?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 2.6.31.8
[linux/fpc-iii.git] / tools / perf / builtin-report.c
blobb8a75f63d5351bdb662092bac1e9042c4eb17c3e
1 /*
2 * builtin-report.c
3 *
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"
9
10 #include "util/util.h"
11
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
21 #include "perf.h"
22 #include "util/header.h"
23
24 #include "util/parse-options.h"
25 #include "util/parse-events.h"
26
27 #define SHOW_KERNEL 1
28 #define SHOW_USER 2
29 #define SHOW_HV 4
30
31 static char const *input_name = "perf.data";
32 static char *vmlinux = NULL;
33
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,
37 *col_width_list_str;
38 static struct strlist *dso_list, *comm_list, *sym_list;
39 static char *field_sep;
40
41 static int force;
42 static int input;
43 static int show_mask = SHOW_KERNEL | SHOW_USER | SHOW_HV;
44
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)
48
49 static int verbose;
50 #define eprintf(x...) do { if (verbose) fprintf(stderr, x); } while (0)
51
52 static int modules;
53
54 static int full_paths;
55 static int show_nr_samples;
56
57 static unsigned long page_size;
58 static unsigned long mmap_window = 32;
59
60 static char default_parent_pattern[] = "^sys_|^do_page_fault";
61 static char *parent_pattern = default_parent_pattern;
62 static regex_t parent_regex;
63
64 static int exclude_other = 1;
65
66 static char callchain_default_opt[] = "fractal,0.5";
67
68 static int callchain;
69
70 static
71 struct callchain_param callchain_param = {
72 .mode = CHAIN_GRAPH_REL,
73 .min_percent = 0.5
74 };
75
76 static u64 sample_type;
77
78 struct ip_event {
79 struct perf_event_header header;
80 u64 ip;
81 u32 pid, tid;
82 unsigned char __more_data[];
83 };
84
85 struct mmap_event {
86 struct perf_event_header header;
87 u32 pid, tid;
88 u64 start;
89 u64 len;
90 u64 pgoff;
91 char filename[PATH_MAX];
92 };
93
94 struct comm_event {
95 struct perf_event_header header;
96 u32 pid, tid;
97 char comm[16];
98 };
99
100 struct fork_event {
101 struct perf_event_header header;
102 u32 pid, ppid;
103 u32 tid, ptid;
104 };
105
106 struct lost_event {
107 struct perf_event_header header;
108 u64 id;
109 u64 lost;
110 };
111
112 struct read_event {
113 struct perf_event_header header;
114 u32 pid,tid;
115 u64 value;
116 u64 time_enabled;
117 u64 time_running;
118 u64 id;
119 };
120
121 typedef union event_union {
122 struct perf_event_header header;
123 struct ip_event ip;
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;
129 } event_t;
130
131 static int repsep_fprintf(FILE *fp, const char *fmt, ...)
132 {
133 int n;
134 va_list ap;
135
136 va_start(ap, fmt);
137 if (!field_sep)
138 n = vfprintf(fp, fmt, ap);
139 else {
140 char *bf = NULL;
141 n = vasprintf(&bf, fmt, ap);
142 if (n > 0) {
143 char *sep = bf;
144 while (1) {
145 sep = strchr(sep, *field_sep);
146 if (sep == NULL)
147 break;
148 *sep = '.';
149 }
150 }
151 fputs(bf, fp);
152 free(bf);
153 }
154 va_end(ap);
155 return n;
156 }
157
158 static LIST_HEAD(dsos);
159 static struct dso *kernel_dso;
160 static struct dso *vdso;
161 static struct dso *hypervisor_dso;
162
163 static void dsos__add(struct dso *dso)
164 {
165 list_add_tail(&dso->node, &dsos);
166 }
167
168 static struct dso *dsos__find(const char *name)
169 {
170 struct dso *pos;
171
172 list_for_each_entry(pos, &dsos, node)
173 if (strcmp(pos->name, name) == 0)
174 return pos;
175 return NULL;
176 }
177
178 static struct dso *dsos__findnew(const char *name)
179 {
180 struct dso *dso = dsos__find(name);
181 int nr;
182
183 if (dso)
184 return dso;
185
186 dso = dso__new(name, 0);
187 if (!dso)
188 goto out_delete_dso;
189
190 nr = dso__load(dso, NULL, verbose);
191 if (nr < 0) {
192 eprintf("Failed to open: %s\n", name);
193 goto out_delete_dso;
194 }
195 if (!nr)
196 eprintf("No symbols found in: %s, maybe install a debug package?\n", name);
197
198 dsos__add(dso);
199
200 return dso;
201
202 out_delete_dso:
203 dso__delete(dso);
204 return NULL;
205 }
206
207 static void dsos__fprintf(FILE *fp)
208 {
209 struct dso *pos;
210
211 list_for_each_entry(pos, &dsos, node)
212 dso__fprintf(pos, fp);
213 }
214
215 static struct symbol *vdso__find_symbol(struct dso *dso, u64 ip)
216 {
217 return dso__find_symbol(dso, ip);
218 }
219
220 static int load_kernel(void)
221 {
222 int err;
223
224 kernel_dso = dso__new("[kernel]", 0);
225 if (!kernel_dso)
226 return -1;
227
228 err = dso__load_kernel(kernel_dso, vmlinux, NULL, verbose, modules);
229 if (err <= 0) {
230 dso__delete(kernel_dso);
231 kernel_dso = NULL;
232 } else
233 dsos__add(kernel_dso);
234
235 vdso = dso__new("[vdso]", 0);
236 if (!vdso)
237 return -1;
238
239 vdso->find_symbol = vdso__find_symbol;
240
241 dsos__add(vdso);
242
243 hypervisor_dso = dso__new("[hypervisor]", 0);
244 if (!hypervisor_dso)
245 return -1;
246 dsos__add(hypervisor_dso);
247
248 return err;
249 }
250
251 static char __cwd[PATH_MAX];
252 static char *cwd = __cwd;
253 static int cwdlen;
254
255 static int strcommon(const char *pathname)
256 {
257 int n = 0;
258
259 while (n < cwdlen && pathname[n] == cwd[n])
260 ++n;
261
262 return n;
263 }
264
265 struct map {
266 struct list_head node;
267 u64 start;
268 u64 end;
269 u64 pgoff;
270 u64 (*map_ip)(struct map *, u64);
271 struct dso *dso;
272 };
273
274 static u64 map__map_ip(struct map *map, u64 ip)
275 {
276 return ip - map->start + map->pgoff;
277 }
278
279 static u64 vdso__map_ip(struct map *map __used, u64 ip)
280 {
281 return ip;
282 }
283
284 static inline int is_anon_memory(const char *filename)
285 {
286 return strcmp(filename, "//anon") == 0;
287 }
288
289 static struct map *map__new(struct mmap_event *event)
290 {
291 struct map *self = malloc(sizeof(*self));
292
293 if (self != NULL) {
294 const char *filename = event->filename;
295 char newfilename[PATH_MAX];
296 int anon;
297
298 if (cwd) {
299 int n = strcommon(filename);
