x86/amd-iommu: Add per IOMMU reference counting
[linux/fpc-iii.git] / tools / perf / builtin-report.c
blob19669c20088e12a256ebdd736efd0c8253fdd73e
1 /*
2 * builtin-report.c
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.
7 */
8 #include "builtin.h"
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"
22 #include "perf.h"
23 #include "util/debug.h"
24 #include "util/header.h"
26 #include "util/parse-options.h"
27 #include "util/parse-events.h"
29 #include "util/thread.h"
31 static char const *input_name = "perf.data";
33 static char default_sort_order[] = "comm,dso,symbol";
34 static char *sort_order = default_sort_order;
35 static char *dso_list_str, *comm_list_str, *sym_list_str,
36 *col_width_list_str;
37 static struct strlist *dso_list, *comm_list, *sym_list;
38 static char *field_sep;
40 static int force;
41 static int input;
42 static int show_mask = SHOW_KERNEL | SHOW_USER | SHOW_HV;
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";
64 static int callchain;
66 static char __cwd[PATH_MAX];
67 static char *cwd = __cwd;
68 static int cwdlen;
70 static struct rb_root threads;
71 static struct thread *last_match;
73 static struct perf_header *header;
75 static
76 struct callchain_param callchain_param = {
77 .mode = CHAIN_GRAPH_REL,
78 .min_percent = 0.5
81 static u64 sample_type;
83 static int repsep_fprintf(FILE *fp, const char *fmt, ...)
85 int n;
86 va_list ap;
88 va_start(ap, fmt);
89 if (!field_sep)
90 n = vfprintf(fp, fmt, ap);
91 else {
92 char *bf = NULL;
93 n = vasprintf(&bf, fmt, ap);
94 if (n > 0) {
95 char *sep = bf;
97 while (1) {
98 sep = strchr(sep, *field_sep);
99 if (sep == NULL)
100 break;
101 *sep = '.';
104 fputs(bf, fp);
105 free(bf);
107 va_end(ap);
108 return n;
111 static unsigned int dsos__col_width,
112 comms__col_width,
113 threads__col_width;
116 * histogram, sorted on item, collects counts
119 static struct rb_root hist;
121 struct hist_entry {
122 struct rb_node rb_node;
124 struct thread *thread;
125 struct map *map;
126 struct dso *dso;
127 struct symbol *sym;
128 struct symbol *parent;
129 u64 ip;
130 char level;
131 struct callchain_node callchain;
132 struct rb_root sorted_chain;
134 u64 count;
138 * configurable sorting bits
141 struct sort_entry {
142 struct list_head list;
144 const char *header;
146 int64_t (*cmp)(struct hist_entry *, struct hist_entry *);
147 int64_t (*collapse)(struct hist_entry *, struct hist_entry *);
148 size_t (*print)(FILE *fp, struct hist_entry *, unsigned int width);
149 unsigned int *width;
150 bool elide;
153 static int64_t cmp_null(void *l, void *r)
155 if (!l && !r)
156 return 0;
157 else if (!l)
158 return -1;
159 else
160 return 1;
163 /* --sort pid */
165 static int64_t
166 sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
168 return right->thread->pid - left->thread->pid;
171 static size_t
172 sort__thread_print(FILE *fp, struct hist_entry *self, unsigned int width)
174 return repsep_fprintf(fp, "%*s:%5d", width - 6,
175 self->thread->comm ?: "", self->thread->pid);
178 static struct sort_entry sort_thread = {
179 .header = "Command: Pid",
180 .cmp = sort__thread_cmp,
181 .print = sort__thread_print,
182 .width = &threads__col_width,
185 /* --sort comm */
187 static int64_t
188 sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
190 return right->thread->pid - left->thread->pid;
193 static int64_t
194 sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
196 char *comm_l = left->thread->comm;
197 char *comm_r = right->thread->comm;
199 if (!comm_l || !comm_r)
200 return cmp_null(comm_l, comm_r);
202 return strcmp(comm_l, comm_r);
205 static size_t
206 sort__comm_print(FILE *fp, struct hist_entry *self, unsigned int width)
208 return repsep_fprintf(fp, "%*s", width, self->thread->comm);
211 static struct sort_entry sort_comm = {
212 .header = "Command",
213 .cmp = sort__comm_cmp,
214 .collapse = sort__comm_collapse,
215 .print = sort__comm_print,
216 .width = &comms__col_width,
219 /* --sort dso */
221 static int64_t
222 sort__dso_cmp(struct hist_entry *left, struct hist_entry *right)
224 struct dso *dso_l = left->dso;
225 struct dso *dso_r = right->dso;
227 if (!dso_l || !dso_r)
228 return cmp_null(dso_l, dso_r);
230 return strcmp(dso_l->name, dso_r->name);
233 static size_t
234 sort__dso_print(FILE *fp, struct hist_entry *self, unsigned int width)
236 if (self->dso)
237 return repsep_fprintf(fp, "%-*s", width, self->dso->name);
239 return repsep_fprintf(fp, "%*llx", width, (u64)self->ip);
242 static struct sort_entry sort_dso = {
243 .header = "Shared Object",
244 .cmp = sort__dso_cmp,
245 .print = sort__dso_print,
246 .width = &dsos__col_width,
249 /* --sort symbol */
251 static int64_t
252 sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
254 u64 ip_l, ip_r;
256 if (left->sym == right->sym)
257 return 0;
259 ip_l = left->sym ? left->sym->start : left->ip;
260 ip_r = right->sym ? right->sym->start : right->ip;
262 return (int64_t)(ip_r - ip_l);
265 static size_t
266 sort__sym_print(FILE *fp, struct hist_entry *self, unsigned int width __used)
268 size_t ret = 0;
270 if (verbose)
271 ret += repsep_fprintf(fp, "%#018llx %c ", (u64)self->ip,
272 dso__symtab_origin(self->dso));
274 ret += repsep_fprintf(fp, "[%c] ", self->level);
