perf report: Allow limiting the number of entries to print in callchains
[deliverable/linux.git] / tools / perf / util / hist.c
1 #include "hist.h"
2 #include "session.h"
3 #include "sort.h"
4 #include <math.h>
5
6 struct callchain_param callchain_param = {
7 .mode = CHAIN_GRAPH_REL,
8 .min_percent = 0.5
9 };
10
11 static void perf_session__add_cpumode_count(struct hist_entry *he,
12 unsigned int cpumode, u64 count)
13 {
14 switch (cpumode) {
15 case PERF_RECORD_MISC_KERNEL:
16 he->count_sys += count;
17 break;
18 case PERF_RECORD_MISC_USER:
19 he->count_us += count;
20 break;
21 case PERF_RECORD_MISC_GUEST_KERNEL:
22 he->count_guest_sys += count;
23 break;
24 case PERF_RECORD_MISC_GUEST_USER:
25 he->count_guest_us += count;
26 break;
27 default:
28 break;
29 }
30 }
31
32 /*
33 * histogram, sorted on item, collects counts
34 */
35
36 static struct hist_entry *hist_entry__new(struct hist_entry *template)
37 {
38 size_t callchain_size = symbol_conf.use_callchain ? sizeof(struct callchain_node) : 0;
39 struct hist_entry *self = malloc(sizeof(*self) + callchain_size);
40
41 if (self != NULL) {
42 *self = *template;
43 if (symbol_conf.use_callchain)
44 callchain_init(self->callchain);
45 }
46
47 return self;
48 }
49
50 struct hist_entry *__perf_session__add_hist_entry(struct rb_root *hists,
51 struct addr_location *al,
52 struct symbol *sym_parent,
53 u64 count)
54 {
55 struct rb_node **p = &hists->rb_node;
56 struct rb_node *parent = NULL;
57 struct hist_entry *he;
58 struct hist_entry entry = {
59 .thread = al->thread,
60 .ms = {
61 .map = al->map,
62 .sym = al->sym,
63 },
64 .ip = al->addr,
65 .level = al->level,
66 .count = count,
67 .parent = sym_parent,
68 };
69 int cmp;
70
71 while (*p != NULL) {
72 parent = *p;
73 he = rb_entry(parent, struct hist_entry, rb_node);
74
75 cmp = hist_entry__cmp(&entry, he);
76
77 if (!cmp) {
78 he->count += count;
79 goto out;
80 }
81
82 if (cmp < 0)
83 p = &(*p)->rb_left;
84 else
85 p = &(*p)->rb_right;
86 }
87
88 he = hist_entry__new(&entry);
89 if (!he)
90 return NULL;
91 rb_link_node(&he->rb_node, parent, p);
92 rb_insert_color(&he->rb_node, hists);
93 out:
94 perf_session__add_cpumode_count(he, al->cpumode, count);
95 return he;
96 }
97
98 int64_t
99 hist_entry__cmp(struct hist_entry *left, struct hist_entry *right)
100 {
101 struct sort_entry *se;
102 int64_t cmp = 0;
103
104 list_for_each_entry(se, &hist_entry__sort_list, list) {
105 cmp = se->se_cmp(left, right);
106 if (cmp)
107 break;
108 }
109
110 return cmp;
111 }
112
113 int64_t
114 hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
115 {
116 struct sort_entry *se;
117 int64_t cmp = 0;
118
119 list_for_each_entry(se, &hist_entry__sort_list, list) {
120 int64_t (*f)(struct hist_entry *, struct hist_entry *);
121
122 f = se->se_collapse ?: se->se_cmp;
123
124 cmp = f(left, right);
125 if (cmp)
126 break;
127 }
128
129 return cmp;
130 }
131
132 void hist_entry__free(struct hist_entry *he)
133 {
134 free(he);
135 }
136
137 /*
138 * collapse the histogram
139 */
140
141 static void collapse__insert_entry(struct rb_root *root, struct hist_entry *he)
142 {
143 struct rb_node **p = &root->rb_node;
144 struct rb_node *parent = NULL;
145 struct hist_entry *iter;
146 int64_t cmp;
147
148 while (*p != NULL) {
149 parent = *p;
150 iter = rb_entry(parent, struct hist_entry, rb_node);
151
152 cmp = hist_entry__collapse(iter, he);
153
154 if (!cmp) {
155 iter->count += he->count;
156 hist_entry__free(he);
157 return;
158 }
159
160 if (cmp < 0)
161 p = &(*p)->rb_left;
162 else
163 p = &(*p)->rb_right;
164 }
165
166 rb_link_node(&he->rb_node, parent, p);
167 rb_insert_color(&he->rb_node, root);
168 }
169
170 void perf_session__collapse_resort(struct rb_root *hists)
171 {
172 struct rb_root tmp;
173 struct rb_node *next;
174 struct hist_entry *n;
175
176 if (!sort__need_collapse)
177 return;
178
179 tmp = RB_ROOT;
180 next = rb_first(hists);
181
182 while (next) {
183 n = rb_entry(next, struct hist_entry, rb_node);
184 next = rb_next(&n->rb_node);
185
186 rb_erase(&n->rb_node, hists);
187 collapse__insert_entry(&tmp, n);
188 }
189
190 *hists = tmp;
191 }
192
193 /*
194 * reverse the map, sort on count.