300
301 if (n == cwdlen) {
302 snprintf(newfilename, sizeof(newfilename),
303 ".%s", filename + n);
304 filename = newfilename;
305 }
306 }
307
308 anon = is_anon_memory(filename);
309
310 if (anon) {
311 snprintf(newfilename, sizeof(newfilename), "/tmp/perf-%d.map", event->pid);
312 filename = newfilename;
313 }
314
315 self->start = event->start;
316 self->end = event->start + event->len;
317 self->pgoff = event->pgoff;
318
319 self->dso = dsos__findnew(filename);
320 if (self->dso == NULL)
321 goto out_delete;
322
323 if (self->dso == vdso || anon)
324 self->map_ip = vdso__map_ip;
325 else
326 self->map_ip = map__map_ip;
327 }
328 return self;
329 out_delete:
330 free(self);
331 return NULL;
332 }
333
334 static struct map *map__clone(struct map *self)
335 {
336 struct map *map = malloc(sizeof(*self));
337
338 if (!map)
339 return NULL;
340
341 memcpy(map, self, sizeof(*self));
342
343 return map;
344 }
345
346 static int map__overlap(struct map *l, struct map *r)
347 {
348 if (l->start > r->start) {
349 struct map *t = l;
350 l = r;
351 r = t;
352 }
353
354 if (l->end > r->start)
355 return 1;
356
357 return 0;
358 }
359
360 static size_t map__fprintf(struct map *self, FILE *fp)
361 {
362 return fprintf(fp, " %Lx-%Lx %Lx %s\n",
363 self->start, self->end, self->pgoff, self->dso->name);
364 }
365
366
367 struct thread {
368 struct rb_node rb_node;
369 struct list_head maps;
370 pid_t pid;
371 char *comm;
372 };
373
374 static struct thread *thread__new(pid_t pid)
375 {
376 struct thread *self = malloc(sizeof(*self));
377
378 if (self != NULL) {
379 self->pid = pid;
380 self->comm = malloc(32);
381 if (self->comm)
382 snprintf(self->comm, 32, ":%d", self->pid);
383 INIT_LIST_HEAD(&self->maps);
384 }
385
386 return self;
387 }
388
389 static unsigned int dsos__col_width,
390 comms__col_width,
391 threads__col_width;
392
393 static int thread__set_comm(struct thread *self, const char *comm)
394 {
395 if (self->comm)
396 free(self->comm);
397 self->comm = strdup(comm);
398 if (!self->comm)
399 return -ENOMEM;
400
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;
407 }
408 }
409
410 return 0;
411 }
412
413 static size_t thread__fprintf(struct thread *self, FILE *fp)
414 {
415 struct map *pos;
416 size_t ret = fprintf(fp, "Thread %d %s\n", self->pid, self->comm);
417
418 list_for_each_entry(pos, &self->maps, node)
419 ret += map__fprintf(pos, fp);
420
421 return ret;
422 }
423
424
425 static struct rb_root threads;
426 static struct thread *last_match;
427
428 static struct thread *threads__findnew(pid_t pid)
429 {
430 struct rb_node **p = &threads.rb_node;
431 struct rb_node *parent = NULL;
432 struct thread *th;
433
434 /*
435 * Font-end cache - PID lookups come in blocks,
436 * so most of the time we dont have to look up
437 * the full rbtree:
438 */
439 if (last_match && last_match->pid == pid)
440 return last_match;
441
442 while (*p != NULL) {
443 parent = *p;
444 th = rb_entry(parent, struct thread, rb_node);
445
446 if (th->pid == pid) {
447 last_match = th;
448 return th;
449 }
450
451 if (pid < th->pid)
452 p = &(*p)->rb_left;
453 else
454 p = &(*p)->rb_right;
455 }
456
457 th = thread__new(pid);
458 if (th != NULL) {
459 rb_link_node(&th->rb_node, parent, p);
460 rb_insert_color(&th->rb_node, &threads);
461 last_match = th;
462 }
463
464 return th;
465 }
466
467 static void thread__insert_map(struct thread *self, struct map *map)
468 {
469 struct map *pos, *tmp;
470
471 list_for_each_entry_safe(pos, tmp, &self->maps, node) {
472 if (map__overlap(pos, map)) {
473 if (verbose >= 2) {
474 printf("overlapping maps:\n");
475 map__fprintf(map, stdout);
476 map__fprintf(pos, stdout);
477 }
478
479 if (map->start <= pos->start && map->end > pos->start)
480 pos->start = map->end;
481
482 if (map->end >= pos->end && map->start < pos->end)
483 pos->end = map->start;
484
485 if (verbose >= 2) {
486 printf("after collision:\n");
487 map__fprintf(pos, stdout);
488 }
489
490 if (pos->start >= pos->end) {
491 list_del_init(&pos->node);
492 free(pos);
493 }
494 }
495 }
496
497 list_add_tail(&map->node, &self->maps);
498 }
499
500 static int thread__fork(struct thread *self, struct thread *parent)
501 {
502 struct map *map;
503
504 if (self->comm)
505 free(self->comm);
506 self->comm = strdup(parent->comm);
507 if (!self->comm)
508 return -ENOMEM;
509
510 list_for_each_entry(map, &parent->maps, node) {
511 struct map *new = map__clone(map);
512 if (!new)
513 return -ENOMEM;
514 thread__insert_map(self, new);
515 }
516
517 return 0;
518 }
519
520 static struct map *thread__find_map(struct thread *self, u64 ip)
521 {
522 struct map *pos;
523
524 if (self == NULL)
525 return NULL;
526
527 list_for_each_entry(pos, &self->maps, node)
528 if (ip >= pos->start && ip <= pos->end)
529 return pos;
530
531 return NULL;
532 }
533
534 static size_t threads__fprintf(FILE *fp)
535 {
536 size_t ret = 0;
537 struct rb_node *nd;
538
539 for (nd = rb_first(&threads); nd; nd = rb_next(nd)) {
540 struct thread *pos = rb_entry(nd, struct thread, rb_node);
541
542 ret += thread__fprintf(pos, fp);
543 }
544
545 return ret;
546 }
547
548 /*
549 * histogram, sorted on item, collects counts
550 */
551
552 static struct rb_root hist;
553
554 struct hist_entry {
555 struct rb_node rb_node;
556
557 struct thread *thread;
558 struct map *map;
559 struct dso *dso;
560 struct symbol *sym;
561 struct symbol *parent;
562 u64 ip;
563 char level;
564 struct callchain_node callchain;
565 struct rb_root sorted_chain;
566
567 u64 count;
568 };
569
570 /*
571 * configurable sorting bits
572 */
573
574 struct sort_entry {
575 struct list_head list;
576
577 char *header;
578
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);
582 unsigned int *width;
583 bool elide;
584 };
585
586 static int64_t cmp_null(void *l, void *r)
587 {
588 if (!l && !r)
589 return 0;
590 else if (!l)
591 return -1;
592 else
593 return 1;
594 }
595
