GNU Linux-libre 4.14.332-gnu1
[releases.git] / tools / perf / util / callchain.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2009-2011, Frederic Weisbecker <fweisbec@gmail.com>
4  *
5  * Handle the callchains from the stream in an ad-hoc radix tree and then
6  * sort them in an rbtree.
7  *
8  * Using a radix for code path provides a fast retrieval and factorizes
9  * memory use. Also that lets us use the paths in a hierarchical graph view.
10  *
11  */
12
13 #include <inttypes.h>
14 #include <stdlib.h>
15 #include <stdio.h>
16 #include <stdbool.h>
17 #include <errno.h>
18 #include <math.h>
19
20 #include "asm/bug.h"
21
22 #include "hist.h"
23 #include "util.h"
24 #include "sort.h"
25 #include "machine.h"
26 #include "callchain.h"
27 #include "branch.h"
28
29 #define CALLCHAIN_PARAM_DEFAULT                 \
30         .mode           = CHAIN_GRAPH_ABS,      \
31         .min_percent    = 0.5,                  \
32         .order          = ORDER_CALLEE,         \
33         .key            = CCKEY_FUNCTION,       \
34         .value          = CCVAL_PERCENT,        \
35
36 struct callchain_param callchain_param = {
37         CALLCHAIN_PARAM_DEFAULT
38 };
39
40 /*
41  * Are there any events usind DWARF callchains?
42  *
43  * I.e.
44  *
45  * -e cycles/call-graph=dwarf/
46  */
47 bool dwarf_callchain_users;
48
49 struct callchain_param callchain_param_default = {
50         CALLCHAIN_PARAM_DEFAULT
51 };
52
53 __thread struct callchain_cursor callchain_cursor;
54
55 int parse_callchain_record_opt(const char *arg, struct callchain_param *param)
56 {
57         return parse_callchain_record(arg, param);
58 }
59
60 static int parse_callchain_mode(const char *value)
61 {
62         if (!strncmp(value, "graph", strlen(value))) {
63                 callchain_param.mode = CHAIN_GRAPH_ABS;
64                 return 0;
65         }
66         if (!strncmp(value, "flat", strlen(value))) {
67                 callchain_param.mode = CHAIN_FLAT;
68                 return 0;
69         }
70         if (!strncmp(value, "fractal", strlen(value))) {
71                 callchain_param.mode = CHAIN_GRAPH_REL;
72                 return 0;
73         }
74         if (!strncmp(value, "folded", strlen(value))) {
75                 callchain_param.mode = CHAIN_FOLDED;
76                 return 0;
77         }
78         return -1;
79 }
80
81 static int parse_callchain_order(const char *value)
82 {
83         if (!strncmp(value, "caller", strlen(value))) {
84                 callchain_param.order = ORDER_CALLER;
85                 callchain_param.order_set = true;
86                 return 0;
87         }
88         if (!strncmp(value, "callee", strlen(value))) {
89                 callchain_param.order = ORDER_CALLEE;
90                 callchain_param.order_set = true;
91                 return 0;
92         }
93         return -1;
94 }
95
96 static int parse_callchain_sort_key(const char *value)
97 {
98         if (!strncmp(value, "function", strlen(value))) {
99                 callchain_param.key = CCKEY_FUNCTION;
100                 return 0;
101         }
102         if (!strncmp(value, "address", strlen(value))) {
103                 callchain_param.key = CCKEY_ADDRESS;
104                 return 0;
105         }
106         if (!strncmp(value, "srcline", strlen(value))) {
107                 callchain_param.key = CCKEY_SRCLINE;
108                 return 0;
109         }
110         if (!strncmp(value, "branch", strlen(value))) {
111                 callchain_param.branch_callstack = 1;
112                 return 0;
113         }
114         return -1;
115 }
116
117 static int parse_callchain_value(const char *value)
118 {
119         if (!strncmp(value, "percent", strlen(value))) {
120                 callchain_param.value = CCVAL_PERCENT;
121                 return 0;
122         }
123         if (!strncmp(value, "period", strlen(value))) {
124                 callchain_param.value = CCVAL_PERIOD;
125                 return 0;
126         }
127         if (!strncmp(value, "count", strlen(value))) {
128                 callchain_param.value = CCVAL_COUNT;
129                 return 0;
130         }
131         return -1;
132 }
133
134 static int get_stack_size(const char *str, unsigned long *_size)
135 {
136         char *endptr;
137         unsigned long size;
138         unsigned long max_size = round_down(USHRT_MAX, sizeof(u64));
139
140         size = strtoul(str, &endptr, 0);
141
142         do {
143                 if (*endptr)
144                         break;
145
146                 size = round_up(size, sizeof(u64));
147                 if (!size || size > max_size)
148                         break;
149
150                 *_size = size;
151                 return 0;
152
153         } while (0);
154
155         pr_err("callchain: Incorrect stack dump size (max %ld): %s\n",
156                max_size, str);
157         return -1;
158 }
159
160 static int
161 __parse_callchain_report_opt(const char *arg, bool allow_record_opt)
162 {
163         char *tok;
164         char *endptr, *saveptr = NULL;
165         bool minpcnt_set = false;
166         bool record_opt_set = false;
167         bool try_stack_size = false;
168
169         callchain_param.enabled = true;
170         symbol_conf.use_callchain = true;
171
172         if (!arg)
173                 return 0;
174
175         while ((tok = strtok_r((char *)arg, ",", &saveptr)) != NULL) {
