perf_counter tools: Standardize color printing
[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
21 #include "util/parse-options.h"
22 #include "util/parse-events.h"
23
24 #define SHOW_KERNEL     1
25 #define SHOW_USER       2
26 #define SHOW_HV         4
27
28 static char             const *input_name = "perf.data";
29 static char             *vmlinux = NULL;
30
31 static char             default_sort_order[] = "comm,dso";
32 static char             *sort_order = default_sort_order;
33
34 static int              input;
35 static int              show_mask = SHOW_KERNEL | SHOW_USER | SHOW_HV;
36
37 static int              dump_trace = 0;
38 #define dprintf(x...)   do { if (dump_trace) printf(x); } while (0)
39
40 static int              verbose;
41 static int              full_paths;
42
43 static unsigned long    page_size;
44 static unsigned long    mmap_window = 32;
45
46 struct ip_event {
47         struct perf_event_header header;
48         __u64 ip;
49         __u32 pid, tid;
50 };
51
52 struct mmap_event {
53         struct perf_event_header header;
54         __u32 pid, tid;
55         __u64 start;
56         __u64 len;
57         __u64 pgoff;
58         char filename[PATH_MAX];
59 };
60
61 struct comm_event {
62         struct perf_event_header header;
63         __u32 pid, tid;
64         char comm[16];
65 };
66
67 struct fork_event {
68         struct perf_event_header header;
69         __u32 pid, ppid;
70 };
71
72 struct period_event {
73         struct perf_event_header header;
74         __u64 time;
75         __u64 id;
76         __u64 sample_period;
77 };
78
79 typedef union event_union {
80         struct perf_event_header        header;
81         struct ip_event                 ip;
82         struct mmap_event               mmap;
83         struct comm_event               comm;
84         struct fork_event               fork;
85         struct period_event             period;
86 } event_t;
87
88 static LIST_HEAD(dsos);
89 static struct dso *kernel_dso;
90 static struct dso *vdso;
91
92 static void dsos__add(struct dso *dso)
93 {
94         list_add_tail(&dso->node, &dsos);
95 }
96
97 static struct dso *dsos__find(const char *name)
98 {
99         struct dso *pos;
100
101         list_for_each_entry(pos, &dsos, node)
102                 if (strcmp(pos->name, name) == 0)
103                         return pos;
104         return NULL;
105 }
106
107 static struct dso *dsos__findnew(const char *name)
108 {
109         struct dso *dso = dsos__find(name);
110         int nr;
111
112         if (dso)
113                 return dso;
114
115         dso = dso__new(name, 0);
116         if (!dso)
117                 goto out_delete_dso;
118
119         nr = dso__load(dso, NULL, verbose);
120         if (nr < 0) {
121                 if (verbose)
122                         fprintf(stderr, "Failed to open: %s\n", name);
123                 goto out_delete_dso;
124         }
125         if (!nr && verbose) {
126                 fprintf(stderr,
127                 "No symbols found in: %s, maybe install a debug package?\n",
128                                 name);
129         }
130
131         dsos__add(dso);
132
133         return dso;
134
135 out_delete_dso:
136         dso__delete(dso);
137         return NULL;
138 }
139
140 static void dsos__fprintf(FILE *fp)
141 {
142         struct dso *pos;
143
144         list_for_each_entry(pos, &dsos, node)
145                 dso__fprintf(pos, fp);
146 }
147
148 static struct symbol *vdso__find_symbol(struct dso *dso, uint64_t ip)
149 {
150         return dso__find_symbol(kernel_dso, ip);
151 }
152
153 static int load_kernel(void)
154 {
155         int err;
156
157         kernel_dso = dso__new("[kernel]", 0);
158         if (!kernel_dso)
159                 return -1;
160
161         err = dso__load_kernel(kernel_dso, vmlinux, NULL, verbose);
162         if (err) {
163                 dso__delete(kernel_dso);
164                 kernel_dso = NULL;
165         } else
166                 dsos__add(kernel_dso);
167
168         vdso = dso__new("[vdso]", 0);
169         if (!vdso)
170                 return -1;
171
172         vdso->find_symbol = vdso__find_symbol;
173
174         dsos__add(vdso);
175
176         return err;
177 }
178
179 static char __cwd[PATH_MAX];
180 static char *cwd = __cwd;
181 static int cwdlen;
182
183 static int strcommon(const char *pathname)
184 {
185         int n = 0;
186
187         while (pathname[n] == cwd[n] && n < cwdlen)
188                 ++n;
189
190         return n;
191 }
192
193 struct map {
