perf hists: Catch and handle out-of-date hist entry maps.
[~shefty/rdma-dev.git] / tools / perf / util / hist.c
1 #include "annotate.h"
2 #include "util.h"
3 #include "build-id.h"
4 #include "hist.h"
5 #include "session.h"
6 #include "sort.h"
7 #include <math.h>
8
9 static bool hists__filter_entry_by_dso(struct hists *hists,
10                                        struct hist_entry *he);
11 static bool hists__filter_entry_by_thread(struct hists *hists,
12                                           struct hist_entry *he);
13 static bool hists__filter_entry_by_symbol(struct hists *hists,
14                                           struct hist_entry *he);
15
16 enum hist_filter {
17         HIST_FILTER__DSO,
18         HIST_FILTER__THREAD,
19         HIST_FILTER__PARENT,
20         HIST_FILTER__SYMBOL,
21 };
22
23 struct callchain_param  callchain_param = {
24         .mode   = CHAIN_GRAPH_REL,
25         .min_percent = 0.5,
26         .order  = ORDER_CALLEE
27 };
28
29 u16 hists__col_len(struct hists *hists, enum hist_column col)
30 {
31         return hists->col_len[col];
32 }
33
34 void hists__set_col_len(struct hists *hists, enum hist_column col, u16 len)
35 {
36         hists->col_len[col] = len;
37 }
38
39 bool hists__new_col_len(struct hists *hists, enum hist_column col, u16 len)
40 {
41         if (len > hists__col_len(hists, col)) {
42                 hists__set_col_len(hists, col, len);
43                 return true;
44         }
45         return false;
46 }
47
48 static void hists__reset_col_len(struct hists *hists)
49 {
50         enum hist_column col;
51
52         for (col = 0; col < HISTC_NR_COLS; ++col)
53                 hists__set_col_len(hists, col, 0);
54 }
55
56 static void hists__set_unres_dso_col_len(struct hists *hists, int dso)
57 {
58         const unsigned int unresolved_col_width = BITS_PER_LONG / 4;
59
60         if (hists__col_len(hists, dso) < unresolved_col_width &&
61             !symbol_conf.col_width_list_str && !symbol_conf.field_sep &&
62             !symbol_conf.dso_list)
63                 hists__set_col_len(hists, dso, unresolved_col_width);
64 }
65
66 static void hists__calc_col_len(struct hists *hists, struct hist_entry *h)
67 {
68         const unsigned int unresolved_col_width = BITS_PER_LONG / 4;
69         u16 len;
70
71         if (h->ms.sym)
72                 hists__new_col_len(hists, HISTC_SYMBOL, h->ms.sym->namelen + 4);
73         else
74                 hists__set_unres_dso_col_len(hists, HISTC_DSO);
75
76         len = thread__comm_len(h->thread);
77         if (hists__new_col_len(hists, HISTC_COMM, len))
78                 hists__set_col_len(hists, HISTC_THREAD, len + 6);
79
80         if (h->ms.map) {
81                 len = dso__name_len(h->ms.map->dso);
82                 hists__new_col_len(hists, HISTC_DSO, len);
83         }
84
85         if (h->branch_info) {
86                 int symlen;
87                 /*
88                  * +4 accounts for '[x] ' priv level info
89                  * +2 account of 0x prefix on raw addresses
90                  */
91                 if (h->branch_info->from.sym) {
92                         symlen = (int)h->branch_info->from.sym->namelen + 4;
93                         hists__new_col_len(hists, HISTC_SYMBOL_FROM, symlen);
94
95                         symlen = dso__name_len(h->branch_info->from.map->dso);
96                         hists__new_col_len(hists, HISTC_DSO_FROM, symlen);
97                 } else {
98                         symlen = unresolved_col_width + 4 + 2;
99                         hists__new_col_len(hists, HISTC_SYMBOL_FROM, symlen);
100                         hists__set_unres_dso_col_len(hists, HISTC_DSO_FROM);
101                 }
102
103                 if (h->branch_info->to.sym) {
104                         symlen = (int)h->branch_info->to.sym->namelen + 4;
105                         hists__new_col_len(hists, HISTC_SYMBOL_TO, symlen);
106
107                         symlen = dso__name_len(h->branch_info->to.map->dso);
108                         hists__new_col_len(hists, HISTC_DSO_TO, symlen);
109                 } else {
110                         symlen = unresolved_col_width + 4 + 2;
111                         hists__new_col_len(hists, HISTC_SYMBOL_TO, symlen);
112                         hists__set_unres_dso_col_len(hists, HISTC_DSO_TO);
113                 }
114         }
115 }
116
117 static void hist_entry__add_cpumode_period(struct hist_entry *he,
118                                            unsigned int cpumode, u64 period)
119 {
120         switch (cpumode) {
121         case PERF_RECORD_MISC_KERNEL:
122                 he->period_sys += period;
123                 break;
124         case PERF_RECORD_MISC_USER:
125                 he->period_us += period;
126                 break;
127         case PERF_RECORD_MISC_GUEST_KERNEL:
128                 he->period_guest_sys += period;
129                 break;
130         case PERF_RECORD_MISC_GUEST_USER:
131                 he->period_guest_us += period;
132                 break;
133         default:
134                 break;
135         }
136 }
137
138 static void hist_entry__decay(struct hist_entry *he)
139 {
140         he->period = (he->period * 7) / 8;
141         he->nr_events = (he->nr_events * 7) / 8;
142 }
143
144 static bool hists__decay_entry(struct hists *hists, struct hist_entry *he)
145 {
146         u64 prev_period = he->period;
147
148         if (prev_period == 0)
149                 return true;
150
151         hist_entry__decay(he);
152
153         if (!he->filtered)
154                 hists->stats.total_period -= prev_period - he->period;
155
156         return he->period == 0;
157 }
158
159 static void __hists__decay_entries(struct hists *hists, bool zap_user,
160                                    bool zap_kernel, bool threaded)
161 {
162         struct rb_node *next = rb_first(&hists->entries);
163         struct hist_entry *n;
164
165         while (next) {
166                 n = rb_entry(next, struct hist_entry, rb_node);
167                 next = rb_next(&n->rb_node);
168                 /*
169                  * We may be annotating this, for instance, so keep it here in
170                  * case some it gets new samples, we'll eventually free it when
171                  * the user stops browsing and it agains gets fully decayed.