195 */
196
197 static void perf_session__insert_output_hist_entry(struct rb_root *root,
198 struct hist_entry *he,
199 u64 min_callchain_hits)
200 {
201 struct rb_node **p = &root->rb_node;
202 struct rb_node *parent = NULL;
203 struct hist_entry *iter;
204
205 if (symbol_conf.use_callchain)
206 callchain_param.sort(&he->sorted_chain, he->callchain,
207 min_callchain_hits, &callchain_param);
208
209 while (*p != NULL) {
210 parent = *p;
211 iter = rb_entry(parent, struct hist_entry, rb_node);
212
213 if (he->count > iter->count)
214 p = &(*p)->rb_left;
215 else
216 p = &(*p)->rb_right;
217 }
218
219 rb_link_node(&he->rb_node, parent, p);
220 rb_insert_color(&he->rb_node, root);
221 }
222
223 u64 perf_session__output_resort(struct rb_root *hists, u64 total_samples)
224 {
225 struct rb_root tmp;
226 struct rb_node *next;
227 struct hist_entry *n;
228 u64 min_callchain_hits;
229 u64 nr_hists = 0;
230
231 min_callchain_hits =
232 total_samples * (callchain_param.min_percent / 100);
233
234 tmp = RB_ROOT;
235 next = rb_first(hists);
236
237 while (next) {
238 n = rb_entry(next, struct hist_entry, rb_node);
239 next = rb_next(&n->rb_node);
240
241 rb_erase(&n->rb_node, hists);
242 perf_session__insert_output_hist_entry(&tmp, n,
243 min_callchain_hits);
244 ++nr_hists;
245 }
246
247 *hists = tmp;
248 return nr_hists;
249 }
250
251 static size_t callchain__fprintf_left_margin(FILE *fp, int left_margin)
252 {
253 int i;
254 int ret = fprintf(fp, " ");
255
256 for (i = 0; i < left_margin; i++)
257 ret += fprintf(fp, " ");
258
259 return ret;
260 }
261
262 static size_t ipchain__fprintf_graph_line(FILE *fp, int depth, int depth_mask,
263 int left_margin)
264 {
265 int i;
266 size_t ret = callchain__fprintf_left_margin(fp, left_margin);
267
268 for (i = 0; i < depth; i++)
269 if (depth_mask & (1 << i))
270 ret += fprintf(fp, "| ");
271 else
272 ret += fprintf(fp, " ");
273
274 ret += fprintf(fp, "\n");
275
276 return ret;
277 }
278
279 static size_t ipchain__fprintf_graph(FILE *fp, struct callchain_list *chain,
280 int depth, int depth_mask, int count,
281 u64 total_samples, int hits,
282 int left_margin)
283 {
284 int i;
285 size_t ret = 0;
286
287 ret += callchain__fprintf_left_margin(fp, left_margin);
288 for (i = 0; i < depth; i++) {
289 if (depth_mask & (1 << i))
290 ret += fprintf(fp, "|");
291 else
292 ret += fprintf(fp, " ");
293 if (!count && i == depth - 1) {
294 double percent;
295
296 percent = hits * 100.0 / total_samples;
297 ret += percent_color_fprintf(fp, "--%2.2f%%-- ", percent);
298 } else
299 ret += fprintf(fp, "%s", " ");
300 }
301 if (chain->ms.sym)
302 ret += fprintf(fp, "%s\n", chain->ms.sym->name);
303 else
304 ret += fprintf(fp, "%p\n", (void *)(long)chain->ip);
305
306 return ret;
307 }
308
309 static struct symbol *rem_sq_bracket;
310 static struct callchain_list rem_hits;
311
312 static void init_rem_hits(void)
313 {
314 rem_sq_bracket = malloc(sizeof(*rem_sq_bracket) + 6);
315 if (!rem_sq_bracket) {
316 fprintf(stderr, "Not enough memory to display remaining hits\n");
317 return;
318 }
319