596 /* --sort pid */
597
598 static int64_t
599 sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
600 {
601 return right->thread->pid - left->thread->pid;
602 }
603
604 static size_t
605 sort__thread_print(FILE *fp, struct hist_entry *self, unsigned int width)
606 {
607 return repsep_fprintf(fp, "%*s:%5d", width - 6,
608 self->thread->comm ?: "", self->thread->pid);
609 }
610
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,
616 };
617
618 /* --sort comm */
619
620 static int64_t
621 sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
622 {
623 return right->thread->pid - left->thread->pid;
624 }
625
626 static int64_t
627 sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
628 {
629 char *comm_l = left->thread->comm;
630 char *comm_r = right->thread->comm;
631
632 if (!comm_l || !comm_r)
633 return cmp_null(comm_l, comm_r);
634
635 return strcmp(comm_l, comm_r);
636 }
637
638 static size_t
639 sort__comm_print(FILE *fp, struct hist_entry *self, unsigned int width)
640 {
641 return repsep_fprintf(fp, "%*s", width, self->thread->comm);
642 }
643
644 static struct sort_entry sort_comm = {
645 .header = "Command",
646 .cmp = sort__comm_cmp,
647 .collapse = sort__comm_collapse,
648 .print = sort__comm_print,
649 .width = &comms__col_width,
650 };
651
652 /* --sort dso */
653
654 static int64_t
655 sort__dso_cmp(struct hist_entry *left, struct hist_entry *right)
656 {
657 struct dso *dso_l = left->dso;
658 struct dso *dso_r = right->dso;
659
660 if (!dso_l || !dso_r)
661 return cmp_null(dso_l, dso_r);
662
663 return strcmp(dso_l->name, dso_r->name);
664 }
665
666 static size_t
667 sort__dso_print(FILE *fp, struct hist_entry *self, unsigned int width)
668 {
669 if (self->dso)
670 return repsep_fprintf(fp, "%-*s", width, self->dso->name);
671
672 return repsep_fprintf(fp, "%*llx", width, (u64)self->ip);
673 }
674
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,
680 };
681
682 /* --sort symbol */
683
684 static int64_t
685 sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
686 {
687 u64 ip_l, ip_r;
688
689 if (left->sym == right->sym)
690 return 0;
691
692 ip_l = left->sym ? left->sym->start : left->ip;
693 ip_r = right->sym ? right->sym->start : right->ip;
694
695 return (int64_t)(ip_r - ip_l);
696 }
697
698 static size_t
699 sort__sym_print(FILE *fp, struct hist_entry *self, unsigned int width __used)
700 {
701 size_t ret = 0;
702
703 if (verbose)
704 ret += repsep_fprintf(fp, "%#018llx %c ", (u64)self->ip,
705 dso__symtab_origin(self->dso));
706
707 ret += repsep_fprintf(fp, "[%c] ", self->level);
708 if (self->sym) {
709 ret += repsep_fprintf(fp, "%s", self->sym->name);
710
711 if (self->sym->module)
712 ret += repsep_fprintf(fp, "\t[%s]",
713 self->sym->module->name);
714 } else {
715 ret += repsep_fprintf(fp, "%#016llx", (u64)self->ip);
716 }
717
718 return ret;
719 }
720
721 static struct sort_entry sort_sym = {
722 .header = "Symbol",
723 .cmp = sort__sym_cmp,
724 .print = sort__sym_print,
725 };
726
727 /* --sort parent */
728
729 static int64_t
730 sort__parent_cmp(struct hist_entry *left, struct hist_entry *right)
731 {
732 struct symbol *sym_l = left->parent;
733 struct symbol *sym_r = right->parent;
734
735 if (!sym_l || !sym_r)
736 return cmp_null(sym_l, sym_r);
737
738 return strcmp(sym_l->name, sym_r->name);
739 }
740
741 static size_t
742 sort__parent_print(FILE *fp, struct hist_entry *self, unsigned int width)
743 {
744 return repsep_fprintf(fp, "%-*s", width,
745 self->parent ? self->parent->name : "[other]");
746 }
747
748 static unsigned int parent_symbol__col_width;
749
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,
755 };
756
757 static int sort__need_collapse = 0;
758 static int sort__has_parent = 0;
759
760 struct sort_dimension {
761 char *name;
762 struct sort_entry *entry;
763 int taken;
764 };
765
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, },
772 };
773
774 static LIST_HEAD(hist_entry__sort_list);
775
776 static int sort_dimension__add(char *tok)
777 {
778 unsigned int i;
779
780 for (i = 0; i < ARRAY_SIZE(sort_dimensions); i++) {
781 struct sort_dimension *sd = &sort_dimensions[i];
782
783 if (sd->taken)
784 continue;
785
786 if (strncasecmp(tok, sd->name, strlen(tok)))
787 continue;
788
789 if (sd->entry->collapse)
790 sort__need_collapse = 1;
791
792 if (sd->entry == &sort_parent) {
793 int ret = regcomp(&parent_regex, parent_pattern, REG_EXTENDED);
794 if (ret) {
795 char err[BUFSIZ];
796
797 regerror(ret, &parent_regex, err, sizeof(err));
798 fprintf(stderr, "Invalid regex: %s\n%s",
799 parent_pattern, err);
800 exit(-1);
801 }
802 sort__has_parent = 1;
803 }
804
805 list_add_tail(&sd->entry->list, &hist_entry__sort_list);
806 sd->taken = 1;
807
808 return 0;
809 }
810
811 return -ESRCH;
812 }
813
814 static int64_t
815 hist_entry__cmp(struct hist_entry *left, struct hist_entry *right)
816 {
817 struct sort_entry *se;
818 int64_t cmp = 0;
819
820 list_for_each_entry(se, &hist_entry__sort_list, list) {
821 cmp = se->cmp(left, right);
822 if (cmp)
823 break;
824 }
825
826 return cmp;
827 }
828
829 static int64_t
830 hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
831 {
832 struct sort_entry *se;
833 int64_t cmp = 0;
834
835 list_for_each_entry(se, &hist_entry__sort_list, list) {
836 int64_t (*f)(struct hist_entry *, struct hist_entry *);
837
838 f = se->collapse ?: se->cmp;
839
840 cmp = f(left, right);
841 if (cmp)
842 break;
843 }
844
845 return cmp;
846 }
847
848 static size_t ipchain__fprintf_graph_line(FILE *fp, int depth, int depth_mask)
849 {
850 int i;
851 size_t ret = 0;
852
853 ret += fprintf(fp, "%s", " ");
854
855 for (i = 0; i < depth; i++)
856 if (depth_mask & (1 << i))
857 ret += fprintf(fp, "| ");
858 else
859 ret += fprintf(fp, " ");
860
861 ret += fprintf(fp, "\n");
862
863 return ret;
864 }
865 static size_t
866 ipchain__fprintf_graph(FILE *fp, struct callchain_list *chain, int depth,