194         struct list_head node;
195         uint64_t         start;
196         uint64_t         end;
197         uint64_t         pgoff;
198         uint64_t         (*map_ip)(struct map *, uint64_t);
199         struct dso       *dso;
200 };
201
202 static uint64_t map__map_ip(struct map *map, uint64_t ip)
203 {
204         return ip - map->start + map->pgoff;
205 }
206
207 static uint64_t vdso__map_ip(struct map *map, uint64_t ip)
208 {
209         return ip;
210 }
211
212 static inline int is_anon_memory(const char *filename)
213 {
214      return strcmp(filename, "//anon") == 0;
215 }
216
217 static struct map *map__new(struct mmap_event *event)
218 {
219         struct map *self = malloc(sizeof(*self));
220
221         if (self != NULL) {
222                 const char *filename = event->filename;
223                 char newfilename[PATH_MAX];
224                 int anon;
225
226                 if (cwd) {
227                         int n = strcommon(filename);
228
229                         if (n == cwdlen) {
230                                 snprintf(newfilename, sizeof(newfilename),
231                                          ".%s", filename + n);
232                                 filename = newfilename;
233                         }
234                 }
235
236                 anon = is_anon_memory(filename);
237
238                 if (anon) {
239                         snprintf(newfilename, sizeof(newfilename), "/tmp/perf-%d.map", event->pid);
240                         filename = newfilename;
241                 }
242
243                 self->start = event->start;
244                 self->end   = event->start + event->len;
245                 self->pgoff = event->pgoff;
246
247                 self->dso = dsos__findnew(filename);
248                 if (self->dso == NULL)
249                         goto out_delete;
250
251                 if (self->dso == vdso || anon)
252                         self->map_ip = vdso__map_ip;
253                 else
254                         self->map_ip = map__map_ip;
255         }
256         return self;
257 out_delete:
258         free(self);
259         return NULL;
260 }
261
262 static struct map *map__clone(struct map *self)
263 {
264         struct map *map = malloc(sizeof(*self));
265
266         if (!map)
267                 return NULL;
268
269         memcpy(map, self, sizeof(*self));
270
271         return map;
272 }
273
274 static int map__overlap(struct map *l, struct map *r)
275 {
276         if (l->start > r->start) {
277                 struct map *t = l;
278                 l = r;
279                 r = t;
280         }
281
282         if (l->end > r->start)
283                 return 1;
284
285         return 0;
286 }
287
288 static size_t map__fprintf(struct map *self, FILE *fp)
289 {
290         return fprintf(fp, " %"PRIx64"-%"PRIx64" %"PRIx64" %s\n",
291                        self->start, self->end, self->pgoff, self->dso->name);
292 }
293
294
295 struct thread {
296         struct rb_node   rb_node;
297         struct list_head maps;
298         pid_t            pid;
299         char             *comm;
300 };
301
302 static struct thread *thread__new(pid_t pid)
303 {
304         struct thread *self = malloc(sizeof(*self));
305
306         if (self != NULL) {
307                 self->pid = pid;
308                 self->comm = malloc(32);
309                 if (self->comm)
310                         snprintf(self->comm, 32, ":%d", self->pid);
311                 INIT_LIST_HEAD(&self->maps);
312         }
313
314         return self;
315 }
316
317 static int thread__set_comm(struct thread *self, const char *comm)
318 {
319         if (self->comm)
320                 free(self->comm);
321         self->comm = strdup(comm);
322         return self->comm ? 0 : -ENOMEM;
323 }
324
325 static size_t thread__fprintf(struct thread *self, FILE *fp)
326 {
327         struct map *pos;
328         size_t ret = fprintf(fp, "Thread %d %s\n", self->pid, self->comm);
329
330         list_for_each_entry(pos, &self->maps, node)
331                 ret += map__fprintf(pos, fp);
332
333         return ret;
334 }
335
336
337 static struct rb_root threads;
338 static struct thread *last_match;
339
340 static struct thread *threads__findnew(pid_t pid)
341 {
342         struct rb_node **p = &threads.rb_node;
343         struct rb_node *parent = NULL;
344         struct thread *th;
345
346         /*
347          * Font-end cache - PID lookups come in blocks,
348          * so most of the time we dont have to look up