320 strcpy(rem_sq_bracket->name, "[...]");
321 rem_hits.ms.sym = rem_sq_bracket;
322 }
323
324 static size_t __callchain__fprintf_graph(FILE *fp, struct callchain_node *self,
325 u64 total_samples, int depth,
326 int depth_mask, int left_margin)
327 {
328 struct rb_node *node, *next;
329 struct callchain_node *child;
330 struct callchain_list *chain;
331 int new_depth_mask = depth_mask;
332 u64 new_total;
333 u64 remaining;
334 size_t ret = 0;
335 int i;
336 uint entries_printed = 0;
337
338 if (callchain_param.mode == CHAIN_GRAPH_REL)
339 new_total = self->children_hit;
340 else
341 new_total = total_samples;
342
343 remaining = new_total;
344
345 node = rb_first(&self->rb_root);
346 while (node) {
347 u64 cumul;
348
349 child = rb_entry(node, struct callchain_node, rb_node);
350 cumul = cumul_hits(child);
351 remaining -= cumul;
352
353 /*
354 * The depth mask manages the output of pipes that show
355 * the depth. We don't want to keep the pipes of the current
356 * level for the last child of this depth.
357 * Except if we have remaining filtered hits. They will
358 * supersede the last child
359 */
360 next = rb_next(node);
361 if (!next && (callchain_param.mode != CHAIN_GRAPH_REL || !remaining))
362 new_depth_mask &= ~(1 << (depth - 1));
363
364 /*
365 * But we keep the older depth mask for the line separator
366 * to keep the level link until we reach the last child
367 */
368 ret += ipchain__fprintf_graph_line(fp, depth, depth_mask,
369 left_margin);
370 i = 0;
371 list_for_each_entry(chain, &child->val, list) {
372 ret += ipchain__fprintf_graph(fp, chain, depth,
373 new_depth_mask, i++,
374 new_total,
375 cumul,
376 left_margin);
377 }
378 ret += __callchain__fprintf_graph(fp, child, new_total,
379 depth + 1,
380 new_depth_mask | (1 << depth),
381 left_margin);
382 node = next;
383 if (++entries_printed == callchain_param.print_limit)
384 break;
385 }
386
387 if (callchain_param.mode == CHAIN_GRAPH_REL &&
388 remaining && remaining != new_total) {
389
390 if (!rem_sq_bracket)
391 return ret;
392
393 new_depth_mask &= ~(1 << (depth - 1));
394
395 ret += ipchain__fprintf_graph(fp, &rem_hits, depth,
396 new_depth_mask, 0, new_total,
397 remaining, left_margin);
398 }
399
400 return ret;
401 }
402
403 static size_t callchain__fprintf_graph(FILE *fp, struct callchain_node *self,
404 u64 total_samples, int left_margin)
405 {
406 struct callchain_list *chain;
407 bool printed = false;
408 int i = 0;
409 int ret = 0;
410 u32 entries_printed = 0;
411
412 list_for_each_entry(chain, &self->val, list) {
413 if (!i++ && sort__first_dimension == SORT_SYM)
414 continue;
415
416 if (!printed) {
417 ret += callchain__fprintf_left_margin(fp, left_margin);
418 ret += fprintf(fp, "|\n");
419 ret += callchain__fprintf_left_margin(fp, left_margin);
420 ret += fprintf(fp, "---");
421
422 left_margin += 3;
423 printed = true;
424 } else
425 ret += callchain__fprintf_left_margin(fp, left_margin);
426
427 if (chain->ms.sym)
428 ret += fprintf(fp, " %s\n", chain->ms.sym->name);
429 else
430 ret += fprintf(fp, " %p\n", (void *)(long)chain->ip);
431