172                  */
173                 if (((zap_user && n->level == '.') ||
174                      (zap_kernel && n->level != '.') ||
175                      hists__decay_entry(hists, n)) &&
176                     !n->used) {
177                         rb_erase(&n->rb_node, &hists->entries);
178
179                         if (sort__need_collapse || threaded)
180                                 rb_erase(&n->rb_node_in, &hists->entries_collapsed);
181
182                         hist_entry__free(n);
183                         --hists->nr_entries;
184                 }
185         }
186 }
187
188 void hists__decay_entries(struct hists *hists, bool zap_user, bool zap_kernel)
189 {
190         return __hists__decay_entries(hists, zap_user, zap_kernel, false);
191 }
192
193 void hists__decay_entries_threaded(struct hists *hists,
194                                    bool zap_user, bool zap_kernel)
195 {
196         return __hists__decay_entries(hists, zap_user, zap_kernel, true);
197 }
198
199 /*
200  * histogram, sorted on item, collects periods
201  */
202
203 static struct hist_entry *hist_entry__new(struct hist_entry *template)
204 {
205         size_t callchain_size = symbol_conf.use_callchain ? sizeof(struct callchain_root) : 0;
206         struct hist_entry *he = malloc(sizeof(*he) + callchain_size);
207
208         if (he != NULL) {
209                 *he = *template;
210                 he->nr_events = 1;
211                 if (he->ms.map)
212                         he->ms.map->referenced = true;
213                 if (symbol_conf.use_callchain)
214                         callchain_init(he->callchain);
215         }
216
217         return he;
218 }
219
220 static void hists__inc_nr_entries(struct hists *hists, struct hist_entry *h)
221 {
222         if (!h->filtered) {
223                 hists__calc_col_len(hists, h);
224                 ++hists->nr_entries;
225                 hists->stats.total_period += h->period;
226         }
227 }
228
229 static u8 symbol__parent_filter(const struct symbol *parent)
230 {
231         if (symbol_conf.exclude_other && parent == NULL)
232                 return 1 << HIST_FILTER__PARENT;
233         return 0;
234 }
235
236 static struct hist_entry *add_hist_entry(struct hists *hists,
237                                       struct hist_entry *entry,
238                                       struct addr_location *al,
239                                       u64 period)
240 {
241         struct rb_node **p;
242         struct rb_node *parent = NULL;
243         struct hist_entry *he;
244         int cmp;
245
246         pthread_mutex_lock(&hists->lock);
247
248         p = &hists->entries_in->rb_node;
249
250         while (*p != NULL) {
251                 parent = *p;
252                 he = rb_entry(parent, struct hist_entry, rb_node_in);
253
254                 cmp = hist_entry__cmp(entry, he);
255
256                 if (!cmp) {
257                         he->period += period;
258                         ++he->nr_events;
259
260                         /* If the map of an existing hist_entry has
261                          * become out-of-date due to an exec() or
262                          * similar, update it.  Otherwise we will
263                          * mis-adjust symbol addresses when computing
264                          * the history counter to increment.
265                          */
266                         if (he->ms.map != entry->ms.map) {
267                                 he->ms.map = entry->ms.map;
268                                 if (he->ms.map)
269                                         he->ms.map->referenced = true;
270                         }
271                         goto out;
272                 }
273
274                 if (cmp < 0)
275                         p = &(*p)->rb_left;
276                 else
277                         p = &(*p)->rb_right;
278         }
279
280         he = hist_entry__new(entry);
281         if (!he)
282                 goto out_unlock;
283
284         rb_link_node(&he->rb_node_in, parent, p);
285         rb_insert_color(&he->rb_node_in, hists->entries_in);
286 out:
287         hist_entry__add_cpumode_period(he, al->cpumode, period);
288 out_unlock:
289         pthread_mutex_unlock(&hists->lock);
290         return he;
291 }
292
293 struct hist_entry *__hists__add_branch_entry(struct hists *self,
294                                              struct addr_location *al,
295                                              struct symbol *sym_parent,
296                                              struct branch_info *bi,
297                                              u64 period)
298 {
299         struct hist_entry entry = {
300                 .thread = al->thread,
301                 .ms = {
302                         .map    = bi->to.map,
303                         .sym    = bi->to.sym,
304                 },
305                 .cpu    = al->cpu,
306                 .ip     = bi->to.addr,
307                 .level  = al->level,
308                 .period = period,
309                 .parent = sym_parent,
310                 .filtered = symbol__parent_filter(sym_parent),
311                 .branch_info = bi,
312         };
313
314         return add_hist_entry(self, &entry, al, period);
315 }
316
317 struct hist_entry *__hists__add_entry(struct hists *self,
318                                       struct addr_location *al,
319                                       struct symbol *sym_parent, u64 period)
320 {
321         struct hist_entry entry = {
322                 .thread = al->thread,
323                 .ms = {
324                         .map    = al->map,
325                         .sym    = al->sym,
326                 },
327                 .cpu    = al->cpu,
328                 .ip     = al->addr,
329                 .level  = al->level,
330                 .period = period,
331                 .parent = sym_parent,
332                 .filtered = symbol__parent_filter(sym_parent),
333         };
334
335         return add_hist_entry(self, &entry, al, period);
336 }
337
338 int64_t
339 hist_entry__cmp(struct hist_entry *left, struct hist_entry *right)
340 {
341         struct sort_entry *se;
342         int64_t cmp = 0;
343
344         list_for_each_entry(se, &hist_entry__sort_list, list) {
345                 cmp = se->se_cmp(left, right);
346                 if (cmp)
347                         break;
348         }
349
350         return cmp;
351 }
352
353 int64_t
354 hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
