fputs(" Samples ", fp);
1414 list_for_each_entry(se, &hist_entry__sort_list, list) {
1418 fprintf(fp, "%c%s", *field_sep, se->header);
1421 width = strlen(se->header);
1423 if (col_width_list_str) {
1425 *se->width = atoi(col_width);
1426 col_width = strchr(col_width, ',');
1431 width = *se->width = max(*se->width, width);
1433 fprintf(fp, " %*s", width, se->header);
1440 fprintf(fp, "# ........");
1441 if (show_nr_samples)
1442 fprintf(fp, " ..........");
1443 list_for_each_entry(se, &hist_entry__sort_list, list) {
1453 width = strlen(se->header);
1454 for (i = 0; i < width; i++)
1462 for (nd = rb_first(&output_hists); nd; nd = rb_next(nd)) {
1463 pos = rb_entry(nd, struct hist_entry, rb_node);
1464 ret += hist_entry__fprintf(fp, pos, total_samples);
1467 if (sort_order == default_sort_order &&
1468 parent_pattern == default_parent_pattern) {
1470 fprintf(fp, "# (For a higher level overview, try: perf report --sort comm,dso)\n");
1475 free(rem_sq_bracket);
1480 static void register_idle_thread(void)
1482 struct thread *thread = threads__findnew(0);
1484 if (thread == NULL ||
1485 thread__set_comm(thread, "[idle]")) {
1486 fprintf(stderr, "problem inserting idle task.\n");
1491 static unsigned long total = 0,
1498 static int validate_chain(struct ip_callchain *chain, event_t *event)
1500 unsigned int chain_size;
1502 chain_size = event->header.size;
1503 chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;
1505 if (chain->nr*sizeof(u64) > chain_size)
1512 process_sample_event(event_t *event, unsigned long offset, unsigned long head)
1516 struct dso *dso = NULL;
1517 struct thread *thread = threads__findnew(event->ip.pid);
1518 u64 ip = event->ip.ip;
1520 struct map *map = NULL;
1521 void *more_data = event->ip.__more_data;
1522 struct ip_callchain *chain = NULL;
1525 if (sample_type & PERF_SAMPLE_PERIOD) {
1526 period = *(u64 *)more_data;
1527 more_data += sizeof(u64);
1530 dprintf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n",
1531 (void *)(offset + head),
1532 (void *)(long)(event->header.size),
1534 event->ip.pid, event->ip.tid,
1538 if (sample_type & PERF_SAMPLE_CALLCHAIN) {
1541 chain = (void *)more_data;
1543 dprintf("... chain: nr:%Lu\n", chain->nr);
1545 if (validate_chain(chain, event) < 0) {
1546 eprintf("call-chain problem with event, skipping it.\n");
1551 for (i = 0; i < chain->nr; i++)
1552 dprintf("..... %2d: %016Lx\n", i, chain->ips[i]);
1556 dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid);
1558 if (thread == NULL) {
1559 eprintf("problem processing %d event, skipping it.\n",
1560 event->header.type);
1564 if (comm_list && !strlist__has_entry(comm_list, thread->comm))
1567 cpumode = event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK;
1569 if (cpumode == PERF_EVENT_MISC_KERNEL) {
1575 dprintf(" ...... dso: %s\n", dso->name);
1577 } else if (cpumode == PERF_EVENT_MISC_USER) {
1586 dso = hypervisor_dso;
1588 dprintf(" ...... dso: [hypervisor]\n");
1591 if (show & show_mask) {
1592 struct symbol *sym = resolve_symbol(thread, &map, &dso, &ip);
1594 if (dso_list && (!dso || !dso->name ||
1595 !strlist__has_entry(dso_list, dso->name)))
1598 if (sym_list && (!sym || !strlist__has_entry(sym_list, sym->name)))
1601 if (hist_entry__add(thread, map, dso, sym, ip, chain, level, period)) {
1602 eprintf("problem incrementing symbol count, skipping event\n");
