xref: /linux/tools/perf/util/hist.c (revision d229807f669ba3dea9f64467ee965051c4366aed)
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 *self,
80 					   unsigned int cpumode, u64 period)
81 {
82 	switch (cpumode) {
83 	case PERF_RECORD_MISC_KERNEL:
84 		self->period_sys += period;
85 		break;
86 	case PERF_RECORD_MISC_USER:
87 		self->period_us += period;
88 		break;
89 	case PERF_RECORD_MISC_GUEST_KERNEL:
90 		self->period_guest_sys += period;
91 		break;
92 	case PERF_RECORD_MISC_GUEST_USER:
93 		self->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 *self = malloc(sizeof(*self) + callchain_size);
169 
170 	if (self != NULL) {
171 		*self = *template;
172 		self->nr_events = 1;
173 		if (self->ms.map)
174 			self->ms.map->referenced = true;
175 		if (symbol_conf.use_callchain)
176 			callchain_init(self->callchain);
177 	}
178 
179 	return self;
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 	hists->stats.total_period = 0;
369 
370 	while (next) {
371 		n = rb_entry(next, struct hist_entry, rb_node_in);
372 		next = rb_next(&n->rb_node_in);
373 
374 		rb_erase(&n->rb_node_in, root);
375 		if (hists__collapse_insert_entry(hists, &hists->entries_collapsed, n)) {
376 			/*
377 			 * If it wasn't combined with one of the entries already
378 			 * collapsed, we need to apply the filters that may have
379 			 * been set by, say, the hist_browser.
380 			 */
381 			hists__apply_filters(hists, n);
382 			hists__inc_nr_entries(hists, n);
383 		}
384 	}
385 }
386 
387 void hists__collapse_resort(struct hists *hists)
388 {
389 	return __hists__collapse_resort(hists, false);
390 }
391 
392 void hists__collapse_resort_threaded(struct hists *hists)
393 {
394 	return __hists__collapse_resort(hists, true);
395 }
396 
397 /*
398  * reverse the map, sort on period.
399  */
400 
401 static void __hists__insert_output_entry(struct rb_root *entries,
402 					 struct hist_entry *he,
403 					 u64 min_callchain_hits)
404 {
405 	struct rb_node **p = &entries->rb_node;
406 	struct rb_node *parent = NULL;
407 	struct hist_entry *iter;
408 
409 	if (symbol_conf.use_callchain)
410 		callchain_param.sort(&he->sorted_chain, he->callchain,
411 				      min_callchain_hits, &callchain_param);
412 
413 	while (*p != NULL) {
414 		parent = *p;
415 		iter = rb_entry(parent, struct hist_entry, rb_node);
416 
417 		if (he->period > iter->period)
418 			p = &(*p)->rb_left;
419 		else
420 			p = &(*p)->rb_right;
421 	}
422 
423 	rb_link_node(&he->rb_node, parent, p);
424 	rb_insert_color(&he->rb_node, entries);
425 }
426 
427 static void __hists__output_resort(struct hists *hists, bool threaded)
428 {
429 	struct rb_root *root;
430 	struct rb_node *next;
431 	struct hist_entry *n;
432 	u64 min_callchain_hits;
433 
434 	min_callchain_hits = hists->stats.total_period * (callchain_param.min_percent / 100);
435 
436 	if (sort__need_collapse || threaded)
437 		root = &hists->entries_collapsed;
438 	else
439 		root = hists->entries_in;
440 
441 	next = rb_first(root);
442 	hists->entries = RB_ROOT;
443 
444 	hists->nr_entries = 0;
445 	hists__reset_col_len(hists);
446 
447 	while (next) {
448 		n = rb_entry(next, struct hist_entry, rb_node_in);
449 		next = rb_next(&n->rb_node_in);
450 
451 		__hists__insert_output_entry(&hists->entries, n, min_callchain_hits);