176                 if (!strncmp(tok, "none", strlen(tok))) {
177                         callchain_param.mode = CHAIN_NONE;
178                         callchain_param.enabled = false;
179                         symbol_conf.use_callchain = false;
180                         return 0;
181                 }
182
183                 if (!parse_callchain_mode(tok) ||
184                     !parse_callchain_order(tok) ||
185                     !parse_callchain_sort_key(tok) ||
186                     !parse_callchain_value(tok)) {
187                         /* parsing ok - move on to the next */
188                         try_stack_size = false;
189                         goto next;
190                 } else if (allow_record_opt && !record_opt_set) {
191                         if (parse_callchain_record(tok, &callchain_param))
192                                 goto try_numbers;
193
194                         /* assume that number followed by 'dwarf' is stack size */
195                         if (callchain_param.record_mode == CALLCHAIN_DWARF)
196                                 try_stack_size = true;
197
198                         record_opt_set = true;
199                         goto next;
200                 }
201
202 try_numbers:
203                 if (try_stack_size) {
204                         unsigned long size = 0;
205
206                         if (get_stack_size(tok, &size) < 0)
207                                 return -1;
208                         callchain_param.dump_size = size;
209                         try_stack_size = false;
210                 } else if (!minpcnt_set) {
211                         /* try to get the min percent */
212                         callchain_param.min_percent = strtod(tok, &endptr);
213                         if (tok == endptr)
214                                 return -1;
215                         minpcnt_set = true;
216                 } else {
217                         /* try print limit at last */
218                         callchain_param.print_limit = strtoul(tok, &endptr, 0);
219                         if (tok == endptr)
220                                 return -1;
221                 }
222 next:
223                 arg = NULL;
224         }
225
226         if (callchain_register_param(&callchain_param) < 0) {
227                 pr_err("Can't register callchain params\n");
228                 return -1;
229         }
230         return 0;
231 }
232
233 int parse_callchain_report_opt(const char *arg)
234 {
235         return __parse_callchain_report_opt(arg, false);
236 }
237
238 int parse_callchain_top_opt(const char *arg)
239 {
240         return __parse_callchain_report_opt(arg, true);
241 }
242
243 int parse_callchain_record(const char *arg, struct callchain_param *param)
244 {
245         char *tok, *name, *saveptr = NULL;
246         char *buf;
247         int ret = -1;
248
249         /* We need buffer that we know we can write to. */
250         buf = malloc(strlen(arg) + 1);
251         if (!buf)
252                 return -ENOMEM;
253
254         strcpy(buf, arg);
255
256         tok = strtok_r((char *)buf, ",", &saveptr);
257         name = tok ? : (char *)buf;
258
259         do {
260                 /* Framepointer style */
261                 if (!strncmp(name, "fp", sizeof("fp"))) {
262                         if (!strtok_r(NULL, ",", &saveptr)) {
263                                 param->record_mode = CALLCHAIN_FP;
264                                 ret = 0;
265                         } else
266                                 pr_err("callchain: No more arguments "
267                                        "needed for --call-graph fp\n");
268                         break;
269
270                 /* Dwarf style */
271                 } else if (!strncmp(name, "dwarf", sizeof("dwarf"))) {
272                         const unsigned long default_stack_dump_size = 8192;
273
274                         ret = 0;
275                         param->record_mode = CALLCHAIN_DWARF;
276                         param->dump_size = default_stack_dump_size;
277                         dwarf_callchain_users = true;
278
279                         tok = strtok_r(NULL, ",", &saveptr);
280                         if (tok) {
281                                 unsigned long size = 0;
282
283                                 ret = get_stack_size(tok, &size);
284                                 param->dump_size = size;
285                         }
286                 } else if (!strncmp(name, "lbr", sizeof("lbr"))) {
287                         if (!strtok_r(NULL, ",", &saveptr)) {
288                                 param->record_mode = CALLCHAIN_LBR;
289                                 ret = 0;
290                         } else
291                                 pr_err("callchain: No more arguments "
292                                         "needed for --call-graph lbr\n");
293                         break;
294                 } else {
295                         pr_err("callchain: Unknown --call-graph option "
296                                "value: %s\n", arg);
297                         break;
298                 }
299
300         } while (0);
301
302         free(buf);
303         return ret;
304 }
305
306 int perf_callchain_config(const char *var, const char *value)
307 {
308         char *endptr;
309
310         if (!strstarts(var, "call-graph."))