432 if (++entries_printed == callchain_param.print_limit)
433 break;
434 }
435
436 ret += __callchain__fprintf_graph(fp, self, total_samples, 1, 1, left_margin);
437
438 return ret;
439 }
440
441 static size_t callchain__fprintf_flat(FILE *fp, struct callchain_node *self,
442 u64 total_samples)
443 {
444 struct callchain_list *chain;
445 size_t ret = 0;
446
447 if (!self)
448 return 0;
449
450 ret += callchain__fprintf_flat(fp, self->parent, total_samples);
451
452
453 list_for_each_entry(chain, &self->val, list) {
454 if (chain->ip >= PERF_CONTEXT_MAX)
455 continue;
456 if (chain->ms.sym)
457 ret += fprintf(fp, " %s\n", chain->ms.sym->name);
458 else
459 ret += fprintf(fp, " %p\n",
460 (void *)(long)chain->ip);
461 }
462
463 return ret;
464 }
465
466 static size_t hist_entry_callchain__fprintf(FILE *fp, struct hist_entry *self,
467 u64 total_samples, int left_margin)
468 {
469 struct rb_node *rb_node;
470 struct callchain_node *chain;
471 size_t ret = 0;
472 u32 entries_printed = 0;
473
474 rb_node = rb_first(&self->sorted_chain);
475 while (rb_node) {
476 double percent;
477
478 chain = rb_entry(rb_node, struct callchain_node, rb_node);
479 percent = chain->hit * 100.0 / total_samples;
480 switch (callchain_param.mode) {
481 case CHAIN_FLAT:
482 ret += percent_color_fprintf(fp, " %6.2f%%\n",
483 percent);
484 ret += callchain__fprintf_flat(fp, chain, total_samples);
485 break;
486 case CHAIN_GRAPH_ABS: /* Falldown */
487 case CHAIN_GRAPH_REL:
488 ret += callchain__fprintf_graph(fp, chain, total_samples,
489 left_margin);
490 case CHAIN_NONE:
491 default:
492 break;
493 }
494 ret += fprintf(fp, "\n");
495 if (++entries_printed == callchain_param.print_limit)
496 break;
497 rb_node = rb_next(rb_node);
498 }
499
500 return ret;
501 }
502
503 int hist_entry__snprintf(struct hist_entry *self,
504 char *s, size_t size,
505 struct perf_session *pair_session,
506 bool show_displacement,
507 long displacement, bool color,
508 u64 session_total)
509 {
510 struct sort_entry *se;
511 u64 count, total, count_sys, count_us, count_guest_sys, count_guest_us;
512 const char *sep = symbol_conf.field_sep;
513 int ret;
514
515 if (symbol_conf.exclude_other && !self->parent)
516 return 0;
517
518 if (pair_session) {
519 count = self->pair ? self->pair->count : 0;
520 total = pair_session->events_stats.total;
521 count_sys = self->pair ? self->pair->count_sys : 0;
522 count_us = self->pair ? self->pair->count_us : 0;
523 count_guest_sys = self->pair ? self->pair->count_guest_sys : 0;
524 count_guest_us = self->pair ? self->pair->count_guest_us : 0;
525 } else {
526 count = self->count;
527 total = session_total;
528 count_sys = self->count_sys;
529 count_us = self->count_us;
530 count_guest_sys = self->count_guest_sys;
531 count_guest_us = self->count_guest_us;
532 }
533
534 if (total) {
535 if (color)
536 ret = percent_color_snprintf(s, size,
537 sep ? "%.2f" : " %6.2f%%",
538 (count * 100.0) / total);
539 else
540 ret = snprintf(s, size, sep ? "%.2f" : " %6.2f%%",
541 (count * 100.0) / total);
542 if (symbol_conf.show_cpu_utilization) {
