perf-report: Add bare minimum PERF_EVENT_READ parsing
[firefly-linux-kernel-4.4.55.git] / tools / perf / builtin-report.c
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 "util/list.h"
14 #include "util/cache.h"
15 #include "util/rbtree.h"
16 #include "util/symbol.h"
17 #include "util/string.h"
18
19 #include "perf.h"
20 #include "util/header.h"
21
22 #include "util/parse-options.h"
23 #include "util/parse-events.h"
24
25 #define SHOW_KERNEL     1
26 #define SHOW_USER       2
27 #define SHOW_HV         4
28
29 static char             const *input_name = "perf.data";
30 static char             *vmlinux = NULL;
31
32 static char             default_sort_order[] = "comm,dso";
33 static char             *sort_order = default_sort_order;
34
35 static int              input;
36 static int              show_mask = SHOW_KERNEL | SHOW_USER | SHOW_HV;
37
38 static int              dump_trace = 0;
39 #define dprintf(x...)   do { if (dump_trace) printf(x); } while (0)
40 #define cdprintf(x...)  do { if (dump_trace) color_fprintf(stdout, color, x); } while (0)
41
42 static int              verbose;
43 #define eprintf(x...)   do { if (verbose) fprintf(stderr, x); } while (0)
44
45 static int              full_paths;
46
47 static unsigned long    page_size;
48 static unsigned long    mmap_window = 32;
49
50 static char             default_parent_pattern[] = "^sys_|^do_page_fault";
51 static char             *parent_pattern = default_parent_pattern;
52 static regex_t          parent_regex;
53
54 static int              exclude_other = 1;
55
56 static u64              sample_type;
57
58 struct ip_event {
59         struct perf_event_header header;
60         u64 ip;
61         u32 pid, tid;
62         unsigned char __more_data[];
63 };
64
65 struct ip_callchain {
66         u64 nr;
67         u64 ips[0];
68 };
69
70 struct mmap_event {
71         struct perf_event_header header;
72         u32 pid, tid;
73         u64 start;
74         u64 len;
75         u64 pgoff;
76         char filename[PATH_MAX];
77 };
78
79 struct comm_event {
80         struct perf_event_header header;
81         u32 pid, tid;
82         char comm[16];
83 };
84
85 struct fork_event {
86         struct perf_event_header header;
87         u32 pid, ppid;
88 };
89
90 struct period_event {
91         struct perf_event_header header;
92         u64 time;
93         u64 id;
94         u64 sample_period;
95 };
96
97 struct lost_event {
98         struct perf_event_header header;
99         u64 id;
100         u64 lost;
101 };
102
103 struct read_event {
104         struct perf_event_header header;
105         u32 pid,tid;
106         u64 value;
107         u64 format[3];
108 };
109
110 typedef union event_union {
111         struct perf_event_header        header;
112         struct ip_event                 ip;
113         struct mmap_event               mmap;
114         struct comm_event               comm;
115         struct fork_event               fork;
116         struct period_event             period;
117         struct lost_event               lost;
118         struct read_event               read;
119 } event_t;
120
121 static LIST_HEAD(dsos);
122 static struct dso *kernel_dso;
123 static struct dso *vdso;
124
125 static void dsos__add(struct dso *dso)
126 {
127         list_add_tail(&dso->node, &dsos);
128 }
129
130 static struct dso *dsos__find(const char *name)
131 {
132         struct dso *pos;
133
134         list_for_each_entry(pos, &dsos, node)
135                 if (strcmp(pos->name, name) == 0)
136                         return pos;
137         return NULL;
138 }
139
140 static struct dso *dsos__findnew(const char *name)
141 {
142         struct dso *dso = dsos__find(name);
143         int nr;
144
145         if (dso)
146                 return dso;
147
148         dso = dso__new(name, 0);
149         if (!dso)
150                 goto out_delete_dso;
151
152         nr = dso__load(dso, NULL, verbose);
153         if (nr < 0) {
154                 eprintf("Failed to open: %s\n", name);
155                 goto out_delete_dso;
156         }
157         if (!nr)
158                 eprintf("No symbols found in: %s, maybe install a debug package?\n", name);
159
160         dsos__add(dso);
161
162         return dso;
163
164 out_delete_dso:
165         dso__delete(dso);
166         return NULL;
167 }
168
169 static void dsos__fprintf(FILE *fp)
170 {
171         struct dso *pos;
172
173         list_for_each_entry(pos, &dsos, node)
174                 dso__fprintf(pos, fp);
175 }
176
177 static struct symbol *vdso__find_symbol(struct dso *dso, u64 ip)
178 {
179         return dso__find_symbol(kernel_dso, ip);
180 }
181
182 static int load_kernel(void)
183 {
184         int err;
185
186         kernel_dso = dso__new("[kernel]", 0);
187         if (!kernel_dso)
188                 return -1;
189
190         err = dso__load_kernel(kernel_dso, vmlinux, NULL, verbose);
191         if (err) {
192                 dso__delete(kernel_dso);
193                 kernel_dso = NULL;
194         } else
195                 dsos__add(kernel_dso);
196
197         vdso = dso__new("[vdso]", 0);
198         if (!vdso)
199                 return -1;
200
201         vdso->find_symbol = vdso__find_symbol;
202
203         dsos__add(vdso);
204
205         return err;
206 }
207
208 static char __cwd[PATH_MAX];
209 static char *cwd = __cwd;
210 static int cwdlen;
211
212 static int strcommon(const char *pathname)
213 {
214         int n = 0;
215
216         while (pathname[n] == cwd[n] && n < cwdlen)
217                 ++n;
218
219         return n;
220 }
221
222 struct map {
223         struct list_head node;
224         u64      start;
225         u64      end;
226         u64      pgoff;
227         u64      (*map_ip)(struct map *, u64);
228         struct dso       *dso;
229 };
230
231 static u64 map__map_ip(struct map *map, u64 ip)
232 {
233         return ip - map->start + map->pgoff;
234 }
235
236 static u64 vdso__map_ip(struct map *map, u64 ip)
237 {
238         return ip;
239 }
240
241 static inline int is_anon_memory(const char *filename)
242 {
243      return strcmp(filename, "//anon") == 0;
244 }
245
246 static struct map *map__new(struct mmap_event *event)
247 {
248         struct map *self = malloc(sizeof(*self));
249
250         if (self != NULL) {