311                 return 0;
312         var += sizeof("call-graph.") - 1;
313
314         if (!strcmp(var, "record-mode"))
315                 return parse_callchain_record_opt(value, &callchain_param);
316         if (!strcmp(var, "dump-size")) {
317                 unsigned long size = 0;
318                 int ret;
319
320                 ret = get_stack_size(value, &size);
321                 callchain_param.dump_size = size;
322
323                 return ret;
324         }
325         if (!strcmp(var, "print-type")){
326                 int ret;
327                 ret = parse_callchain_mode(value);
328                 if (ret == -1)
329                         pr_err("Invalid callchain mode: %s\n", value);
330                 return ret;
331         }
332         if (!strcmp(var, "order")){
333                 int ret;
334                 ret = parse_callchain_order(value);
335                 if (ret == -1)
336                         pr_err("Invalid callchain order: %s\n", value);
337                 return ret;
338         }
339         if (!strcmp(var, "sort-key")){
340                 int ret;
341                 ret = parse_callchain_sort_key(value);
342                 if (ret == -1)
343                         pr_err("Invalid callchain sort key: %s\n", value);
344                 return ret;
345         }
346         if (!strcmp(var, "threshold")) {
347                 callchain_param.min_percent = strtod(value, &endptr);
348                 if (value == endptr) {
349                         pr_err("Invalid callchain threshold: %s\n", value);
350                         return -1;
351                 }
352         }
353         if (!strcmp(var, "print-limit")) {
354                 callchain_param.print_limit = strtod(value, &endptr);
355                 if (value == endptr) {
356                         pr_err("Invalid callchain print limit: %s\n", value);
357                         return -1;
358                 }
359         }
360
361         return 0;
362 }
363
364 static void
365 rb_insert_callchain(struct rb_root *root, struct callchain_node *chain,
366                     enum chain_mode mode)
367 {
368         struct rb_node **p = &root->rb_node;
369         struct rb_node *parent = NULL;
370         struct callchain_node *rnode;
371         u64 chain_cumul = callchain_cumul_hits(chain);
372
373         while (*p) {
374                 u64 rnode_cumul;
375
376                 parent = *p;
377                 rnode = rb_entry(parent, struct callchain_node, rb_node);
378                 rnode_cumul = callchain_cumul_hits(rnode);
379
380                 switch (mode) {
381                 case CHAIN_FLAT:
382                 case CHAIN_FOLDED:
383                         if (rnode->hit < chain->hit)
384                                 p = &(*p)->rb_left;
385                         else
386                                 p = &(*p)->rb_right;
387                         break;
388                 case CHAIN_GRAPH_ABS: /* Falldown */
389                 case CHAIN_GRAPH_REL:
390                         if (rnode_cumul < chain_cumul)
391                                 p = &(*p)->rb_left;
392                         else
393                                 p = &(*p)->rb_right;
394                         break;
395                 case CHAIN_NONE:
396                 default:
397                         break;
398                 }
399         }
400
401         rb_link_node(&chain->rb_node, parent, p);
402         rb_insert_color(&chain->rb_node, root);
403 }
404
405 static void
406 __sort_chain_flat(struct rb_root *rb_root, struct callchain_node *node,
407                   u64 min_hit)
408 {
409         struct rb_node *n;
410         struct callchain_node *child;
411
412         n = rb_first(&node->rb_root_in);
413         while (n) {
414                 child = rb_entry(n, struct callchain_node, rb_node_in);
415                 n = rb_next(n);
416
417                 __sort_chain_flat(rb_root, child, min_hit);
418         }
419
420         if (node->hit && node->hit >= min_hit)
421                 rb_insert_callchain(rb_root, node, CHAIN_FLAT);
422 }
423
424 /*
425  * Once we get every callchains from the stream, we can now
426  * sort them by hit
427  */
428 static void
429 sort_chain_flat(struct rb_root *rb_root, struct callchain_root *root,
430                 u64 min_hit, struct callchain_param *param __maybe_unused)
431 {
432         *rb_root = RB_ROOT;
433         __sort_chain_flat(rb_root, &root->node, min_hit);
434 }
435
436 static void __sort_chain_graph_abs(struct callchain_node *node,
437                                    u64 min_hit)
438 {
439         struct rb_node *n;
440         struct callchain_node *child;
441
442         node->rb_root = RB_ROOT;
443         n = rb_first(&node->rb_root_in);
444
445         while (n) {
446                 child = rb_entry(n, struct callchain_node, rb_node_in);
447                 n = rb_next(n);
448
449                 __sort_chain_graph_abs(child, min_hit);
450                 if (callchain_cumul_hits(child) >= min_hit)
451                         rb_insert_callchain(&node->rb_root, child,
452                                             CHAIN_GRAPH_ABS);
453         }
454 }
455
456 static void
457 sort_chain_graph_abs(struct rb_root *rb_root, struct callchain_root *chain_root,
458                      u64 min_hit, struct callchain_param *param __maybe_unused)
459 {
460         __sort_chain_graph_abs(&chain_root->node, min_hit);
461         rb_root->rb_node = chain_root->node.rb_root.rb_node;
462 }
463
464 static void __sort_chain_graph_rel(struct callchain_node *node,
465                                    double min_percent)
466 {
467         struct rb_node *n;
468         struct callchain_node *child;
469         u64 min_hit;
470
471         node->rb_root = RB_ROOT;
472         min_hit = ceil(node->children_hit * min_percent);
473
474         n = rb_first(&node->rb_root_in);
475         while (n) {
476                 child = rb_entry(n, struct callchain_node, rb_node_in);
477                 n = rb_next(n);
478
479                 __sort_chain_graph_rel(child, min_percent);
480                 if (callchain_cumul_hits(child) >= min_hit)
481                         rb_insert_callchain(&node->rb_root, child,
