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;
42 static int show_mask = SHOW_KERNEL | SHOW_USER | SHOW_HV;
44 static int dump_trace = 0;
45 #define dprintf(x...) do { if (dump_trace) printf(x); } while (0)
46 #define cdprintf(x...) do { if (dump_trace) color_fprintf(stdout, color, x); } while (0)
49 #define eprintf(x...) do { if (verbose) fprintf(stderr, x); } while (0)
53 static int full_paths;
54 static int show_nr_samples;
56 static unsigned long page_size;
57 static unsigned long mmap_window = 32;
59 static char default_parent_pattern[] = "^sys_|^do_page_fault";
60 static char *parent_pattern = default_parent_pattern;
61 static regex_t parent_regex;
63 static int exclude_other = 1;
65 static char callchain_default_opt[] = "fractal,0.5";
70 struct callchain_param callchain_param = {
71 .mode = CHAIN_GRAPH_ABS,
75 static u64 sample_type;
78 struct perf_event_header header;
81 unsigned char __more_data[];
85 struct perf_event_header header;
90 char filename[PATH_MAX];
94 struct perf_event_header header;
100 struct perf_event_header header;
106 struct perf_event_header header;
112 struct perf_event_header header;
120 typedef union event_union {
121 struct perf_event_header header;
123 struct mmap_event mmap;
124 struct comm_event comm;
125 struct fork_event fork;
126 struct lost_event lost;
127 struct read_event read;
130 static int repsep_fprintf(FILE *fp, const char *fmt, ...)
137 n = vfprintf(fp, fmt, ap);
140 n = vasprintf(&bf, fmt, ap);
144 sep = strchr(sep, *field_sep);
157 static LIST_HEAD(dsos);
158 static struct dso *kernel_dso;
159 static struct dso *vdso;
160 static struct dso *hypervisor_dso;
162 static void dsos__add(struct dso *dso)
164 list_add_tail(&dso->node, &dsos);
167 static struct dso *dsos__find(const char *name)
171 list_for_each_entry(pos, &dsos, node)
172 if (strcmp(pos->name, name) == 0)
177 static struct dso *dsos__findnew(const char *name)
179 struct dso *dso = dsos__find(name);
185 dso = dso__new(name, 0);
189 nr = dso__load(dso, NULL, verbose);
191 eprintf("Failed to open: %s\n", name);
195 eprintf("No symbols found in: %s, maybe install a debug package?\n", name);
206 static void dsos__fprintf(FILE *fp)
210 list_for_each_entry(pos, &dsos, node)
211 dso__fprintf(pos, fp);
214 static struct symbol *vdso__find_symbol(struct dso *dso, u64 ip)
216 return dso__find_symbol(dso, ip);
219 static int load_kernel(void)
223 kernel_dso = dso__new("[kernel]", 0);
227 err = dso__load_kernel(kernel_dso, vmlinux, NULL, verbose, modules);
229 dso__delete(kernel_dso);
232 dsos__add(kernel_dso);
234 vdso = dso__new("[vdso]", 0);
238 vdso->find_symbol = vdso__find_symbol;
242 hypervisor_dso = dso__new("[hypervisor]", 0);
245 dsos__add(hypervisor_dso);
250 static char __cwd[PATH_MAX];
251 static char *cwd = __cwd;
254 static int strcommon(const char *pathname)
258 while (n < cwdlen && pathname[n] == cwd[n])
265 struct list_head node;
269 u64 (*map_ip)(struct map *, u64);
273 static u64 map__map_ip(struct map *map, u64 ip)
275 return ip - map->start + map->pgoff;
278 static u64 vdso__map_ip(struct map *map __used, u64 ip)
283 static inline int is_anon_memory(const char *filename)
285 return strcmp(filename, "//anon") == 0;
288 static struct map *map__new(struct mmap_event *event)
290 struct map *self = malloc(sizeof(*self));
293 const char *filename = event->filename;
294 char newfilename[PATH_MAX];
298 int n = strcommon(filename);
301 snprintf(newfilename, sizeof(newfilename),
302 ".%s", filename + n);
303 filename = newfilename;
307 anon = is_anon_memory(filename);
310 snprintf(newfilename, sizeof(newfilename), "/tmp/perf-%d.map", event->pid);
311 filename = newfilename;
314 self->start = event->start;
315 self->end = event->start + event->len;
316 self->pgoff = event->pgoff;
318 self->dso = dsos__findnew(filename);
319 if (self->dso == NULL)
322 if (self->dso == vdso || anon)
323 self->map_ip = vdso__map_ip;
325 self->map_ip = map__map_ip;
333 static struct map *map__clone(struct map *self)
335 struct map *map = malloc(sizeof(*self));
340 memcpy(map, self, sizeof(*self));
345 static int map__overlap(struct map *l, struct map *r)