1612 process_mmap_event(event_t *event, unsigned long offset, unsigned long head)
1614 struct thread *thread = threads__findnew(event->mmap.pid);
1615 struct map *map = map__new(&event->mmap);
1617 dprintf("%p [%p]: PERF_EVENT_MMAP %d/%d: [%p(%p) @ %p]: %s\n",
1618 (void *)(offset + head),
1619 (void *)(long)(event->header.size),
1622 (void *)(long)event->mmap.start,
1623 (void *)(long)event->mmap.len,
1624 (void *)(long)event->mmap.pgoff,
1625 event->mmap.filename);
1627 if (thread == NULL || map == NULL) {
1628 dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1632 thread__insert_map(thread, map);
1639 process_comm_event(event_t *event, unsigned long offset, unsigned long head)
1641 struct thread *thread = threads__findnew(event->comm.pid);
1643 dprintf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
1644 (void *)(offset + head),
1645 (void *)(long)(event->header.size),
1646 event->comm.comm, event->comm.pid);
1648 if (thread == NULL ||
1649 thread__set_comm(thread, event->comm.comm)) {
1650 dprintf("problem processing PERF_EVENT_COMM, skipping event.\n");
1659 process_task_event(event_t *event, unsigned long offset, unsigned long head)
1661 struct thread *thread = threads__findnew(event->fork.pid);
1662 struct thread *parent = threads__findnew(event->fork.ppid);
1664 dprintf("%p [%p]: PERF_EVENT_%s: (%d:%d):(%d:%d)\n",
1665 (void *)(offset + head),
1666 (void *)(long)(event->header.size),
1667 event->header.type == PERF_EVENT_FORK ? "FORK" : "EXIT",
1668 event->fork.pid, event->fork.tid,
1669 event->fork.ppid, event->fork.ptid);
1672 * A thread clone will have the same PID for both
1675 if (thread == parent)
1678 if (event->header.type == PERF_EVENT_EXIT)
1681 if (!thread || !parent || thread__fork(thread, parent)) {
1682 dprintf("problem processing PERF_EVENT_FORK, skipping event.\n");
1691 process_lost_event(event_t *event, unsigned long offset, unsigned long head)
1693 dprintf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n",
1694 (void *)(offset + head),
1695 (void *)(long)(event->header.size),
1699 total_lost += event->lost.lost;
1704 static void trace_event(event_t *event)
1706 unsigned char *raw_event = (void *)event;
1707 char *color = PERF_COLOR_BLUE;
1714 cdprintf("\n. ... raw event: size %d bytes\n", event->header.size);
1716 for (i = 0; i < event->header.size; i++) {
1717 if ((i & 15) == 0) {
1719 cdprintf(" %04x: ", i);
1722 cdprintf(" %02x", raw_event[i]);
1724 if (((i & 15) == 15) || i == event->header.size-1) {
1726 for (j = 0; j < 15-(i & 15); j++)
1728 for (j = 0; j < (i & 15); j++) {
1729 if (isprint(raw_event[i-15+j]))
1730 cdprintf("%c", raw_event[i-15+j]);
1740 static struct perf_header *header;
1742 static struct perf_counter_attr *perf_header__find_attr(u64 id)
1746 for (i = 0; i < header->attrs; i++) {
1747 struct perf_header_attr *attr = header->attr[i];
1750 for (j = 0; j < attr->ids; j++) {
1751 if (attr->id[j] == id)
1760 process_read_event(event_t *event, unsigned long offset, unsigned long head)
1762 struct perf_counter_attr *attr = perf_header__find_attr(event->read.id);
1764 dprintf("%p [%p]: PERF_EVENT_READ: %d %d %s %Lu\n",
1765 (void *)(offset + head),
1766 (void *)(long)(event->header.size),
1769 attr ? __event_name(attr->type, attr->config)
1777 process_event(event_t *event, unsigned long offset, unsigned long head)
1781 switch (event->header.type) {
1782 case PERF_EVENT_SAMPLE:
1783 return process_sample_event(event, offset, head);
1785 case PERF_EVENT_MMAP:
1786 return process_mmap_event(event, offset, head);
1788 case PERF_EVENT_COMM:
1789 return process_comm_event(event, offset, head);
1791 case PERF_EVENT_FORK:
1792 case PERF_EVENT_EXIT:
1793 return process_task_event(event, offset, head);
1795 case PERF_EVENT_LOST:
1796 return process_lost_event(event, offset, head);
1798 case PERF_EVENT_READ:
1799 return process_read_event(event, offset, head);
1802 * We dont process them right now but they are fine:
1805 case PERF_EVENT_THROTTLE:
1806 case PERF_EVENT_UNTHROTTLE:
1816 static u64 perf_header__sample_type(void)
1818 u64 sample_type = 0;
1821 for (i = 0; i < header->attrs; i++) {
1822 struct perf_header_attr *attr = header->attr[i];
1825 sample_type = attr->attr.sample_type;
1826 else if (sample_type != attr->attr.sample_type)
1827 die("non matching sample_type");
1833 static int __cmd_report(void)
1835 int ret, rc = EXIT_FAILURE;
1836 unsigned long offset = 0;
1837 unsigned long head, shift;
1843 register_idle_thread();
1845 input = open(input_name, O_RDONLY);
1847 fprintf(stderr, " failed to open file: %s", input_name);
1848 if (!strcmp(input_name, "perf.data"))
1849 fprintf(stderr, " (try 'perf record' first)");
1850 fprintf(stderr, "\n");
1854 ret = fstat(input, &stat);
1856 perror("failed to stat file");
1860 if (!force && (stat.st_uid != geteuid())) {
1861 fprintf(stderr, "file: %s not owned by current user\n", input_name);
1865 if (!stat.st_size) {
1866 fprintf(stderr, "zero-sized file, nothing to do!\n");
1870 header = perf_header__read(input);
1871 head = header->data_offset;
1873 sample_type = perf_header__sample_type();
1875 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
1876 if (sort__has_parent) {
1877 fprintf(stderr, "selected --sort parent, but no"
1878 " callchain data. Did you call"
1879 " perf record without -g?\n");
1883 fprintf(stderr, "selected -c but no callchain data."
1884 " Did you call perf record without"
1888 } else if (callchain_param.mode != CHAIN_NONE && !callchain) {
1890 if (register_callchain_param(&callchain_param) < 0) {
1891 fprintf(stderr, "Can't register callchain"
1897 if (load_kernel() < 0) {
1898 perror("failed to load kernel symbols");
1899 return EXIT_FAILURE;
1903 if (getcwd(__cwd, sizeof(__cwd)) == NULL) {
1904 perror("failed to get the current directory");
1905 return EXIT_FAILURE;
1907 cwdlen = strlen(cwd);
1913 shift = page_size * (head / page_size);
1918 buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
1919 MAP_SHARED, input, offset);
1920 if (buf == MAP_FAILED) {
1921 perror("failed to mmap file");
1926 event = (event_t *)(buf + head);
1928 size = event->header.size;
1932 if (head + event->header.size >= page_size * mmap_window) {
1935 shift = page_size * (head / page_size);
1937 ret = munmap(buf, page_size * mmap_window);
1945 size = event->header.size;
1947 dprintf("\n%p [%p]: event: %d\n",
1948 (void *)(offset + head),
1949 (void *)(long)event->header.size,
1950 event->header.type);
1952 if (!size || process_event(event, offset, head) < 0) {
1954 dprintf("%p [%p]: skipping unknown header type: %d\n",
1955 (void *)(offset + head),
1956 (void *)(long)(event->header.size),
1957 event->header.type);
1962 * assume we lost track of the stream, check alignment, and
1963 * increment a single u64 in the hope to catch on again 'soon'.