349          * the full rbtree:
350          */
351         if (last_match && last_match->pid == pid)
352                 return last_match;
353
354         while (*p != NULL) {
355                 parent = *p;
356                 th = rb_entry(parent, struct thread, rb_node);
357
358                 if (th->pid == pid) {
359                         last_match = th;
360                         return th;
361                 }
362
363                 if (pid < th->pid)
364                         p = &(*p)->rb_left;
365                 else
366                         p = &(*p)->rb_right;
367         }
368
369         th = thread__new(pid);
370         if (th != NULL) {
371                 rb_link_node(&th->rb_node, parent, p);
372                 rb_insert_color(&th->rb_node, &threads);
373                 last_match = th;
374         }
375
376         return th;
377 }
378
379 static void thread__insert_map(struct thread *self, struct map *map)
380 {
381         struct map *pos, *tmp;
382
383         list_for_each_entry_safe(pos, tmp, &self->maps, node) {
384                 if (map__overlap(pos, map)) {
385                         list_del_init(&pos->node);
386                         /* XXX leaks dsos */
387                         free(pos);
388                 }
389         }
390
391         list_add_tail(&map->node, &self->maps);
392 }
393
394 static int thread__fork(struct thread *self, struct thread *parent)
395 {
396         struct map *map;
397
398         if (self->comm)
399                 free(self->comm);
400         self->comm = strdup(parent->comm);
401         if (!self->comm)
402                 return -ENOMEM;
403
404         list_for_each_entry(map, &parent->maps, node) {
405                 struct map *new = map__clone(map);
406                 if (!new)
407                         return -ENOMEM;
408                 thread__insert_map(self, new);
409         }
410
411         return 0;
412 }
413
414 static struct map *thread__find_map(struct thread *self, uint64_t ip)
415 {
416         struct map *pos;
417
418         if (self == NULL)
419                 return NULL;
420
421         list_for_each_entry(pos, &self->maps, node)
422                 if (ip >= pos->start && ip <= pos->end)
423                         return pos;
424
425         return NULL;
426 }
427
428 static size_t threads__fprintf(FILE *fp)
429 {
430         size_t ret = 0;
431         struct rb_node *nd;
432
433         for (nd = rb_first(&threads); nd; nd = rb_next(nd)) {
434                 struct thread *pos = rb_entry(nd, struct thread, rb_node);
435
436                 ret += thread__fprintf(pos, fp);
437         }
438
439         return ret;
440 }
441
442 /*
443  * histogram, sorted on item, collects counts
444  */
445
446 static struct rb_root hist;
447
448 struct hist_entry {
449         struct rb_node   rb_node;
450
451         struct thread    *thread;
452         struct map       *map;
453         struct dso       *dso;
454         struct symbol    *sym;
455         uint64_t         ip;
456         char             level;
457
458         uint32_t         count;
459 };
460
461 /*
462  * configurable sorting bits
463  */
464
465 struct sort_entry {
466         struct list_head list;
467
468         char *header;
469
470         int64_t (*cmp)(struct hist_entry *, struct hist_entry *);
471         int64_t (*collapse)(struct hist_entry *, struct hist_entry *);
472         size_t  (*print)(FILE *fp, struct hist_entry *);
473 };
474
475 /* --sort pid */
476
477 static int64_t
478 sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
479 {
480         return right->thread->pid - left->thread->pid;
481 }
482
483 static size_t
484 sort__thread_print(FILE *fp, struct hist_entry *self)
485 {
486         return fprintf(fp, "%16s:%5d", self->thread->comm ?: "", self->thread->pid);
487 }
488
489 static struct sort_entry sort_thread = {
490         .header = "         Command:  Pid",
491         .cmp    = sort__thread_cmp,
492         .print  = sort__thread_print,
493 };
494
495 /* --sort comm */
496
497 static int64_t
498 sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
499 {
500         return right->thread->pid - left->thread->pid;
501 }
502
503 static int64_t
504 sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
505 {
506         char *comm_l = left->thread->comm;
507         char *comm_r = right->thread->comm;
508
509         if (!comm_l || !comm_r) {
510                 if (!comm_l && !comm_r)
511                         return 0;
512                 else if (!comm_l)
513                         return -1;