347 if (l->start > r->start) {
353 if (l->end > r->start)
359 static size_t map__fprintf(struct map *self, FILE *fp)
361 return fprintf(fp, " %Lx-%Lx %Lx %s\n",
362 self->start, self->end, self->pgoff, self->dso->name);
367 struct rb_node rb_node;
368 struct list_head maps;
373 static struct thread *thread__new(pid_t pid)
375 struct thread *self = malloc(sizeof(*self));
379 self->comm = malloc(32);
381 snprintf(self->comm, 32, ":%d", self->pid);
382 INIT_LIST_HEAD(&self->maps);
388 static unsigned int dsos__col_width,
392 static int thread__set_comm(struct thread *self, const char *comm)
396 self->comm = strdup(comm);
400 if (!col_width_list_str && !field_sep &&
401 (!comm_list || strlist__has_entry(comm_list, comm))) {
402 unsigned int slen = strlen(comm);
403 if (slen > comms__col_width) {
404 comms__col_width = slen;
405 threads__col_width = slen + 6;
412 static size_t thread__fprintf(struct thread *self, FILE *fp)
415 size_t ret = fprintf(fp, "Thread %d %s\n", self->pid, self->comm);
417 list_for_each_entry(pos, &self->maps, node)
418 ret += map__fprintf(pos, fp);
424 static struct rb_root threads;
425 static struct thread *last_match;
427 static struct thread *threads__findnew(pid_t pid)
429 struct rb_node **p = &threads.rb_node;
430 struct rb_node *parent = NULL;
434 * Font-end cache - PID lookups come in blocks,
435 * so most of the time we dont have to look up
438 if (last_match && last_match->pid == pid)
443 th = rb_entry(parent, struct thread, rb_node);
445 if (th->pid == pid) {
456 th = thread__new(pid);
458 rb_link_node(&th->rb_node, parent, p);
459 rb_insert_color(&th->rb_node, &threads);
466 static void thread__insert_map(struct thread *self, struct map *map)
468 struct map *pos, *tmp;
470 list_for_each_entry_safe(pos, tmp, &self->maps, node) {
471 if (map__overlap(pos, map)) {
473 printf("overlapping maps:\n");
474 map__fprintf(map, stdout);
475 map__fprintf(pos, stdout);
478 if (map->start <= pos->start && map->end > pos->start)
479 pos->start = map->end;
481 if (map->end >= pos->end && map->start < pos->end)
482 pos->end = map->start;
485 printf("after collision:\n");
486 map__fprintf(pos, stdout);
489 if (pos->start >= pos->end) {
490 list_del_init(&pos->node);
496 list_add_tail(&map->node, &self->maps);
499 static int thread__fork(struct thread *self, struct thread *parent)
505 self->comm = strdup(parent->comm);
509 list_for_each_entry(map, &parent->maps, node) {
510 struct map *new = map__clone(map);
513 thread__insert_map(self, new);
519 static struct map *thread__find_map(struct thread *self, u64 ip)
526 list_for_each_entry(pos, &self->maps, node)
527 if (ip >= pos->start && ip <= pos->end)
533 static size_t threads__fprintf(FILE *fp)
538 for (nd = rb_first(&threads); nd; nd = rb_next(nd)) {
539 struct thread *pos = rb_entry(nd, struct thread, rb_node);
541 ret += thread__fprintf(pos, fp);
548 * histogram, sorted on item, collects counts
551 static struct rb_root hist;
554 struct rb_node rb_node;
556 struct thread *thread;
560 struct symbol *parent;
563 struct callchain_node callchain;
564 struct rb_root sorted_chain;
570 * configurable sorting bits
574 struct list_head list;
578 int64_t (*cmp)(struct hist_entry *, struct hist_entry *);
579 int64_t (*collapse)(struct hist_entry *, struct hist_entry *);
580 size_t (*print)(FILE *fp, struct hist_entry *, unsigned int width);
585 static int64_t cmp_null(void *l, void *r)
598 sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
600 return right->thread->pid - left->thread->pid;
604 sort__thread_print(FILE *fp, struct hist_entry *self, unsigned int width)
606 return repsep_fprintf(fp, "%*s:%5d", width - 6,
607 self->thread->comm ?: "", self->thread->pid);
610 static struct sort_entry sort_thread = {
611 .header = "Command: Pid",
612 .cmp = sort__thread_cmp,
613 .print = sort__thread_print,
614 .width = &threads__col_width,
620 sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
622 return right->thread->pid - left->thread->pid;
626 sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
628 char *comm_l = left->thread->comm;
629 char *comm_r = right->thread->comm;
631 if (!comm_l || !comm_r)