355 {
356         struct sort_entry *se;
357         int64_t cmp = 0;
358
359         list_for_each_entry(se, &hist_entry__sort_list, list) {
360                 int64_t (*f)(struct hist_entry *, struct hist_entry *);
361
362                 f = se->se_collapse ?: se->se_cmp;
363
364                 cmp = f(left, right);
365                 if (cmp)
366                         break;
367         }
368
369         return cmp;
370 }
371
372 void hist_entry__free(struct hist_entry *he)
373 {
374         free(he);
375 }
376
377 /*
378  * collapse the histogram
379  */
380
381 static bool hists__collapse_insert_entry(struct hists *hists,
382                                          struct rb_root *root,
383                                          struct hist_entry *he)
384 {
385         struct rb_node **p = &root->rb_node;
386         struct rb_node *parent = NULL;
387         struct hist_entry *iter;
388         int64_t cmp;
389
390         while (*p != NULL) {
391                 parent = *p;
392                 iter = rb_entry(parent, struct hist_entry, rb_node_in);
393
394                 cmp = hist_entry__collapse(iter, he);
395
396                 if (!cmp) {
397                         iter->period += he->period;
398                         iter->nr_events += he->nr_events;
399                         if (symbol_conf.use_callchain) {
400                                 callchain_cursor_reset(&hists->callchain_cursor);
401                                 callchain_merge(&hists->callchain_cursor, iter->callchain,
402                                                 he->callchain);
403                         }
404                         hist_entry__free(he);
405                         return false;
406                 }
407
408                 if (cmp < 0)
409                         p = &(*p)->rb_left;
410                 else
411                         p = &(*p)->rb_right;
412         }
413
414         rb_link_node(&he->rb_node_in, parent, p);
415         rb_insert_color(&he->rb_node_in, root);
416         return true;
417 }
418
419 static struct rb_root *hists__get_rotate_entries_in(struct hists *hists)
420 {
421         struct rb_root *root;
422
423         pthread_mutex_lock(&hists->lock);
424
425         root = hists->entries_in;
426         if (++hists->entries_in > &hists->entries_in_array[1])
427                 hists->entries_in = &hists->entries_in_array[0];
428
429         pthread_mutex_unlock(&hists->lock);
430
431         return root;
432 }
433
434 static void hists__apply_filters(struct hists *hists, struct hist_entry *he)
435 {
436         hists__filter_entry_by_dso(hists, he);
437         hists__filter_entry_by_thread(hists, he);
438         hists__filter_entry_by_symbol(hists, he);
439 }
440
441 static void __hists__collapse_resort(struct hists *hists, bool threaded)
442 {
443         struct rb_root *root;
444         struct rb_node *next;
445         struct hist_entry *n;
446
447         if (!sort__need_collapse && !threaded)
448                 return;
449
450         root = hists__get_rotate_entries_in(hists);
451         next = rb_first(root);
452
453         while (next) {
454                 n = rb_entry(next, struct hist_entry, rb_node_in);
455                 next = rb_next(&n->rb_node_in);
456
457                 rb_erase(&n->rb_node_in, root);
458                 if (hists__collapse_insert_entry(hists, &hists->entries_collapsed, n)) {
459                         /*
460                          * If it wasn't combined with one of the entries already
461                          * collapsed, we need to apply the filters that may have
462                          * been set by, say, the hist_browser.
463                          */
464                         hists__apply_filters(hists, n);
465                 }
466         }
467 }
468
469 void hists__collapse_resort(struct hists *hists)
470 {
471         return __hists__collapse_resort(hists, false);
472 }
473
474 void hists__collapse_resort_threaded(struct hists *hists)
475 {
476         return __hists__collapse_resort(hists, true);
477 }
478
479 /*
480  * reverse the map, sort on period.
481  */
482
483 static void __hists__insert_output_entry(struct rb_root *entries,
484                                          struct hist_entry *he,
485                                          u64 min_callchain_hits)
486 {
487         struct rb_node **p = &entries->rb_node;
488         struct rb_node *parent = NULL;
489         struct hist_entry *iter;
490
491         if (symbol_conf.use_callchain)
492                 callchain_param.sort(&he->sorted_chain, he->callchain,
493                                       min_callchain_hits, &callchain_param);
494
495         while (*p != NULL) {
496                 parent = *p;
497                 iter = rb_entry(parent, struct hist_entry, rb_node);
498
499                 if (he->period > iter->period)
500                         p = &(*p)->rb_left;
501                 else
502                         p = &(*p)->rb_right;
503         }
504
505         rb_link_node(&he->rb_node, parent, p);
506         rb_insert_color(&he->rb_node, entries);
507 }
508
509 static void __hists__output_resort(struct hists *hists, bool threaded)
510 {
511         struct rb_root *root;
512         struct rb_node *next;
513         struct hist_entry *n;
514         u64 min_callchain_hits;
515
516         min_callchain_hits = hists->stats.total_period * (callchain_param.min_percent / 100);
517
518         if (sort__need_collapse || threaded)
519                 root = &hists->entries_collapsed;
520         else
521                 root = hists->entries_in;
522
523         next = rb_first(root);
524         hists->entries = RB_ROOT;
525
526         hists->nr_entries = 0;
527         hists->stats.total_period = 0;
528         hists__reset_col_len(hists);
529
530         while (next) {
531                 n = rb_entry(next, struct hist_entry, rb_node_in);
532                 next = rb_next(&n->rb_node_in);
533
534                 __hists__insert_output_entry(&hists->entries, n, min_callchain_hits);
535                 hists__inc_nr_entries(hists, n);
536         }
537 }
538
539 void hists__output_resort(struct hists *hists)
540 {
541         return __hists__output_resort(hists, false);
542 }
543
544 void hists__output_resort_threaded(struct hists *hists)
545 {
546         return __hists__output_resort(hists, true);
547 }
548
549 static size_t callchain__fprintf_left_margin(FILE *fp, int left_margin)
550 {
551         int i;
552         int ret = fprintf(fp, "            ");