251                 const char *filename = event->filename;
252                 char newfilename[PATH_MAX];
253                 int anon;
254
255                 if (cwd) {
256                         int n = strcommon(filename);
257
258                         if (n == cwdlen) {
259                                 snprintf(newfilename, sizeof(newfilename),
260                                          ".%s", filename + n);
261                                 filename = newfilename;
262                         }
263                 }
264
265                 anon = is_anon_memory(filename);
266
267                 if (anon) {
268                         snprintf(newfilename, sizeof(newfilename), "/tmp/perf-%d.map", event->pid);
269                         filename = newfilename;
270                 }
271
272                 self->start = event->start;
273                 self->end   = event->start + event->len;
274                 self->pgoff = event->pgoff;
275
276                 self->dso = dsos__findnew(filename);
277                 if (self->dso == NULL)
278                         goto out_delete;
279
280                 if (self->dso == vdso || anon)
281                         self->map_ip = vdso__map_ip;
282                 else
283                         self->map_ip = map__map_ip;
284         }
285         return self;
286 out_delete:
287         free(self);
288         return NULL;
289 }
290
291 static struct map *map__clone(struct map *self)
292 {
293         struct map *map = malloc(sizeof(*self));
294
295         if (!map)
296                 return NULL;
297
298         memcpy(map, self, sizeof(*self));
299
300         return map;
301 }
302
303 static int map__overlap(struct map *l, struct map *r)
304 {
305         if (l->start > r->start) {
306                 struct map *t = l;
307                 l = r;
308                 r = t;
309         }
310
311         if (l->end > r->start)
312                 return 1;
313
314         return 0;
315 }
316
317 static size_t map__fprintf(struct map *self, FILE *fp)
318 {
319         return fprintf(fp, " %Lx-%Lx %Lx %s\n",
320                        self->start, self->end, self->pgoff, self->dso->name);
321 }
322
323
324 struct thread {
325         struct rb_node   rb_node;
326         struct list_head maps;
327         pid_t            pid;
328         char             *comm;
329 };
330
331 static struct thread *thread__new(pid_t pid)
332 {
333         struct thread *self = malloc(sizeof(*self));
334
335         if (self != NULL) {
336                 self->pid = pid;
337                 self->comm = malloc(32);
338                 if (self->comm)
339                         snprintf(self->comm, 32, ":%d", self->pid);
340                 INIT_LIST_HEAD(&self->maps);
341         }
342
343         return self;
344 }
345
346 static int thread__set_comm(struct thread *self, const char *comm)
347 {
348         if (self->comm)
349                 free(self->comm);
350         self->comm = strdup(comm);
351         return self->comm ? 0 : -ENOMEM;
352 }
353
354 static size_t thread__fprintf(struct thread *self, FILE *fp)
355 {
356         struct map *pos;
357         size_t ret = fprintf(fp, "Thread %d %s\n", self->pid, self->comm);
358
359         list_for_each_entry(pos, &self->maps, node)
360                 ret += map__fprintf(pos, fp);
361
362         return ret;
363 }
364
365
366 static struct rb_root threads;
367 static struct thread *last_match;
368
369 static struct thread *threads__findnew(pid_t pid)
370 {
371         struct rb_node **p = &threads.rb_node;
372         struct rb_node *parent = NULL;
373         struct thread *th;
374
375         /*
376          * Font-end cache - PID lookups come in blocks,
377          * so most of the time we dont have to look up
378          * the full rbtree:
379          */
380         if (last_match && last_match->pid == pid)
381                 return last_match;
382
383         while (*p != NULL) {
384                 parent = *p;
385                 th = rb_entry(parent, struct thread, rb_node);
386
387                 if (th->pid == pid) {
388                         last_match = th;
389                         return th;
390                 }
391
392                 if (pid < th->pid)
393                         p = &(*p)->rb_left;
394                 else
395                         p = &(*p)->rb_right;
396         }
397
398         th = thread__new(pid);
399         if (th != NULL) {
400                 rb_link_node(&th->rb_node, parent, p);
401                 rb_insert_color(&th->rb_node, &threads);
402                 last_match = th;
403         }
404
405         return th;
406 }
407
408 static void thread__insert_map(struct thread *self, struct map *map)
409 {
410         struct map *pos, *tmp;
411
412         list_for_each_entry_safe(pos, tmp, &self->maps, node) {
413                 if (map__overlap(pos, map)) {
414                         if (verbose >= 2) {
415                                 printf("overlapping maps:\n");
416                                 map__fprintf(map, stdout);
417                                 map__fprintf(pos, stdout);
418                         }
419
420                         if (map->start <= pos->start && map->end > pos->start)
421                                 pos->start = map->end;
422
423                         if (map->end >= pos->end && map->start < pos->end)
424                                 pos->end = map->start;
425
426                         if (verbose >= 2) {
427                                 printf("after collision:\n");
428                                 map__fprintf(pos, stdout);
429                         }
430
431                         if (pos->start >= pos->end) {
432                                 list_del_init(&pos->node);
433                                 free(pos);
434                         }
435                 }
436         }
437
438         list_add_tail(&map->node, &self->maps);
439 }
440
441 static int thread__fork(struct thread *self, struct thread *parent)
442 {
443         struct map *map;
444
445         if (self->comm)
446                 free(self->comm);
447         self->comm = strdup(parent->comm);
448         if (!self->comm)
449                 return -ENOMEM;
450
451         list_for_each_entry(map, &parent->maps, node) {