632 return cmp_null(comm_l, comm_r);
634 return strcmp(comm_l, comm_r);
638 sort__comm_print(FILE *fp, struct hist_entry *self, unsigned int width)
640 return repsep_fprintf(fp, "%*s", width, self->thread->comm);
643 static struct sort_entry sort_comm = {
645 .cmp = sort__comm_cmp,
646 .collapse = sort__comm_collapse,
647 .print = sort__comm_print,
648 .width = &comms__col_width,
654 sort__dso_cmp(struct hist_entry *left, struct hist_entry *right)
656 struct dso *dso_l = left->dso;
657 struct dso *dso_r = right->dso;
659 if (!dso_l || !dso_r)
660 return cmp_null(dso_l, dso_r);
662 return strcmp(dso_l->name, dso_r->name);
666 sort__dso_print(FILE *fp, struct hist_entry *self, unsigned int width)
669 return repsep_fprintf(fp, "%-*s", width, self->dso->name);
671 return repsep_fprintf(fp, "%*llx", width, (u64)self->ip);
674 static struct sort_entry sort_dso = {
675 .header = "Shared Object",
676 .cmp = sort__dso_cmp,
677 .print = sort__dso_print,
678 .width = &dsos__col_width,
684 sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
688 if (left->sym == right->sym)
691 ip_l = left->sym ? left->sym->start : left->ip;
692 ip_r = right->sym ? right->sym->start : right->ip;
694 return (int64_t)(ip_r - ip_l);
698 sort__sym_print(FILE *fp, struct hist_entry *self, unsigned int width __used)
703 ret += repsep_fprintf(fp, "%#018llx ", (u64)self->ip);
705 ret += repsep_fprintf(fp, "[%c] ", self->level);
707 ret += repsep_fprintf(fp, "%s", self->sym->name);
709 if (self->sym->module)
710 ret += repsep_fprintf(fp, "\t[%s]",
711 self->sym->module->name);
713 ret += repsep_fprintf(fp, "%#016llx", (u64)self->ip);
719 static struct sort_entry sort_sym = {
721 .cmp = sort__sym_cmp,
722 .print = sort__sym_print,
728 sort__parent_cmp(struct hist_entry *left, struct hist_entry *right)
730 struct symbol *sym_l = left->parent;
731 struct symbol *sym_r = right->parent;
733 if (!sym_l || !sym_r)
734 return cmp_null(sym_l, sym_r);
736 return strcmp(sym_l->name, sym_r->name);
740 sort__parent_print(FILE *fp, struct hist_entry *self, unsigned int width)
742 return repsep_fprintf(fp, "%-*s", width,
743 self->parent ? self->parent->name : "[other]");
746 static unsigned int parent_symbol__col_width;
748 static struct sort_entry sort_parent = {
749 .header = "Parent symbol",
750 .cmp = sort__parent_cmp,
751 .print = sort__parent_print,
752 .width = &parent_symbol__col_width,
755 static int sort__need_collapse = 0;
756 static int sort__has_parent = 0;
758 struct sort_dimension {
760 struct sort_entry *entry;
764 static struct sort_dimension sort_dimensions[] = {
765 { .name = "pid", .entry = &sort_thread, },
766 { .name = "comm", .entry = &sort_comm, },
767 { .name = "dso", .entry = &sort_dso, },
768 { .name = "symbol", .entry = &sort_sym, },
769 { .name = "parent", .entry = &sort_parent, },
772 static LIST_HEAD(hist_entry__sort_list);
774 static int sort_dimension__add(char *tok)
778 for (i = 0; i < ARRAY_SIZE(sort_dimensions); i++) {
779 struct sort_dimension *sd = &sort_dimensions[i];
784 if (strncasecmp(tok, sd->name, strlen(tok)))
787 if (sd->entry->collapse)
788 sort__need_collapse = 1;
790 if (sd->entry == &sort_parent) {
791 int ret = regcomp(&parent_regex, parent_pattern, REG_EXTENDED);
795 regerror(ret, &parent_regex, err, sizeof(err));
796 fprintf(stderr, "Invalid regex: %s\n%s",
797 parent_pattern, err);
800 sort__has_parent = 1;
803 list_add_tail(&sd->entry->list, &hist_entry__sort_list);
813 hist_entry__cmp(struct hist_entry *left, struct hist_entry *right)
815 struct sort_entry *se;
818 list_for_each_entry(se, &hist_entry__sort_list, list) {
819 cmp = se->cmp(left, right);
828 hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
830 struct sort_entry *se;
833 list_for_each_entry(se, &hist_entry__sort_list, list) {
834 int64_t (*f)(struct hist_entry *, struct hist_entry *);
836 f = se->collapse ?: se->cmp;
838 cmp = f(left, right);
846 static size_t ipchain__fprintf_graph_line(FILE *fp, int depth, int depth_mask)
851 ret += fprintf(fp, "%s", " ");
853 for (i = 0; i < depth; i++)
854 if (depth_mask & (1 << i))
855 ret += fprintf(fp, "| ");
857 ret += fprintf(fp, " ");
859 ret += fprintf(fp, "\n");