543 ret += percent_color_snprintf(s + ret, size - ret,
544 sep ? "%.2f" : " %6.2f%%",
545 (count_sys * 100.0) / total);
546 ret += percent_color_snprintf(s + ret, size - ret,
547 sep ? "%.2f" : " %6.2f%%",
548 (count_us * 100.0) / total);
549 if (perf_guest) {
550 ret += percent_color_snprintf(s + ret,
551 size - ret,
552 sep ? "%.2f" : " %6.2f%%",
553 (count_guest_sys * 100.0) /
554 total);
555 ret += percent_color_snprintf(s + ret,
556 size - ret,
557 sep ? "%.2f" : " %6.2f%%",
558 (count_guest_us * 100.0) /
559 total);
560 }
561 }
562 } else
563 ret = snprintf(s, size, sep ? "%lld" : "%12lld ", count);
564
565 if (symbol_conf.show_nr_samples) {
566 if (sep)
567 ret += snprintf(s + ret, size - ret, "%c%lld", *sep, count);
568 else
569 ret += snprintf(s + ret, size - ret, "%11lld", count);
570 }
571
572 if (pair_session) {
573 char bf[32];
574 double old_percent = 0, new_percent = 0, diff;
575
576 if (total > 0)
577 old_percent = (count * 100.0) / total;
578 if (session_total > 0)
579 new_percent = (self->count * 100.0) / session_total;
580
581 diff = new_percent - old_percent;
582
583 if (fabs(diff) >= 0.01)
584 snprintf(bf, sizeof(bf), "%+4.2F%%", diff);
585 else
586 snprintf(bf, sizeof(bf), " ");
587
588 if (sep)
589 ret += snprintf(s + ret, size - ret, "%c%s", *sep, bf);
590 else
591 ret += snprintf(s + ret, size - ret, "%11.11s", bf);
592
593 if (show_displacement) {
594 if (displacement)
595 snprintf(bf, sizeof(bf), "%+4ld", displacement);
596 else
597 snprintf(bf, sizeof(bf), " ");
598
599 if (sep)
600 ret += snprintf(s + ret, size - ret, "%c%s", *sep, bf);
601 else
602 ret += snprintf(s + ret, size - ret, "%6.6s", bf);
603 }
604 }
605
606 list_for_each_entry(se, &hist_entry__sort_list, list) {
607 if (se->elide)
608 continue;
609
610 ret += snprintf(s + ret, size - ret, "%s", sep ?: " ");
611 ret += se->se_snprintf(self, s + ret, size - ret,
612 se->se_width ? *se->se_width : 0);
613 }
614
615 return ret;
616 }
617
618 int hist_entry__fprintf(struct hist_entry *self,
619 struct perf_session *pair_session,
620 bool show_displacement,
621 long displacement, FILE *fp,
622 u64 session_total)
623 {
624 char bf[512];
625 hist_entry__snprintf(self, bf, sizeof(bf), pair_session,
626 show_displacement, displacement,
627 true, session_total);
628 return fprintf(fp, "%s\n", bf);
629 }
630
631 static size_t hist_entry__fprintf_callchain(struct hist_entry *self, FILE *fp,
632 u64 session_total)
633 {
634 int left_margin = 0;
635
636 if (sort__first_dimension == SORT_COMM) {
637 struct sort_entry *se = list_first_entry(&hist_entry__sort_list,
638 typeof(*se), list);
639 left_margin = se->se_width ? *se->se_width : 0;
640 left_margin -= thread__comm_len(self->thread);
641 }
642
643 return hist_entry_callchain__fprintf(fp, self, session_total,
644 left_margin);
645 }
646
647 size_t perf_session__fprintf_hists(struct rb_root *hists,
648 struct perf_session *pair,
649 bool show_displacement, FILE *fp,
650 u64 session_total)
651 {
652 struct sort_entry *se;
653 struct rb_node *nd;
654 size_t ret = 0;
655 unsigned long position = 1;