452                 struct map *new = map__clone(map);
453                 if (!new)
454                         return -ENOMEM;
455                 thread__insert_map(self, new);
456         }
457
458         return 0;
459 }
460
461 static struct map *thread__find_map(struct thread *self, u64 ip)
462 {
463         struct map *pos;
464
465         if (self == NULL)
466                 return NULL;
467
468         list_for_each_entry(pos, &self->maps, node)
469                 if (ip >= pos->start && ip <= pos->end)
470                         return pos;
471
472         return NULL;
473 }
474
475 static size_t threads__fprintf(FILE *fp)
476 {
477         size_t ret = 0;
478         struct rb_node *nd;
479
480         for (nd = rb_first(&threads); nd; nd = rb_next(nd)) {
481                 struct thread *pos = rb_entry(nd, struct thread, rb_node);
482
483                 ret += thread__fprintf(pos, fp);
484         }
485
486         return ret;
487 }
488
489 /*
490  * histogram, sorted on item, collects counts
491  */
492
493 static struct rb_root hist;
494
495 struct hist_entry {
496         struct rb_node   rb_node;
497
498         struct thread    *thread;
499         struct map       *map;
500         struct dso       *dso;
501         struct symbol    *sym;
502         struct symbol    *parent;
503         u64              ip;
504         char             level;
505
506         u64              count;
507 };
508
509 /*
510  * configurable sorting bits
511  */
512
513 struct sort_entry {
514         struct list_head list;
515
516         char *header;
517
518         int64_t (*cmp)(struct hist_entry *, struct hist_entry *);
519         int64_t (*collapse)(struct hist_entry *, struct hist_entry *);
520         size_t  (*print)(FILE *fp, struct hist_entry *);
521 };
522
523 static int64_t cmp_null(void *l, void *r)
524 {
525         if (!l && !r)
526                 return 0;
527         else if (!l)
528                 return -1;
529         else
530                 return 1;
531 }
532
533 /* --sort pid */
534
535 static int64_t
536 sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
537 {
538         return right->thread->pid - left->thread->pid;
539 }
540
541 static size_t
542 sort__thread_print(FILE *fp, struct hist_entry *self)
543 {
544         return fprintf(fp, "%16s:%5d", self->thread->comm ?: "", self->thread->pid);
545 }
546
547 static struct sort_entry sort_thread = {
548         .header = "         Command:  Pid",
549         .cmp    = sort__thread_cmp,
550         .print  = sort__thread_print,
551 };
552
553 /* --sort comm */
554
555 static int64_t
556 sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
557 {
558         return right->thread->pid - left->thread->pid;
559 }
560
561 static int64_t
562 sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
563 {
564         char *comm_l = left->thread->comm;
565         char *comm_r = right->thread->comm;
566
567         if (!comm_l || !comm_r)
568                 return cmp_null(comm_l, comm_r);
569
570         return strcmp(comm_l, comm_r);
571 }
572
573 static size_t
574 sort__comm_print(FILE *fp, struct hist_entry *self)
575 {
576         return fprintf(fp, "%16s", self->thread->comm);
577 }
578
579 static struct sort_entry sort_comm = {
580         .header         = "         Command",
581         .cmp            = sort__comm_cmp,
582         .collapse       = sort__comm_collapse,
583         .print          = sort__comm_print,
584 };
585
586 /* --sort dso */
587
588 static int64_t
589 sort__dso_cmp(struct hist_entry *left, struct hist_entry *right)
590 {
591         struct dso *dso_l = left->dso;
592         struct dso *dso_r = right->dso;
593
594         if (!dso_l || !dso_r)
595                 return cmp_null(dso_l, dso_r);
596
597         return strcmp(dso_l->name, dso_r->name);
598 }
599
600 static size_t
601 sort__dso_print(FILE *fp, struct hist_entry *self)
602 {
603         if (self->dso)
604                 return fprintf(fp, "%-25s", self->dso->name);
605
606         return fprintf(fp, "%016llx         ", (u64)self->ip);
607 }
608
609 static struct sort_entry sort_dso = {
610         .header = "Shared Object            ",
611         .cmp    = sort__dso_cmp,
612         .print  = sort__dso_print,
613 };
614
615 /* --sort symbol */
616
617 static int64_t
618 sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
619 {
620         u64 ip_l, ip_r;
621
622         if (left->sym == right->sym)
623                 return 0;
624
625         ip_l = left->sym ? left->sym->start : left->ip;
626         ip_r = right->sym ? right->sym->start : right->ip;
627
628         return (int64_t)(ip_r - ip_l);
629 }
630
631 static size_t
632 sort__sym_print(FILE *fp, struct hist_entry *self)
633 {
634         size_t ret = 0;
635
636         if (verbose)
637                 ret += fprintf(fp, "%#018llx  ", (u64)self->ip);
638
639         if (self->sym) {
640                 ret += fprintf(fp, "[%c] %s",
641                         self->dso == kernel_dso ? 'k' : '.', self->sym->name);
642         } else {
643                 ret += fprintf(fp, "%#016llx", (u64)self->ip);
644         }
645
646         return ret;
647 }
648
649 static struct sort_entry sort_sym = {
650         .header = "Symbol",
651         .cmp    = sort__sym_cmp,
652         .print  = sort__sym_print,
653 };
654
655 /* --sort parent */
656
657 static int64_t
658 sort__parent_cmp(struct hist_entry *left, struct hist_entry *right)
659 {
660         struct symbol *sym_l = left->parent;
661         struct symbol *sym_r = right->parent;
662
663         if (!sym_l || !sym_r)
664                 return cmp_null(sym_l, sym_r);
665
666         return strcmp(sym_l->name, sym_r->name);
667 }
668
669 static size_t
670 sort__parent_print(FILE *fp, struct hist_entry *self)
671 {
672         size_t ret = 0;
673
674         ret += fprintf(fp, "%-20s", self->parent ? self->parent->name : "[other]");
675
676         return ret;
677 }
678
679 static struct sort_entry sort_parent = {
680         .header = "Parent symbol       ",
681         .cmp    = sort__parent_cmp,
682         .print  = sort__parent_print,
683 };
684
685 static int sort__need_collapse = 0;
686 static int sort__has_parent = 0;
687
688 struct sort_dimension {
689         char                    *name;
690         struct sort_entry       *entry;