864 ipchain__fprintf_graph(FILE *fp, struct callchain_list *chain, int depth,
865 int depth_mask, int count, u64 total_samples,
871 ret += fprintf(fp, "%s", " ");
872 for (i = 0; i < depth; i++) {
873 if (depth_mask & (1 << i))
874 ret += fprintf(fp, "|");
876 ret += fprintf(fp, " ");
877 if (!count && i == depth - 1) {
880 percent = hits * 100.0 / total_samples;
881 ret += percent_color_fprintf(fp, "--%2.2f%%-- ", percent);
883 ret += fprintf(fp, "%s", " ");
886 ret += fprintf(fp, "%s\n", chain->sym->name);
888 ret += fprintf(fp, "%p\n", (void *)(long)chain->ip);
894 callchain__fprintf_graph(FILE *fp, struct callchain_node *self,
895 u64 total_samples, int depth, int depth_mask)
897 struct rb_node *node, *next;
898 struct callchain_node *child;
899 struct callchain_list *chain;
900 int new_depth_mask = depth_mask;
905 if (callchain_param.mode == CHAIN_GRAPH_REL)
906 new_total = self->children_hit;
908 new_total = total_samples;
910 node = rb_first(&self->rb_root);
912 child = rb_entry(node, struct callchain_node, rb_node);
915 * The depth mask manages the output of pipes that show
916 * the depth. We don't want to keep the pipes of the current
917 * level for the last child of this depth
919 next = rb_next(node);
921 new_depth_mask &= ~(1 << (depth - 1));
924 * But we keep the older depth mask for the line seperator
925 * to keep the level link until we reach the last child
927 ret += ipchain__fprintf_graph_line(fp, depth, depth_mask);
929 list_for_each_entry(chain, &child->val, list) {
930 if (chain->ip >= PERF_CONTEXT_MAX)
932 ret += ipchain__fprintf_graph(fp, chain, depth,
937 ret += callchain__fprintf_graph(fp, child, new_total,
939 new_depth_mask | (1 << depth));
947 callchain__fprintf_flat(FILE *fp, struct callchain_node *self,
950 struct callchain_list *chain;
956 ret += callchain__fprintf_flat(fp, self->parent, total_samples);
959 list_for_each_entry(chain, &self->val, list) {
960 if (chain->ip >= PERF_CONTEXT_MAX)
963 ret += fprintf(fp, " %s\n", chain->sym->name);
965 ret += fprintf(fp, " %p\n",
966 (void *)(long)chain->ip);
973 hist_entry_callchain__fprintf(FILE *fp, struct hist_entry *self,
976 struct rb_node *rb_node;
977 struct callchain_node *chain;
980 rb_node = rb_first(&self->sorted_chain);
984 chain = rb_entry(rb_node, struct callchain_node, rb_node);
985 percent = chain->hit * 100.0 / total_samples;
986 switch (callchain_param.mode) {
988 ret += percent_color_fprintf(fp, " %6.2f%%\n",
990 ret += callchain__fprintf_flat(fp, chain, total_samples);
992 case CHAIN_GRAPH_ABS: /* Falldown */
993 case CHAIN_GRAPH_REL:
994 ret += callchain__fprintf_graph(fp, chain,
995 total_samples, 1, 1);
999 ret += fprintf(fp, "\n");
1000 rb_node = rb_next(rb_node);
1008 hist_entry__fprintf(FILE *fp, struct hist_entry *self, u64 total_samples)
1010 struct sort_entry *se;
1013 if (exclude_other && !self->parent)
1017 ret = percent_color_fprintf(fp,
1018 field_sep ? "%.2f" : " %6.2f%%",
1019 (self->count * 100.0) / total_samples);
1021 ret = fprintf(fp, field_sep ? "%lld" : "%12lld ", self->count);
1023 if (show_nr_samples) {
1025 fprintf(fp, "%c%lld", *field_sep, self->count);
1027 fprintf(fp, "%11lld", self->count);
1030 list_for_each_entry(se, &hist_entry__sort_list, list) {
1034 fprintf(fp, "%s", field_sep ?: " ");
1035 ret += se->print(fp, self, se->width ? *se->width : 0);
1038 ret += fprintf(fp, "\n");
1041 hist_entry_callchain__fprintf(fp, self, total_samples);
1050 static void dso__calc_col_width(struct dso *self)
1052 if (!col_width_list_str && !field_sep &&
1053 (!dso_list || strlist__has_entry(dso_list, self->name))) {
1054 unsigned int slen = strlen(self->name);
1055 if (slen > dsos__col_width)
1056 dsos__col_width = slen;
1059 self->slen_calculated = 1;
1062 static struct symbol *
1063 resolve_symbol(struct thread *thread, struct map **mapp,
1064 struct dso **dsop, u64 *ipp)
1066 struct dso *dso = dsop ? *dsop : NULL;
1067 struct map *map = mapp ? *mapp : NULL;
1079 map = thread__find_map(thread, ip);
1082 * We have to do this here as we may have a dso
1083 * with no symbol hit that has a name longer than
1084 * the ones with symbols sampled.