656 long displacement = 0;
657 unsigned int width;
658 const char *sep = symbol_conf.field_sep;
659 char *col_width = symbol_conf.col_width_list_str;
660
661 init_rem_hits();
662
663 fprintf(fp, "# %s", pair ? "Baseline" : "Overhead");
664
665 if (symbol_conf.show_nr_samples) {
666 if (sep)
667 fprintf(fp, "%cSamples", *sep);
668 else
669 fputs(" Samples ", fp);
670 }
671
672 if (symbol_conf.show_cpu_utilization) {
673 if (sep) {
674 ret += fprintf(fp, "%csys", *sep);
675 ret += fprintf(fp, "%cus", *sep);
676 if (perf_guest) {
677 ret += fprintf(fp, "%cguest sys", *sep);
678 ret += fprintf(fp, "%cguest us", *sep);
679 }
680 } else {
681 ret += fprintf(fp, " sys ");
682 ret += fprintf(fp, " us ");
683 if (perf_guest) {
684 ret += fprintf(fp, " guest sys ");
685 ret += fprintf(fp, " guest us ");
686 }
687 }
688 }
689
690 if (pair) {
691 if (sep)
692 ret += fprintf(fp, "%cDelta", *sep);
693 else
694 ret += fprintf(fp, " Delta ");
695
696 if (show_displacement) {
697 if (sep)
698 ret += fprintf(fp, "%cDisplacement", *sep);
699 else
700 ret += fprintf(fp, " Displ");
701 }
702 }
703
704 list_for_each_entry(se, &hist_entry__sort_list, list) {
705 if (se->elide)
706 continue;
707 if (sep) {
708 fprintf(fp, "%c%s", *sep, se->se_header);
709 continue;
710 }
711 width = strlen(se->se_header);
712 if (se->se_width) {
713 if (symbol_conf.col_width_list_str) {
714 if (col_width) {
715 *se->se_width = atoi(col_width);
716 col_width = strchr(col_width, ',');
717 if (col_width)
718 ++col_width;
719 }
720 }
721 width = *se->se_width = max(*se->se_width, width);
722 }
723 fprintf(fp, " %*s", width, se->se_header);
724 }
725 fprintf(fp, "\n");
726
727 if (sep)
728 goto print_entries;
729
730 fprintf(fp, "# ........");
731 if (symbol_conf.show_nr_samples)
732 fprintf(fp, " ..........");
733 if (pair) {
734 fprintf(fp, " ..........");
735 if (show_displacement)
736 fprintf(fp, " .....");
737 }
738 list_for_each_entry(se, &hist_entry__sort_list, list) {
739 unsigned int i;
740
741 if (se->elide)
742 continue;
743
744 fprintf(fp, " ");
745 if (se->se_width)
746 width = *se->se_width;
747 else
748 width = strlen(se->se_header);
749 for (i = 0; i < width; i++)
750 fprintf(fp, ".");
751 }
752
753 fprintf(fp, "\n#\n");
754
755 print_entries:
756 for (nd = rb_first(hists); nd; nd = rb_next(nd)) {
757 struct hist_entry *h = rb_entry(nd, struct hist_entry, rb_node);
758
759 if (show_displacement) {
760 if (h->pair != NULL)
761 displacement = ((long)h->pair->position -
762 (long)position);
763 else
764 displacement = 0;
765 ++position;
766 }
767 ret += hist_entry__fprintf(h, pair, show_displacement,
768 displacement, fp, session_total);
769
770 if (symbol_conf.use_callchain)
771 ret += hist_entry__fprintf_callchain(h, fp, session_total);
772
773 if (h->ms.map == NULL && verbose > 1) {
774 __map_groups__fprintf_maps(&h->thread->mg,
775 MAP__FUNCTION, verbose, fp);
776 fprintf(fp, "%.10s end\n", graph_dotted_line);
777 }
778 }
779
780 free(rem_sq_bracket);
781
782 return ret;
783 }
This page took 0.055824 seconds and 5 git commands to generate.