452 		hists__inc_nr_entries(hists, n);
453 	}
454 }
455 
456 void hists__output_resort(struct hists *hists)
457 {
458 	return __hists__output_resort(hists, false);
459 }
460 
461 void hists__output_resort_threaded(struct hists *hists)
462 {
463 	return __hists__output_resort(hists, true);
464 }
465 
466 static size_t callchain__fprintf_left_margin(FILE *fp, int left_margin)
467 {
468 	int i;
469 	int ret = fprintf(fp, "            ");
470 
471 	for (i = 0; i < left_margin; i++)
472 		ret += fprintf(fp, " ");
473 
474 	return ret;
475 }
476 
477 static size_t ipchain__fprintf_graph_line(FILE *fp, int depth, int depth_mask,
478 					  int left_margin)
479 {
480 	int i;
481 	size_t ret = callchain__fprintf_left_margin(fp, left_margin);
482 
483 	for (i = 0; i < depth; i++)
484 		if (depth_mask & (1 << i))
485 			ret += fprintf(fp, "|          ");
486 		else
487 			ret += fprintf(fp, "           ");
488 
489 	ret += fprintf(fp, "\n");
490 
491 	return ret;
492 }
493 
494 static size_t ipchain__fprintf_graph(FILE *fp, struct callchain_list *chain,
495 				     int depth, int depth_mask, int period,
496 				     u64 total_samples, u64 hits,
497 				     int left_margin)
498 {
499 	int i;
500 	size_t ret = 0;
501 
502 	ret += callchain__fprintf_left_margin(fp, left_margin);
503 	for (i = 0; i < depth; i++) {
504 		if (depth_mask & (1 << i))
505 			ret += fprintf(fp, "|");
506 		else
507 			ret += fprintf(fp, " ");
508 		if (!period && i == depth - 1) {
509 			double percent;
510 
511 			percent = hits * 100.0 / total_samples;
512 			ret += percent_color_fprintf(fp, "--%2.2f%%-- ", percent);
513 		} else
514 			ret += fprintf(fp, "%s", "          ");
515 	}
516 	if (chain->ms.sym)
517 		ret += fprintf(fp, "%s\n", chain->ms.sym->name);
518 	else
519 		ret += fprintf(fp, "%p\n", (void *)(long)chain->ip);
520 
521 	return ret;
522 }
523 
524 static struct symbol *rem_sq_bracket;
525 static struct callchain_list rem_hits;
526 
527 static void init_rem_hits(void)
528 {
529 	rem_sq_bracket = malloc(sizeof(*rem_sq_bracket) + 6);
530 	if (!rem_sq_bracket) {
531 		fprintf(stderr, "Not enough memory to display remaining hits\n");
532 		return;
533 	}
534 
535 	strcpy(rem_sq_bracket->name, "[...]");
536 	rem_hits.ms.sym = rem_sq_bracket;
537 }
538 
539 static size_t __callchain__fprintf_graph(FILE *fp, struct callchain_node *self,
540 					 u64 total_samples, int depth,
541 					 int depth_mask, int left_margin)
542 {
543 	struct rb_node *node, *next;
544 	struct callchain_node *child;
545 	struct callchain_list *chain;
546 	int new_depth_mask = depth_mask;
547 	u64 new_total;
548 	u64 remaining;
549 	size_t ret = 0;
550 	int i;
551 	uint entries_printed = 0;
552 
553 	if (callchain_param.mode == CHAIN_GRAPH_REL)
554 		new_total = self->children_hit;
555 	else
556 		new_total = total_samples;
557 
558 	remaining = new_total;
559 
560 	node = rb_first(&self->rb_root);
561 	while (node) {
562 		u64 cumul;
563 
564 		child = rb_entry(node, struct callchain_node, rb_node);
565 		cumul = callchain_cumul_hits(child);
566 		remaining -= cumul;
567 
568 		/*
569 		 * The depth mask manages the output of pipes that show
570 		 * the depth. We don't want to keep the pipes of the current
571 		 * level for the last child of this depth.