482                                             CHAIN_GRAPH_REL);
483         }
484 }
485
486 static void
487 sort_chain_graph_rel(struct rb_root *rb_root, struct callchain_root *chain_root,
488                      u64 min_hit __maybe_unused, struct callchain_param *param)
489 {
490         __sort_chain_graph_rel(&chain_root->node, param->min_percent / 100.0);
491         rb_root->rb_node = chain_root->node.rb_root.rb_node;
492 }
493
494 int callchain_register_param(struct callchain_param *param)
495 {
496         switch (param->mode) {
497         case CHAIN_GRAPH_ABS:
498                 param->sort = sort_chain_graph_abs;
499                 break;
500         case CHAIN_GRAPH_REL:
501                 param->sort = sort_chain_graph_rel;
502                 break;
503         case CHAIN_FLAT:
504         case CHAIN_FOLDED:
505                 param->sort = sort_chain_flat;
506                 break;
507         case CHAIN_NONE:
508         default:
509                 return -1;
510         }
511         return 0;
512 }
513
514 /*
515  * Create a child for a parent. If inherit_children, then the new child
516  * will become the new parent of it's parent children
517  */
518 static struct callchain_node *
519 create_child(struct callchain_node *parent, bool inherit_children)
520 {
521         struct callchain_node *new;
522
523         new = zalloc(sizeof(*new));
524         if (!new) {
525                 perror("not enough memory to create child for code path tree");
526                 return NULL;
527         }
528         new->parent = parent;
529         INIT_LIST_HEAD(&new->val);
530         INIT_LIST_HEAD(&new->parent_val);
531
532         if (inherit_children) {
533                 struct rb_node *n;
534                 struct callchain_node *child;
535
536                 new->rb_root_in = parent->rb_root_in;
537                 parent->rb_root_in = RB_ROOT;
538
539                 n = rb_first(&new->rb_root_in);
540                 while (n) {
541                         child = rb_entry(n, struct callchain_node, rb_node_in);
542                         child->parent = new;
543                         n = rb_next(n);
544                 }
545
546                 /* make it the first child */
547                 rb_link_node(&new->rb_node_in, NULL, &parent->rb_root_in.rb_node);
548                 rb_insert_color(&new->rb_node_in, &parent->rb_root_in);
549         }
550
551         return new;
552 }
553
554
555 /*
556  * Fill the node with callchain values
557  */
558 static int
559 fill_node(struct callchain_node *node, struct callchain_cursor *cursor)
560 {
561         struct callchain_cursor_node *cursor_node;
562
563         node->val_nr = cursor->nr - cursor->pos;
564         if (!node->val_nr)
565                 pr_warning("Warning: empty node in callchain tree\n");
566
567         cursor_node = callchain_cursor_current(cursor);
568
569         while (cursor_node) {
570                 struct callchain_list *call;
571
572                 call = zalloc(sizeof(*call));
573                 if (!call) {
574                         perror("not enough memory for the code path tree");
575                         return -1;
576                 }
577                 call->ip = cursor_node->ip;
578                 call->ms.sym = cursor_node->sym;
579                 call->ms.map = map__get(cursor_node->map);
580
581                 if (cursor_node->branch) {
582                         call->branch_count = 1;
583
584                         if (cursor_node->branch_from) {
585                                 /*
586                                  * branch_from is set with value somewhere else
587                                  * to imply it's "to" of a branch.
588                                  */
589                                 call->brtype_stat.branch_to = true;
590
591                                 if (cursor_node->branch_flags.predicted)
592                                         call->predicted_count = 1;
593
594                                 if (cursor_node->branch_flags.abort)
595                                         call->abort_count = 1;
596
597                                 branch_type_count(&call->brtype_stat,
598                                                   &cursor_node->branch_flags,
599                                                   cursor_node->branch_from,
600                                                   cursor_node->ip);
601                         } else {
602                                 /*
603                                  * It's "from" of a branch
604                                  */
605                                 call->brtype_stat.branch_to = false;
606                                 call->cycles_count =
607                                         cursor_node->branch_flags.cycles;
608                                 call->iter_count = cursor_node->nr_loop_iter;
609                                 call->iter_cycles = cursor_node->iter_cycles;
610                         }
611                 }
612
613                 list_add_tail(&call->list, &node->val);
614
615                 callchain_cursor_advance(cursor);
616                 cursor_node = callchain_cursor_current(cursor);
617         }
618         return 0;
619 }
620
621 static struct callchain_node *
622 add_child(struct callchain_node *parent,
623           struct callchain_cursor *cursor,
624           u64 period)
625 {
626         struct callchain_node *new;
627
628         new = create_child(parent, false);
629         if (new == NULL)
630                 return NULL;
631
632         if (fill_node(new, cursor) < 0) {
633                 struct callchain_list *call, *tmp;
634
635                 list_for_each_entry_safe(call, tmp, &new->val, list) {
636                         list_del(&call->list);
637                         map__zput(call->ms.map);
638                         free(call);
639                 }
640                 free(new);
641                 return NULL;
642         }
643
644         new->children_hit = 0;
645         new->hit = period;
646         new->children_count = 0;
647         new->count = 1;
648         return new;
649 }
650
651 enum match_result {
652         MATCH_ERROR  = -1,
653         MATCH_EQ,
654         MATCH_LT,
655         MATCH_GT,
656 };
657