691         int                     taken;
692 };
693
694 static struct sort_dimension sort_dimensions[] = {
695         { .name = "pid",        .entry = &sort_thread,  },
696         { .name = "comm",       .entry = &sort_comm,    },
697         { .name = "dso",        .entry = &sort_dso,     },
698         { .name = "symbol",     .entry = &sort_sym,     },
699         { .name = "parent",     .entry = &sort_parent,  },
700 };
701
702 static LIST_HEAD(hist_entry__sort_list);
703
704 static int sort_dimension__add(char *tok)
705 {
706         int i;
707
708         for (i = 0; i < ARRAY_SIZE(sort_dimensions); i++) {
709                 struct sort_dimension *sd = &sort_dimensions[i];
710
711                 if (sd->taken)
712                         continue;
713
714                 if (strncasecmp(tok, sd->name, strlen(tok)))
715                         continue;
716
717                 if (sd->entry->collapse)
718                         sort__need_collapse = 1;
719
720                 if (sd->entry == &sort_parent) {
721                         int ret = regcomp(&parent_regex, parent_pattern, REG_EXTENDED);
722                         if (ret) {
723                                 char err[BUFSIZ];
724
725                                 regerror(ret, &parent_regex, err, sizeof(err));
726                                 fprintf(stderr, "Invalid regex: %s\n%s",
727                                         parent_pattern, err);
728                                 exit(-1);
729                         }
730                         sort__has_parent = 1;
731                 }
732
733                 list_add_tail(&sd->entry->list, &hist_entry__sort_list);
734                 sd->taken = 1;
735
736                 return 0;
737         }
738
739         return -ESRCH;
740 }
741
742 static int64_t
743 hist_entry__cmp(struct hist_entry *left, struct hist_entry *right)
744 {
745         struct sort_entry *se;
746         int64_t cmp = 0;
747
748         list_for_each_entry(se, &hist_entry__sort_list, list) {
749                 cmp = se->cmp(left, right);
750                 if (cmp)
751                         break;
752         }
753
754         return cmp;
755 }
756
757 static int64_t
758 hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
759 {
760         struct sort_entry *se;
761         int64_t cmp = 0;
762
763         list_for_each_entry(se, &hist_entry__sort_list, list) {
764                 int64_t (*f)(struct hist_entry *, struct hist_entry *);
765
766                 f = se->collapse ?: se->cmp;
767
768                 cmp = f(left, right);
769                 if (cmp)
770                         break;
771         }
772
773         return cmp;
774 }
775
776 static size_t
777 hist_entry__fprintf(FILE *fp, struct hist_entry *self, u64 total_samples)
778 {
779         struct sort_entry *se;
780         size_t ret;
781
782         if (exclude_other && !self->parent)
783                 return 0;
784
785         if (total_samples) {
786                 double percent = self->count * 100.0 / total_samples;
787                 char *color = PERF_COLOR_NORMAL;
788
789                 /*
790                  * We color high-overhead entries in red, mid-overhead
791                  * entries in green - and keep the low overhead places
792                  * normal:
793                  */
794                 if (percent >= 5.0) {
795                         color = PERF_COLOR_RED;
796                 } else {
797                         if (percent >= 0.5)
798                                 color = PERF_COLOR_GREEN;
799                 }
800
801                 ret = color_fprintf(fp, color, "   %6.2f%%",
802                                 (self->count * 100.0) / total_samples);
803         } else
804                 ret = fprintf(fp, "%12Ld ", self->count);
805
806         list_for_each_entry(se, &hist_entry__sort_list, list) {
807                 if (exclude_other && (se == &sort_parent))
808                         continue;
809
810                 fprintf(fp, "  ");
811                 ret += se->print(fp, self);
812         }
813
814         ret += fprintf(fp, "\n");
815
816         return ret;
817 }
818
819 /*
820  *
821  */
822
823 static struct symbol *
824 resolve_symbol(struct thread *thread, struct map **mapp,
825                struct dso **dsop, u64 *ipp)
826 {
827         struct dso *dso = dsop ? *dsop : NULL;
828         struct map *map = mapp ? *mapp : NULL;
829         u64 ip = *ipp;
830
831         if (!thread)
832                 return NULL;
833
834         if (dso)
835                 goto got_dso;
836
837         if (map)
838                 goto got_map;
839
840         map = thread__find_map(thread, ip);
841         if (map != NULL) {
842                 if (mapp)
843                         *mapp = map;
844 got_map:
845                 ip = map->map_ip(map, ip);
846
847                 dso = map->dso;
848         } else {
849                 /*
850                  * If this is outside of all known maps,
851                  * and is a negative address, try to look it
852                  * up in the kernel dso, as it might be a
853                  * vsyscall (which executes in user-mode):
854                  */
855                 if ((long long)ip < 0)
856                 dso = kernel_dso;
857         }
858         dprintf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
859         dprintf(" ...... map: %Lx -> %Lx\n", *ipp, ip);
860         *ipp  = ip;
861
862         if (dsop)
863                 *dsop = dso;
864
865         if (!dso)
866                 return NULL;
867 got_dso:
868         return dso->find_symbol(dso, ip);
869 }
870
871 static int call__match(struct symbol *sym)
872 {
873         if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
874                 return 1;
875
876         return 0;
877 }
878
879 /*
880  * collect histogram counts
881  */
882
883 static int
884 hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
885                 struct symbol *sym, u64 ip, struct ip_callchain *chain,
886                 char level, u64 count)
887 {
888         struct rb_node **p = &hist.rb_node;
889         struct rb_node *parent = NULL;
890         struct hist_entry *he;
891         struct hist_entry entry = {
892                 .thread = thread,
893                 .map    = map,