572 		 * Except if we have remaining filtered hits. They will
573 		 * supersede the last child
574 		 */
575 		next = rb_next(node);
576 		if (!next && (callchain_param.mode != CHAIN_GRAPH_REL || !remaining))
577 			new_depth_mask &= ~(1 << (depth - 1));
578 
579 		/*
580 		 * But we keep the older depth mask for the line separator
581 		 * to keep the level link until we reach the last child
582 		 */
583 		ret += ipchain__fprintf_graph_line(fp, depth, depth_mask,
584 						   left_margin);
585 		i = 0;
586 		list_for_each_entry(chain, &child->val, list) {
587 			ret += ipchain__fprintf_graph(fp, chain, depth,
588 						      new_depth_mask, i++,
589 						      new_total,
590 						      cumul,
591 						      left_margin);
592 		}
593 		ret += __callchain__fprintf_graph(fp, child, new_total,
594 						  depth + 1,
595 						  new_depth_mask | (1 << depth),
596 						  left_margin);
597 		node = next;
598 		if (++entries_printed == callchain_param.print_limit)
599 			break;
600 	}
601 
602 	if (callchain_param.mode == CHAIN_GRAPH_REL &&
603 		remaining && remaining != new_total) {
604 
605 		if (!rem_sq_bracket)
606 			return ret;
607 
608 		new_depth_mask &= ~(1 << (depth - 1));
609 
610 		ret += ipchain__fprintf_graph(fp, &rem_hits, depth,
611 					      new_depth_mask, 0, new_total,
612 					      remaining, left_margin);
613 	}
614 
615 	return ret;
616 }
617 
618 static size_t callchain__fprintf_graph(FILE *fp, struct callchain_node *self,
619 				       u64 total_samples, int left_margin)
620 {
621 	struct callchain_list *chain;
622 	bool printed = false;
623 	int i = 0;
624 	int ret = 0;
625 	u32 entries_printed = 0;
626 
627 	list_for_each_entry(chain, &self->val, list) {
628 		if (!i++ && sort__first_dimension == SORT_SYM)
629 			continue;
630 
631 		if (!printed) {
632 			ret += callchain__fprintf_left_margin(fp, left_margin);
633 			ret += fprintf(fp, "|\n");
634 			ret += callchain__fprintf_left_margin(fp, left_margin);
635 			ret += fprintf(fp, "---");
636 
637 			left_margin += 3;
638 			printed = true;
639 		} else
640 			ret += callchain__fprintf_left_margin(fp, left_margin);
641 
642 		if (chain->ms.sym)
643 			ret += fprintf(fp, " %s\n", chain->ms.sym->name);
644 		else
645 			ret += fprintf(fp, " %p\n", (void *)(long)chain->ip);
646 
647 		if (++entries_printed == callchain_param.print_limit)
648 			break;
649 	}
650 
651 	ret += __callchain__fprintf_graph(fp, self, total_samples, 1, 1, left_margin);
652 
653 	return ret;
654 }
655 
656 static size_t callchain__fprintf_flat(FILE *fp, struct callchain_node *self,
657 				      u64 total_samples)
658 {
659 	struct callchain_list *chain;
660 	size_t ret = 0;
661 
662 	if (!self)
663 		return 0;
664 
665 	ret += callchain__fprintf_flat(fp, self->parent, total_samples);
666 
667 
668 	list_for_each_entry(chain, &self->val, list) {
669 		if (chain->ip >= PERF_CONTEXT_MAX)
670 			continue;
671 		if (chain->ms.sym)
672 			ret += fprintf(fp, "                %s\n", chain->ms.sym->name);
673 		else
674 			ret += fprintf(fp, "                %p\n",
675 					(void *)(long)chain->ip);
676 	}
677 
678 	return ret;
679 }
680 
681 static size_t hist_entry_callchain__fprintf(FILE *fp, struct hist_entry *self,
682 					    u64 total_samples, int left_margin)
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(&self->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 *self, char *s,
735 				     size_t size, struct hists *pair_hists,
736 				     bool show_displacement, long displacement,
737 				     bool color, u64 session_total)
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 && !self->parent)