1086 if (!sort_dso.elide && !map->dso->slen_calculated)
1087 dso__calc_col_width(map->dso);
1092 ip = map->map_ip(map, ip);
1097 * If this is outside of all known maps,
1098 * and is a negative address, try to look it
1099 * up in the kernel dso, as it might be a
1100 * vsyscall (which executes in user-mode):
1102 if ((long long)ip < 0)
1105 dprintf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
1106 dprintf(" ...... map: %Lx -> %Lx\n", *ipp, ip);
1115 return dso->find_symbol(dso, ip);
1118 static int call__match(struct symbol *sym)
1120 if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
1126 static struct symbol **
1127 resolve_callchain(struct thread *thread, struct map *map __used,
1128 struct ip_callchain *chain, struct hist_entry *entry)
1130 u64 context = PERF_CONTEXT_MAX;
1131 struct symbol **syms = NULL;
1135 syms = calloc(chain->nr, sizeof(*syms));
1137 fprintf(stderr, "Can't allocate memory for symbols\n");
1142 for (i = 0; i < chain->nr; i++) {
1143 u64 ip = chain->ips[i];
1144 struct dso *dso = NULL;
1147 if (ip >= PERF_CONTEXT_MAX) {
1153 case PERF_CONTEXT_HV:
1154 dso = hypervisor_dso;
1156 case PERF_CONTEXT_KERNEL:
1163 sym = resolve_symbol(thread, NULL, &dso, &ip);
1166 if (sort__has_parent && call__match(sym) &&
1168 entry->parent = sym;
1179 * collect histogram counts
1183 hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
1184 struct symbol *sym, u64 ip, struct ip_callchain *chain,
1185 char level, u64 count)
1187 struct rb_node **p = &hist.rb_node;
1188 struct rb_node *parent = NULL;
1189 struct hist_entry *he;
1190 struct symbol **syms = NULL;
1191 struct hist_entry entry = {
1200 .sorted_chain = RB_ROOT
1204 if ((sort__has_parent || callchain) && chain)
1205 syms = resolve_callchain(thread, map, chain, &entry);
1207 while (*p != NULL) {
1209 he = rb_entry(parent, struct hist_entry, rb_node);
1211 cmp = hist_entry__cmp(&entry, he);
1216 append_chain(&he->callchain, chain, syms);
1225 p = &(*p)->rb_right;
1228 he = malloc(sizeof(*he));
1233 callchain_init(&he->callchain);
1234 append_chain(&he->callchain, chain, syms);
1237 rb_link_node(&he->rb_node, parent, p);
1238 rb_insert_color(&he->rb_node, &hist);
1243 static void hist_entry__free(struct hist_entry *he)
1249 * collapse the histogram
1252 static struct rb_root collapse_hists;
1254 static void collapse__insert_entry(struct hist_entry *he)
1256 struct rb_node **p = &collapse_hists.rb_node;
1257 struct rb_node *parent = NULL;
1258 struct hist_entry *iter;
1261 while (*p != NULL) {
1263 iter = rb_entry(parent, struct hist_entry, rb_node);
1265 cmp = hist_entry__collapse(iter, he);
1268 iter->count += he->count;
1269 hist_entry__free(he);
1276 p = &(*p)->rb_right;
1279 rb_link_node(&he->rb_node, parent, p);
1280 rb_insert_color(&he->rb_node, &collapse_hists);
1283 static void collapse__resort(void)
1285 struct rb_node *next;
1286 struct hist_entry *n;
1288 if (!sort__need_collapse)
1291 next = rb_first(&hist);
1293 n = rb_entry(next, struct hist_entry, rb_node);
1294 next = rb_next(&n->rb_node);
1296 rb_erase(&n->rb_node, &hist);
1297 collapse__insert_entry(n);
1302 * reverse the map, sort on count.
1305 static struct rb_root output_hists;
1307 static void output__insert_entry(struct hist_entry *he, u64 min_callchain_hits)
1309 struct rb_node **p = &output_hists.rb_node;
1310 struct rb_node *parent = NULL;
1311 struct hist_entry *iter;
1314 callchain_param.sort(&he->sorted_chain, &he->callchain,
1315 min_callchain_hits, &callchain_param);
1317 while (*p != NULL) {
1319 iter = rb_entry(parent, struct hist_entry, rb_node);
1321 if (he->count > iter->count)
1324 p = &(*p)->rb_right;
1327 rb_link_node(&he->rb_node, parent, p);
1328 rb_insert_color(&he->rb_node, &output_hists);
1331 static void output__resort(u64 total_samples)
1333 struct rb_node *next;
1334 struct hist_entry *n;
1335 struct rb_root *tree = &hist;
1336 u64 min_callchain_hits;
1338 min_callchain_hits = total_samples * (callchain_param.min_percent / 100);
1340 if (sort__need_collapse)
1341 tree = &collapse_hists;
1343 next = rb_first(tree);
1346 n = rb_entry(next, struct hist_entry, rb_node);
1347 next = rb_next(&n->rb_node);
1349 rb_erase(&n->rb_node, tree);
1350 output__insert_entry(n, min_callchain_hits);
1354 static size_t output__fprintf(FILE *fp, u64 total_samples)
1356 struct hist_entry *pos;
1357 struct sort_entry *se;
1361 char *col_width = col_width_list_str;
1363 fprintf(fp, "# Samples: %Ld\n", (u64)total_samples);
1366 fprintf(fp, "# Overhead");
1367 if (show_nr_samples) {
1369 fprintf(fp, "%cSamples", *field_sep);
1371 fputs(" Samples ", fp);
1373 list_for_each_entry(se, &hist_entry__sort_list, list) {
1377 fprintf(fp, "%c%s", *field_sep, se->header);
1380 width = strlen(se->header);
1382 if (col_width_list_str) {