1966 if (unlikely(head & 7))
1974 if (offset + head >= header->data_offset + header->data_size)
1977 if (offset + head < (unsigned long)stat.st_size)
1984 dprintf(" IP events: %10ld\n", total);
1985 dprintf(" mmap events: %10ld\n", total_mmap);
1986 dprintf(" comm events: %10ld\n", total_comm);
1987 dprintf(" fork events: %10ld\n", total_fork);
1988 dprintf(" lost events: %10ld\n", total_lost);
1989 dprintf(" unknown events: %10ld\n", total_unknown);
1995 threads__fprintf(stdout);
1998 dsos__fprintf(stdout);
2001 output__resort(total);
2002 output__fprintf(stdout, total);
2008 parse_callchain_opt(const struct option *opt __used, const char *arg,
2019 tok = strtok((char *)arg, ",");
2023 /* get the output mode */
2024 if (!strncmp(tok, "graph", strlen(arg)))
2025 callchain_param.mode = CHAIN_GRAPH_ABS;
2027 else if (!strncmp(tok, "flat", strlen(arg)))
2028 callchain_param.mode = CHAIN_FLAT;
2030 else if (!strncmp(tok, "fractal", strlen(arg)))
2031 callchain_param.mode = CHAIN_GRAPH_REL;
2033 else if (!strncmp(tok, "none", strlen(arg))) {
2034 callchain_param.mode = CHAIN_NONE;
2043 /* get the min percentage */
2044 tok = strtok(NULL, ",");
2048 callchain_param.min_percent = strtod(tok, &endptr);
2053 if (register_callchain_param(&callchain_param) < 0) {
2054 fprintf(stderr, "Can't register callchain params\n");
2060 static const char * const report_usage[] = {
2061 "perf report [<options>] <command>",
2065 static const struct option options[] = {
2066 OPT_STRING('i', "input", &input_name, "file",
2068 OPT_BOOLEAN('v', "verbose", &verbose,
2069 "be more verbose (show symbol address, etc)"),
2070 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
2071 "dump raw trace in ASCII"),
2072 OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"),
2073 OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
2074 OPT_BOOLEAN('m', "modules", &modules,
2075 "load module symbols - WARNING: use only with -k and LIVE kernel"),
2076 OPT_BOOLEAN('n', "show-nr-samples", &show_nr_samples,
2077 "Show a column with the number of samples"),
2078 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
2079 "sort by key(s): pid, comm, dso, symbol, parent"),
2080 OPT_BOOLEAN('P', "full-paths", &full_paths,
2081 "Don't shorten the pathnames taking into account the cwd"),
2082 OPT_STRING('p', "parent", &parent_pattern, "regex",
2083 "regex filter to identify parent, see: '--sort parent'"),
2084 OPT_BOOLEAN('x', "exclude-other", &exclude_other,
2085 "Only display entries with parent-match"),
2086 OPT_CALLBACK_DEFAULT('g', "call-graph", NULL, "output_type,min_percent",
2087 "Display callchains using output_type and min percent threshold. "
2088 "Default: fractal,0.5", &parse_callchain_opt, callchain_default_opt),
2089 OPT_STRING('d', "dsos", &dso_list_str, "dso[,dso...]",
2090 "only consider symbols in these dsos"),
2091 OPT_STRING('C', "comms", &comm_list_str, "comm[,comm...]",
2092 "only consider symbols in these comms"),
2093 OPT_STRING('S', "symbols", &sym_list_str, "symbol[,symbol...]",
2094 "only consider these symbols"),
2095 OPT_STRING('w', "column-widths", &col_width_list_str,
2097 "don't try to adjust column width, use these fixed values"),
2098 OPT_STRING('t', "field-separator", &field_sep, "separator",
2099 "separator for columns, no spaces will be added between "
2100 "columns '.' is reserved."),
2104 static void setup_sorting(void)
2106 char *tmp, *tok, *str = strdup(sort_order);
2108 for (tok = strtok_r(str, ", ", &tmp);
2109 tok; tok = strtok_r(NULL, ", ", &tmp)) {
2110 if (sort_dimension__add(tok) < 0) {
2111 error("Unknown --sort key: `%s'", tok);
2112 usage_with_options(report_usage, options);
2119 static void setup_list(struct strlist **list, const char *list_str,
2120 struct sort_entry *se, const char *list_name,
2124 *list = strlist__new(true, list_str);
2126 fprintf(stderr, "problems parsing %s list\n",
2130 if (strlist__nr_entries(*list) == 1) {
2131 fprintf(fp, "# %s: %s\n", list_name,
2132 strlist__entry(*list, 0)->s);
2138 int cmd_report(int argc, const char **argv, const char *prefix __used)
2142 page_size = getpagesize();
2144 argc = parse_options(argc, argv, options, report_usage, 0);
2148 if (parent_pattern != default_parent_pattern) {
2149 sort_dimension__add("parent");
2150 sort_parent.elide = 1;
2155 * Any (unrecognized) arguments left?
2158 usage_with_options(report_usage, options);
2162 setup_list(&dso_list, dso_list_str, &sort_dso, "dso", stdout);
2163 setup_list(&comm_list, comm_list_str, &sort_comm, "comm", stdout);
2164 setup_list(&sym_list, sym_list_str, &sort_sym, "symbol", stdout);
2166 if (field_sep && *field_sep == '.') {
2167 fputs("'.' is the only non valid --field-separator argument\n",
2172 return __cmd_report();