553
554         for (i = 0; i < left_margin; i++)
555                 ret += fprintf(fp, " ");
556
557         return ret;
558 }
559
560 static size_t ipchain__fprintf_graph_line(FILE *fp, int depth, int depth_mask,
561                                           int left_margin)
562 {
563         int i;
564         size_t ret = callchain__fprintf_left_margin(fp, left_margin);
565
566         for (i = 0; i < depth; i++)
567                 if (depth_mask & (1 << i))
568                         ret += fprintf(fp, "|          ");
569                 else
570                         ret += fprintf(fp, "           ");
571
572         ret += fprintf(fp, "\n");
573
574         return ret;
575 }
576
577 static size_t ipchain__fprintf_graph(FILE *fp, struct callchain_list *chain,
578                                      int depth, int depth_mask, int period,
579                                      u64 total_samples, u64 hits,
580                                      int left_margin)
581 {
582         int i;
583         size_t ret = 0;
584
585         ret += callchain__fprintf_left_margin(fp, left_margin);
586         for (i = 0; i < depth; i++) {
587                 if (depth_mask & (1 << i))
588                         ret += fprintf(fp, "|");
589                 else
590                         ret += fprintf(fp, " ");
591                 if (!period && i == depth - 1) {
592                         double percent;
593
594                         percent = hits * 100.0 / total_samples;
595                         ret += percent_color_fprintf(fp, "--%2.2f%%-- ", percent);
596                 } else
597                         ret += fprintf(fp, "%s", "          ");
598         }
599         if (chain->ms.sym)
600                 ret += fprintf(fp, "%s\n", chain->ms.sym->name);
601         else
602                 ret += fprintf(fp, "%p\n", (void *)(long)chain->ip);
603
604         return ret;
605 }
606
607 static struct symbol *rem_sq_bracket;
608 static struct callchain_list rem_hits;
609
610 static void init_rem_hits(void)
611 {
612         rem_sq_bracket = malloc(sizeof(*rem_sq_bracket) + 6);
613         if (!rem_sq_bracket) {
614                 fprintf(stderr, "Not enough memory to display remaining hits\n");
615                 return;
616         }
617
618         strcpy(rem_sq_bracket->name, "[...]");
619         rem_hits.ms.sym = rem_sq_bracket;
620 }
621
622 static size_t __callchain__fprintf_graph(FILE *fp, struct rb_root *root,
623                                          u64 total_samples, int depth,
624                                          int depth_mask, int left_margin)
625 {
626         struct rb_node *node, *next;
627         struct callchain_node *child;
628         struct callchain_list *chain;
629         int new_depth_mask = depth_mask;
630         u64 remaining;
631         size_t ret = 0;
632         int i;
633         uint entries_printed = 0;
634
635         remaining = total_samples;
636
637         node = rb_first(root);
638         while (node) {
639                 u64 new_total;
640                 u64 cumul;
641
642                 child = rb_entry(node, struct callchain_node, rb_node);
643                 cumul = callchain_cumul_hits(child);
644                 remaining -= cumul;
645
646                 /*
647                  * The depth mask manages the output of pipes that show
648                  * the depth. We don't want to keep the pipes of the current
649                  * level for the last child of this depth.
650                  * Except if we have remaining filtered hits. They will
651                  * supersede the last child
652                  */
653                 next = rb_next(node);
654                 if (!next && (callchain_param.mode != CHAIN_GRAPH_REL || !remaining))
655                         new_depth_mask &= ~(1 << (depth - 1));
656
657                 /*
658                  * But we keep the older depth mask for the line separator
659                  * to keep the level link until we reach the last child
660                  */
661                 ret += ipchain__fprintf_graph_line(fp, depth, depth_mask,
662                                                    left_margin);
663                 i = 0;
664                 list_for_each_entry(chain, &child->val, list) {
665                         ret += ipchain__fprintf_graph(fp, chain, depth,
666                                                       new_depth_mask, i++,
667                                                       total_samples,
668                                                       cumul,
669                                                       left_margin);
670                 }
671
672                 if (callchain_param.mode == CHAIN_GRAPH_REL)
673                         new_total = child->children_hit;
674                 else
675                         new_total = total_samples;
676
677                 ret += __callchain__fprintf_graph(fp, &child->rb_root, new_total,
678                                                   depth + 1,
679                                                   new_depth_mask | (1 << depth),
680                                                   left_margin);
681                 node = next;
682                 if (++entries_printed == callchain_param.print_limit)
683                         break;
684         }
685
686         if (callchain_param.mode == CHAIN_GRAPH_REL &&
687                 remaining && remaining != total_samples) {
688
689                 if (!rem_sq_bracket)
690                         return ret;
691
692                 new_depth_mask &= ~(1 << (depth - 1));
693                 ret += ipchain__fprintf_graph(fp, &rem_hits, depth,
694                                               new_depth_mask, 0, total_samples,
695                                               remaining, left_margin);
696         }
697
698         return ret;
699 }
700
701 static size_t callchain__fprintf_graph(FILE *fp, struct rb_root *root,
702                                        u64 total_samples, int left_margin)
703 {
704         struct callchain_node *cnode;
705         struct callchain_list *chain;
706         u32 entries_printed = 0;
707         bool printed = false;
708         struct rb_node *node;
709         int i = 0;
710         int ret;
711
712         /*
713          * If have one single callchain root, don't bother printing
714          * its percentage (100 % in fractal mode and the same percentage
715          * than the hist in graph mode). This also avoid one level of column.