658 static enum match_result match_chain_srcline(struct callchain_cursor_node *node,
659                                              struct callchain_list *cnode)
660 {
661         char *left = NULL;
662         char *right = NULL;
663         enum match_result ret = MATCH_EQ;
664         int cmp;
665
666         if (cnode->ms.map)
667                 left = get_srcline(cnode->ms.map->dso,
668                                  map__rip_2objdump(cnode->ms.map, cnode->ip),
669                                  cnode->ms.sym, true, false);
670         if (node->map)
671                 right = get_srcline(node->map->dso,
672                                   map__rip_2objdump(node->map, node->ip),
673                                   node->sym, true, false);
674
675         if (left && right)
676                 cmp = strcmp(left, right);
677         else if (!left && right)
678                 cmp = 1;
679         else if (left && !right)
680                 cmp = -1;
681         else if (cnode->ip == node->ip)
682                 cmp = 0;
683         else
684                 cmp = (cnode->ip < node->ip) ? -1 : 1;
685
686         if (cmp != 0)
687                 ret = cmp < 0 ? MATCH_LT : MATCH_GT;
688
689         free_srcline(left);
690         free_srcline(right);
691         return ret;
692 }
693
694 static enum match_result match_chain(struct callchain_cursor_node *node,
695                                      struct callchain_list *cnode)
696 {
697         struct symbol *sym = node->sym;
698         u64 left, right;
699         struct dso *left_dso = NULL;
700         struct dso *right_dso = NULL;
701
702         if (callchain_param.key == CCKEY_SRCLINE) {
703                 enum match_result match = match_chain_srcline(node, cnode);
704
705                 if (match != MATCH_ERROR)
706                         return match;
707         }
708
709         if (cnode->ms.sym && sym && callchain_param.key == CCKEY_FUNCTION) {
710                 left = cnode->ms.sym->start;
711                 right = sym->start;
712                 left_dso = cnode->ms.map->dso;
713                 right_dso = node->map->dso;
714         } else {
715                 left = cnode->ip;
716                 right = node->ip;
717         }
718
719         if (left == right && left_dso == right_dso) {
720                 if (node->branch) {
721                         cnode->branch_count++;
722
723                         if (node->branch_from) {
724                                 /*
725                                  * It's "to" of a branch
726                                  */
727                                 cnode->brtype_stat.branch_to = true;
728
729                                 if (node->branch_flags.predicted)
730                                         cnode->predicted_count++;
731
732                                 if (node->branch_flags.abort)
733                                         cnode->abort_count++;
734
735                                 branch_type_count(&cnode->brtype_stat,
736                                                   &node->branch_flags,
737                                                   node->branch_from,
738                                                   node->ip);
739                         } else {
740                                 /*
741                                  * It's "from" of a branch
742                                  */
743                                 cnode->brtype_stat.branch_to = false;
744                                 cnode->cycles_count +=
745                                         node->branch_flags.cycles;
746                                 cnode->iter_count += node->nr_loop_iter;
747                                 cnode->iter_cycles += node->iter_cycles;
748                         }
749                 }
750
751                 return MATCH_EQ;
752         }
753
754         return left > right ? MATCH_GT : MATCH_LT;
755 }
756
757 /*
758  * Split the parent in two parts (a new child is created) and
759  * give a part of its callchain to the created child.
760  * Then create another child to host the given callchain of new branch
761  */
762 static int
763 split_add_child(struct callchain_node *parent,
764                 struct callchain_cursor *cursor,
765                 struct callchain_list *to_split,
766                 u64 idx_parents, u64 idx_local, u64 period)
767 {
768         struct callchain_node *new;
769         struct list_head *old_tail;
770         unsigned int idx_total = idx_parents + idx_local;
771
772         /* split */
773         new = create_child(parent, true);
774         if (new == NULL)
775                 return -1;
776
777         /* split the callchain and move a part to the new child */
778         old_tail = parent->val.prev;
779         list_del_range(&to_split->list, old_tail);
780         new->val.next = &to_split->list;
781         new->val.prev = old_tail;
782         to_split->list.prev = &new->val;
783         old_tail->next = &new->val;
784
785         /* split the hits */
786         new->hit = parent->hit;
787         new->children_hit = parent->children_hit;
788         parent->children_hit = callchain_cumul_hits(new);
789         new->val_nr = parent->val_nr - idx_local;
790         parent->val_nr = idx_local;
791         new->count = parent->count;
792         new->children_count = parent->children_count;
793         parent->children_count = callchain_cumul_counts(new);
794
795         /* create a new child for the new branch if any */
796         if (idx_total < cursor->nr) {
797                 struct callchain_node *first;
798                 struct callchain_list *cnode;
799                 struct callchain_cursor_node *node;
800                 struct rb_node *p, **pp;
801
802                 parent->hit = 0;
803                 parent->children_hit += period;
804                 parent->count = 0;
805                 parent->children_count += 1;
806
807                 node = callchain_cursor_current(cursor);
808                 new = add_child(parent, cursor, period);
809                 if (new == NULL)
810                         return -1;
811
812                 /*
813                  * This is second child since we moved parent's children
814                  * to new (first) child above.