514                 else
515                         return 1;
516         }
517
518         return strcmp(comm_l, comm_r);
519 }
520
521 static size_t
522 sort__comm_print(FILE *fp, struct hist_entry *self)
523 {
524         return fprintf(fp, "%16s", self->thread->comm);
525 }
526
527 static struct sort_entry sort_comm = {
528         .header         = "         Command",
529         .cmp            = sort__comm_cmp,
530         .collapse       = sort__comm_collapse,
531         .print          = sort__comm_print,
532 };
533
534 /* --sort dso */
535
536 static int64_t
537 sort__dso_cmp(struct hist_entry *left, struct hist_entry *right)
538 {
539         struct dso *dso_l = left->dso;
540         struct dso *dso_r = right->dso;
541
542         if (!dso_l || !dso_r) {
543                 if (!dso_l && !dso_r)
544                         return 0;
545                 else if (!dso_l)
546                         return -1;
547                 else
548                         return 1;
549         }
550
551         return strcmp(dso_l->name, dso_r->name);
552 }
553
554 static size_t
555 sort__dso_print(FILE *fp, struct hist_entry *self)
556 {
557         if (self->dso)
558                 return fprintf(fp, "%-25s", self->dso->name);
559
560         return fprintf(fp, "%016llx         ", (__u64)self->ip);
561 }
562
563 static struct sort_entry sort_dso = {
564         .header = "Shared Object            ",
565         .cmp    = sort__dso_cmp,
566         .print  = sort__dso_print,
567 };
568
569 /* --sort symbol */
570
571 static int64_t
572 sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
573 {
574         uint64_t ip_l, ip_r;
575
576         if (left->sym == right->sym)
577                 return 0;
578
579         ip_l = left->sym ? left->sym->start : left->ip;
580         ip_r = right->sym ? right->sym->start : right->ip;
581
582         return (int64_t)(ip_r - ip_l);
583 }
584
585 static size_t
586 sort__sym_print(FILE *fp, struct hist_entry *self)
587 {
588         size_t ret = 0;
589
590         if (verbose)
591                 ret += fprintf(fp, "%#018llx  ", (__u64)self->ip);
592
593         if (self->sym) {
594                 ret += fprintf(fp, "[%c] %s",
595                         self->dso == kernel_dso ? 'k' : '.', self->sym->name);
596         } else {
597                 ret += fprintf(fp, "%#016llx", (__u64)self->ip);
598         }
599
600         return ret;
601 }
602
603 static struct sort_entry sort_sym = {
604         .header = "Symbol",
605         .cmp    = sort__sym_cmp,
606         .print  = sort__sym_print,
607 };
608
609 static int sort__need_collapse = 0;
610
611 struct sort_dimension {
612         char                    *name;
613         struct sort_entry       *entry;
614         int                     taken;
615 };
616
617 static struct sort_dimension sort_dimensions[] = {
618         { .name = "pid",        .entry = &sort_thread,  },
619         { .name = "comm",       .entry = &sort_comm,    },
620         { .name = "dso",        .entry = &sort_dso,     },
621         { .name = "symbol",     .entry = &sort_sym,     },
622 };
623
624 static LIST_HEAD(hist_entry__sort_list);
625
626 static int sort_dimension__add(char *tok)
627 {
628         int i;
629
630         for (i = 0; i < ARRAY_SIZE(sort_dimensions); i++) {
631                 struct sort_dimension *sd = &sort_dimensions[i];
632
633                 if (sd->taken)
634                         continue;
635
636                 if (strncasecmp(tok, sd->name, strlen(tok)))
637                         continue;
638
639                 if (sd->entry->collapse)
640                         sort__need_collapse = 1;
641
642                 list_add_tail(&sd->entry->list, &hist_entry__sort_list);
643                 sd->taken = 1;
644
645                 return 0;
646         }
647
648         return -ESRCH;
649 }
650
651 static int64_t
652 hist_entry__cmp(struct hist_entry *left, struct hist_entry *right)
653 {
654         struct sort_entry *se;
655         int64_t cmp = 0;
656
657         list_for_each_entry(se, &hist_entry__sort_list, list) {
658                 cmp = se->cmp(left, right);
659                 if (cmp)
660                         break;
661         }
662
663         return cmp;
664 }
665
666 static int64_t
667 hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
668 {
669         struct sort_entry *se;
670         int64_t cmp = 0;
671
672         list_for_each_entry(se, &hist_entry__sort_list, list) {
673                 int64_t (*f)(struct hist_entry *, struct hist_entry *);
674