867 int depth_mask, int count, u64 total_samples,
868 int hits)
869 {
870 int i;
871 size_t ret = 0;
872
873 ret += fprintf(fp, "%s", " ");
874 for (i = 0; i < depth; i++) {
875 if (depth_mask & (1 << i))
876 ret += fprintf(fp, "|");
877 else
878 ret += fprintf(fp, " ");
879 if (!count && i == depth - 1) {
880 double percent;
881
882 percent = hits * 100.0 / total_samples;
883 ret += percent_color_fprintf(fp, "--%2.2f%%-- ", percent);
884 } else
885 ret += fprintf(fp, "%s", " ");
886 }
887 if (chain->sym)
888 ret += fprintf(fp, "%s\n", chain->sym->name);
889 else
890 ret += fprintf(fp, "%p\n", (void *)(long)chain->ip);
891
892 return ret;
893 }
894
895 static struct symbol *rem_sq_bracket;
896 static struct callchain_list rem_hits;
897
898 static void init_rem_hits(void)
899 {
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");
903 return;
904 }
905
906 strcpy(rem_sq_bracket->name, "[...]");
907 rem_hits.sym = rem_sq_bracket;
908 }
909
910 static size_t
911 callchain__fprintf_graph(FILE *fp, struct callchain_node *self,
912 u64 total_samples, int depth, int depth_mask)
913 {
914 struct rb_node *node, *next;
915 struct callchain_node *child;
916 struct callchain_list *chain;
917 int new_depth_mask = depth_mask;
918 u64 new_total;
919 u64 remaining;
920 size_t ret = 0;
921 int i;
922
923 if (callchain_param.mode == CHAIN_GRAPH_REL)
924 new_total = self->children_hit;
925 else
926 new_total = total_samples;
927
928 remaining = new_total;
929
930 node = rb_first(&self->rb_root);
931 while (node) {
932 u64 cumul;
933
934 child = rb_entry(node, struct callchain_node, rb_node);
935 cumul = cumul_hits(child);
936 remaining -= cumul;
937
938 /*
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
944 */
945 next = rb_next(node);
946 if (!next && (callchain_param.mode != CHAIN_GRAPH_REL || !remaining))
947 new_depth_mask &= ~(1 << (depth - 1));
948
949 /*
950 * But we keep the older depth mask for the line seperator
951 * to keep the level link until we reach the last child
952 */
953 ret += ipchain__fprintf_graph_line(fp, depth, depth_mask);
954 i = 0;
955 list_for_each_entry(chain, &child->val, list) {
956 if (chain->ip >= PERF_CONTEXT_MAX)
957 continue;
958 ret += ipchain__fprintf_graph(fp, chain, depth,
959 new_depth_mask, i++,
960 new_total,
961 cumul);
962 }
963 ret += callchain__fprintf_graph(fp, child, new_total,
964 depth + 1,
965 new_depth_mask | (1 << depth));
966 node = next;
967 }
968
969 if (callchain_param.mode == CHAIN_GRAPH_REL &&
970 remaining && remaining != new_total) {
971
972 if (!rem_sq_bracket)
973 return ret;
974
975 new_depth_mask &= ~(1 << (depth - 1));
976
977 ret += ipchain__fprintf_graph(fp, &rem_hits, depth,
978 new_depth_mask, 0, new_total,
979 remaining);
980 }
981
982 return ret;
983 }
984
985 static size_t
986 callchain__fprintf_flat(FILE *fp, struct callchain_node *self,
987 u64 total_samples)
988 {
989 struct callchain_list *chain;
990 size_t ret = 0;
991
992 if (!self)
993 return 0;
994
995 ret += callchain__fprintf_flat(fp, self->parent, total_samples);
996
997
998 list_for_each_entry(chain, &self->val, list) {
999 if (chain->ip >= PERF_CONTEXT_MAX)
1000 continue;
1001 if (chain->sym)
1002 ret += fprintf(fp, " %s\n", chain->sym->name);
1003 else
1004 ret += fprintf(fp, " %p\n",
1005 (void *)(long)chain->ip);
1006 }
1007
1008 return ret;
1009 }
1010
1011 static size_t
1012 hist_entry_callchain__fprintf(FILE *fp, struct hist_entry *self,
1013 u64 total_samples)
1014 {
1015 struct rb_node *rb_node;
1016 struct callchain_node *chain;
1017 size_t ret = 0;
1018
1019 rb_node = rb_first(&self->sorted_chain);
1020 while (rb_node) {
1021 double percent;
1022
1023 chain = rb_entry(rb_node, struct callchain_node, rb_node);
1024 percent = chain->hit * 100.0 / total_samples;
1025 switch (callchain_param.mode) {
1026 case CHAIN_FLAT:
1027 ret += percent_color_fprintf(fp, " %6.2f%%\n",
1028 percent);
1029 ret += callchain__fprintf_flat(fp, chain, total_samples);
1030 break;
1031 case CHAIN_GRAPH_ABS: /* Falldown */
1032 case CHAIN_GRAPH_REL:
1033 ret += callchain__fprintf_graph(fp, chain,
1034 total_samples, 1, 1);
1035 default:
1036 break;
1037 }
1038 ret += fprintf(fp, "\n");
1039 rb_node = rb_next(rb_node);
1040 }
1041
1042 return ret;
1043 }
1044
1045
1046 static size_t
1047 hist_entry__fprintf(FILE *fp, struct hist_entry *self, u64 total_samples)
1048 {
1049 struct sort_entry *se;
1050 size_t ret;
1051
1052 if (exclude_other && !self->parent)
1053 return 0;
1054
1055 if (total_samples)
1056 ret = percent_color_fprintf(fp,
1057 field_sep ? "%.2f" : " %6.2f%%",
1058 (self->count * 100.0) / total_samples);
1059 else
1060 ret = fprintf(fp, field_sep ? "%lld" : "%12lld ", self->count);
1061
1062 if (show_nr_samples) {
1063 if (field_sep)
1064 fprintf(fp, "%c%lld", *field_sep, self->count);
1065 else
1066 fprintf(fp, "%11lld", self->count);
1067 }
1068
1069 list_for_each_entry(se, &hist_entry__sort_list, list) {
1070 if (se->elide)
1071 continue;
1072
1073 fprintf(fp, "%s", field_sep ?: " ");
1074 ret += se->print(fp, self, se->width ? *se->width : 0);
1075 }
1076
1077 ret += fprintf(fp, "\n");
1078
1079 if (callchain)
1080 hist_entry_callchain__fprintf(fp, self, total_samples);
1081
1082 return ret;
1083 }
1084
1085 /*
1086 *
1087 */
1088
1089 static void dso__calc_col_width(struct dso *self)
1090 {
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;
1096 }
1097
1098 self->slen_calculated = 1;
1099 }
1100
1101 static struct symbol *
1102 resolve_symbol(struct thread *thread, struct map **mapp,
1103 struct dso **dsop, u64 *ipp)
1104 {
1105 struct dso *dso = dsop ? *dsop : NULL;
1106 struct map *map = mapp ? *mapp : NULL;
1107 u64 ip = *ipp;
1108
1109 if (!thread)
1110 return NULL;
1111
1112 if (dso)
1113 goto got_dso;
1114
1115 if (map)
1116 goto got_map;
1117
1118 map = thread__find_map(thread, ip);
1119 if (map != NULL) {
1120 /*
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.