745 		return 0;
746 
747 	if (pair_hists) {
748 		period = self->pair ? self->pair->period : 0;
749 		nr_events = self->pair ? self->pair->nr_events : 0;
750 		total = pair_hists->stats.total_period;
751 		period_sys = self->pair ? self->pair->period_sys : 0;
752 		period_us = self->pair ? self->pair->period_us : 0;
753 		period_guest_sys = self->pair ? self->pair->period_guest_sys : 0;
754 		period_guest_us = self->pair ? self->pair->period_guest_us : 0;
755 	} else {
756 		period = self->period;
757 		nr_events = self->nr_events;
758 		total = session_total;
759 		period_sys = self->period_sys;
760 		period_us = self->period_us;
761 		period_guest_sys = self->period_guest_sys;
762 		period_guest_us = self->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 (session_total > 0)
817 			new_percent = (self->period * 100.0) / session_total;
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 int hist_entry__fprintf(struct hist_entry *he, size_t size, struct hists *hists,
867 			struct hists *pair_hists, bool show_displacement,
868 			long displacement, FILE *fp, u64 session_total)
869 {
870 	char bf[512];
871 	int ret;
872 
873 	if (size == 0 || size > sizeof(bf))
874 		size = sizeof(bf);
875 
876 	ret = hist_entry__pcnt_snprintf(he, bf, size, pair_hists,
877 					show_displacement, displacement,
878 					true, session_total);
879 	hist_entry__snprintf(he, bf + ret, size - ret, hists);
880 	return fprintf(fp, "%s\n", bf);
881 }
882 
883 static size_t hist_entry__fprintf_callchain(struct hist_entry *self,
884 					    struct hists *hists, FILE *fp,
885 					    u64 session_total)
886 {
887 	int left_margin = 0;
888 
889 	if (sort__first_dimension == SORT_COMM) {
890 		struct sort_entry *se = list_first_entry(&hist_entry__sort_list,
891 							 typeof(*se), list);
892 		left_margin = hists__col_len(hists, se->se_width_idx);
893 		left_margin -= thread__comm_len(self->thread);
894 	}
895 
896 	return hist_entry_callchain__fprintf(fp, self, session_total,
897 					     left_margin);
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 	unsigned long position = 1;
908 	long displacement = 0;
909 	unsigned int width;
910 	const char *sep = symbol_conf.field_sep;
911 	const char *col_width = symbol_conf.col_width_list_str;
912 	int nr_rows = 0;
913 
914 	init_rem_hits();
915 
916 	if (!show_header)
917 		goto print_entries;
918 
919 	fprintf(fp, "# %s", pair ? "Baseline" : "Overhead");
920 
921 	if (symbol_conf.show_nr_samples) {
922 		if (sep)
923 			fprintf(fp, "%cSamples", *sep);
924 		else
925 			fputs("  Samples  ", fp);
926 	}
927 
928 	if (symbol_conf.show_total_period) {
929 		if (sep)
930 			ret += fprintf(fp, "%cPeriod", *sep);
931 		else
932 			ret += fprintf(fp, "   Period    ");
933 	}
934 
935 	if (symbol_conf.show_cpu_utilization) {
936 		if (sep) {
937 			ret += fprintf(fp, "%csys", *sep);
938 			ret += fprintf(fp, "%cus", *sep);
939 			if (perf_guest) {
940 				ret += fprintf(fp, "%cguest sys", *sep);
941 				ret += fprintf(fp, "%cguest us", *sep);
942 			}
943 		} else {
944 			ret += fprintf(fp, "  sys  ");
945 			ret += fprintf(fp, "  us  ");
946 			if (perf_guest) {
947 				ret += fprintf(fp, "  guest sys  ");
948 				ret += fprintf(fp, "  guest us  ");
949 			}
950 		}
951 	}
952 
953 	if (pair) {
954 		if (sep)
955 			ret += fprintf(fp, "%cDelta", *sep);
956 		else
957 			ret += fprintf(fp, "  Delta    ");
958 
959 		if (show_displacement) {
960 			if (sep)
961 				ret += fprintf(fp, "%cDisplacement", *sep);
962 			else
963 				ret += fprintf(fp, " Displ");
964 		}
965 	}
966 