275 if (self->sym) {
276 ret += repsep_fprintf(fp, "%s", self->sym->name);
278 if (self->sym->module)
279 ret += repsep_fprintf(fp, "\t[%s]",
280 self->sym->module->name);
281 } else {
282 ret += repsep_fprintf(fp, "%#016llx", (u64)self->ip);
285 return ret;
288 static struct sort_entry sort_sym = {
289 .header = "Symbol",
290 .cmp = sort__sym_cmp,
291 .print = sort__sym_print,
294 /* --sort parent */
296 static int64_t
297 sort__parent_cmp(struct hist_entry *left, struct hist_entry *right)
299 struct symbol *sym_l = left->parent;
300 struct symbol *sym_r = right->parent;
302 if (!sym_l || !sym_r)
303 return cmp_null(sym_l, sym_r);
305 return strcmp(sym_l->name, sym_r->name);
308 static size_t
309 sort__parent_print(FILE *fp, struct hist_entry *self, unsigned int width)
311 return repsep_fprintf(fp, "%-*s", width,
312 self->parent ? self->parent->name : "[other]");
315 static unsigned int parent_symbol__col_width;
317 static struct sort_entry sort_parent = {
318 .header = "Parent symbol",
319 .cmp = sort__parent_cmp,
320 .print = sort__parent_print,
321 .width = &parent_symbol__col_width,
324 static int sort__need_collapse = 0;
325 static int sort__has_parent = 0;
327 struct sort_dimension {
328 const char *name;
329 struct sort_entry *entry;
330 int taken;
333 static struct sort_dimension sort_dimensions[] = {
334 { .name = "pid", .entry = &sort_thread, },
335 { .name = "comm", .entry = &sort_comm, },
336 { .name = "dso", .entry = &sort_dso, },
337 { .name = "symbol", .entry = &sort_sym, },
338 { .name = "parent", .entry = &sort_parent, },
341 static LIST_HEAD(hist_entry__sort_list);
343 static int sort_dimension__add(const char *tok)
345 unsigned int i;
347 for (i = 0; i < ARRAY_SIZE(sort_dimensions); i++) {
348 struct sort_dimension *sd = &sort_dimensions[i];
350 if (sd->taken)
351 continue;
353 if (strncasecmp(tok, sd->name, strlen(tok)))
354 continue;
356 if (sd->entry->collapse)
357 sort__need_collapse = 1;
359 if (sd->entry == &sort_parent) {
360 int ret = regcomp(&parent_regex, parent_pattern, REG_EXTENDED);
361 if (ret) {
362 char err[BUFSIZ];
364 regerror(ret, &parent_regex, err, sizeof(err));
365 fprintf(stderr, "Invalid regex: %s\n%s",
366 parent_pattern, err);
367 exit(-1);
369 sort__has_parent = 1;
372 list_add_tail(&sd->entry->list, &hist_entry__sort_list);
373 sd->taken = 1;
375 return 0;
378 return -ESRCH;
381 static int64_t
382 hist_entry__cmp(struct hist_entry *left, struct hist_entry *right)
384 struct sort_entry *se;
385 int64_t cmp = 0;
387 list_for_each_entry(se, &hist_entry__sort_list, list) {
388 cmp = se->cmp(left, right);
389 if (cmp)
390 break;
393 return cmp;
396 static int64_t
397 hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
399 struct sort_entry *se;
400 int64_t cmp = 0;
402 list_for_each_entry(se, &hist_entry__sort_list, list) {
403 int64_t (*f)(struct hist_entry *, struct hist_entry *);
405 f = se->collapse ?: se->cmp;
407 cmp = f(left, right);
408 if (cmp)
409 break;
412 return cmp;
415 static size_t ipchain__fprintf_graph_line(FILE *fp, int depth, int depth_mask)
417 int i;
418 size_t ret = 0;
420 ret += fprintf(fp, "%s", " ");
422 for (i = 0; i < depth; i++)
423 if (depth_mask & (1 << i))
424 ret += fprintf(fp, "| ");
425 else
426 ret += fprintf(fp, " ");
428 ret += fprintf(fp, "\n");
430 return ret;
432 static size_t
433 ipchain__fprintf_graph(FILE *fp, struct callchain_list *chain, int depth,
434 int depth_mask, int count, u64 total_samples,
435 int hits)
437 int i;
438 size_t ret = 0;
440 ret += fprintf(fp, "%s", " ");
441 for (i = 0; i < depth; i++) {
442 if (depth_mask & (1 << i))
443 ret += fprintf(fp, "|");
444 else
445 ret += fprintf(fp, " ");
446 if (!count && i == depth - 1) {
447 double percent;
449 percent = hits * 100.0 / total_samples;
450 ret += percent_color_fprintf(fp, "--%2.2f%%-- ", percent);
451 } else
452 ret += fprintf(fp, "%s", " ");
454 if (chain->sym)
455 ret += fprintf(fp, "%s\n", chain->sym->name);
456 else
457 ret += fprintf(fp, "%p\n", (void *)(long)chain->ip);
459 return ret;
462 static struct symbol *rem_sq_bracket;
463 static struct callchain_list rem_hits;
465 static void init_rem_hits(void)
467 rem_sq_bracket = malloc(sizeof(*rem_sq_bracket) + 6);
468 if (!rem_sq_bracket) {
469 fprintf(stderr, "Not enough memory to display remaining hits\n");
470 return;
473 strcpy(rem_sq_bracket->name, "[...]");
474 rem_hits.sym = rem_sq_bracket;
477 static size_t
478 callchain__fprintf_graph(FILE *fp, struct callchain_node *self,
479 u64 total_samples, int depth, int depth_mask)
481 struct rb_node *node, *next;
482 struct callchain_node *child;
483 struct callchain_list *chain;
484 int new_depth_mask = depth_mask;
485 u64 new_total;
486 u64 remaining;
487 size_t ret = 0;
488 int i;
490 if (callchain_param.mode == CHAIN_GRAPH_REL)
491 new_total = self->children_hit;
492 else
493 new_total = total_samples;
495 remaining = new_total;
497 node = rb_first(&self->rb_root);
498 while (node) {
499 u64 cumul;
501 child = rb_entry(node, struct callchain_node, rb_node);
502 cumul = cumul_hits(child);
503 remaining -= cumul;
506 * The depth mask manages the output of pipes that show
507 * the depth. We don't want to keep the pipes of the current
508 * level for the last child of this depth.