1124 */
1125 if (!sort_dso.elide && !map->dso->slen_calculated)
1126 dso__calc_col_width(map->dso);
1127
1128 if (mapp)
1129 *mapp = map;
1130 got_map:
1131 ip = map->map_ip(map, ip);
1132
1133 dso = map->dso;
1134 } else {
1135 /*
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):
1140 */
1141 if ((long long)ip < 0)
1142 dso = kernel_dso;
1143 }
1144 dprintf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
1145 dprintf(" ...... map: %Lx -> %Lx\n", *ipp, ip);
1146 *ipp = ip;
1147
1148 if (dsop)
1149 *dsop = dso;
1150
1151 if (!dso)
1152 return NULL;
1153 got_dso:
1154 return dso->find_symbol(dso, ip);
1155 }
1156
1157 static int call__match(struct symbol *sym)
1158 {
1159 if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
1160 return 1;
1161
1162 return 0;
1163 }
1164
1165 static struct symbol **
1166 resolve_callchain(struct thread *thread, struct map *map __used,
1167 struct ip_callchain *chain, struct hist_entry *entry)
1168 {
1169 u64 context = PERF_CONTEXT_MAX;
1170 struct symbol **syms = NULL;
1171 unsigned int i;
1172
1173 if (callchain) {
1174 syms = calloc(chain->nr, sizeof(*syms));
1175 if (!syms) {
1176 fprintf(stderr, "Can't allocate memory for symbols\n");
1177 exit(-1);
1178 }
1179 }
1180
1181 for (i = 0; i < chain->nr; i++) {
1182 u64 ip = chain->ips[i];
1183 struct dso *dso = NULL;
1184 struct symbol *sym;
1185
1186 if (ip >= PERF_CONTEXT_MAX) {
1187 context = ip;
1188 continue;
1189 }
1190
1191 switch (context) {
1192 case PERF_CONTEXT_HV:
1193 dso = hypervisor_dso;
1194 break;
1195 case PERF_CONTEXT_KERNEL:
1196 dso = kernel_dso;
1197 break;
1198 default:
1199 break;
1200 }
1201
1202 sym = resolve_symbol(thread, NULL, &dso, &ip);
1203
1204 if (sym) {
1205 if (sort__has_parent && call__match(sym) &&
1206 !entry->parent)
1207 entry->parent = sym;
1208 if (!callchain)
1209 break;
1210 syms[i] = sym;
1211 }
1212 }
1213
1214 return syms;
1215 }
1216
1217 /*
1218 * collect histogram counts
1219 */
1220
1221 static int
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)
1225 {
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 = {
1231 .thread = thread,
1232 .map = map,
1233 .dso = dso,
1234 .sym = sym,
1235 .ip = ip,
1236 .level = level,
1237 .count = count,
1238 .parent = NULL,
1239 .sorted_chain = RB_ROOT
1240 };
1241 int cmp;
1242
1243 if ((sort__has_parent || callchain) && chain)
1244 syms = resolve_callchain(thread, map, chain, &entry);
1245
1246 while (*p != NULL) {
1247 parent = *p;
1248 he = rb_entry(parent, struct hist_entry, rb_node);
1249
1250 cmp = hist_entry__cmp(&entry, he);
1251
1252 if (!cmp) {
1253 he->count += count;
1254 if (callchain) {
1255 append_chain(&he->callchain, chain, syms);
1256 free(syms);
1257 }
1258 return 0;
1259 }
1260
1261 if (cmp < 0)
1262 p = &(*p)->rb_left;
1263 else
1264 p = &(*p)->rb_right;
1265 }
1266
1267 he = malloc(sizeof(*he));
1268 if (!he)
1269 return -ENOMEM;
1270 *he = entry;
1271 if (callchain) {
1272 callchain_init(&he->callchain);
1273 append_chain(&he->callchain, chain, syms);
1274 free(syms);
1275 }
1276 rb_link_node(&he->rb_node, parent, p);
1277 rb_insert_color(&he->rb_node, &hist);
1278
1279 return 0;
1280 }
1281
1282 static void hist_entry__free(struct hist_entry *he)
1283 {
1284 free(he);
1285 }
1286
1287 /*
1288 * collapse the histogram
1289 */
1290
1291 static struct rb_root collapse_hists;
1292
1293 static void collapse__insert_entry(struct hist_entry *he)
1294 {
1295 struct rb_node **p = &collapse_hists.rb_node;
1296 struct rb_node *parent = NULL;
1297 struct hist_entry *iter;
1298 int64_t cmp;
1299
1300 while (*p != NULL) {
1301 parent = *p;
1302 iter = rb_entry(parent, struct hist_entry, rb_node);
1303
1304 cmp = hist_entry__collapse(iter, he);
1305
1306 if (!cmp) {
1307 iter->count += he->count;
1308 hist_entry__free(he);
1309 return;
1310 }
1311
1312 if (cmp < 0)
1313 p = &(*p)->rb_left;
1314 else
1315 p = &(*p)->rb_right;
1316 }
1317
1318 rb_link_node(&he->rb_node, parent, p);
1319 rb_insert_color(&he->rb_node, &collapse_hists);
1320 }
1321
1322 static void collapse__resort(void)
1323 {
1324 struct rb_node *next;
1325 struct hist_entry *n;
1326
1327 if (!sort__need_collapse)
1328 return;
1329
1330 next = rb_first(&hist);
1331 while (next) {
1332 n = rb_entry(next, struct hist_entry, rb_node);
1333 next = rb_next(&n->rb_node);
1334
1335 rb_erase(&n->rb_node, &hist);
1336 collapse__insert_entry(n);
1337 }
1338 }
1339
1340 /*
1341 * reverse the map, sort on count.