716          */
717         node = rb_first(root);
718         if (node && !rb_next(node)) {
719                 cnode = rb_entry(node, struct callchain_node, rb_node);
720                 list_for_each_entry(chain, &cnode->val, list) {
721                         /*
722                          * If we sort by symbol, the first entry is the same than
723                          * the symbol. No need to print it otherwise it appears as
724                          * displayed twice.
725                          */
726                         if (!i++ && sort__first_dimension == SORT_SYM)
727                                 continue;
728                         if (!printed) {
729                                 ret += callchain__fprintf_left_margin(fp, left_margin);
730                                 ret += fprintf(fp, "|\n");
731                                 ret += callchain__fprintf_left_margin(fp, left_margin);
732                                 ret += fprintf(fp, "---");
733                                 left_margin += 3;
734                                 printed = true;
735                         } else
736                                 ret += callchain__fprintf_left_margin(fp, left_margin);
737
738                         if (chain->ms.sym)
739                                 ret += fprintf(fp, " %s\n", chain->ms.sym->name);
740                         else
741                                 ret += fprintf(fp, " %p\n", (void *)(long)chain->ip);
742
743                         if (++entries_printed == callchain_param.print_limit)
744                                 break;
745                 }
746                 root = &cnode->rb_root;
747         }
748
749         return __callchain__fprintf_graph(fp, root, total_samples,
750                                           1, 1, left_margin);
751 }
752
753 static size_t __callchain__fprintf_flat(FILE *fp,
754                                         struct callchain_node *self,
755                                         u64 total_samples)
756 {
757         struct callchain_list *chain;
758         size_t ret = 0;
759
760         if (!self)
761                 return 0;
762
763         ret += __callchain__fprintf_flat(fp, self->parent, total_samples);
764
765
766         list_for_each_entry(chain, &self->val, list) {
767                 if (chain->ip >= PERF_CONTEXT_MAX)
768                         continue;
769                 if (chain->ms.sym)
770                         ret += fprintf(fp, "                %s\n", chain->ms.sym->name);
771                 else
772                         ret += fprintf(fp, "                %p\n",
773                                         (void *)(long)chain->ip);
774         }
775
776         return ret;
777 }
778
779 static size_t callchain__fprintf_flat(FILE *fp, struct rb_root *self,
780                                       u64 total_samples)
781 {
782         size_t ret = 0;
783         u32 entries_printed = 0;
784         struct rb_node *rb_node;
785         struct callchain_node *chain;
786
787         rb_node = rb_first(self);
788         while (rb_node) {
789                 double percent;
790
791                 chain = rb_entry(rb_node, struct callchain_node, rb_node);
792                 percent = chain->hit * 100.0 / total_samples;
793
794                 ret = percent_color_fprintf(fp, "           %6.2f%%\n", percent);
795                 ret += __callchain__fprintf_flat(fp, chain, total_samples);
796                 ret += fprintf(fp, "\n");
797                 if (++entries_printed == callchain_param.print_limit)
798                         break;
799
800                 rb_node = rb_next(rb_node);
801         }
802
803         return ret;
804 }
805
806 static size_t hist_entry_callchain__fprintf(struct hist_entry *he,
807                                             u64 total_samples, int left_margin,
808                                             FILE *fp)
809 {
810         switch (callchain_param.mode) {
811         case CHAIN_GRAPH_REL:
812                 return callchain__fprintf_graph(fp, &he->sorted_chain, he->period,
813                                                 left_margin);
814                 break;
815         case CHAIN_GRAPH_ABS:
816                 return callchain__fprintf_graph(fp, &he->sorted_chain, total_samples,
817                                                 left_margin);
818                 break;
819         case CHAIN_FLAT:
820                 return callchain__fprintf_flat(fp, &he->sorted_chain, total_samples);
821                 break;
822         case CHAIN_NONE:
823                 break;
824         default:
825                 pr_err("Bad callchain mode\n");
826         }
827
828         return 0;
829 }
830
831 void hists__output_recalc_col_len(struct hists *hists, int max_rows)
832 {
833         struct rb_node *next = rb_first(&hists->entries);
834         struct hist_entry *n;
835         int row = 0;
836
837         hists__reset_col_len(hists);
838
839         while (next && row++ < max_rows) {
840                 n = rb_entry(next, struct hist_entry, rb_node);
841                 if (!n->filtered)
842                         hists__calc_col_len(hists, n);
843                 next = rb_next(&n->rb_node);
844         }
845 }
846
847 static int hist_entry__pcnt_snprintf(struct hist_entry *he, char *s,
848                                      size_t size, struct hists *pair_hists,
849                                      bool show_displacement, long displacement,
850                                      bool color, u64 total_period)
851 {
852         u64 period, total, period_sys, period_us, period_guest_sys, period_guest_us;
853         u64 nr_events;
854         const char *sep = symbol_conf.field_sep;
855         int ret;
856
857         if (symbol_conf.exclude_other && !he->parent)
858                 return 0;
859
860         if (pair_hists) {
861                 period = he->pair ? he->pair->period : 0;
862                 nr_events = he->pair ? he->pair->nr_events : 0;
863                 total = pair_hists->stats.total_period;
864                 period_sys = he->pair ? he->pair->period_sys : 0;
865                 period_us = he->pair ? he->pair->period_us : 0;
866                 period_guest_sys = he->pair ? he->pair->period_guest_sys : 0;
867                 period_guest_us = he->pair ? he->pair->period_guest_us : 0;
868         } else {
869                 period = he->period;
870                 nr_events = he->nr_events;