894                 .dso    = dso,
895                 .sym    = sym,
896                 .ip     = ip,
897                 .level  = level,
898                 .count  = count,
899                 .parent = NULL,
900         };
901         int cmp;
902
903         if (sort__has_parent && chain) {
904                 u64 context = PERF_CONTEXT_MAX;
905                 int i;
906
907                 for (i = 0; i < chain->nr; i++) {
908                         u64 ip = chain->ips[i];
909                         struct dso *dso = NULL;
910                         struct symbol *sym;
911
912                         if (ip >= PERF_CONTEXT_MAX) {
913                                 context = ip;
914                                 continue;
915                         }
916
917                         switch (context) {
918                         case PERF_CONTEXT_KERNEL:
919                                 dso = kernel_dso;
920                                 break;
921                         default:
922                                 break;
923                         }
924
925                         sym = resolve_symbol(thread, NULL, &dso, &ip);
926
927                         if (sym && call__match(sym)) {
928                                 entry.parent = sym;
929                                 break;
930                         }
931                 }
932         }
933
934         while (*p != NULL) {
935                 parent = *p;
936                 he = rb_entry(parent, struct hist_entry, rb_node);
937
938                 cmp = hist_entry__cmp(&entry, he);
939
940                 if (!cmp) {
941                         he->count += count;
942                         return 0;
943                 }
944
945                 if (cmp < 0)
946                         p = &(*p)->rb_left;
947                 else
948                         p = &(*p)->rb_right;
949         }
950
951         he = malloc(sizeof(*he));
952         if (!he)
953                 return -ENOMEM;
954         *he = entry;
955         rb_link_node(&he->rb_node, parent, p);
956         rb_insert_color(&he->rb_node, &hist);
957
958         return 0;
959 }
960
961 static void hist_entry__free(struct hist_entry *he)
962 {
963         free(he);
964 }
965
966 /*
967  * collapse the histogram
968  */
969
970 static struct rb_root collapse_hists;
971
972 static void collapse__insert_entry(struct hist_entry *he)
973 {
974         struct rb_node **p = &collapse_hists.rb_node;
975         struct rb_node *parent = NULL;
976         struct hist_entry *iter;
977         int64_t cmp;
978
979         while (*p != NULL) {
980                 parent = *p;
981                 iter = rb_entry(parent, struct hist_entry, rb_node);
982
983                 cmp = hist_entry__collapse(iter, he);
984
985                 if (!cmp) {
986                         iter->count += he->count;
987                         hist_entry__free(he);
988                         return;
989                 }
990
991                 if (cmp < 0)
992                         p = &(*p)->rb_left;
993                 else
994                         p = &(*p)->rb_right;
995         }
996
997         rb_link_node(&he->rb_node, parent, p);
998         rb_insert_color(&he->rb_node, &collapse_hists);
999 }
1000
1001 static void collapse__resort(void)
1002 {
1003         struct rb_node *next;
1004         struct hist_entry *n;
1005
1006         if (!sort__need_collapse)
1007                 return;
1008
1009         next = rb_first(&hist);
1010         while (next) {
1011                 n = rb_entry(next, struct hist_entry, rb_node);
1012                 next = rb_next(&n->rb_node);
1013
1014                 rb_erase(&n->rb_node, &hist);
1015                 collapse__insert_entry(n);
1016         }
1017 }
1018
1019 /*
1020  * reverse the map, sort on count.
1021  */
1022
1023 static struct rb_root output_hists;
1024
1025 static void output__insert_entry(struct hist_entry *he)
1026 {
1027         struct rb_node **p = &output_hists.rb_node;
1028         struct rb_node *parent = NULL;
1029         struct hist_entry *iter;
1030
1031         while (*p != NULL) {
1032                 parent = *p;
1033                 iter = rb_entry(parent, struct hist_entry, rb_node);
1034
1035                 if (he->count > iter->count)
1036                         p = &(*p)->rb_left;
1037                 else
1038                         p = &(*p)->rb_right;
1039         }
1040
1041         rb_link_node(&he->rb_node, parent, p);
1042         rb_insert_color(&he->rb_node, &output_hists);
1043 }
1044
1045 static void output__resort(void)
1046 {
1047         struct rb_node *next;
1048         struct hist_entry *n;
1049         struct rb_root *tree = &hist;
1050
1051         if (sort__need_collapse)
1052                 tree = &collapse_hists;
1053
1054         next = rb_first(tree);
1055
1056         while (next) {
1057                 n = rb_entry(next, struct hist_entry, rb_node);
1058                 next = rb_next(&n->rb_node);
1059
1060                 rb_erase(&n->rb_node, tree);
1061                 output__insert_entry(n);
1062         }
1063 }
1064
1065 static size_t output__fprintf(FILE *fp, u64 total_samples)
1066 {
1067         struct hist_entry *pos;
1068         struct sort_entry *se;
1069         struct rb_node *nd;
1070         size_t ret = 0;
1071
1072         fprintf(fp, "\n");
1073         fprintf(fp, "#\n");
1074         fprintf(fp, "# (%Ld samples)\n", (u64)total_samples);
1075         fprintf(fp, "#\n");
1076
1077         fprintf(fp, "# Overhead");
1078         list_for_each_entry(se, &hist_entry__sort_list, list) {
1079                 if (exclude_other && (se == &sort_parent))
1080                         continue;
1081                 fprintf(fp, "  %s", se->header);
1082         }
1083         fprintf(fp, "\n");
1084
1085         fprintf(fp, "# ........");
1086         list_for_each_entry(se, &hist_entry__sort_list, list) {
1087                 int i;
1088
1089                 if (exclude_other && (se == &sort_parent))
1090                         continue;
1091
1092                 fprintf(fp, "  ");
1093                 for (i = 0; i < strlen(se->header); i++)
1094                         fprintf(fp, ".");
1095         }
1096         fprintf(fp, "\n");
1097
1098         fprintf(fp, "#\n");