1342 */
1343
1344 static struct rb_root output_hists;
1345
1346 static void output__insert_entry(struct hist_entry *he, u64 min_callchain_hits)
1347 {
1348 struct rb_node **p = &output_hists.rb_node;
1349 struct rb_node *parent = NULL;
1350 struct hist_entry *iter;
1351
1352 if (callchain)
1353 callchain_param.sort(&he->sorted_chain, &he->callchain,
1354 min_callchain_hits, &callchain_param);
1355
1356 while (*p != NULL) {
1357 parent = *p;
1358 iter = rb_entry(parent, struct hist_entry, rb_node);
1359
1360 if (he->count > iter->count)
1361 p = &(*p)->rb_left;
1362 else
1363 p = &(*p)->rb_right;
1364 }
1365
1366 rb_link_node(&he->rb_node, parent, p);
1367 rb_insert_color(&he->rb_node, &output_hists);
1368 }
1369
1370 static void output__resort(u64 total_samples)
1371 {
1372 struct rb_node *next;
1373 struct hist_entry *n;
1374 struct rb_root *tree = &hist;
1375 u64 min_callchain_hits;
1376
1377 min_callchain_hits = total_samples * (callchain_param.min_percent / 100);
1378
1379 if (sort__need_collapse)
1380 tree = &collapse_hists;
1381
1382 next = rb_first(tree);
1383
1384 while (next) {
1385 n = rb_entry(next, struct hist_entry, rb_node);
1386 next = rb_next(&n->rb_node);
1387
1388 rb_erase(&n->rb_node, tree);
1389 output__insert_entry(n, min_callchain_hits);
1390 }
1391 }
1392
1393 static size_t output__fprintf(FILE *fp, u64 total_samples)
1394 {
1395 struct hist_entry *pos;
1396 struct sort_entry *se;
1397 struct rb_node *nd;
1398 size_t ret = 0;
1399 unsigned int width;
1400 char *col_width = col_width_list_str;
1401
1402 init_rem_hits();
1403
1404 fprintf(fp, "# Samples: %Ld\n", (u64)total_samples);
1405 fprintf(fp, "#\n");
1406
1407 fprintf(fp, "# Overhead");
1408 if (show_nr_samples) {
1409 if (field_sep)
1410 fprintf(fp, "%cSamples", *field_sep);
1411 else
1412 fputs(" Samples ", fp);
1413 }
1414 list_for_each_entry(se, &hist_entry__sort_list, list) {
1415 if (se->elide)
1416 continue;
1417 if (field_sep) {
1418 fprintf(fp, "%c%s", *field_sep, se->header);
1419 continue;
1420 }
1421 width = strlen(se->header);
1422 if (se->width) {
1423 if (col_width_list_str) {
1424 if (col_width) {
1425 *se->width = atoi(col_width);
1426 col_width = strchr(col_width, ',');
1427 if (col_width)
1428 ++col_width;
1429 }
1430 }
1431 width = *se->width = max(*se->width, width);
1432 }
1433 fprintf(fp, " %*s", width, se->header);
1434 }
1435 fprintf(fp, "\n");
1436
1437 if (field_sep)
1438 goto print_entries;
1439
1440 fprintf(fp, "# ........");
1441 if (show_nr_samples)
1442 fprintf(fp, " ..........");
1443 list_for_each_entry(se, &hist_entry__sort_list, list) {
1444 unsigned int i;
1445
1446 if (se->elide)
1447 continue;
1448
1449 fprintf(fp, " ");
1450 if (se->width)
1451 width = *se->width;
1452 else
1453 width = strlen(se->header);
1454 for (i = 0; i < width; i++)
1455 fprintf(fp, ".");
1456 }
1457 fprintf(fp, "\n");
1458
1459 fprintf(fp, "#\n");
1460
1461 print_entries:
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);
1465 }
1466
1467 if (sort_order == default_sort_order &&
1468 parent_pattern == default_parent_pattern) {
1469 fprintf(fp, "#\n");
1470 fprintf(fp, "# (For a higher level overview, try: perf report --sort comm,dso)\n");
1471 fprintf(fp, "#\n");
1472 }
1473 fprintf(fp, "\n");
1474
1475 free(rem_sq_bracket);
1476
1477 return ret;
1478 }
1479
1480 static void register_idle_thread(void)
1481 {
1482 struct thread *thread = threads__findnew(0);
1483
1484 if (thread == NULL ||
1485 thread__set_comm(thread, "[idle]")) {
1486 fprintf(stderr, "problem inserting idle task.\n");
1487 exit(-1);
1488 }
1489 }
1490
1491 static unsigned long total = 0,
1492 total_mmap = 0,
1493 total_comm = 0,
1494 total_fork = 0,
1495 total_unknown = 0,
1496 total_lost = 0;
1497
1498 static int validate_chain(struct ip_callchain *chain, event_t *event)
1499 {
1500 unsigned int chain_size;
1501
1502 chain_size = event->header.size;
1503 chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;
1504
1505 if (chain->nr*sizeof(u64) > chain_size)
1506 return -1;
1507
1508 return 0;
1509 }
1510
1511 static int
1512 process_sample_event(event_t *event, unsigned long offset, unsigned long head)
1513 {
1514 char level;
1515 int show = 0;
1516 struct dso *dso = NULL;
1517 struct thread *thread = threads__findnew(event->ip.pid);
1518 u64 ip = event->ip.ip;
1519 u64 period = 1;
1520 struct map *map = NULL;
1521 void *more_data = event->ip.__more_data;
1522 struct ip_callchain *chain = NULL;
1523 int cpumode;
1524
1525 if (sample_type & PERF_SAMPLE_PERIOD) {
1526 period = *(u64 *)more_data;
1527 more_data += sizeof(u64);
1528 }
1529
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),
1533 event->header.misc,
1534 event->ip.pid, event->ip.tid,
1535 (void *)(long)ip,
1536 (long long)period);
1537
1538 if (sample_type & PERF_SAMPLE_CALLCHAIN) {
1539 unsigned int i;
1540
1541 chain = (void *)more_data;
1542
1543 dprintf("... chain: nr:%Lu\n", chain->nr);
1544
1545 if (validate_chain(chain, event) < 0) {
1546 eprintf("call-chain problem with event, skipping it.\n");
1547 return 0;
1548 }
1549
1550 if (dump_trace) {
1551 for (i = 0; i < chain->nr; i++)
1552 dprintf("..... %2d: %016Lx\n", i, chain->ips[i]);
1553 }
1554 }
1555
1556 dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid);
1557
1558 if (thread == NULL) {
1559 eprintf("problem processing %d event, skipping it.\n",
1560 event->header.type);
1561 return -1;
1562 }
1563
1564 if (comm_list && !strlist__has_entry(comm_list, thread->comm))
1565 return 0;
1566
1567 cpumode = event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK;
1568
1569 if (cpumode == PERF_EVENT_MISC_KERNEL) {
1570 show = SHOW_KERNEL;
1571 level = 'k';
1572
1573 dso = kernel_dso;
1574
1575 dprintf(" ...... dso: %s\n", dso->name);
1576
1577 } else if (cpumode == PERF_EVENT_MISC_USER) {
1578
1579 show = SHOW_USER;
1580 level = '.';
1581
1582 } else {
1583 show = SHOW_HV;
1584 level = 'H';
1585
1586 dso = hypervisor_dso;
1587
1588 dprintf(" ...... dso: [hypervisor]\n");
1589 }
1590
1591 if (show & show_mask) {
1592 struct symbol *sym = resolve_symbol(thread, &map, &dso, &ip);
1593
1594 if (dso_list && (!dso || !dso->name ||
1595 !strlist__has_entry(dso_list, dso->name)))
1596 return 0;
1597
1598 if (sym_list && (!sym || !strlist__has_entry(sym_list, sym->name)))
1599 return 0;
1600
1601 if (hist_entry__add(thread, map, dso, sym, ip, chain, level, period)) {
1602 eprintf("problem incrementing symbol count, skipping event\n");
1603 return -1;
1604 }
1605 }
1606 total += period;
1607
1608 return 0;
1609 }
1610
1611 static int
1612 process_mmap_event(event_t *event, unsigned long offset, unsigned long head)
1613 {
1614 struct thread *thread = threads__findnew(event->mmap.pid);
1615 struct map *map = map__new(&event->mmap);
1616
1617 dprintf("%p [%p]: PERF_EVENT_MMAP %d/%d: [%p(%p) @ %p]: %s\n",
1618 (void *)(offset + head),
1619 (void *)(long)(event->header.size),
1620 event->mmap.pid,
1621 event->mmap.tid,
1622 (void *)(long)event->mmap.start,
1623 (void *)(long)event->mmap.len,
1624 (void *)(long)event->mmap.pgoff,
1625 event->mmap.filename);
1626
1627 if (thread == NULL || map == NULL) {
1628 dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1629 return 0;
1630 }
1631
1632 thread__insert_map(thread, map);
1633 total_mmap++;
1634
1635 return 0;
1636 }
1637
1638 static int
1639 process_comm_event(event_t *event, unsigned long offset, unsigned long head)
1640 {
1641 struct thread *thread = threads__findnew(event->comm.pid);
1642
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);
1647
1648 if (thread == NULL ||
1649 thread__set_comm(thread, event->comm.comm)) {
1650 dprintf("problem processing PERF_EVENT_COMM, skipping event.\n");
1651 return -1;
1652 }
1653 total_comm++;
1654
1655 return 0;
1656 }
1657
1658 static int
1659 process_task_event(event_t *event, unsigned long offset, unsigned long head)
1660 {
1661 struct thread *thread = threads__findnew(event->fork.pid);
1662 struct thread *parent = threads__findnew(event->fork.ppid);
1663
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);
1670
1671 /*
1672 * A thread clone will have the same PID for both
1673 * parent and child.