1384 *se->width = atoi(col_width);
1385 col_width = strchr(col_width, ',');
1390 width = *se->width = max(*se->width, width);
1392 fprintf(fp, " %*s", width, se->header);
1399 fprintf(fp, "# ........");
1400 if (show_nr_samples)
1401 fprintf(fp, " ..........");
1402 list_for_each_entry(se, &hist_entry__sort_list, list) {
1412 width = strlen(se->header);
1413 for (i = 0; i < width; i++)
1421 for (nd = rb_first(&output_hists); nd; nd = rb_next(nd)) {
1422 pos = rb_entry(nd, struct hist_entry, rb_node);
1423 ret += hist_entry__fprintf(fp, pos, total_samples);
1426 if (sort_order == default_sort_order &&
1427 parent_pattern == default_parent_pattern) {
1429 fprintf(fp, "# (For a higher level overview, try: perf report --sort comm,dso)\n");
1437 static void register_idle_thread(void)
1439 struct thread *thread = threads__findnew(0);
1441 if (thread == NULL ||
1442 thread__set_comm(thread, "[idle]")) {
1443 fprintf(stderr, "problem inserting idle task.\n");
1448 static unsigned long total = 0,
1455 static int validate_chain(struct ip_callchain *chain, event_t *event)
1457 unsigned int chain_size;
1459 chain_size = event->header.size;
1460 chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;
1462 if (chain->nr*sizeof(u64) > chain_size)
1469 process_sample_event(event_t *event, unsigned long offset, unsigned long head)
1473 struct dso *dso = NULL;
1474 struct thread *thread = threads__findnew(event->ip.pid);
1475 u64 ip = event->ip.ip;
1477 struct map *map = NULL;
1478 void *more_data = event->ip.__more_data;
1479 struct ip_callchain *chain = NULL;
1482 if (sample_type & PERF_SAMPLE_PERIOD) {
1483 period = *(u64 *)more_data;
1484 more_data += sizeof(u64);
1487 dprintf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d: %p period: %Ld\n",
1488 (void *)(offset + head),
1489 (void *)(long)(event->header.size),
1495 if (sample_type & PERF_SAMPLE_CALLCHAIN) {
1498 chain = (void *)more_data;
1500 dprintf("... chain: nr:%Lu\n", chain->nr);
1502 if (validate_chain(chain, event) < 0) {
1503 eprintf("call-chain problem with event, skipping it.\n");
1508 for (i = 0; i < chain->nr; i++)
1509 dprintf("..... %2d: %016Lx\n", i, chain->ips[i]);
1513 dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid);
1515 if (thread == NULL) {
1516 eprintf("problem processing %d event, skipping it.\n",
1517 event->header.type);
1521 if (comm_list && !strlist__has_entry(comm_list, thread->comm))
1524 cpumode = event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK;
1526 if (cpumode == PERF_EVENT_MISC_KERNEL) {
1532 dprintf(" ...... dso: %s\n", dso->name);
1534 } else if (cpumode == PERF_EVENT_MISC_USER) {
1543 dso = hypervisor_dso;
1545 dprintf(" ...... dso: [hypervisor]\n");
1548 if (show & show_mask) {
1549 struct symbol *sym = resolve_symbol(thread, &map, &dso, &ip);
1551 if (dso_list && dso && dso->name && !strlist__has_entry(dso_list, dso->name))
1554 if (sym_list && sym && !strlist__has_entry(sym_list, sym->name))
1557 if (hist_entry__add(thread, map, dso, sym, ip, chain, level, period)) {
1558 eprintf("problem incrementing symbol count, skipping event\n");
1568 process_mmap_event(event_t *event, unsigned long offset, unsigned long head)
1570 struct thread *thread = threads__findnew(event->mmap.pid);
1571 struct map *map = map__new(&event->mmap);
1573 dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n",
1574 (void *)(offset + head),
1575 (void *)(long)(event->header.size),
1577 (void *)(long)event->mmap.start,
1578 (void *)(long)event->mmap.len,
1579 (void *)(long)event->mmap.pgoff,
1580 event->mmap.filename);
1582 if (thread == NULL || map == NULL) {
1583 dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1587 thread__insert_map(thread, map);
1594 process_comm_event(event_t *event, unsigned long offset, unsigned long head)
1596 struct thread *thread = threads__findnew(event->comm.pid);
1598 dprintf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
1599 (void *)(offset + head),
1600 (void *)(long)(event->header.size),
1601 event->comm.comm, event->comm.pid);
1603 if (thread == NULL ||
1604 thread__set_comm(thread, event->comm.comm)) {
1605 dprintf("problem processing PERF_EVENT_COMM, skipping event.\n");
1614 process_task_event(event_t *event, unsigned long offset, unsigned long head)
1616 struct thread *thread = threads__findnew(event->fork.pid);
1617 struct thread *parent = threads__findnew(event->fork.ppid);
1619 dprintf("%p [%p]: PERF_EVENT_%s: (%d:%d):(%d:%d)\n",
1620 (void *)(offset + head),
1621 (void *)(long)(event->header.size),
1622 event->header.type == PERF_EVENT_FORK ? "FORK" : "EXIT",
1623 event->fork.pid, event->fork.tid,
1624 event->fork.ppid, event->fork.ptid);
1627 * A thread clone will have the same PID for both
1630 if (thread == parent)
1633 if (event->header.type == PERF_EVENT_EXIT)
