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