1674 */
1675 if (thread == parent)
1676 return 0;
1677
1678 if (event->header.type == PERF_EVENT_EXIT)
1679 return 0;
1680
1681 if (!thread || !parent || thread__fork(thread, parent)) {
1682 dprintf("problem processing PERF_EVENT_FORK, skipping event.\n");
1683 return -1;
1684 }
1685 total_fork++;
1686
1687 return 0;
1688 }
1689
1690 static int
1691 process_lost_event(event_t *event, unsigned long offset, unsigned long head)
1692 {
1693 dprintf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n",
1694 (void *)(offset + head),
1695 (void *)(long)(event->header.size),
1696 event->lost.id,
1697 event->lost.lost);
1698
1699 total_lost += event->lost.lost;
1700
1701 return 0;
1702 }
1703
1704 static void trace_event(event_t *event)
1705 {
1706 unsigned char *raw_event = (void *)event;
1707 char *color = PERF_COLOR_BLUE;
1708 int i, j;
1709
1710 if (!dump_trace)
1711 return;
1712
1713 dprintf(".");
1714 cdprintf("\n. ... raw event: size %d bytes\n", event->header.size);
1715
1716 for (i = 0; i < event->header.size; i++) {
1717 if ((i & 15) == 0) {
1718 dprintf(".");
1719 cdprintf(" %04x: ", i);
1720 }
1721
1722 cdprintf(" %02x", raw_event[i]);
1723
1724 if (((i & 15) == 15) || i == event->header.size-1) {
1725 cdprintf(" ");
1726 for (j = 0; j < 15-(i & 15); j++)
1727 cdprintf(" ");
1728 for (j = 0; j < (i & 15); j++) {
1729 if (isprint(raw_event[i-15+j]))
1730 cdprintf("%c", raw_event[i-15+j]);
1731 else
1732 cdprintf(".");
1733 }
1734 cdprintf("\n");
1735 }
1736 }
1737 dprintf(".\n");
1738 }
1739
1740 static struct perf_header *header;
1741
1742 static struct perf_counter_attr *perf_header__find_attr(u64 id)
1743 {
1744 int i;
1745
1746 for (i = 0; i < header->attrs; i++) {
1747 struct perf_header_attr *attr = header->attr[i];
1748 int j;
1749
1750 for (j = 0; j < attr->ids; j++) {
1751 if (attr->id[j] == id)
1752 return &attr->attr;
1753 }
1754 }
1755
1756 return NULL;
1757 }
1758
1759 static int
1760 process_read_event(event_t *event, unsigned long offset, unsigned long head)
1761 {
1762 struct perf_counter_attr *attr = perf_header__find_attr(event->read.id);
1763
1764 dprintf("%p [%p]: PERF_EVENT_READ: %d %d %s %Lu\n",
1765 (void *)(offset + head),
1766 (void *)(long)(event->header.size),
1767 event->read.pid,
1768 event->read.tid,
1769 attr ? __event_name(attr->type, attr->config)
1770 : "FAIL",
1771 event->read.value);
1772
1773 return 0;
1774 }
1775
1776 static int
1777 process_event(event_t *event, unsigned long offset, unsigned long head)
1778 {
1779 trace_event(event);
1780
1781 switch (event->header.type) {
1782 case PERF_EVENT_SAMPLE:
1783 return process_sample_event(event, offset, head);
1784
1785 case PERF_EVENT_MMAP:
1786 return process_mmap_event(event, offset, head);
1787
1788 case PERF_EVENT_COMM:
1789 return process_comm_event(event, offset, head);
1790
1791 case PERF_EVENT_FORK:
1792 case PERF_EVENT_EXIT:
1793 return process_task_event(event, offset, head);
1794
1795 case PERF_EVENT_LOST:
1796 return process_lost_event(event, offset, head);
1797
1798 case PERF_EVENT_READ:
1799 return process_read_event(event, offset, head);
1800
1801 /*
1802 * We dont process them right now but they are fine:
1803 */
1804
1805 case PERF_EVENT_THROTTLE:
1806 case PERF_EVENT_UNTHROTTLE:
1807 return 0;
1808
1809 default:
1810 return -1;
1811 }
1812
1813 return 0;
1814 }
1815
1816 static u64 perf_header__sample_type(void)
1817 {
1818 u64 sample_type = 0;
1819 int i;
1820
1821 for (i = 0; i < header->attrs; i++) {
1822 struct perf_header_attr *attr = header->attr[i];
1823
1824 if (!sample_type)
1825 sample_type = attr->attr.sample_type;
1826 else if (sample_type != attr->attr.sample_type)
1827 die("non matching sample_type");
1828 }
1829
1830 return sample_type;
1831 }
1832
1833 static int __cmd_report(void)
1834 {
1835 int ret, rc = EXIT_FAILURE;
1836 unsigned long offset = 0;
1837 unsigned long head, shift;
1838 struct stat stat;
1839 event_t *event;
1840 uint32_t size;
1841 char *buf;
1842
1843 register_idle_thread();
1844
1845 input = open(input_name, O_RDONLY);
1846 if (input < 0) {
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");
1851 exit(-1);
1852 }
1853
1854 ret = fstat(input, &stat);
1855 if (ret < 0) {
1856 perror("failed to stat file");
1857 exit(-1);
1858 }
1859
1860 if (!force && stat.st_uid && (stat.st_uid != geteuid())) {
1861 fprintf(stderr, "file: %s not owned by current user or root\n", input_name);
1862 exit(-1);
1863 }
1864
1865 if (!stat.st_size) {
1866 fprintf(stderr, "zero-sized file, nothing to do!\n");
1867 exit(0);
1868 }
1869
1870 header = perf_header__read(input);
1871 head = header->data_offset;
1872
1873 sample_type = perf_header__sample_type();
1874
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");
1880 exit(-1);
1881 }
1882 if (callchain) {
1883 fprintf(stderr, "selected -c but no callchain data."