1099
1100         for (nd = rb_first(&output_hists); nd; nd = rb_next(nd)) {
1101                 pos = rb_entry(nd, struct hist_entry, rb_node);
1102                 ret += hist_entry__fprintf(fp, pos, total_samples);
1103         }
1104
1105         if (sort_order == default_sort_order &&
1106                         parent_pattern == default_parent_pattern) {
1107                 fprintf(fp, "#\n");
1108                 fprintf(fp, "# (For more details, try: perf report --sort comm,dso,symbol)\n");
1109                 fprintf(fp, "#\n");
1110         }
1111         fprintf(fp, "\n");
1112
1113         return ret;
1114 }
1115
1116 static void register_idle_thread(void)
1117 {
1118         struct thread *thread = threads__findnew(0);
1119
1120         if (thread == NULL ||
1121                         thread__set_comm(thread, "[idle]")) {
1122                 fprintf(stderr, "problem inserting idle task.\n");
1123                 exit(-1);
1124         }
1125 }
1126
1127 static unsigned long total = 0,
1128                      total_mmap = 0,
1129                      total_comm = 0,
1130                      total_fork = 0,
1131                      total_unknown = 0,
1132                      total_lost = 0;
1133
1134 static int validate_chain(struct ip_callchain *chain, event_t *event)
1135 {
1136         unsigned int chain_size;
1137
1138         chain_size = event->header.size;
1139         chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;
1140
1141         if (chain->nr*sizeof(u64) > chain_size)
1142                 return -1;
1143
1144         return 0;
1145 }
1146
1147 static int
1148 process_sample_event(event_t *event, unsigned long offset, unsigned long head)
1149 {
1150         char level;
1151         int show = 0;
1152         struct dso *dso = NULL;
1153         struct thread *thread = threads__findnew(event->ip.pid);
1154         u64 ip = event->ip.ip;
1155         u64 period = 1;
1156         struct map *map = NULL;
1157         void *more_data = event->ip.__more_data;
1158         struct ip_callchain *chain = NULL;
1159
1160         if (sample_type & PERF_SAMPLE_PERIOD) {
1161                 period = *(u64 *)more_data;
1162                 more_data += sizeof(u64);
1163         }
1164
1165         dprintf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d: %p period: %Ld\n",
1166                 (void *)(offset + head),
1167                 (void *)(long)(event->header.size),
1168                 event->header.misc,
1169                 event->ip.pid,
1170                 (void *)(long)ip,
1171                 (long long)period);
1172
1173         if (sample_type & PERF_SAMPLE_CALLCHAIN) {
1174                 int i;
1175
1176                 chain = (void *)more_data;
1177
1178                 dprintf("... chain: nr:%Lu\n", chain->nr);
1179
1180                 if (validate_chain(chain, event) < 0) {
1181                         eprintf("call-chain problem with event, skipping it.\n");
1182                         return 0;
1183                 }
1184
1185                 if (dump_trace) {
1186                         for (i = 0; i < chain->nr; i++)
1187                                 dprintf("..... %2d: %016Lx\n", i, chain->ips[i]);
1188                 }
1189         }
1190
1191         dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid);
1192
1193         if (thread == NULL) {
1194                 eprintf("problem processing %d event, skipping it.\n",
1195                         event->header.type);
1196                 return -1;
1197         }
1198
1199         if (event->header.misc & PERF_EVENT_MISC_KERNEL) {
1200                 show = SHOW_KERNEL;
1201                 level = 'k';
1202
1203                 dso = kernel_dso;
1204
1205                 dprintf(" ...... dso: %s\n", dso->name);
1206
1207         } else if (event->header.misc & PERF_EVENT_MISC_USER) {
1208
1209                 show = SHOW_USER;
1210                 level = '.';
1211
1212         } else {
1213                 show = SHOW_HV;
1214                 level = 'H';
1215                 dprintf(" ...... dso: [hypervisor]\n");
1216         }
1217
1218         if (show & show_mask) {
1219                 struct symbol *sym = resolve_symbol(thread, &map, &dso, &ip);
1220
1221                 if (hist_entry__add(thread, map, dso, sym, ip, chain, level, period)) {
1222                         eprintf("problem incrementing symbol count, skipping event\n");
1223                         return -1;
1224                 }
1225         }
1226         total += period;
1227
1228         return 0;
1229 }
1230
1231 static int
1232 process_mmap_event(event_t *event, unsigned long offset, unsigned long head)
1233 {
1234         struct thread *thread = threads__findnew(event->mmap.pid);
1235         struct map *map = map__new(&event->mmap);
1236
1237         dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n",
1238                 (void *)(offset + head),
1239                 (void *)(long)(event->header.size),
1240                 event->mmap.pid,
1241                 (void *)(long)event->mmap.start,
1242                 (void *)(long)event->mmap.len,
1243                 (void *)(long)event->mmap.pgoff,
1244                 event->mmap.filename);
1245
1246         if (thread == NULL || map == NULL) {
1247                 dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1248                 return 0;
1249         }
1250
1251         thread__insert_map(thread, map);
1252         total_mmap++;
1253
1254         return 0;
1255 }
1256
1257 static int
1258 process_comm_event(event_t *event, unsigned long offset, unsigned long head)
1259 {
1260         struct thread *thread = threads__findnew(event->comm.pid);
1261
1262         dprintf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
1263                 (void *)(offset + head),
1264                 (void *)(long)(event->header.size),
1265                 event->comm.comm, event->comm.pid);
1266
1267         if (thread == NULL ||
1268             thread__set_comm(thread, event->comm.comm)) {
1269                 dprintf("problem processing PERF_EVENT_COMM, skipping event.\n");
1270                 return -1;
1271         }
1272         total_comm++;
1273
1274         return 0;
1275 }
1276
1277 static int
1278 process_fork_event(event_t *event, unsigned long offset, unsigned long head)