675                 f = se->collapse ?: se->cmp;
676
677                 cmp = f(left, right);
678                 if (cmp)
679                         break;
680         }
681
682         return cmp;
683 }
684
685 static size_t
686 hist_entry__fprintf(FILE *fp, struct hist_entry *self, uint64_t total_samples)
687 {
688         struct sort_entry *se;
689         size_t ret;
690
691         if (total_samples) {
692                 double percent = self->count * 100.0 / total_samples;
693                 char *color = PERF_COLOR_NORMAL;
694
695                 /*
696                  * We color high-overhead entries in red, mid-overhead
697                  * entries in green - and keep the low overhead places
698                  * normal:
699                  */
700                 if (percent >= 5.0) {
701                         color = PERF_COLOR_RED;
702                 } else {
703                         if (percent >= 0.5)
704                                 color = PERF_COLOR_GREEN;
705                 }
706
707                 ret = color_fprintf(fp, color, "   %6.2f%%",
708                                 (self->count * 100.0) / total_samples);
709         } else
710                 ret = fprintf(fp, "%12d ", self->count);
711
712         list_for_each_entry(se, &hist_entry__sort_list, list) {
713                 fprintf(fp, "  ");
714                 ret += se->print(fp, self);
715         }
716
717         ret += fprintf(fp, "\n");
718
719         return ret;
720 }
721
722 /*
723  * collect histogram counts
724  */
725
726 static int
727 hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
728                 struct symbol *sym, uint64_t ip, char level)
729 {
730         struct rb_node **p = &hist.rb_node;
731         struct rb_node *parent = NULL;
732         struct hist_entry *he;
733         struct hist_entry entry = {
734                 .thread = thread,
735                 .map    = map,
736                 .dso    = dso,
737                 .sym    = sym,
738                 .ip     = ip,
739                 .level  = level,
740                 .count  = 1,
741         };
742         int cmp;
743
744         while (*p != NULL) {
745                 parent = *p;
746                 he = rb_entry(parent, struct hist_entry, rb_node);
747
748                 cmp = hist_entry__cmp(&entry, he);
749
750                 if (!cmp) {
751                         he->count++;
752                         return 0;
753                 }
754
755                 if (cmp < 0)
756                         p = &(*p)->rb_left;
757                 else
758                         p = &(*p)->rb_right;
759         }
760
761         he = malloc(sizeof(*he));
762         if (!he)
763                 return -ENOMEM;
764         *he = entry;
765         rb_link_node(&he->rb_node, parent, p);
766         rb_insert_color(&he->rb_node, &hist);
767
768         return 0;
769 }
770
771 static void hist_entry__free(struct hist_entry *he)
772 {
773         free(he);
774 }
775
776 /*
777  * collapse the histogram
778  */
779
780 static struct rb_root collapse_hists;
781
782 static void collapse__insert_entry(struct hist_entry *he)
783 {
784         struct rb_node **p = &collapse_hists.rb_node;
785         struct rb_node *parent = NULL;
786         struct hist_entry *iter;
787         int64_t cmp;
788
789         while (*p != NULL) {
790                 parent = *p;
791                 iter = rb_entry(parent, struct hist_entry, rb_node);
792
793                 cmp = hist_entry__collapse(iter, he);
794
795                 if (!cmp) {
796                         iter->count += he->count;
797                         hist_entry__free(he);
798                         return;
799                 }
800
801                 if (cmp < 0)
802                         p = &(*p)->rb_left;
803                 else
804                         p = &(*p)->rb_right;
805         }
806
807         rb_link_node(&he->rb_node, parent, p);
808         rb_insert_color(&he->rb_node, &collapse_hists);
809 }
810
811 static void collapse__resort(void)
812 {
813         struct rb_node *next;
814         struct hist_entry *n;
815
816         if (!sort__need_collapse)
817                 return;
818
819         next = rb_first(&hist);
820         while (next) {
821                 n = rb_entry(next, struct hist_entry, rb_node);
822                 next = rb_next(&n->rb_node);
823
824                 rb_erase(&n->rb_node, &hist);
825                 collapse__insert_entry(n);
826         }
827 }
828
829 /*
830  * reverse the map, sort on count.