509 * Except if we have remaining filtered hits. They will
510 * supersede the last child
512 next = rb_next(node);
513 if (!next && (callchain_param.mode != CHAIN_GRAPH_REL || !remaining))
514 new_depth_mask &= ~(1 << (depth - 1));
517 * But we keep the older depth mask for the line seperator
518 * to keep the level link until we reach the last child
520 ret += ipchain__fprintf_graph_line(fp, depth, depth_mask);
521 i = 0;
522 list_for_each_entry(chain, &child->val, list) {
523 if (chain->ip >= PERF_CONTEXT_MAX)
524 continue;
525 ret += ipchain__fprintf_graph(fp, chain, depth,
526 new_depth_mask, i++,
527 new_total,
528 cumul);
530 ret += callchain__fprintf_graph(fp, child, new_total,
531 depth + 1,
532 new_depth_mask | (1 << depth));
533 node = next;
536 if (callchain_param.mode == CHAIN_GRAPH_REL &&
537 remaining && remaining != new_total) {
539 if (!rem_sq_bracket)
540 return ret;
542 new_depth_mask &= ~(1 << (depth - 1));
544 ret += ipchain__fprintf_graph(fp, &rem_hits, depth,
545 new_depth_mask, 0, new_total,
546 remaining);
549 return ret;
552 static size_t
553 callchain__fprintf_flat(FILE *fp, struct callchain_node *self,
554 u64 total_samples)
556 struct callchain_list *chain;
557 size_t ret = 0;
559 if (!self)
560 return 0;
562 ret += callchain__fprintf_flat(fp, self->parent, total_samples);
565 list_for_each_entry(chain, &self->val, list) {
566 if (chain->ip >= PERF_CONTEXT_MAX)
567 continue;
568 if (chain->sym)
569 ret += fprintf(fp, " %s\n", chain->sym->name);
570 else
571 ret += fprintf(fp, " %p\n",
572 (void *)(long)chain->ip);
575 return ret;
578 static size_t
579 hist_entry_callchain__fprintf(FILE *fp, struct hist_entry *self,
580 u64 total_samples)
582 struct rb_node *rb_node;
583 struct callchain_node *chain;
584 size_t ret = 0;
586 rb_node = rb_first(&self->sorted_chain);
587 while (rb_node) {
588 double percent;
590 chain = rb_entry(rb_node, struct callchain_node, rb_node);
591 percent = chain->hit * 100.0 / total_samples;
592 switch (callchain_param.mode) {
593 case CHAIN_FLAT:
594 ret += percent_color_fprintf(fp, " %6.2f%%\n",
595 percent);
596 ret += callchain__fprintf_flat(fp, chain, total_samples);
597 break;
598 case CHAIN_GRAPH_ABS: /* Falldown */
599 case CHAIN_GRAPH_REL:
600 ret += callchain__fprintf_graph(fp, chain,
601 total_samples, 1, 1);
602 case CHAIN_NONE:
603 default:
604 break;
606 ret += fprintf(fp, "\n");
607 rb_node = rb_next(rb_node);
610 return ret;
614 static size_t
615 hist_entry__fprintf(FILE *fp, struct hist_entry *self, u64 total_samples)
617 struct sort_entry *se;
618 size_t ret;
620 if (exclude_other && !self->parent)
621 return 0;
623 if (total_samples)
624 ret = percent_color_fprintf(fp,
625 field_sep ? "%.2f" : " %6.2f%%",
626 (self->count * 100.0) / total_samples);
627 else
628 ret = fprintf(fp, field_sep ? "%lld" : "%12lld ", self->count);
630 if (show_nr_samples) {
631 if (field_sep)
632 fprintf(fp, "%c%lld", *field_sep, self->count);
633 else
634 fprintf(fp, "%11lld", self->count);
637 list_for_each_entry(se, &hist_entry__sort_list, list) {
638 if (se->elide)
639 continue;
641 fprintf(fp, "%s", field_sep ?: " ");
642 ret += se->print(fp, self, se->width ? *se->width : 0);
645 ret += fprintf(fp, "\n");
647 if (callchain)
648 hist_entry_callchain__fprintf(fp, self, total_samples);
650 return ret;
657 static void dso__calc_col_width(struct dso *self)
659 if (!col_width_list_str && !field_sep &&
660 (!dso_list || strlist__has_entry(dso_list, self->name))) {
661 unsigned int slen = strlen(self->name);
662 if (slen > dsos__col_width)
663 dsos__col_width = slen;
666 self->slen_calculated = 1;
669 static void thread__comm_adjust(struct thread *self)
671 char *comm = self->comm;
673 if (!col_width_list_str && !field_sep &&
674 (!comm_list || strlist__has_entry(comm_list, comm))) {
675 unsigned int slen = strlen(comm);
677 if (slen > comms__col_width) {
678 comms__col_width = slen;
679 threads__col_width = slen + 6;
684 static int thread__set_comm_adjust(struct thread *self, const char *comm)
686 int ret = thread__set_comm(self, comm);
688 if (ret)
689 return ret;
691 thread__comm_adjust(self);
693 return 0;
697 static struct symbol *
698 resolve_symbol(struct thread *thread, struct map **mapp,
699 struct dso **dsop, u64 *ipp)
701 struct dso *dso = dsop ? *dsop : NULL;
702 struct map *map = mapp ? *mapp : NULL;
703 u64 ip = *ipp;
705 if (!thread)
706 return NULL;
708 if (dso)
709 goto got_dso;
711 if (map)
712 goto got_map;
714 map = thread__find_map(thread, ip);
715 if (map != NULL) {
717 * We have to do this here as we may have a dso
718 * with no symbol hit that has a name longer than
719 * the ones with symbols sampled.