815                  */
816                 p = parent->rb_root_in.rb_node;
817                 first = rb_entry(p, struct callchain_node, rb_node_in);
818                 cnode = list_first_entry(&first->val, struct callchain_list,
819                                          list);
820
821                 if (match_chain(node, cnode) == MATCH_LT)
822                         pp = &p->rb_left;
823                 else
824                         pp = &p->rb_right;
825
826                 rb_link_node(&new->rb_node_in, p, pp);
827                 rb_insert_color(&new->rb_node_in, &parent->rb_root_in);
828         } else {
829                 parent->hit = period;
830                 parent->count = 1;
831         }
832         return 0;
833 }
834
835 static enum match_result
836 append_chain(struct callchain_node *root,
837              struct callchain_cursor *cursor,
838              u64 period);
839
840 static int
841 append_chain_children(struct callchain_node *root,
842                       struct callchain_cursor *cursor,
843                       u64 period)
844 {
845         struct callchain_node *rnode;
846         struct callchain_cursor_node *node;
847         struct rb_node **p = &root->rb_root_in.rb_node;
848         struct rb_node *parent = NULL;
849
850         node = callchain_cursor_current(cursor);
851         if (!node)
852                 return -1;
853
854         /* lookup in childrens */
855         while (*p) {
856                 enum match_result ret;
857
858                 parent = *p;
859                 rnode = rb_entry(parent, struct callchain_node, rb_node_in);
860
861                 /* If at least first entry matches, rely to children */
862                 ret = append_chain(rnode, cursor, period);
863                 if (ret == MATCH_EQ)
864                         goto inc_children_hit;
865                 if (ret == MATCH_ERROR)
866                         return -1;
867
868                 if (ret == MATCH_LT)
869                         p = &parent->rb_left;
870                 else
871                         p = &parent->rb_right;
872         }
873         /* nothing in children, add to the current node */
874         rnode = add_child(root, cursor, period);
875         if (rnode == NULL)
876                 return -1;
877
878         rb_link_node(&rnode->rb_node_in, parent, p);
879         rb_insert_color(&rnode->rb_node_in, &root->rb_root_in);
880
881 inc_children_hit:
882         root->children_hit += period;
883         root->children_count++;
884         return 0;
885 }
886
887 static enum match_result
888 append_chain(struct callchain_node *root,
889              struct callchain_cursor *cursor,
890              u64 period)
891 {
892         struct callchain_list *cnode;
893         u64 start = cursor->pos;
894         bool found = false;
895         u64 matches;
896         enum match_result cmp = MATCH_ERROR;
897
898         /*
899          * Lookup in the current node
900          * If we have a symbol, then compare the start to match
901          * anywhere inside a function, unless function
902          * mode is disabled.
903          */
904         list_for_each_entry(cnode, &root->val, list) {
905                 struct callchain_cursor_node *node;
906
907                 node = callchain_cursor_current(cursor);
908                 if (!node)
909                         break;
910
911                 cmp = match_chain(node, cnode);
912                 if (cmp != MATCH_EQ)
913                         break;
914
915                 found = true;
916
917                 callchain_cursor_advance(cursor);
918         }
919
920         /* matches not, relay no the parent */
921         if (!found) {
922                 WARN_ONCE(cmp == MATCH_ERROR, "Chain comparison error\n");
923                 return cmp;
924         }
925
926         matches = cursor->pos - start;
927
928         /* we match only a part of the node. Split it and add the new chain */
929         if (matches < root->val_nr) {
930                 if (split_add_child(root, cursor, cnode, start, matches,
931                                     period) < 0)
932                         return MATCH_ERROR;
933
934                 return MATCH_EQ;
935         }
936
937         /* we match 100% of the path, increment the hit */
938         if (matches == root->val_nr && cursor->pos == cursor->nr) {
939                 root->hit += period;
940                 root->count++;
941                 return MATCH_EQ;
942         }
943
944         /* We match the node and still have a part remaining */
945         if (append_chain_children(root, cursor, period) < 0)
946                 return MATCH_ERROR;
947
948         return MATCH_EQ;
949 }
950
951 int callchain_append(struct callchain_root *root,
952                      struct callchain_cursor *cursor,
953                      u64 period)
954 {
955         if (!cursor->nr)
956                 return 0;
957
958         callchain_cursor_commit(cursor);
959
960         if (append_chain_children(&root->node, cursor, period) < 0)
961                 return -1;
962
963         if (cursor->nr > root->max_depth)
964                 root->max_depth = cursor->nr;
965
966         return 0;
967 }
968
969 static int
970 merge_chain_branch(struct callchain_cursor *cursor,
971                    struct callchain_node *dst, struct callchain_node *src)
972 {
973         struct callchain_cursor_node **old_last = cursor->last;
974         struct callchain_node *child;
975         struct callchain_list *list, *next_list;
976         struct rb_node *n;
977         int old_pos = cursor->nr;
978         int err = 0;
979
980         list_for_each_entry_safe(list, next_list, &src->val, list) {
981                 callchain_cursor_append(cursor, list->ip,
982                                         list->ms.map, list->ms.sym,
983                                         false, NULL, 0, 0, 0);
984                 list_del(&list->list);
985                 map__zput(list->ms.map);
986                 free(list);
987         }
988
989         if (src->hit) {
990                 callchain_cursor_commit(cursor);