831  */
832
833 static struct rb_root output_hists;
834
835 static void output__insert_entry(struct hist_entry *he)
836 {
837         struct rb_node **p = &output_hists.rb_node;
838         struct rb_node *parent = NULL;
839         struct hist_entry *iter;
840
841         while (*p != NULL) {
842                 parent = *p;
843                 iter = rb_entry(parent, struct hist_entry, rb_node);
844
845                 if (he->count > iter->count)
846                         p = &(*p)->rb_left;
847                 else
848                         p = &(*p)->rb_right;
849         }
850
851         rb_link_node(&he->rb_node, parent, p);
852         rb_insert_color(&he->rb_node, &output_hists);
853 }
854
855 static void output__resort(void)
856 {
857         struct rb_node *next;
858         struct hist_entry *n;
859         struct rb_root *tree = &hist;
860
861         if (sort__need_collapse)
862                 tree = &collapse_hists;
863
864         next = rb_first(tree);
865
866         while (next) {
867                 n = rb_entry(next, struct hist_entry, rb_node);
868                 next = rb_next(&n->rb_node);
869
870                 rb_erase(&n->rb_node, tree);
871                 output__insert_entry(n);
872         }
873 }
874
875 static size_t output__fprintf(FILE *fp, uint64_t total_samples)
876 {
877         struct hist_entry *pos;
878         struct sort_entry *se;
879         struct rb_node *nd;
880         size_t ret = 0;
881
882         fprintf(fp, "\n");
883         fprintf(fp, "#\n");
884         fprintf(fp, "# (%Ld samples)\n", (__u64)total_samples);
885         fprintf(fp, "#\n");
886
887         fprintf(fp, "# Overhead");
888         list_for_each_entry(se, &hist_entry__sort_list, list)
889                 fprintf(fp, "  %s", se->header);
890         fprintf(fp, "\n");
891
892         fprintf(fp, "# ........");
893         list_for_each_entry(se, &hist_entry__sort_list, list) {
894                 int i;
895
896                 fprintf(fp, "  ");
897                 for (i = 0; i < strlen(se->header); i++)
898                         fprintf(fp, ".");
899         }
900         fprintf(fp, "\n");
901
902         fprintf(fp, "#\n");
903
904         for (nd = rb_first(&output_hists); nd; nd = rb_next(nd)) {
905                 pos = rb_entry(nd, struct hist_entry, rb_node);
906                 ret += hist_entry__fprintf(fp, pos, total_samples);
907         }
908
909         if (!strcmp(sort_order, default_sort_order)) {
910                 fprintf(fp, "#\n");
911                 fprintf(fp, "# (For more details, try: perf report --sort comm,dso,symbol)\n");
912                 fprintf(fp, "#\n");
913         }
914         fprintf(fp, "\n");
915
916         return ret;
917 }
918
919 static void register_idle_thread(void)
920 {
921         struct thread *thread = threads__findnew(0);
922
923         if (thread == NULL ||
924                         thread__set_comm(thread, "[idle]")) {
925                 fprintf(stderr, "problem inserting idle task.\n");
926                 exit(-1);
927         }
928 }
929
930 static unsigned long total = 0,
931                      total_mmap = 0,
932                      total_comm = 0,
933                      total_fork = 0,
934                      total_unknown = 0;
935
936 static int
937 process_overflow_event(event_t *event, unsigned long offset, unsigned long head)
938 {
939         char level;
940         int show = 0;
941         struct dso *dso = NULL;
942         struct thread *thread = threads__findnew(event->ip.pid);
943         uint64_t ip = event->ip.ip;
944         struct map *map = NULL;
945
946         dprintf("%p [%p]: PERF_EVENT (IP, %d): %d: %p\n",
947                 (void *)(offset + head),
948                 (void *)(long)(event->header.size),
949                 event->header.misc,
950                 event->ip.pid,
951                 (void *)(long)ip);
952
953         dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid);
954
955         if (thread == NULL) {
956                 fprintf(stderr, "problem processing %d event, skipping it.\n",
957                         event->header.type);
958                 return -1;
959         }
960
961         if (event->header.misc & PERF_EVENT_MISC_KERNEL) {
962                 show = SHOW_KERNEL;
963                 level = 'k';
964
965                 dso = kernel_dso;
966
967                 dprintf(" ...... dso: %s\n", dso->name);
968
969         } else if (event->header.misc & PERF_EVENT_MISC_USER) {
970
971                 show = SHOW_USER;
972                 level = '.';
973
974                 map = thread__find_map(thread, ip);
975                 if (map != NULL) {
976                         ip = map->map_ip(map, ip);
977                         dso = map->dso;
978                 } else {
979                         /*
980                          * If this is outside of all known maps,
981                          * and is a negative address, try to look it
982                          * up in the kernel dso, as it might be a
983                          * vsyscall (which executes in user-mode):
984                          */
985                         if ((long long)ip < 0)
986                                 dso = kernel_dso;
987                 }
988                 dprintf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
989
990         } else {
991                 show = SHOW_HV;