721 if (!sort_dso.elide && !map->dso->slen_calculated)
722 dso__calc_col_width(map->dso);
724 if (mapp)
725 *mapp = map;
726 got_map:
727 ip = map->map_ip(map, ip);
729 dso = map->dso;
730 } else {
732 * If this is outside of all known maps,
733 * and is a negative address, try to look it
734 * up in the kernel dso, as it might be a
735 * vsyscall (which executes in user-mode):
737 if ((long long)ip < 0)
738 dso = kernel_dso;
740 dump_printf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
741 dump_printf(" ...... map: %Lx -> %Lx\n", *ipp, ip);
742 *ipp = ip;
744 if (dsop)
745 *dsop = dso;
747 if (!dso)
748 return NULL;
749 got_dso:
750 return dso->find_symbol(dso, ip);
753 static int call__match(struct symbol *sym)
755 if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
756 return 1;
758 return 0;
761 static struct symbol **
762 resolve_callchain(struct thread *thread, struct map *map __used,
763 struct ip_callchain *chain, struct hist_entry *entry)
765 u64 context = PERF_CONTEXT_MAX;
766 struct symbol **syms = NULL;
767 unsigned int i;
769 if (callchain) {
770 syms = calloc(chain->nr, sizeof(*syms));
771 if (!syms) {
772 fprintf(stderr, "Can't allocate memory for symbols\n");
773 exit(-1);
777 for (i = 0; i < chain->nr; i++) {
778 u64 ip = chain->ips[i];
779 struct dso *dso = NULL;
780 struct symbol *sym;
782 if (ip >= PERF_CONTEXT_MAX) {
783 context = ip;
784 continue;
787 switch (context) {
788 case PERF_CONTEXT_HV:
789 dso = hypervisor_dso;
790 break;
791 case PERF_CONTEXT_KERNEL:
792 dso = kernel_dso;
793 break;
794 default:
795 break;
798 sym = resolve_symbol(thread, NULL, &dso, &ip);
800 if (sym) {
801 if (sort__has_parent && call__match(sym) &&
802 !entry->parent)
803 entry->parent = sym;
804 if (!callchain)
805 break;
806 syms[i] = sym;
810 return syms;
814 * collect histogram counts
817 static int
818 hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
819 struct symbol *sym, u64 ip, struct ip_callchain *chain,
820 char level, u64 count)
822 struct rb_node **p = &hist.rb_node;
823 struct rb_node *parent = NULL;
824 struct hist_entry *he;
825 struct symbol **syms = NULL;
826 struct hist_entry entry = {
827 .thread = thread,
828 .map = map,
829 .dso = dso,
830 .sym = sym,
831 .ip = ip,
832 .level = level,
833 .count = count,
834 .parent = NULL,
835 .sorted_chain = RB_ROOT
837 int cmp;
839 if ((sort__has_parent || callchain) && chain)
840 syms = resolve_callchain(thread, map, chain, &entry);
842 while (*p != NULL) {
843 parent = *p;
844 he = rb_entry(parent, struct hist_entry, rb_node);
846 cmp = hist_entry__cmp(&entry, he);
848 if (!cmp) {
849 he->count += count;
850 if (callchain) {
851 append_chain(&he->callchain, chain, syms);
852 free(syms);
854 return 0;
857 if (cmp < 0)
858 p = &(*p)->rb_left;
859 else
860 p = &(*p)->rb_right;
863 he = malloc(sizeof(*he));
864 if (!he)
865 return -ENOMEM;
866 *he = entry;
867 if (callchain) {
868 callchain_init(&he->callchain);
869 append_chain(&he->callchain, chain, syms);
870 free(syms);
872 rb_link_node(&he->rb_node, parent, p);
873 rb_insert_color(&he->rb_node, &hist);
875 return 0;
878 static void hist_entry__free(struct hist_entry *he)
880 free(he);
884 * collapse the histogram
887 static struct rb_root collapse_hists;
889 static void collapse__insert_entry(struct hist_entry *he)
891 struct rb_node **p = &collapse_hists.rb_node;
892 struct rb_node *parent = NULL;
893 struct hist_entry *iter;
894 int64_t cmp;
896 while (*p != NULL) {
897 parent = *p;
898 iter = rb_entry(parent, struct hist_entry, rb_node);
900 cmp = hist_entry__collapse(iter, he);
902 if (!cmp) {
903 iter->count += he->count;
904 hist_entry__free(he);
905 return;
908 if (cmp < 0)
909 p = &(*p)->rb_left;
910 else
911 p = &(*p)->rb_right;
914 rb_link_node(&he->rb_node, parent, p);
915 rb_insert_color(&he->rb_node, &collapse_hists);
918 static void collapse__resort(void)
920 struct rb_node *next;
921 struct hist_entry *n;
923 if (!sort__need_collapse)
924 return;
926 next = rb_first(&hist);
927 while (next) {
928 n = rb_entry(next, struct hist_entry, rb_node);
929 next = rb_next(&n->rb_node);
931 rb_erase(&n->rb_node, &hist);
932 collapse__insert_entry(n);
937 * reverse the map, sort on count.