1636 if (!thread || !parent || thread__fork(thread, parent)) {
1637 dprintf("problem processing PERF_EVENT_FORK, skipping event.\n");
1646 process_lost_event(event_t *event, unsigned long offset, unsigned long head)
1648 dprintf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n",
1649 (void *)(offset + head),
1650 (void *)(long)(event->header.size),
1654 total_lost += event->lost.lost;
1659 static void trace_event(event_t *event)
1661 unsigned char *raw_event = (void *)event;
1662 char *color = PERF_COLOR_BLUE;
1669 cdprintf("\n. ... raw event: size %d bytes\n", event->header.size);
1671 for (i = 0; i < event->header.size; i++) {
1672 if ((i & 15) == 0) {
1674 cdprintf(" %04x: ", i);
1677 cdprintf(" %02x", raw_event[i]);
1679 if (((i & 15) == 15) || i == event->header.size-1) {
1681 for (j = 0; j < 15-(i & 15); j++)
1683 for (j = 0; j < (i & 15); j++) {
1684 if (isprint(raw_event[i-15+j]))
1685 cdprintf("%c", raw_event[i-15+j]);
1695 static struct perf_header *header;
1697 static struct perf_counter_attr *perf_header__find_attr(u64 id)
1701 for (i = 0; i < header->attrs; i++) {
1702 struct perf_header_attr *attr = header->attr[i];
1705 for (j = 0; j < attr->ids; j++) {
1706 if (attr->id[j] == id)
1715 process_read_event(event_t *event, unsigned long offset, unsigned long head)
1717 struct perf_counter_attr *attr = perf_header__find_attr(event->read.id);
1719 dprintf("%p [%p]: PERF_EVENT_READ: %d %d %s %Lu\n",
1720 (void *)(offset + head),
1721 (void *)(long)(event->header.size),
1724 attr ? __event_name(attr->type, attr->config)
1732 process_event(event_t *event, unsigned long offset, unsigned long head)
1736 switch (event->header.type) {
1737 case PERF_EVENT_SAMPLE:
1738 return process_sample_event(event, offset, head);
1740 case PERF_EVENT_MMAP:
1741 return process_mmap_event(event, offset, head);
1743 case PERF_EVENT_COMM:
1744 return process_comm_event(event, offset, head);
1746 case PERF_EVENT_FORK:
1747 case PERF_EVENT_EXIT:
1748 return process_task_event(event, offset, head);
1750 case PERF_EVENT_LOST:
1751 return process_lost_event(event, offset, head);
1753 case PERF_EVENT_READ:
1754 return process_read_event(event, offset, head);
1757 * We dont process them right now but they are fine:
1760 case PERF_EVENT_THROTTLE:
1761 case PERF_EVENT_UNTHROTTLE:
1771 static u64 perf_header__sample_type(void)
1773 u64 sample_type = 0;
1776 for (i = 0; i < header->attrs; i++) {
1777 struct perf_header_attr *attr = header->attr[i];
1780 sample_type = attr->attr.sample_type;
1781 else if (sample_type != attr->attr.sample_type)
1782 die("non matching sample_type");
1788 static int __cmd_report(void)
1790 int ret, rc = EXIT_FAILURE;
1791 unsigned long offset = 0;
1792 unsigned long head, shift;
1798 register_idle_thread();
1800 input = open(input_name, O_RDONLY);
1802 fprintf(stderr, " failed to open file: %s", input_name);
1803 if (!strcmp(input_name, "perf.data"))
1804 fprintf(stderr, " (try 'perf record' first)");
1805 fprintf(stderr, "\n");
1809 ret = fstat(input, &stat);
1811 perror("failed to stat file");
1815 if (!stat.st_size) {
1816 fprintf(stderr, "zero-sized file, nothing to do!\n");
1820 header = perf_header__read(input);
1821 head = header->data_offset;
1823 sample_type = perf_header__sample_type();
1825 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
1826 if (sort__has_parent) {
1827 fprintf(stderr, "selected --sort parent, but no"
1828 " callchain data. Did you call"
1829 " perf record without -g?\n");
1833 fprintf(stderr, "selected -c but no callchain data."
1834 " Did you call perf record without"
1840 if (load_kernel() < 0) {
1841 perror("failed to load kernel symbols");
1842 return EXIT_FAILURE;
1846 if (getcwd(__cwd, sizeof(__cwd)) == NULL) {
1847 perror("failed to get the current directory");
1848 return EXIT_FAILURE;
1850 cwdlen = strlen(cwd);
1856 shift = page_size * (head / page_size);
1861 buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
1862 MAP_SHARED, input, offset);
1863 if (buf == MAP_FAILED) {
1864 perror("failed to mmap file");
1869 event = (event_t *)(buf + head);
1871 size = event->header.size;
1875 if (head + event->header.size >= page_size * mmap_window) {
1878 shift = page_size * (head / page_size);
1880 ret = munmap(buf, page_size * mmap_window);
1888 size = event->header.size;
1890 dprintf("\n%p [%p]: event: %d\n",
1891 (void *)(offset + head),
1892 (void *)(long)event->header.size,
1893 event->header.type);
1895 if (!size || process_event(event, offset, head) < 0) {
1897 dprintf("%p [%p]: skipping unknown header type: %d\n",
1898 (void *)(offset + head),
1899 (void *)(long)(event->header.size),
1900 event->header.type);
1905 * assume we lost track of the stream, check alignment, and
1906 * increment a single u64 in the hope to catch on again 'soon'.