992                 level = 'H';
993                 dprintf(" ...... dso: [hypervisor]\n");
994         }
995
996         if (show & show_mask) {
997                 struct symbol *sym = NULL;
998
999                 if (dso)
1000                         sym = dso->find_symbol(dso, ip);
1001
1002                 if (hist_entry__add(thread, map, dso, sym, ip, level)) {
1003                         fprintf(stderr,
1004                 "problem incrementing symbol count, skipping event\n");
1005                         return -1;
1006                 }
1007         }
1008         total++;
1009
1010         return 0;
1011 }
1012
1013 static int
1014 process_mmap_event(event_t *event, unsigned long offset, unsigned long head)
1015 {
1016         struct thread *thread = threads__findnew(event->mmap.pid);
1017         struct map *map = map__new(&event->mmap);
1018
1019         dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n",
1020                 (void *)(offset + head),
1021                 (void *)(long)(event->header.size),
1022                 event->mmap.pid,
1023                 (void *)(long)event->mmap.start,
1024                 (void *)(long)event->mmap.len,
1025                 (void *)(long)event->mmap.pgoff,
1026                 event->mmap.filename);
1027
1028         if (thread == NULL || map == NULL) {
1029                 dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1030                 return 0;
1031         }
1032
1033         thread__insert_map(thread, map);
1034         total_mmap++;
1035
1036         return 0;
1037 }
1038
1039 static int
1040 process_comm_event(event_t *event, unsigned long offset, unsigned long head)
1041 {
1042         struct thread *thread = threads__findnew(event->comm.pid);
1043
1044         dprintf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
1045                 (void *)(offset + head),
1046                 (void *)(long)(event->header.size),
1047                 event->comm.comm, event->comm.pid);
1048
1049         if (thread == NULL ||
1050             thread__set_comm(thread, event->comm.comm)) {
1051                 dprintf("problem processing PERF_EVENT_COMM, skipping event.\n");
1052                 return -1;
1053         }
1054         total_comm++;
1055
1056         return 0;
1057 }
1058
1059 static int
1060 process_fork_event(event_t *event, unsigned long offset, unsigned long head)
1061 {
1062         struct thread *thread = threads__findnew(event->fork.pid);
1063         struct thread *parent = threads__findnew(event->fork.ppid);
1064
1065         dprintf("%p [%p]: PERF_EVENT_FORK: %d:%d\n",
1066                 (void *)(offset + head),
1067                 (void *)(long)(event->header.size),
1068                 event->fork.pid, event->fork.ppid);
1069
1070         if (!thread || !parent || thread__fork(thread, parent)) {
1071                 dprintf("problem processing PERF_EVENT_FORK, skipping event.\n");
1072                 return -1;
1073         }
1074         total_fork++;
1075
1076         return 0;
1077 }
1078
1079 static int
1080 process_period_event(event_t *event, unsigned long offset, unsigned long head)
1081 {
1082         dprintf("%p [%p]: PERF_EVENT_PERIOD: time:%Ld, id:%Ld: period:%Ld\n",
1083                 (void *)(offset + head),
1084                 (void *)(long)(event->header.size),
1085                 event->period.time,
1086                 event->period.id,
1087                 event->period.sample_period);
1088
1089         return 0;
1090 }
1091
1092 static int
1093 process_event(event_t *event, unsigned long offset, unsigned long head)
1094 {
1095         if (event->header.misc & PERF_EVENT_MISC_OVERFLOW)
1096                 return process_overflow_event(event, offset, head);
1097
1098         switch (event->header.type) {
1099         case PERF_EVENT_MMAP:
1100                 return process_mmap_event(event, offset, head);
1101
1102         case PERF_EVENT_COMM:
1103                 return process_comm_event(event, offset, head);
1104
1105         case PERF_EVENT_FORK:
1106                 return process_fork_event(event, offset, head);
1107
1108         case PERF_EVENT_PERIOD:
1109                 return process_period_event(event, offset, head);
1110         /*
1111          * We dont process them right now but they are fine:
1112          */
1113
1114         case PERF_EVENT_THROTTLE:
1115         case PERF_EVENT_UNTHROTTLE:
1116                 return 0;
1117
1118         default:
1119                 return -1;
1120         }
1121
1122         return 0;
1123 }
1124
1125 static int __cmd_report(void)
1126 {
1127         int ret, rc = EXIT_FAILURE;
1128         unsigned long offset = 0;
1129         unsigned long head = 0;
1130         struct stat stat;
1131         event_t *event;
1132         uint32_t size;
1133         char *buf;
1134
1135         register_idle_thread();
1136
1137         input = open(input_name, O_RDONLY);
1138         if (input < 0) {
1139                 fprintf(stderr, " failed to open file: %s", input_name);
1140                 if (!strcmp(input_name, "perf.data"))
1141                         fprintf(stderr, "  (try 'perf record' first)");
1142                 fprintf(stderr, "\n");
1143                 exit(-1);
1144         }
1145
1146         ret = fstat(input, &stat);
1147         if (ret < 0) {
1148                 perror("failed to stat file");
1149                 exit(-1);