940 static struct rb_root output_hists;
942 static void output__insert_entry(struct hist_entry *he, u64 min_callchain_hits)
944 struct rb_node **p = &output_hists.rb_node;
945 struct rb_node *parent = NULL;
946 struct hist_entry *iter;
948 if (callchain)
949 callchain_param.sort(&he->sorted_chain, &he->callchain,
950 min_callchain_hits, &callchain_param);
952 while (*p != NULL) {
953 parent = *p;
954 iter = rb_entry(parent, struct hist_entry, rb_node);
956 if (he->count > iter->count)
957 p = &(*p)->rb_left;
958 else
959 p = &(*p)->rb_right;
962 rb_link_node(&he->rb_node, parent, p);
963 rb_insert_color(&he->rb_node, &output_hists);
966 static void output__resort(u64 total_samples)
968 struct rb_node *next;
969 struct hist_entry *n;
970 struct rb_root *tree = &hist;
971 u64 min_callchain_hits;
973 min_callchain_hits = total_samples * (callchain_param.min_percent / 100);
975 if (sort__need_collapse)
976 tree = &collapse_hists;
978 next = rb_first(tree);
980 while (next) {
981 n = rb_entry(next, struct hist_entry, rb_node);
982 next = rb_next(&n->rb_node);
984 rb_erase(&n->rb_node, tree);
985 output__insert_entry(n, min_callchain_hits);
989 static size_t output__fprintf(FILE *fp, u64 total_samples)
991 struct hist_entry *pos;
992 struct sort_entry *se;
993 struct rb_node *nd;
994 size_t ret = 0;
995 unsigned int width;
996 char *col_width = col_width_list_str;
997 int raw_printing_style;
999 raw_printing_style = !strcmp(pretty_printing_style, "raw");
1001 init_rem_hits();
1003 fprintf(fp, "# Samples: %Ld\n", (u64)total_samples);
1004 fprintf(fp, "#\n");
1006 fprintf(fp, "# Overhead");
1007 if (show_nr_samples) {
1008 if (field_sep)
1009 fprintf(fp, "%cSamples", *field_sep);
1010 else
1011 fputs(" Samples ", fp);
1013 list_for_each_entry(se, &hist_entry__sort_list, list) {
1014 if (se->elide)
1015 continue;
1016 if (field_sep) {
1017 fprintf(fp, "%c%s", *field_sep, se->header);
1018 continue;
1020 width = strlen(se->header);
1021 if (se->width) {
1022 if (col_width_list_str) {
1023 if (col_width) {
1024 *se->width = atoi(col_width);
1025 col_width = strchr(col_width, ',');
1026 if (col_width)
1027 ++col_width;
1030 width = *se->width = max(*se->width, width);
1032 fprintf(fp, " %*s", width, se->header);
1034 fprintf(fp, "\n");
1036 if (field_sep)
1037 goto print_entries;
1039 fprintf(fp, "# ........");
1040 if (show_nr_samples)
1041 fprintf(fp, " ..........");
1042 list_for_each_entry(se, &hist_entry__sort_list, list) {
1043 unsigned int i;
1045 if (se->elide)
1046 continue;
1048 fprintf(fp, " ");
1049 if (se->width)
1050 width = *se->width;
1051 else
1052 width = strlen(se->header);
1053 for (i = 0; i < width; i++)
1054 fprintf(fp, ".");
1056 fprintf(fp, "\n");
1058 fprintf(fp, "#\n");
1060 print_entries:
1061 for (nd = rb_first(&output_hists); nd; nd = rb_next(nd)) {
1062 pos = rb_entry(nd, struct hist_entry, rb_node);
1063 ret += hist_entry__fprintf(fp, pos, total_samples);
1066 if (sort_order == default_sort_order &&
1067 parent_pattern == default_parent_pattern) {
1068 fprintf(fp, "#\n");
1069 fprintf(fp, "# (For a higher level overview, try: perf report --sort comm,dso)\n");
1070 fprintf(fp, "#\n");
1072 fprintf(fp, "\n");
1074 free(rem_sq_bracket);
1076 if (show_threads)
1077 perf_read_values_display(fp, &show_threads_values,
1078 raw_printing_style);
1080 return ret;
1083 static unsigned long total = 0,
1084 total_mmap = 0,
1085 total_comm = 0,
1086 total_fork = 0,
1087 total_unknown = 0,
1088 total_lost = 0;
1090 static int validate_chain(struct ip_callchain *chain, event_t *event)
1092 unsigned int chain_size;
1094 chain_size = event->header.size;
1095 chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;
1097 if (chain->nr*sizeof(u64) > chain_size)
1098 return -1;
1100 return 0;
1103 static int
1104 process_sample_event(event_t *event, unsigned long offset, unsigned long head)
1106 char level;
1107 int show = 0;
1108 struct dso *dso = NULL;
1109 struct thread *thread;
1110 u64 ip = event->ip.ip;
1111 u64 period = 1;
1112 struct map *map = NULL;
1113 void *more_data = event->ip.__more_data;
1114 struct ip_callchain *chain = NULL;
1115 int cpumode;
1117 thread = threads__findnew(event->ip.pid, &threads, &last_match);
1119 if (sample_type & PERF_SAMPLE_PERIOD) {
1120 period = *(u64 *)more_data;
1121 more_data += sizeof(u64);
1124 dump_printf("%p [%p]: PERF_RECORD_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n",
1125 (void *)(offset + head),
1126 (void *)(long)(event->header.size),
1127 event->header.misc,
1128 event->ip.pid, event->ip.tid,
1129 (void *)(long)ip,