1279 {
1280         struct thread *thread = threads__findnew(event->fork.pid);
1281         struct thread *parent = threads__findnew(event->fork.ppid);
1282
1283         dprintf("%p [%p]: PERF_EVENT_FORK: %d:%d\n",
1284                 (void *)(offset + head),
1285                 (void *)(long)(event->header.size),
1286                 event->fork.pid, event->fork.ppid);
1287
1288         if (!thread || !parent || thread__fork(thread, parent)) {
1289                 dprintf("problem processing PERF_EVENT_FORK, skipping event.\n");
1290                 return -1;
1291         }
1292         total_fork++;
1293
1294         return 0;
1295 }
1296
1297 static int
1298 process_period_event(event_t *event, unsigned long offset, unsigned long head)
1299 {
1300         dprintf("%p [%p]: PERF_EVENT_PERIOD: time:%Ld, id:%Ld: period:%Ld\n",
1301                 (void *)(offset + head),
1302                 (void *)(long)(event->header.size),
1303                 event->period.time,
1304                 event->period.id,
1305                 event->period.sample_period);
1306
1307         return 0;
1308 }
1309
1310 static int
1311 process_lost_event(event_t *event, unsigned long offset, unsigned long head)
1312 {
1313         dprintf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n",
1314                 (void *)(offset + head),
1315                 (void *)(long)(event->header.size),
1316                 event->lost.id,
1317                 event->lost.lost);
1318
1319         total_lost += event->lost.lost;
1320
1321         return 0;
1322 }
1323
1324 static void trace_event(event_t *event)
1325 {
1326         unsigned char *raw_event = (void *)event;
1327         char *color = PERF_COLOR_BLUE;
1328         int i, j;
1329
1330         if (!dump_trace)
1331                 return;
1332
1333         dprintf(".");
1334         cdprintf("\n. ... raw event: size %d bytes\n", event->header.size);
1335
1336         for (i = 0; i < event->header.size; i++) {
1337                 if ((i & 15) == 0) {
1338                         dprintf(".");
1339                         cdprintf("  %04x: ", i);
1340                 }
1341
1342                 cdprintf(" %02x", raw_event[i]);
1343
1344                 if (((i & 15) == 15) || i == event->header.size-1) {
1345                         cdprintf("  ");
1346                         for (j = 0; j < 15-(i & 15); j++)
1347                                 cdprintf("   ");
1348                         for (j = 0; j < (i & 15); j++) {
1349                                 if (isprint(raw_event[i-15+j]))
1350                                         cdprintf("%c", raw_event[i-15+j]);
1351                                 else
1352                                         cdprintf(".");
1353                         }
1354                         cdprintf("\n");
1355                 }
1356         }
1357         dprintf(".\n");
1358 }
1359
1360 static int
1361 process_read_event(event_t *event, unsigned long offset, unsigned long head)
1362 {
1363         dprintf("%p [%p]: PERF_EVENT_READ: %d %d %Lu\n",
1364                         (void *)(offset + head),
1365                         (void *)(long)(event->header.size),
1366                         event->read.pid,
1367                         event->read.tid,
1368                         event->read.value);
1369
1370         return 0;
1371 }
1372
1373 static int
1374 process_event(event_t *event, unsigned long offset, unsigned long head)
1375 {
1376         trace_event(event);
1377
1378         switch (event->header.type) {
1379         case PERF_EVENT_SAMPLE:
1380                 return process_sample_event(event, offset, head);
1381
1382         case PERF_EVENT_MMAP:
1383                 return process_mmap_event(event, offset, head);
1384
1385         case PERF_EVENT_COMM:
1386                 return process_comm_event(event, offset, head);
1387
1388         case PERF_EVENT_FORK:
1389                 return process_fork_event(event, offset, head);
1390
1391         case PERF_EVENT_PERIOD:
1392                 return process_period_event(event, offset, head);
1393
1394         case PERF_EVENT_LOST:
1395                 return process_lost_event(event, offset, head);
1396
1397         case PERF_EVENT_READ:
1398                 return process_read_event(event, offset, head);
1399
1400         /*
1401          * We dont process them right now but they are fine:
1402          */
1403
1404         case PERF_EVENT_THROTTLE:
1405         case PERF_EVENT_UNTHROTTLE:
1406                 return 0;
1407
1408         default:
1409                 return -1;
1410         }
1411
1412         return 0;
1413 }
1414
1415 static struct perf_header       *header;
1416
1417 static u64 perf_header__sample_type(void)
1418 {
1419         u64 sample_type = 0;
1420         int i;
1421
1422         for (i = 0; i < header->attrs; i++) {
1423                 struct perf_header_attr *attr = header->attr[i];
1424
1425                 if (!sample_type)
1426                         sample_type = attr->attr.sample_type;
1427                 else if (sample_type != attr->attr.sample_type)
1428                         die("non matching sample_type");
1429         }
1430
1431         return sample_type;
1432 }
1433
1434 static int __cmd_report(void)
1435 {
1436         int ret, rc = EXIT_FAILURE;
1437         unsigned long offset = 0;
1438         unsigned long head, shift;
1439         struct stat stat;
1440         event_t *event;
1441         uint32_t size;
1442         char *buf;
1443
1444         register_idle_thread();
1445
1446         input = open(input_name, O_RDONLY);
1447         if (input < 0) {
1448                 fprintf(stderr, " failed to open file: %s", input_name);
1449                 if (!strcmp(input_name, "perf.data"))
1450                         fprintf(stderr, "  (try 'perf record' first)");
1451                 fprintf(stderr, "\n");
1452                 exit(-1);
1453         }
1454
1455         ret = fstat(input, &stat);
1456         if (ret < 0) {
1457                 perror("failed to stat file");
1458                 exit(-1);
1459         }
1460
1461         if (!stat.st_size) {
1462                 fprintf(stderr, "zero-sized file, nothing to do!\n");
1463                 exit(0);
1464         }
1465