991                 if (append_chain_children(dst, cursor, src->hit) < 0)
992                         return -1;
993         }
994
995         n = rb_first(&src->rb_root_in);
996         while (n) {
997                 child = container_of(n, struct callchain_node, rb_node_in);
998                 n = rb_next(n);
999                 rb_erase(&child->rb_node_in, &src->rb_root_in);
1000
1001                 err = merge_chain_branch(cursor, dst, child);
1002                 if (err)
1003                         break;
1004
1005                 free(child);
1006         }
1007
1008         cursor->nr = old_pos;
1009         cursor->last = old_last;
1010
1011         return err;
1012 }
1013
1014 int callchain_merge(struct callchain_cursor *cursor,
1015                     struct callchain_root *dst, struct callchain_root *src)
1016 {
1017         return merge_chain_branch(cursor, &dst->node, &src->node);
1018 }
1019
1020 int callchain_cursor_append(struct callchain_cursor *cursor,
1021                             u64 ip, struct map *map, struct symbol *sym,
1022                             bool branch, struct branch_flags *flags,
1023                             int nr_loop_iter, u64 iter_cycles, u64 branch_from)
1024 {
1025         struct callchain_cursor_node *node = *cursor->last;
1026
1027         if (!node) {
1028                 node = calloc(1, sizeof(*node));
1029                 if (!node)
1030                         return -ENOMEM;
1031
1032                 *cursor->last = node;
1033         }
1034
1035         node->ip = ip;
1036         map__zput(node->map);
1037         node->map = map__get(map);
1038         node->sym = sym;
1039         node->branch = branch;
1040         node->nr_loop_iter = nr_loop_iter;
1041         node->iter_cycles = iter_cycles;
1042
1043         if (flags)
1044                 memcpy(&node->branch_flags, flags,
1045                         sizeof(struct branch_flags));
1046
1047         node->branch_from = branch_from;
1048         cursor->nr++;
1049
1050         cursor->last = &node->next;
1051
1052         return 0;
1053 }
1054
1055 int sample__resolve_callchain(struct perf_sample *sample,
1056                               struct callchain_cursor *cursor, struct symbol **parent,
1057                               struct perf_evsel *evsel, struct addr_location *al,
1058                               int max_stack)
1059 {
1060         if (sample->callchain == NULL && !symbol_conf.show_branchflag_count)
1061                 return 0;
1062
1063         if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain ||
1064             perf_hpp_list.parent || symbol_conf.show_branchflag_count) {
1065                 return thread__resolve_callchain(al->thread, cursor, evsel, sample,
1066                                                  parent, al, max_stack);
1067         }
1068         return 0;
1069 }
1070
1071 int hist_entry__append_callchain(struct hist_entry *he, struct perf_sample *sample)
1072 {
1073         if ((!symbol_conf.use_callchain || sample->callchain == NULL) &&
1074                 !symbol_conf.show_branchflag_count)
1075                 return 0;
1076         return callchain_append(he->callchain, &callchain_cursor, sample->period);
1077 }
1078
1079 int fill_callchain_info(struct addr_location *al, struct callchain_cursor_node *node,
1080                         bool hide_unresolved)
1081 {
1082         al->map = node->map;
1083         al->sym = node->sym;
1084         if (node->map)
1085                 al->addr = node->map->map_ip(node->map, node->ip);
1086         else
1087                 al->addr = node->ip;
1088
1089         if (al->sym == NULL) {
1090                 if (hide_unresolved)
1091                         return 0;
1092                 if (al->map == NULL)
1093                         goto out;
1094         }
1095
1096         if (al->map->groups == &al->machine->kmaps) {
1097                 if (machine__is_host(al->machine)) {
1098                         al->cpumode = PERF_RECORD_MISC_KERNEL;
1099                         al->level = 'k';
1100                 } else {
1101                         al->cpumode = PERF_RECORD_MISC_GUEST_KERNEL;
1102                         al->level = 'g';
1103                 }
1104         } else {
1105                 if (machine__is_host(al->machine)) {
1106                         al->cpumode = PERF_RECORD_MISC_USER;
1107                         al->level = '.';
1108                 } else if (perf_guest) {
1109                         al->cpumode = PERF_RECORD_MISC_GUEST_USER;
1110                         al->level = 'u';
1111                 } else {
1112                         al->cpumode = PERF_RECORD_MISC_HYPERVISOR;
1113                         al->level = 'H';
1114                 }
1115         }
1116
1117 out:
1118         return 1;
1119 }
1120
1121 char *callchain_list__sym_name(struct callchain_list *cl,
1122                                char *bf, size_t bfsize, bool show_dso)
1123 {
1124         bool show_addr = callchain_param.key == CCKEY_ADDRESS;
1125         bool show_srcline = show_addr || callchain_param.key == CCKEY_SRCLINE;
1126         int printed;
1127
1128         if (cl->ms.sym) {
1129                 if (show_srcline && cl->ms.map && !cl->srcline)
1130                         cl->srcline = get_srcline(cl->ms.map->dso,
1131                                                   map__rip_2objdump(cl->ms.map,
1132                                                                     cl->ip),
1133                                                   cl->ms.sym, false, show_addr);
1134                 if (cl->srcline)
1135                         printed = scnprintf(bf, bfsize, "%s %s",
1136                                         cl->ms.sym->name, cl->srcline);
1137                 else
1138                         printed = scnprintf(bf, bfsize, "%s", cl->ms.sym->name);
1139         } else
1140                 printed = scnprintf(bf, bfsize, "%#" PRIx64, cl->ip);
1141
1142         if (show_dso)
1143                 scnprintf(bf + printed, bfsize - printed, " %s",
1144                           cl->ms.map ?