967 	list_for_each_entry(se, &hist_entry__sort_list, list) {
968 		if (se->elide)
969 			continue;
970 		if (sep) {
971 			fprintf(fp, "%c%s", *sep, se->se_header);
972 			continue;
973 		}
974 		width = strlen(se->se_header);
975 		if (symbol_conf.col_width_list_str) {
976 			if (col_width) {
977 				hists__set_col_len(hists, se->se_width_idx,
978 						   atoi(col_width));
979 				col_width = strchr(col_width, ',');
980 				if (col_width)
981 					++col_width;
982 			}
983 		}
984 		if (!hists__new_col_len(hists, se->se_width_idx, width))
985 			width = hists__col_len(hists, se->se_width_idx);
986 		fprintf(fp, "  %*s", width, se->se_header);
987 	}
988 
989 	fprintf(fp, "\n");
990 	if (max_rows && ++nr_rows >= max_rows)
991 		goto out;
992 
993 	if (sep)
994 		goto print_entries;
995 
996 	fprintf(fp, "# ........");
997 	if (symbol_conf.show_nr_samples)
998 		fprintf(fp, " ..........");
999 	if (symbol_conf.show_total_period)
1000 		fprintf(fp, " ............");
1001 	if (pair) {
1002 		fprintf(fp, " ..........");
1003 		if (show_displacement)
1004 			fprintf(fp, " .....");
1005 	}
1006 	list_for_each_entry(se, &hist_entry__sort_list, list) {
1007 		unsigned int i;
1008 
1009 		if (se->elide)
1010 			continue;
1011 
1012 		fprintf(fp, "  ");
1013 		width = hists__col_len(hists, se->se_width_idx);
1014 		if (width == 0)
1015 			width = strlen(se->se_header);
1016 		for (i = 0; i < width; i++)
1017 			fprintf(fp, ".");
1018 	}
1019 
1020 	fprintf(fp, "\n");
1021 	if (max_rows && ++nr_rows >= max_rows)
1022 		goto out;
1023 
1024 	fprintf(fp, "#\n");
1025 	if (max_rows && ++nr_rows >= max_rows)
1026 		goto out;
1027 
1028 print_entries:
1029 	for (nd = rb_first(&hists->entries); nd; nd = rb_next(nd)) {
1030 		struct hist_entry *h = rb_entry(nd, struct hist_entry, rb_node);
1031 
1032 		if (h->filtered)
1033 			continue;
1034 
1035 		if (show_displacement) {
1036 			if (h->pair != NULL)
1037 				displacement = ((long)h->pair->position -
1038 					        (long)position);
1039 			else
1040 				displacement = 0;
1041 			++position;
1042 		}
1043 		ret += hist_entry__fprintf(h, max_cols, hists, pair, show_displacement,
1044 					   displacement, fp, hists->stats.total_period);
1045 
1046 		if (symbol_conf.use_callchain)
1047 			ret += hist_entry__fprintf_callchain(h, hists, fp,
1048 							     hists->stats.total_period);
1049 		if (max_rows && ++nr_rows >= max_rows)
1050 			goto out;
1051 
1052 		if (h->ms.map == NULL && verbose > 1) {
1053 			__map_groups__fprintf_maps(&h->thread->mg,
1054 						   MAP__FUNCTION, verbose, fp);
1055 			fprintf(fp, "%.10s end\n", graph_dotted_line);
1056 		}
1057 	}
1058 out:
1059 	free(rem_sq_bracket);
1060 
1061 	return ret;
1062 }
1063 
1064 /*
1065  * See hists__fprintf to match the column widths
1066  */
1067 unsigned int hists__sort_list_width(struct hists *hists)
1068 {
1069 	struct sort_entry *se;
1070 	int ret = 9; /* total % */
1071 
1072 	if (symbol_conf.show_cpu_utilization) {
1073 		ret += 7; /* count_sys % */
1074 		ret += 6; /* count_us % */
1075 		if (perf_guest) {
1076 			ret += 13; /* count_guest_sys % */
1077 			ret += 12; /* count_guest_us % */
1078 		}
1079 	}
1080 
1081 	if (symbol_conf.show_nr_samples)
1082 		ret += 11;
1083 
1084 	if (symbol_conf.show_total_period)
1085 		ret += 13;
1086 
1087 	list_for_each_entry(se, &hist_entry__sort_list, list)
1088 		if (!se->elide)
1089 			ret += 2 + hists__col_len(hists, se->se_width_idx);
1090 
1091 	if (verbose) /* Addr + origin */
1092 		ret += 3 + BITS_PER_LONG / 4;
1093 
1094 	return ret;
1095 }
1096 
1097 static void hists__remove_entry_filter(struct hists *hists, struct hist_entry *h,