1466         header = perf_header__read(input);
1467         head = header->data_offset;
1468
1469         sample_type = perf_header__sample_type();
1470
1471         if (sort__has_parent && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
1472                 fprintf(stderr, "selected --sort parent, but no callchain data\n");
1473                 exit(-1);
1474         }
1475
1476         if (load_kernel() < 0) {
1477                 perror("failed to load kernel symbols");
1478                 return EXIT_FAILURE;
1479         }
1480
1481         if (!full_paths) {
1482                 if (getcwd(__cwd, sizeof(__cwd)) == NULL) {
1483                         perror("failed to get the current directory");
1484                         return EXIT_FAILURE;
1485                 }
1486                 cwdlen = strlen(cwd);
1487         } else {
1488                 cwd = NULL;
1489                 cwdlen = 0;
1490         }
1491
1492         shift = page_size * (head / page_size);
1493         offset += shift;
1494         head -= shift;
1495
1496 remap:
1497         buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
1498                            MAP_SHARED, input, offset);
1499         if (buf == MAP_FAILED) {
1500                 perror("failed to mmap file");
1501                 exit(-1);
1502         }
1503
1504 more:
1505         event = (event_t *)(buf + head);
1506
1507         size = event->header.size;
1508         if (!size)
1509                 size = 8;
1510
1511         if (head + event->header.size >= page_size * mmap_window) {
1512                 int ret;
1513
1514                 shift = page_size * (head / page_size);
1515
1516                 ret = munmap(buf, page_size * mmap_window);
1517                 assert(ret == 0);
1518
1519                 offset += shift;
1520                 head -= shift;
1521                 goto remap;
1522         }
1523
1524         size = event->header.size;
1525
1526         dprintf("\n%p [%p]: event: %d\n",
1527                         (void *)(offset + head),
1528                         (void *)(long)event->header.size,
1529                         event->header.type);
1530
1531         if (!size || process_event(event, offset, head) < 0) {
1532
1533                 dprintf("%p [%p]: skipping unknown header type: %d\n",
1534                         (void *)(offset + head),
1535                         (void *)(long)(event->header.size),
1536                         event->header.type);
1537
1538                 total_unknown++;
1539
1540                 /*
1541                  * assume we lost track of the stream, check alignment, and
1542                  * increment a single u64 in the hope to catch on again 'soon'.
1543                  */
1544
1545                 if (unlikely(head & 7))
1546                         head &= ~7ULL;
1547
1548                 size = 8;
1549         }
1550
1551         head += size;
1552
1553         if (offset + head >= header->data_offset + header->data_size)
1554                 goto done;
1555
1556         if (offset + head < stat.st_size)
1557                 goto more;
1558
1559 done:
1560         rc = EXIT_SUCCESS;
1561         close(input);
1562
1563         dprintf("      IP events: %10ld\n", total);
1564         dprintf("    mmap events: %10ld\n", total_mmap);
1565         dprintf("    comm events: %10ld\n", total_comm);
1566         dprintf("    fork events: %10ld\n", total_fork);
1567         dprintf("    lost events: %10ld\n", total_lost);
1568         dprintf(" unknown events: %10ld\n", total_unknown);
1569
1570         if (dump_trace)
1571                 return 0;
1572
1573         if (verbose >= 3)
1574                 threads__fprintf(stdout);
1575
1576         if (verbose >= 2)
1577                 dsos__fprintf(stdout);
1578
1579         collapse__resort();
1580         output__resort();
1581         output__fprintf(stdout, total);
1582
1583         return rc;
1584 }
1585
1586 static const char * const report_usage[] = {
1587         "perf report [<options>] <command>",
1588         NULL
1589 };
1590
1591 static const struct option options[] = {
1592         OPT_STRING('i', "input", &input_name, "file",
1593                     "input file name"),
1594         OPT_BOOLEAN('v', "verbose", &verbose,
1595                     "be more verbose (show symbol address, etc)"),
1596         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1597                     "dump raw trace in ASCII"),
1598         OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"),
1599         OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1600                    "sort by key(s): pid, comm, dso, symbol, parent"),
1601         OPT_BOOLEAN('P', "full-paths", &full_paths,
1602                     "Don't shorten the pathnames taking into account the cwd"),
1603         OPT_STRING('p', "parent", &parent_pattern, "regex",
1604                    "regex filter to identify parent, see: '--sort parent'"),
1605         OPT_BOOLEAN('x', "exclude-other", &exclude_other,
1606                     "Only display entries with parent-match"),
1607         OPT_END()
1608 };
1609
1610 static void setup_sorting(void)
1611 {
1612         char *tmp, *tok, *str = strdup(sort_order);
1613
1614         for (tok = strtok_r(str, ", ", &tmp);
1615                         tok; tok = strtok_r(NULL, ", ", &tmp)) {
1616                 if (sort_dimension__add(tok) < 0) {
1617                         error("Unknown --sort key: `%s'", tok);
1618                         usage_with_options(report_usage, options);
1619                 }
1620         }
1621
1622         free(str);
1623 }
1624
1625 int cmd_report(int argc, const char **argv, const char *prefix)
1626 {
1627         symbol__init();
1628
1629         page_size = getpagesize();
1630
1631         argc = parse_options(argc, argv, options, report_usage, 0);
1632
1633         setup_sorting();
1634
1635         if (parent_pattern != default_parent_pattern)
1636                 sort_dimension__add("parent");
1637         else
1638                 exclude_other = 0;
1639
1640         /*
1641          * Any (unrecognized) arguments left?
1642          */
1643         if (argc)
1644                 usage_with_options(report_usage, options);
1645
1646         setup_pager();
1647
1648         return __cmd_report();
1649 }