871                 total = total_period;
872                 period_sys = he->period_sys;
873                 period_us = he->period_us;
874                 period_guest_sys = he->period_guest_sys;
875                 period_guest_us = he->period_guest_us;
876         }
877
878         if (total) {
879                 if (color)
880                         ret = percent_color_snprintf(s, size,
881                                                      sep ? "%.2f" : "   %6.2f%%",
882                                                      (period * 100.0) / total);
883                 else
884                         ret = scnprintf(s, size, sep ? "%.2f" : "   %6.2f%%",
885                                        (period * 100.0) / total);
886                 if (symbol_conf.show_cpu_utilization) {
887                         ret += percent_color_snprintf(s + ret, size - ret,
888                                         sep ? "%.2f" : "   %6.2f%%",
889                                         (period_sys * 100.0) / total);
890                         ret += percent_color_snprintf(s + ret, size - ret,
891                                         sep ? "%.2f" : "   %6.2f%%",
892                                         (period_us * 100.0) / total);
893                         if (perf_guest) {
894                                 ret += percent_color_snprintf(s + ret,
895                                                 size - ret,
896                                                 sep ? "%.2f" : "   %6.2f%%",
897                                                 (period_guest_sys * 100.0) /
898                                                                 total);
899                                 ret += percent_color_snprintf(s + ret,
900                                                 size - ret,
901                                                 sep ? "%.2f" : "   %6.2f%%",
902                                                 (period_guest_us * 100.0) /
903                                                                 total);
904                         }
905                 }
906         } else
907                 ret = scnprintf(s, size, sep ? "%" PRIu64 : "%12" PRIu64 " ", period);
908
909         if (symbol_conf.show_nr_samples) {
910                 if (sep)
911                         ret += scnprintf(s + ret, size - ret, "%c%" PRIu64, *sep, nr_events);
912                 else
913                         ret += scnprintf(s + ret, size - ret, "%11" PRIu64, nr_events);
914         }
915
916         if (symbol_conf.show_total_period) {
917                 if (sep)
918                         ret += scnprintf(s + ret, size - ret, "%c%" PRIu64, *sep, period);
919                 else
920                         ret += scnprintf(s + ret, size - ret, " %12" PRIu64, period);
921         }
922
923         if (pair_hists) {
924                 char bf[32];
925                 double old_percent = 0, new_percent = 0, diff;
926
927                 if (total > 0)
928                         old_percent = (period * 100.0) / total;
929                 if (total_period > 0)
930                         new_percent = (he->period * 100.0) / total_period;
931
932                 diff = new_percent - old_percent;
933
934                 if (fabs(diff) >= 0.01)
935                         scnprintf(bf, sizeof(bf), "%+4.2F%%", diff);
936                 else
937                         scnprintf(bf, sizeof(bf), " ");
938
939                 if (sep)
940                         ret += scnprintf(s + ret, size - ret, "%c%s", *sep, bf);
941                 else
942                         ret += scnprintf(s + ret, size - ret, "%11.11s", bf);
943
944                 if (show_displacement) {
945                         if (displacement)
946                                 scnprintf(bf, sizeof(bf), "%+4ld", displacement);
947                         else
948                                 scnprintf(bf, sizeof(bf), " ");
949
950                         if (sep)
951                                 ret += scnprintf(s + ret, size - ret, "%c%s", *sep, bf);
952                         else
953                                 ret += scnprintf(s + ret, size - ret, "%6.6s", bf);
954                 }
955         }
956
957         return ret;
958 }
959
960 int hist_entry__snprintf(struct hist_entry *he, char *s, size_t size,
961                          struct hists *hists)
962 {
963         const char *sep = symbol_conf.field_sep;
964         struct sort_entry *se;
965         int ret = 0;
966
967         list_for_each_entry(se, &hist_entry__sort_list, list) {
968                 if (se->elide)
969                         continue;
970
971                 ret += scnprintf(s + ret, size - ret, "%s", sep ?: "  ");
972                 ret += se->se_snprintf(he, s + ret, size - ret,
973                                        hists__col_len(hists, se->se_width_idx));
974         }
975
976         return ret;
977 }
978
979 static int hist_entry__fprintf(struct hist_entry *he, size_t size,
980                                struct hists *hists, struct hists *pair_hists,
981                                bool show_displacement, long displacement,
982                                u64 total_period, FILE *fp)
983 {
984         char bf[512];
985         int ret;
986
987         if (size == 0 || size > sizeof(bf))
988                 size = sizeof(bf);
989
990         ret = hist_entry__pcnt_snprintf(he, bf, size, pair_hists,
991                                         show_displacement, displacement,
992                                         true, total_period);
993         hist_entry__snprintf(he, bf + ret, size - ret, hists);
994         return fprintf(fp, "%s\n", bf);
995 }
996
997 static size_t hist_entry__fprintf_callchain(struct hist_entry *he,
998                                             struct hists *hists,
999                                             u64 total_period, FILE *fp)
1000 {
1001         int left_margin = 0;
1002
1003         if (sort__first_dimension == SORT_COMM) {
1004                 struct sort_entry *se = list_first_entry(&hist_entry__sort_list,
1005                                                          typeof(*se), list);
1006                 left_margin = hists__col_len(hists, se->se_width_idx);
1007                 left_margin -= thread__comm_len(he->thread);
1008         }
1009
1010         return hist_entry_callchain__fprintf(he, total_period, left_margin, fp);
1011 }
1012