1130 (long long)period);
1132 if (sample_type & PERF_SAMPLE_CALLCHAIN) {
1133 unsigned int i;
1135 chain = (void *)more_data;
1137 dump_printf("... chain: nr:%Lu\n", chain->nr);
1139 if (validate_chain(chain, event) < 0) {
1140 eprintf("call-chain problem with event, skipping it.\n");
1141 return 0;
1144 if (dump_trace) {
1145 for (i = 0; i < chain->nr; i++)
1146 dump_printf("..... %2d: %016Lx\n", i, chain->ips[i]);
1150 dump_printf(" ... thread: %s:%d\n", thread->comm, thread->pid);
1152 if (thread == NULL) {
1153 eprintf("problem processing %d event, skipping it.\n",
1154 event->header.type);
1155 return -1;
1158 if (comm_list && !strlist__has_entry(comm_list, thread->comm))
1159 return 0;
1161 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
1163 if (cpumode == PERF_RECORD_MISC_KERNEL) {
1164 show = SHOW_KERNEL;
1165 level = 'k';
1167 dso = kernel_dso;
1169 dump_printf(" ...... dso: %s\n", dso->name);
1171 } else if (cpumode == PERF_RECORD_MISC_USER) {
1173 show = SHOW_USER;
1174 level = '.';
1176 } else {
1177 show = SHOW_HV;
1178 level = 'H';
1180 dso = hypervisor_dso;
1182 dump_printf(" ...... dso: [hypervisor]\n");
1185 if (show & show_mask) {
1186 struct symbol *sym = resolve_symbol(thread, &map, &dso, &ip);
1188 if (dso_list && (!dso || !dso->name ||
1189 !strlist__has_entry(dso_list, dso->name)))
1190 return 0;
1192 if (sym_list && (!sym || !strlist__has_entry(sym_list, sym->name)))
1193 return 0;
1195 if (hist_entry__add(thread, map, dso, sym, ip, chain, level, period)) {
1196 eprintf("problem incrementing symbol count, skipping event\n");
1197 return -1;
1200 total += period;
1202 return 0;
1205 static int
1206 process_mmap_event(event_t *event, unsigned long offset, unsigned long head)
1208 struct thread *thread;
1209 struct map *map = map__new(&event->mmap, cwd, cwdlen);
1211 thread = threads__findnew(event->mmap.pid, &threads, &last_match);
1213 dump_printf("%p [%p]: PERF_RECORD_MMAP %d/%d: [%p(%p) @ %p]: %s\n",
1214 (void *)(offset + head),
1215 (void *)(long)(event->header.size),
1216 event->mmap.pid,
1217 event->mmap.tid,
1218 (void *)(long)event->mmap.start,
1219 (void *)(long)event->mmap.len,
1220 (void *)(long)event->mmap.pgoff,
1221 event->mmap.filename);
1223 if (thread == NULL || map == NULL) {
1224 dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1225 return 0;
1228 thread__insert_map(thread, map);
1229 total_mmap++;
1231 return 0;
1234 static int
1235 process_comm_event(event_t *event, unsigned long offset, unsigned long head)
1237 struct thread *thread;
1239 thread = threads__findnew(event->comm.pid, &threads, &last_match);
1241 dump_printf("%p [%p]: PERF_RECORD_COMM: %s:%d\n",
1242 (void *)(offset + head),
1243 (void *)(long)(event->header.size),
1244 event->comm.comm, event->comm.pid);
1246 if (thread == NULL ||
1247 thread__set_comm_adjust(thread, event->comm.comm)) {
1248 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
1249 return -1;
1251 total_comm++;
1253 return 0;
1256 static int
1257 process_task_event(event_t *event, unsigned long offset, unsigned long head)
1259 struct thread *thread;
1260 struct thread *parent;
1262 thread = threads__findnew(event->fork.pid, &threads, &last_match);
1263 parent = threads__findnew(event->fork.ppid, &threads, &last_match);
1265 dump_printf("%p [%p]: PERF_RECORD_%s: (%d:%d):(%d:%d)\n",
1266 (void *)(offset + head),
1267 (void *)(long)(event->header.size),
1268 event->header.type == PERF_RECORD_FORK ? "FORK" : "EXIT",
1269 event->fork.pid, event->fork.tid,
1270 event->fork.ppid, event->fork.ptid);
1273 * A thread clone will have the same PID for both
1274 * parent and child.
1276 if (thread == parent)
1277 return 0;
1279 if (event->header.type == PERF_RECORD_EXIT)
1280 return 0;
1282 if (!thread || !parent || thread__fork(thread, parent)) {
1283 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1284 return -1;
1286 total_fork++;
1288 return 0;
1291 static int
1292 process_lost_event(event_t *event, unsigned long offset, unsigned long head)
1294 dump_printf("%p [%p]: PERF_RECORD_LOST: id:%Ld: lost:%Ld\n",
1295 (void *)(offset + head),
1296 (void *)(long)(event->header.size),
1297 event->lost.id,
1298 event->lost.lost);
1300 total_lost += event->lost.lost;
1302 return 0;
1305 static int
1306 process_read_event(event_t *event, unsigned long offset, unsigned long head)
1308 struct perf_event_attr *attr;
1310 attr = perf_header__find_attr(event->read.id, header);
1312 if (show_threads) {
1313 const char *name = attr ? __event_name(attr->type, attr->config)
1314 : "unknown";
1315 perf_read_values_add_value(&show_threads_values,