1884 " Did you call perf record without"
1885 " -g?\n");
1886 exit(-1);
1887 }
1888 } else if (callchain_param.mode != CHAIN_NONE && !callchain) {
1889 callchain = 1;
1890 if (register_callchain_param(&callchain_param) < 0) {
1891 fprintf(stderr, "Can't register callchain"
1892 " params\n");
1893 exit(-1);
1894 }
1895 }
1896
1897 if (load_kernel() < 0) {
1898 perror("failed to load kernel symbols");
1899 return EXIT_FAILURE;
1900 }
1901
1902 if (!full_paths) {
1903 if (getcwd(__cwd, sizeof(__cwd)) == NULL) {
1904 perror("failed to get the current directory");
1905 return EXIT_FAILURE;
1906 }
1907 cwdlen = strlen(cwd);
1908 } else {
1909 cwd = NULL;
1910 cwdlen = 0;
1911 }
1912
1913 shift = page_size * (head / page_size);
1914 offset += shift;
1915 head -= shift;
1916
1917 remap:
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");
1922 exit(-1);
1923 }
1924
1925 more:
1926 event = (event_t *)(buf + head);
1927
1928 size = event->header.size;
1929 if (!size)
1930 size = 8;
1931
1932 if (head + event->header.size >= page_size * mmap_window) {
1933 int ret;
1934
1935 shift = page_size * (head / page_size);
1936
1937 ret = munmap(buf, page_size * mmap_window);
1938 assert(ret == 0);
1939
1940 offset += shift;
1941 head -= shift;
1942 goto remap;
1943 }
1944
1945 size = event->header.size;
1946
1947 dprintf("\n%p [%p]: event: %d\n",
1948 (void *)(offset + head),
1949 (void *)(long)event->header.size,
1950 event->header.type);
1951
1952 if (!size || process_event(event, offset, head) < 0) {
1953
1954 dprintf("%p [%p]: skipping unknown header type: %d\n",
1955 (void *)(offset + head),
1956 (void *)(long)(event->header.size),
1957 event->header.type);
1958
1959 total_unknown++;
1960
1961 /*
1962 * assume we lost track of the stream, check alignment, and
1963 * increment a single u64 in the hope to catch on again 'soon'.
1964 */
1965
1966 if (unlikely(head & 7))
1967 head &= ~7ULL;
1968
1969 size = 8;
1970 }
1971
1972 head += size;
1973
1974 if (offset + head >= header->data_offset + header->data_size)
1975 goto done;
1976
1977 if (offset + head < (unsigned long)stat.st_size)
1978 goto more;
1979
1980 done:
1981 rc = EXIT_SUCCESS;
1982 close(input);
1983
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);
1990
1991 if (dump_trace)
1992 return 0;
1993
1994 if (verbose >= 3)
1995 threads__fprintf(stdout);
1996
1997 if (verbose >= 2)
1998 dsos__fprintf(stdout);
1999
2000 collapse__resort();
2001 output__resort(total);
2002 output__fprintf(stdout, total);
2003
2004 return rc;
2005 }
2006
2007 static int
2008 parse_callchain_opt(const struct option *opt __used, const char *arg,
2009 int unset __used)
2010 {
2011 char *tok;
2012 char *endptr;
2013
2014 callchain = 1;
2015
2016 if (!arg)
2017 return 0;
2018
2019 tok = strtok((char *)arg, ",");
2020 if (!tok)
2021 return -1;
2022
2023 /* get the output mode */
2024 if (!strncmp(tok, "graph", strlen(arg)))
2025 callchain_param.mode = CHAIN_GRAPH_ABS;
2026
2027 else if (!strncmp(tok, "flat", strlen(arg)))
2028 callchain_param.mode = CHAIN_FLAT;
2029
2030 else if (!strncmp(tok, "fractal", strlen(arg)))
2031 callchain_param.mode = CHAIN_GRAPH_REL;
2032
2033 else if (!strncmp(tok, "none", strlen(arg))) {
2034 callchain_param.mode = CHAIN_NONE;
2035 callchain = 0;
2036
2037 return 0;
2038 }
2039
2040 else
2041 return -1;
2042
2043 /* get the min percentage */
2044 tok = strtok(NULL, ",");
2045 if (!tok)
2046 goto setup;
2047
2048 callchain_param.min_percent = strtod(tok, &endptr);
2049 if (tok == endptr)
2050 return -1;
2051
2052 setup:
2053 if (register_callchain_param(&callchain_param) < 0) {
2054 fprintf(stderr, "Can't register callchain params\n");
2055 return -1;
2056 }
2057 return 0;
2058 }
2059
2060 static const char * const report_usage[] = {
2061 "perf report [<options>] <command>",
2062 NULL
2063 };
2064
2065 static const struct option options[] = {
2066 OPT_STRING('i', "input", &input_name, "file",
2067 "input file name"),
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,
2096 "width[,width...]",
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."),
2101 OPT_END()
2102 };
2103
2104 static void setup_sorting(void)
2105 {
2106 char *tmp, *tok, *str = strdup(sort_order);
2107
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);
2113 }
2114 }
2115
2116 free(str);
2117 }
2118
2119 static void setup_list(struct strlist **list, const char *list_str,
2120 struct sort_entry *se, const char *list_name,
2121 FILE *fp)
2122 {
2123 if (list_str) {
2124 *list = strlist__new(true, list_str);
2125 if (!*list) {
2126 fprintf(stderr, "problems parsing %s list\n",
2127 list_name);
2128 exit(129);
2129 }
2130 if (strlist__nr_entries(*list) == 1) {
2131 fprintf(fp, "# %s: %s\n", list_name,
2132 strlist__entry(*list, 0)->s);
2133 se->elide = true;
2134 }
2135 }
2136 }
2137
2138 int cmd_report(int argc, const char **argv, const char *prefix __used)
2139 {
2140 symbol__init();
2141
2142 page_size = getpagesize();
2143
2144 argc = parse_options(argc, argv, options, report_usage, 0);
2145
2146 setup_sorting();
2147
2148 if (parent_pattern != default_parent_pattern) {
2149 sort_dimension__add("parent");
2150 sort_parent.elide = 1;
2151 } else
2152 exclude_other = 0;
2153
2154 /*
2155 * Any (unrecognized) arguments left?
2156 */
2157 if (argc)
2158 usage_with_options(report_usage, options);
2159
2160 setup_pager();
2161
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);
2165
2166 if (field_sep && *field_sep == '.') {
2167 fputs("'.' is the only non valid --field-separator argument\n",
2168 stderr);
2169 exit(129);
2170 }
2171
2172 return __cmd_report();
2173 }
Linux with added FPC-III support