1013 size_t hists__fprintf(struct hists *hists, struct hists *pair,
1014                       bool show_displacement, bool show_header, int max_rows,
1015                       int max_cols, FILE *fp)
1016 {
1017         struct sort_entry *se;
1018         struct rb_node *nd;
1019         size_t ret = 0;
1020         u64 total_period;
1021         unsigned long position = 1;
1022         long displacement = 0;
1023         unsigned int width;
1024         const char *sep = symbol_conf.field_sep;
1025         const char *col_width = symbol_conf.col_width_list_str;
1026         int nr_rows = 0;
1027
1028         init_rem_hits();
1029
1030         if (!show_header)
1031                 goto print_entries;
1032
1033         fprintf(fp, "# %s", pair ? "Baseline" : "Overhead");
1034
1035         if (symbol_conf.show_cpu_utilization) {
1036                 if (sep) {
1037                         ret += fprintf(fp, "%csys", *sep);
1038                         ret += fprintf(fp, "%cus", *sep);
1039                         if (perf_guest) {
1040                                 ret += fprintf(fp, "%cguest sys", *sep);
1041                                 ret += fprintf(fp, "%cguest us", *sep);
1042                         }
1043                 } else {
1044                         ret += fprintf(fp, "     sys  ");
1045                         ret += fprintf(fp, "      us  ");
1046                         if (perf_guest) {
1047                                 ret += fprintf(fp, "  guest sys  ");
1048                                 ret += fprintf(fp, "  guest us  ");
1049                         }
1050                 }
1051         }
1052
1053         if (symbol_conf.show_nr_samples) {
1054                 if (sep)
1055                         fprintf(fp, "%cSamples", *sep);
1056                 else
1057                         fputs("  Samples  ", fp);
1058         }
1059
1060         if (symbol_conf.show_total_period) {
1061                 if (sep)
1062                         ret += fprintf(fp, "%cPeriod", *sep);
1063                 else
1064                         ret += fprintf(fp, "   Period    ");
1065         }
1066
1067         if (pair) {
1068                 if (sep)
1069                         ret += fprintf(fp, "%cDelta", *sep);
1070                 else
1071                         ret += fprintf(fp, "  Delta    ");
1072
1073                 if (show_displacement) {
1074                         if (sep)
1075                                 ret += fprintf(fp, "%cDisplacement", *sep);
1076                         else
1077                                 ret += fprintf(fp, " Displ");
1078                 }
1079         }
1080
1081         list_for_each_entry(se, &hist_entry__sort_list, list) {
1082                 if (se->elide)
1083                         continue;
1084                 if (sep) {
1085                         fprintf(fp, "%c%s", *sep, se->se_header);
1086                         continue;
1087                 }
1088                 width = strlen(se->se_header);
1089                 if (symbol_conf.col_width_list_str) {
1090                         if (col_width) {
1091                                 hists__set_col_len(hists, se->se_width_idx,
1092                                                    atoi(col_width));
1093                                 col_width = strchr(col_width, ',');
1094                                 if (col_width)
1095                                         ++col_width;
1096                         }
1097                 }
1098                 if (!hists__new_col_len(hists, se->se_width_idx, width))
1099                         width = hists__col_len(hists, se->se_width_idx);
1100                 fprintf(fp, "  %*s", width, se->se_header);
1101         }
1102
1103         fprintf(fp, "\n");
1104         if (max_rows && ++nr_rows >= max_rows)
1105                 goto out;
1106
1107         if (sep)
1108                 goto print_entries;
1109
1110         fprintf(fp, "# ........");
1111         if (symbol_conf.show_cpu_utilization)
1112                 fprintf(fp, "   .......   .......");
1113         if (symbol_conf.show_nr_samples)
1114                 fprintf(fp, " ..........");
1115         if (symbol_conf.show_total_period)
1116                 fprintf(fp, " ............");
1117         if (pair) {
1118                 fprintf(fp, " ..........");
1119                 if (show_displacement)
1120                         fprintf(fp, " .....");
1121         }
1122         list_for_each_entry(se, &hist_entry__sort_list, list) {
1123                 unsigned int i;
1124
1125                 if (se->elide)
1126                         continue;
1127
1128                 fprintf(fp, "  ");
1129                 width = hists__col_len(hists, se->se_width_idx);
1130                 if (width == 0)
1131                         width = strlen(se->se_header);
1132                 for (i = 0; i < width; i++)
1133                         fprintf(fp, ".");
1134         }
1135
1136         fprintf(fp, "\n");
1137         if (max_rows && ++nr_rows >= max_rows)
1138                 goto out;
1139
1140         fprintf(fp, "#\n");
1141         if (max_rows && ++nr_rows >= max_rows)
1142                 goto out;
1143
1144 print_entries:
1145         total_period = hists->stats.total_period;
1146
1147         for (nd = rb_first(&hists->entries); nd; nd = rb_next(nd)) {
1148                 struct hist_entry *h = rb_entry(nd, struct hist_entry, rb_node);
1149
1150                 if (h->filtered)
1151                         continue;
1152
1153                 if (show_displacement) {
1154                         if (h->pair != NULL)
1155                                 displacement = ((long)h->pair->position -
1156                                                 (long)position);
1157                         else
1158                                 displacement = 0;
1159                         ++position;
1160                 }
1161                 ret += hist_entry__fprintf(h, max_cols, hists, pair, show_displacement,
1162                                            displacement, total_period, fp);
1163
1164                 if (symbol_conf.use_callchain)
1165                         ret += hist_entry__fprintf_callchain(h, hists, total_period, fp);
1166                 if (max_rows && ++nr_rows >= max_rows)
1167                         goto out;
1168
1169                 if (h->ms.map == NULL && verbose > 1) {
1170                         __map_groups__fprintf_maps(&h->thread->mg,