1909 if (unlikely(head & 7))
1917 if (offset + head >= header->data_offset + header->data_size)
1920 if (offset + head < (unsigned long)stat.st_size)
1927 dprintf(" IP events: %10ld\n", total);
1928 dprintf(" mmap events: %10ld\n", total_mmap);
1929 dprintf(" comm events: %10ld\n", total_comm);
1930 dprintf(" fork events: %10ld\n", total_fork);
1931 dprintf(" lost events: %10ld\n", total_lost);
1932 dprintf(" unknown events: %10ld\n", total_unknown);
1938 threads__fprintf(stdout);
1941 dsos__fprintf(stdout);
1944 output__resort(total);
1945 output__fprintf(stdout, total);
1951 parse_callchain_opt(const struct option *opt __used, const char *arg,
1962 tok = strtok((char *)arg, ",");
1966 /* get the output mode */
1967 if (!strncmp(tok, "graph", strlen(arg)))
1968 callchain_param.mode = CHAIN_GRAPH_ABS;
1970 else if (!strncmp(tok, "flat", strlen(arg)))
1971 callchain_param.mode = CHAIN_FLAT;
1973 else if (!strncmp(tok, "fractal", strlen(arg)))
1974 callchain_param.mode = CHAIN_GRAPH_REL;
1979 /* get the min percentage */
1980 tok = strtok(NULL, ",");
1984 callchain_param.min_percent = strtod(tok, &endptr);
1989 if (register_callchain_param(&callchain_param) < 0) {
1990 fprintf(stderr, "Can't register callchain params\n");
1996 static const char * const report_usage[] = {
1997 "perf report [<options>] <command>",
2001 static const struct option options[] = {
2002 OPT_STRING('i', "input", &input_name, "file",
2004 OPT_BOOLEAN('v', "verbose", &verbose,
2005 "be more verbose (show symbol address, etc)"),
2006 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
2007 "dump raw trace in ASCII"),
2008 OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"),
2009 OPT_BOOLEAN('m', "modules", &modules,
2010 "load module symbols - WARNING: use only with -k and LIVE kernel"),
2011 OPT_BOOLEAN('n', "show-nr-samples", &show_nr_samples,
2012 "Show a column with the number of samples"),
2013 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
2014 "sort by key(s): pid, comm, dso, symbol, parent"),
2015 OPT_BOOLEAN('P', "full-paths", &full_paths,
2016 "Don't shorten the pathnames taking into account the cwd"),
2017 OPT_STRING('p', "parent", &parent_pattern, "regex",
2018 "regex filter to identify parent, see: '--sort parent'"),
2019 OPT_BOOLEAN('x', "exclude-other", &exclude_other,
2020 "Only display entries with parent-match"),
2021 OPT_CALLBACK_DEFAULT('g', "call-graph", NULL, "output_type,min_percent",
2022 "Display callchains using output_type and min percent threshold. "
2023 "Default: fractal,0.5", &parse_callchain_opt, callchain_default_opt),
2024 OPT_STRING('d', "dsos", &dso_list_str, "dso[,dso...]",
2025 "only consider symbols in these dsos"),
2026 OPT_STRING('C', "comms", &comm_list_str, "comm[,comm...]",
2027 "only consider symbols in these comms"),
2028 OPT_STRING('S', "symbols", &sym_list_str, "symbol[,symbol...]",
2029 "only consider these symbols"),
2030 OPT_STRING('w', "column-widths", &col_width_list_str,
2032 "don't try to adjust column width, use these fixed values"),
2033 OPT_STRING('t', "field-separator", &field_sep, "separator",
2034 "separator for columns, no spaces will be added between "
2035 "columns '.' is reserved."),
2039 static void setup_sorting(void)
2041 char *tmp, *tok, *str = strdup(sort_order);
2043 for (tok = strtok_r(str, ", ", &tmp);
2044 tok; tok = strtok_r(NULL, ", ", &tmp)) {
2045 if (sort_dimension__add(tok) < 0) {
2046 error("Unknown --sort key: `%s'", tok);
2047 usage_with_options(report_usage, options);
2054 static void setup_list(struct strlist **list, const char *list_str,
2055 struct sort_entry *se, const char *list_name,
2059 *list = strlist__new(true, list_str);
2061 fprintf(stderr, "problems parsing %s list\n",
2065 if (strlist__nr_entries(*list) == 1) {
2066 fprintf(fp, "# %s: %s\n", list_name,
2067 strlist__entry(*list, 0)->s);
2073 int cmd_report(int argc, const char **argv, const char *prefix __used)
2077 page_size = getpagesize();
2079 argc = parse_options(argc, argv, options, report_usage, 0);
2083 if (parent_pattern != default_parent_pattern) {
2084 sort_dimension__add("parent");
2085 sort_parent.elide = 1;
2090 * Any (unrecognized) arguments left?
2093 usage_with_options(report_usage, options);
2097 setup_list(&dso_list, dso_list_str, &sort_dso, "dso", stdout);
2098 setup_list(&comm_list, comm_list_str, &sort_comm, "comm", stdout);
2099 setup_list(&sym_list, sym_list_str, &sort_sym, "symbol", stdout);
2101 if (field_sep && *field_sep == '.') {
2102 fputs("'.' is the only non valid --field-separator argument\n",
2107 return __cmd_report();