1098 				       enum hist_filter filter)
1099 {
1100 	h->filtered &= ~(1 << filter);
1101 	if (h->filtered)
1102 		return;
1103 
1104 	++hists->nr_entries;
1105 	if (h->ms.unfolded)
1106 		hists->nr_entries += h->nr_rows;
1107 	h->row_offset = 0;
1108 	hists->stats.total_period += h->period;
1109 	hists->stats.nr_events[PERF_RECORD_SAMPLE] += h->nr_events;
1110 
1111 	hists__calc_col_len(hists, h);
1112 }
1113 
1114 
1115 static bool hists__filter_entry_by_dso(struct hists *hists,
1116 				       struct hist_entry *he)
1117 {
1118 	if (hists->dso_filter != NULL &&
1119 	    (he->ms.map == NULL || he->ms.map->dso != hists->dso_filter)) {
1120 		he->filtered |= (1 << HIST_FILTER__DSO);
1121 		return true;
1122 	}
1123 
1124 	return false;
1125 }
1126 
1127 void hists__filter_by_dso(struct hists *hists)
1128 {
1129 	struct rb_node *nd;
1130 
1131 	hists->nr_entries = hists->stats.total_period = 0;
1132 	hists->stats.nr_events[PERF_RECORD_SAMPLE] = 0;
1133 	hists__reset_col_len(hists);
1134 
1135 	for (nd = rb_first(&hists->entries); nd; nd = rb_next(nd)) {
1136 		struct hist_entry *h = rb_entry(nd, struct hist_entry, rb_node);
1137 
1138 		if (symbol_conf.exclude_other && !h->parent)
1139 			continue;
1140 
1141 		if (hists__filter_entry_by_dso(hists, h))
1142 			continue;
1143 
1144 		hists__remove_entry_filter(hists, h, HIST_FILTER__DSO);
1145 	}
1146 }
1147 
1148 static bool hists__filter_entry_by_thread(struct hists *hists,
1149 					  struct hist_entry *he)
1150 {
1151 	if (hists->thread_filter != NULL &&
1152 	    he->thread != hists->thread_filter) {
1153 		he->filtered |= (1 << HIST_FILTER__THREAD);
1154 		return true;
1155 	}
1156 
1157 	return false;
1158 }
1159 
1160 void hists__filter_by_thread(struct hists *hists)
1161 {
1162 	struct rb_node *nd;
1163 
1164 	hists->nr_entries = hists->stats.total_period = 0;
1165 	hists->stats.nr_events[PERF_RECORD_SAMPLE] = 0;
1166 	hists__reset_col_len(hists);
1167 
1168 	for (nd = rb_first(&hists->entries); nd; nd = rb_next(nd)) {
1169 		struct hist_entry *h = rb_entry(nd, struct hist_entry, rb_node);
1170 
1171 		if (hists__filter_entry_by_thread(hists, h))
1172 			continue;
1173 
1174 		hists__remove_entry_filter(hists, h, HIST_FILTER__THREAD);
1175 	}
1176 }
1177 
1178 int hist_entry__inc_addr_samples(struct hist_entry *he, int evidx, u64 ip)
1179 {
1180 	return symbol__inc_addr_samples(he->ms.sym, he->ms.map, evidx, ip);
1181 }
1182 
1183 int hist_entry__annotate(struct hist_entry *he, size_t privsize)
1184 {
1185 	return symbol__annotate(he->ms.sym, he->ms.map, privsize);
1186 }
1187 
1188 void hists__inc_nr_events(struct hists *hists, u32 type)
1189 {
1190 	++hists->stats.nr_events[0];
1191 	++hists->stats.nr_events[type];
1192 }
1193 
1194 size_t hists__fprintf_nr_events(struct hists *hists, FILE *fp)
1195 {
1196 	int i;
1197 	size_t ret = 0;
1198 
1199 	for (i = 0; i < PERF_RECORD_HEADER_MAX; ++i) {
1200 		const char *name;
1201 
1202 		if (hists->stats.nr_events[i] == 0)
1203 			continue;
1204 
1205 		name = perf_event__name(i);
1206 		if (!strcmp(name, "UNKNOWN"))
1207 			continue;
1208 
1209 		ret += fprintf(fp, "%16s events: %10d\n", name,
1210 			       hists->stats.nr_events[i]);
1211 	}
1212 
1213 	return ret;
1214 }
1215 
1216 void hists__init(struct hists *hists)
1217 {
1218 	memset(hists, 0, sizeof(*hists));
1219 	hists->entries_in_array[0] = hists->entries_in_array[1] = RB_ROOT;
1220 	hists->entries_in = &hists->entries_in_array[0];
1221 	hists->entries_collapsed = RB_ROOT;
1222 	hists->entries = RB_ROOT;
1223 	pthread_mutex_init(&hists->lock, NULL);
1224 }
1225