1171                                                    MAP__FUNCTION, verbose, fp);
1172                         fprintf(fp, "%.10s end\n", graph_dotted_line);
1173                 }
1174         }
1175 out:
1176         free(rem_sq_bracket);
1177
1178         return ret;
1179 }
1180
1181 /*
1182  * See hists__fprintf to match the column widths
1183  */
1184 unsigned int hists__sort_list_width(struct hists *hists)
1185 {
1186         struct sort_entry *se;
1187         int ret = 9; /* total % */
1188
1189         if (symbol_conf.show_cpu_utilization) {
1190                 ret += 7; /* count_sys % */
1191                 ret += 6; /* count_us % */
1192                 if (perf_guest) {
1193                         ret += 13; /* count_guest_sys % */
1194                         ret += 12; /* count_guest_us % */
1195                 }
1196         }
1197
1198         if (symbol_conf.show_nr_samples)
1199                 ret += 11;
1200
1201         if (symbol_conf.show_total_period)
1202                 ret += 13;
1203
1204         list_for_each_entry(se, &hist_entry__sort_list, list)
1205                 if (!se->elide)
1206                         ret += 2 + hists__col_len(hists, se->se_width_idx);
1207
1208         if (verbose) /* Addr + origin */
1209                 ret += 3 + BITS_PER_LONG / 4;
1210
1211         return ret;
1212 }
1213
1214 static void hists__remove_entry_filter(struct hists *hists, struct hist_entry *h,
1215                                        enum hist_filter filter)
1216 {
1217         h->filtered &= ~(1 << filter);
1218         if (h->filtered)
1219                 return;
1220
1221         ++hists->nr_entries;
1222         if (h->ms.unfolded)
1223                 hists->nr_entries += h->nr_rows;
1224         h->row_offset = 0;
1225         hists->stats.total_period += h->period;
1226         hists->stats.nr_events[PERF_RECORD_SAMPLE] += h->nr_events;
1227
1228         hists__calc_col_len(hists, h);
1229 }
1230
1231
1232 static bool hists__filter_entry_by_dso(struct hists *hists,
1233                                        struct hist_entry *he)
1234 {
1235         if (hists->dso_filter != NULL &&
1236             (he->ms.map == NULL || he->ms.map->dso != hists->dso_filter)) {
1237                 he->filtered |= (1 << HIST_FILTER__DSO);
1238                 return true;
1239         }
1240
1241         return false;
1242 }
1243
1244 void hists__filter_by_dso(struct hists *hists)
1245 {
1246         struct rb_node *nd;
1247
1248         hists->nr_entries = hists->stats.total_period = 0;
1249         hists->stats.nr_events[PERF_RECORD_SAMPLE] = 0;
1250         hists__reset_col_len(hists);
1251
1252         for (nd = rb_first(&hists->entries); nd; nd = rb_next(nd)) {
1253                 struct hist_entry *h = rb_entry(nd, struct hist_entry, rb_node);
1254
1255                 if (symbol_conf.exclude_other && !h->parent)
1256                         continue;
1257
1258                 if (hists__filter_entry_by_dso(hists, h))
1259                         continue;
1260
1261                 hists__remove_entry_filter(hists, h, HIST_FILTER__DSO);
1262         }
1263 }
1264
1265 static bool hists__filter_entry_by_thread(struct hists *hists,
1266                                           struct hist_entry *he)
1267 {
1268         if (hists->thread_filter != NULL &&
1269             he->thread != hists->thread_filter) {
1270                 he->filtered |= (1 << HIST_FILTER__THREAD);
1271                 return true;
1272         }
1273
1274         return false;
1275 }
1276
1277 void hists__filter_by_thread(struct hists *hists)
1278 {
1279         struct rb_node *nd;
1280
1281         hists->nr_entries = hists->stats.total_period = 0;
1282         hists->stats.nr_events[PERF_RECORD_SAMPLE] = 0;
1283         hists__reset_col_len(hists);
1284
1285         for (nd = rb_first(&hists->entries); nd; nd = rb_next(nd)) {
1286                 struct hist_entry *h = rb_entry(nd, struct hist_entry, rb_node);
1287
1288                 if (hists__filter_entry_by_thread(hists, h))
1289                         continue;
1290
1291                 hists__remove_entry_filter(hists, h, HIST_FILTER__THREAD);
1292         }
1293 }
1294
1295 static bool hists__filter_entry_by_symbol(struct hists *hists,
1296                                           struct hist_entry *he)
1297 {
1298         if (hists->symbol_filter_str != NULL &&
1299             (!he->ms.sym || strstr(he->ms.sym->name,
1300                                    hists->symbol_filter_str) == NULL)) {
1301                 he->filtered |= (1 << HIST_FILTER__SYMBOL);
1302                 return true;
1303         }
1304
1305         return false;
1306 }
1307
1308 void hists__filter_by_symbol(struct hists *hists)
1309 {
1310         struct rb_node *nd;
1311
1312         hists->nr_entries = hists->stats.total_period = 0;
1313         hists->stats.nr_events[PERF_RECORD_SAMPLE] = 0;
1314         hists__reset_col_len(hists);
1315
1316         for (nd = rb_first(&hists->entries); nd; nd = rb_next(nd)) {
1317                 struct hist_entry *h = rb_entry(nd, struct hist_entry, rb_node);
1318
1319                 if (hists__filter_entry_by_symbol(hists, h))
1320                         continue;
1321
1322                 hists__remove_entry_filter(hists, h, HIST_FILTER__SYMBOL);
1323         }
1324 }
1325
1326 int hist_entry__inc_addr_samples(struct hist_entry *he, int evidx, u64 ip)
1327 {
1328         return symbol__inc_addr_samples(he->ms.sym, he->ms.map, evidx, ip);
1329 }
1330
1331 int hist_entry__annotate(struct hist_entry *he, size_t privsize)
1332 {
1333         return symbol__annotate(he->ms.sym, he->ms.map, privsize);
1334 }
1335
1336 void hists__inc_nr_events(struct hists *hists, u32 type)
1337 {
1338         ++hists->stats.nr_events[0];
1339         ++hists->stats.nr_events[type];
1340 }
1341
1342 size_t hists__fprintf_nr_events(struct hists *hists, FILE *fp)
1343 {
1344         int i;
1345         size_t ret = 0;
1346
1347         for (i = 0; i < PERF_RECORD_HEADER_MAX; ++i) {
1348                 const char *name;
1349
1350                 if (hists->stats.nr_events[i] == 0)
1351                         continue;
1352
1353                 name = perf_event__name(i);
1354                 if (!strcmp(name, "UNKNOWN"))
1355                         continue;
1356
1357                 ret += fprintf(fp, "%16s events: %10d\n", name,
1358                                hists->stats.nr_events[i]);
1359         }
1360
1361         return ret;
1362 }