1316 event->read.pid, event->read.tid,
1317 event->read.id,
1318 name,
1319 event->read.value);
1322 dump_printf("%p [%p]: PERF_RECORD_READ: %d %d %s %Lu\n",
1323 (void *)(offset + head),
1324 (void *)(long)(event->header.size),
1325 event->read.pid,
1326 event->read.tid,
1327 attr ? __event_name(attr->type, attr->config)
1328 : "FAIL",
1329 event->read.value);
1331 return 0;
1334 static int
1335 process_event(event_t *event, unsigned long offset, unsigned long head)
1337 trace_event(event);
1339 switch (event->header.type) {
1340 case PERF_RECORD_SAMPLE:
1341 return process_sample_event(event, offset, head);
1343 case PERF_RECORD_MMAP:
1344 return process_mmap_event(event, offset, head);
1346 case PERF_RECORD_COMM:
1347 return process_comm_event(event, offset, head);
1349 case PERF_RECORD_FORK:
1350 case PERF_RECORD_EXIT:
1351 return process_task_event(event, offset, head);
1353 case PERF_RECORD_LOST:
1354 return process_lost_event(event, offset, head);
1356 case PERF_RECORD_READ:
1357 return process_read_event(event, offset, head);
1360 * We dont process them right now but they are fine:
1363 case PERF_RECORD_THROTTLE:
1364 case PERF_RECORD_UNTHROTTLE:
1365 return 0;
1367 default:
1368 return -1;
1371 return 0;
1374 static int __cmd_report(void)
1376 int ret, rc = EXIT_FAILURE;
1377 unsigned long offset = 0;
1378 unsigned long head, shift;
1379 struct stat input_stat;
1380 struct thread *idle;
1381 event_t *event;
1382 uint32_t size;
1383 char *buf;
1385 idle = register_idle_thread(&threads, &last_match);
1386 thread__comm_adjust(idle);
1388 if (show_threads)
1389 perf_read_values_init(&show_threads_values);
1391 input = open(input_name, O_RDONLY);
1392 if (input < 0) {
1393 fprintf(stderr, " failed to open file: %s", input_name);
1394 if (!strcmp(input_name, "perf.data"))
1395 fprintf(stderr, " (try 'perf record' first)");
1396 fprintf(stderr, "\n");
1397 exit(-1);
1400 ret = fstat(input, &input_stat);
1401 if (ret < 0) {
1402 perror("failed to stat file");
1403 exit(-1);
1406 if (!force && input_stat.st_uid && (input_stat.st_uid != geteuid())) {
1407 fprintf(stderr, "file: %s not owned by current user or root\n", input_name);
1408 exit(-1);
1411 if (!input_stat.st_size) {
1412 fprintf(stderr, "zero-sized file, nothing to do!\n");
1413 exit(0);
1416 header = perf_header__read(input);
1417 head = header->data_offset;
1419 sample_type = perf_header__sample_type(header);
1421 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
1422 if (sort__has_parent) {
1423 fprintf(stderr, "selected --sort parent, but no"
1424 " callchain data. Did you call"
1425 " perf record without -g?\n");
1426 exit(-1);
1428 if (callchain) {
1429 fprintf(stderr, "selected -g but no callchain data."
1430 " Did you call perf record without"
1431 " -g?\n");
1432 exit(-1);
1434 } else if (callchain_param.mode != CHAIN_NONE && !callchain) {
1435 callchain = 1;
1436 if (register_callchain_param(&callchain_param) < 0) {
1437 fprintf(stderr, "Can't register callchain"
1438 " params\n");
1439 exit(-1);
1443 if (load_kernel() < 0) {
1444 perror("failed to load kernel symbols");
1445 return EXIT_FAILURE;
1448 if (!full_paths) {
1449 if (getcwd(__cwd, sizeof(__cwd)) == NULL) {
1450 perror("failed to get the current directory");
1451 return EXIT_FAILURE;
1453 cwdlen = strlen(cwd);
1454 } else {
1455 cwd = NULL;
1456 cwdlen = 0;
1459 shift = page_size * (head / page_size);
1460 offset += shift;
1461 head -= shift;
1463 remap:
1464 buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
1465 MAP_SHARED, input, offset);
1466 if (buf == MAP_FAILED) {
1467 perror("failed to mmap file");
1468 exit(-1);
1471 more:
1472 event = (event_t *)(buf + head);
1474 size = event->header.size;
1475 if (!size)
1476 size = 8;
1478 if (head + event->header.size >= page_size * mmap_window) {
1479 int munmap_ret;
1481 shift = page_size * (head / page_size);
1483 munmap_ret = munmap(buf, page_size * mmap_window);
1484 assert(munmap_ret == 0);
1486 offset += shift;
1487 head -= shift;
1488 goto remap;
1491 size = event->header.size;
1493 dump_printf("\n%p [%p]: event: %d\n",
1494 (void *)(offset + head),
1495 (void *)(long)event->header.size,
1496 event->header.type);
1498 if (!size || process_event(event, offset, head) < 0) {
1500 dump_printf("%p [%p]: skipping unknown header type: %d\n",
1501 (void *)(offset + head),
1502 (void *)(long)(event->header.size),
1503 event->header.type);
1505 total_unknown++;
1508 * assume we lost track of the stream, check alignment, and
1509 * increment a single u64 in the hope to catch on again 'soon'.