1150         }
1151
1152         if (!stat.st_size) {
1153                 fprintf(stderr, "zero-sized file, nothing to do!\n");
1154                 exit(0);
1155         }
1156
1157         if (load_kernel() < 0) {
1158                 perror("failed to load kernel symbols");
1159                 return EXIT_FAILURE;
1160         }
1161
1162         if (!full_paths) {
1163                 if (getcwd(__cwd, sizeof(__cwd)) == NULL) {
1164                         perror("failed to get the current directory");
1165                         return EXIT_FAILURE;
1166                 }
1167                 cwdlen = strlen(cwd);
1168         } else {
1169                 cwd = NULL;
1170                 cwdlen = 0;
1171         }
1172 remap:
1173         buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
1174                            MAP_SHARED, input, offset);
1175         if (buf == MAP_FAILED) {
1176                 perror("failed to mmap file");
1177                 exit(-1);
1178         }
1179
1180 more:
1181         event = (event_t *)(buf + head);
1182
1183         size = event->header.size;
1184         if (!size)
1185                 size = 8;
1186
1187         if (head + event->header.size >= page_size * mmap_window) {
1188                 unsigned long shift = page_size * (head / page_size);
1189                 int ret;
1190
1191                 ret = munmap(buf, page_size * mmap_window);
1192                 assert(ret == 0);
1193
1194                 offset += shift;
1195                 head -= shift;
1196                 goto remap;
1197         }
1198
1199         size = event->header.size;
1200
1201         dprintf("%p [%p]: event: %d\n",
1202                         (void *)(offset + head),
1203                         (void *)(long)event->header.size,
1204                         event->header.type);
1205
1206         if (!size || process_event(event, offset, head) < 0) {
1207
1208                 dprintf("%p [%p]: skipping unknown header type: %d\n",
1209                         (void *)(offset + head),
1210                         (void *)(long)(event->header.size),
1211                         event->header.type);
1212
1213                 total_unknown++;
1214
1215                 /*
1216                  * assume we lost track of the stream, check alignment, and
1217                  * increment a single u64 in the hope to catch on again 'soon'.
1218                  */
1219
1220                 if (unlikely(head & 7))
1221                         head &= ~7ULL;
1222
1223                 size = 8;
1224         }
1225
1226         head += size;
1227
1228         if (offset + head < stat.st_size)
1229                 goto more;
1230
1231         rc = EXIT_SUCCESS;
1232         close(input);
1233
1234         dprintf("      IP events: %10ld\n", total);
1235         dprintf("    mmap events: %10ld\n", total_mmap);
1236         dprintf("    comm events: %10ld\n", total_comm);
1237         dprintf("    fork events: %10ld\n", total_fork);
1238         dprintf(" unknown events: %10ld\n", total_unknown);
1239
1240         if (dump_trace)
1241                 return 0;
1242
1243         if (verbose >= 3)
1244                 threads__fprintf(stdout);
1245
1246         if (verbose >= 2)
1247                 dsos__fprintf(stdout);
1248
1249         collapse__resort();
1250         output__resort();
1251         output__fprintf(stdout, total);
1252
1253         return rc;
1254 }
1255
1256 static const char * const report_usage[] = {
1257         "perf report [<options>] <command>",
1258         NULL
1259 };
1260
1261 static const struct option options[] = {
1262         OPT_STRING('i', "input", &input_name, "file",
1263                     "input file name"),
1264         OPT_BOOLEAN('v', "verbose", &verbose,
1265                     "be more verbose (show symbol address, etc)"),
1266         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1267                     "dump raw trace in ASCII"),
1268         OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"),
1269         OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1270                    "sort by key(s): pid, comm, dso, symbol. Default: pid,symbol"),
1271         OPT_BOOLEAN('P', "full-paths", &full_paths,
1272                     "Don't shorten the pathnames taking into account the cwd"),
1273         OPT_END()
1274 };
1275
1276 static void setup_sorting(void)
1277 {
1278         char *tmp, *tok, *str = strdup(sort_order);
1279
1280         for (tok = strtok_r(str, ", ", &tmp);
1281                         tok; tok = strtok_r(NULL, ", ", &tmp)) {
1282                 if (sort_dimension__add(tok) < 0) {
1283                         error("Unknown --sort key: `%s'", tok);
1284                         usage_with_options(report_usage, options);
1285                 }
1286         }
1287
1288         free(str);
1289 }
1290
1291 int cmd_report(int argc, const char **argv, const char *prefix)
1292 {
1293         symbol__init();
1294
1295         page_size = getpagesize();
1296
1297         argc = parse_options(argc, argv, options, report_usage, 0);
1298
1299         setup_sorting();
1300
1301         /*
1302          * Any (unrecognized) arguments left?
1303          */
1304         if (argc)
1305                 usage_with_options(report_usage, options);
1306
1307         setup_pager();
1308
1309         return __cmd_report();
1310 }