1512 if (unlikely(head & 7))
1513 head &= ~7ULL;
1515 size = 8;
1518 head += size;
1520 if (offset + head >= header->data_offset + header->data_size)
1521 goto done;
1523 if (offset + head < (unsigned long)input_stat.st_size)
1524 goto more;
1526 done:
1527 rc = EXIT_SUCCESS;
1528 close(input);
1530 dump_printf(" IP events: %10ld\n", total);
1531 dump_printf(" mmap events: %10ld\n", total_mmap);
1532 dump_printf(" comm events: %10ld\n", total_comm);
1533 dump_printf(" fork events: %10ld\n", total_fork);
1534 dump_printf(" lost events: %10ld\n", total_lost);
1535 dump_printf(" unknown events: %10ld\n", total_unknown);
1537 if (dump_trace)
1538 return 0;
1540 if (verbose >= 3)
1541 threads__fprintf(stdout, &threads);
1543 if (verbose >= 2)
1544 dsos__fprintf(stdout);
1546 collapse__resort();
1547 output__resort(total);
1548 output__fprintf(stdout, total);
1550 if (show_threads)
1551 perf_read_values_destroy(&show_threads_values);
1553 return rc;
1556 static int
1557 parse_callchain_opt(const struct option *opt __used, const char *arg,
1558 int unset __used)
1560 char *tok;
1561 char *endptr;
1563 callchain = 1;
1565 if (!arg)
1566 return 0;
1568 tok = strtok((char *)arg, ",");
1569 if (!tok)
1570 return -1;
1572 /* get the output mode */
1573 if (!strncmp(tok, "graph", strlen(arg)))
1574 callchain_param.mode = CHAIN_GRAPH_ABS;
1576 else if (!strncmp(tok, "flat", strlen(arg)))
1577 callchain_param.mode = CHAIN_FLAT;
1579 else if (!strncmp(tok, "fractal", strlen(arg)))
1580 callchain_param.mode = CHAIN_GRAPH_REL;
1582 else if (!strncmp(tok, "none", strlen(arg))) {
1583 callchain_param.mode = CHAIN_NONE;
1584 callchain = 0;
1586 return 0;
1589 else
1590 return -1;
1592 /* get the min percentage */
1593 tok = strtok(NULL, ",");
1594 if (!tok)
1595 goto setup;
1597 callchain_param.min_percent = strtod(tok, &endptr);
1598 if (tok == endptr)
1599 return -1;
1601 setup:
1602 if (register_callchain_param(&callchain_param) < 0) {
1603 fprintf(stderr, "Can't register callchain params\n");
1604 return -1;
1606 return 0;
1609 static const char * const report_usage[] = {
1610 "perf report [<options>] <command>",
1611 NULL
1614 static const struct option options[] = {
1615 OPT_STRING('i', "input", &input_name, "file",
1616 "input file name"),
1617 OPT_BOOLEAN('v', "verbose", &verbose,
1618 "be more verbose (show symbol address, etc)"),
1619 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1620 "dump raw trace in ASCII"),
1621 OPT_STRING('k', "vmlinux", &vmlinux_name, "file", "vmlinux pathname"),
1622 OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
1623 OPT_BOOLEAN('m', "modules", &modules,
1624 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1625 OPT_BOOLEAN('n', "show-nr-samples", &show_nr_samples,
1626 "Show a column with the number of samples"),
1627 OPT_BOOLEAN('T', "threads", &show_threads,
1628 "Show per-thread event counters"),
1629 OPT_STRING(0, "pretty", &pretty_printing_style, "key",
1630 "pretty printing style key: normal raw"),
1631 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1632 "sort by key(s): pid, comm, dso, symbol, parent"),
1633 OPT_BOOLEAN('P', "full-paths", &full_paths,
1634 "Don't shorten the pathnames taking into account the cwd"),
1635 OPT_STRING('p', "parent", &parent_pattern, "regex",
1636 "regex filter to identify parent, see: '--sort parent'"),
1637 OPT_BOOLEAN('x', "exclude-other", &exclude_other,
1638 "Only display entries with parent-match"),
1639 OPT_CALLBACK_DEFAULT('g', "call-graph", NULL, "output_type,min_percent",
1640 "Display callchains using output_type and min percent threshold. "
1641 "Default: fractal,0.5", &parse_callchain_opt, callchain_default_opt),
1642 OPT_STRING('d', "dsos", &dso_list_str, "dso[,dso...]",
1643 "only consider symbols in these dsos"),
1644 OPT_STRING('C', "comms", &comm_list_str, "comm[,comm...]",
1645 "only consider symbols in these comms"),
1646 OPT_STRING('S', "symbols", &sym_list_str, "symbol[,symbol...]",
1647 "only consider these symbols"),
1648 OPT_STRING('w', "column-widths", &col_width_list_str,
1649 "width[,width...]",
1650 "don't try to adjust column width, use these fixed values"),
1651 OPT_STRING('t', "field-separator", &field_sep, "separator",
1652 "separator for columns, no spaces will be added between "
1653 "columns '.' is reserved."),
1654 OPT_END()
1657 static void setup_sorting(void)
1659 char *tmp, *tok, *str = strdup(sort_order);
1661 for (tok = strtok_r(str, ", ", &tmp);
1662 tok; tok = strtok_r(NULL, ", ", &tmp)) {
1663 if (sort_dimension__add(tok) < 0) {
1664 error("Unknown --sort key: `%s'", tok);
1665 usage_with_options(report_usage, options);
1669 free(str);
1672 static void setup_list(struct strlist **list, const char *list_str,
1673 struct sort_entry *se, const char *list_name,
1674 FILE *fp)
1676 if (list_str) {
1677 *list = strlist__new(true, list_str);
1678 if (!*list) {
1679 fprintf(stderr, "problems parsing %s list\n",
1680 list_name);
1681 exit(129);
1683 if (strlist__nr_entries(*list) == 1) {
1684 fprintf(fp, "# %s: %s\n", list_name,
1685 strlist__entry(*list, 0)->s);
1686 se->elide = true;
1691 int cmd_report(int argc, const char **argv, const char *prefix __used)
1693 symbol__init();
1695 page_size = getpagesize();
1697 argc = parse_options(argc, argv, options, report_usage, 0);
1699 setup_sorting();
1701 if (parent_pattern != default_parent_pattern) {
1702 sort_dimension__add("parent");
1703 sort_parent.elide = 1;
1704 } else
1705 exclude_other = 0;
1708 * Any (unrecognized) arguments left?
1710 if (argc)
1711 usage_with_options(report_usage, options);
1713 setup_pager();
1715 setup_list(&dso_list, dso_list_str, &sort_dso, "dso", stdout);
1716 setup_list(&comm_list, comm_list_str, &sort_comm, "comm", stdout);
1717 setup_list(&sym_list, sym_list_str, &sort_sym, "symbol", stdout);
1719 if (field_sep && *field_sep == '.') {
1720 fputs("'.' is the only non valid --field-separator argument\n",
1721 stderr);
1722 exit(129);
1725 return __cmd_report();