1145                           cl->ms.map->dso->short_name :
1146                           "unknown");
1147
1148         return bf;
1149 }
1150
1151 char *callchain_node__scnprintf_value(struct callchain_node *node,
1152                                       char *bf, size_t bfsize, u64 total)
1153 {
1154         double percent = 0.0;
1155         u64 period = callchain_cumul_hits(node);
1156         unsigned count = callchain_cumul_counts(node);
1157
1158         if (callchain_param.mode == CHAIN_FOLDED) {
1159                 period = node->hit;
1160                 count = node->count;
1161         }
1162
1163         switch (callchain_param.value) {
1164         case CCVAL_PERIOD:
1165                 scnprintf(bf, bfsize, "%"PRIu64, period);
1166                 break;
1167         case CCVAL_COUNT:
1168                 scnprintf(bf, bfsize, "%u", count);
1169                 break;
1170         case CCVAL_PERCENT:
1171         default:
1172                 if (total)
1173                         percent = period * 100.0 / total;
1174                 scnprintf(bf, bfsize, "%.2f%%", percent);
1175                 break;
1176         }
1177         return bf;
1178 }
1179
1180 int callchain_node__fprintf_value(struct callchain_node *node,
1181                                  FILE *fp, u64 total)
1182 {
1183         double percent = 0.0;
1184         u64 period = callchain_cumul_hits(node);
1185         unsigned count = callchain_cumul_counts(node);
1186
1187         if (callchain_param.mode == CHAIN_FOLDED) {
1188                 period = node->hit;
1189                 count = node->count;
1190         }
1191
1192         switch (callchain_param.value) {
1193         case CCVAL_PERIOD:
1194                 return fprintf(fp, "%"PRIu64, period);
1195         case CCVAL_COUNT:
1196                 return fprintf(fp, "%u", count);
1197         case CCVAL_PERCENT:
1198         default:
1199                 if (total)
1200                         percent = period * 100.0 / total;
1201                 return percent_color_fprintf(fp, "%.2f%%", percent);
1202         }
1203         return 0;
1204 }
1205
1206 static void callchain_counts_value(struct callchain_node *node,
1207                                    u64 *branch_count, u64 *predicted_count,
1208                                    u64 *abort_count, u64 *cycles_count)
1209 {
1210         struct callchain_list *clist;
1211
1212         list_for_each_entry(clist, &node->val, list) {
1213                 if (branch_count)
1214                         *branch_count += clist->branch_count;
1215
1216                 if (predicted_count)
1217                         *predicted_count += clist->predicted_count;
1218
1219                 if (abort_count)
1220                         *abort_count += clist->abort_count;
1221
1222                 if (cycles_count)
1223                         *cycles_count += clist->cycles_count;
1224         }
1225 }
1226
1227 static int callchain_node_branch_counts_cumul(struct callchain_node *node,
1228                                               u64 *branch_count,
1229                                               u64 *predicted_count,
1230                                               u64 *abort_count,
1231                                               u64 *cycles_count)
1232 {
1233         struct callchain_node *child;
1234         struct rb_node *n;
1235
1236         n = rb_first(&node->rb_root_in);
1237         while (n) {
1238                 child = rb_entry(n, struct callchain_node, rb_node_in);
1239                 n = rb_next(n);
1240
1241                 callchain_node_branch_counts_cumul(child, branch_count,
1242                                                    predicted_count,
1243                                                    abort_count,
1244                                                    cycles_count);
1245
1246                 callchain_counts_value(child, branch_count,
1247                                        predicted_count, abort_count,
1248                                        cycles_count);
1249         }
1250
1251         return 0;
1252 }
1253
1254 int callchain_branch_counts(struct callchain_root *root,
1255                             u64 *branch_count, u64 *predicted_count,
1256                             u64 *abort_count, u64 *cycles_count)
1257 {
1258         if (branch_count)
1259                 *branch_count = 0;
1260
1261         if (predicted_count)
1262                 *predicted_count = 0;
1263
1264         if (abort_count)
1265                 *abort_count = 0;
1266
1267         if (cycles_count)
1268                 *cycles_count = 0;
1269
1270         return callchain_node_branch_counts_cumul(&root->node,
1271                                                   branch_count,
1272                                                   predicted_count,
1273                                                   abort_count,
1274                                                   cycles_count);
1275 }
1276
1277 static int count_pri64_printf(int idx, const char *str, u64 value, char *bf, int bfsize)
1278 {
1279         int printed;
1280
1281         printed = scnprintf(bf, bfsize, "%s%s:%" PRId64 "", (idx) ? " " : " (", str, value);
1282
1283         return printed;
1284 }
1285
1286 static int count_float_printf(int idx, const char *str, float value,
1287                               char *bf, int bfsize, float threshold)
1288 {
1289         int printed;
1290
1291         if (threshold != 0.0 && value < threshold)
1292                 return 0;
1293
1294         printed = scnprintf(bf, bfsize, "%s%s:%.1f%%", (idx) ? " " : " (", str, value);
1295
1296         return printed;
1297 }
1298
1299 static int branch_to_str(char *bf, int bfsize,
1300                          u64 branch_count, u64 predicted_count,
1301                          u64 abort_count,
1302                          struct branch_type_stat *brtype_stat)
1303 {
1304         int printed, i = 0;
1305
1306         printed = branch_type_str(brtype_stat, bf, bfsize);
1307         if (printed)
1308                 i++;
1309
1310         if (predicted_count < branch_count) {
1311                 printed += count_float_printf(i++, "predicted",
1312                                 predicted_count * 100.0 / branch_count,
1313                                 bf + printed, bfsize - printed, 0.0);
1314         }
1315
1316         if (abort_count) {
1317                 printed += count_float_printf(i++, "abort",
1318                                 abort_count * 100.0 / branch_count,
1319                                 bf + printed, bfsize - printed, 0.1);
1320         }
1321
1322         if (i)
1323                 printed += scnprintf(bf + printed, bfsize - printed, ")");
1324
1325         return printed;
1326 }
1327
1328 static int branch_from_str(char *bf, int bfsize,
1329                            u64 branch_count,
1330                            u64 cycles_count, u64 iter_count,
1331                            u64 iter_cycles)
1332 {
1333         int printed = 0, i = 0;
1334         u64 cycles;
1335
1336         cycles = cycles_count / branch_count;
1337         if (cycles) {
1338                 printed += count_pri64_printf(i++, "cycles",
1339                                 cycles,
1340                                 bf + printed, bfsize - printed);
1341         }
1342
1343         if (iter_count) {
1344                 printed += count_pri64_printf(i++, "iter",
1345                                 iter_count,
1346                                 bf + printed, bfsize - printed);
1347
1348                 printed += count_pri64_printf(i++, "avg_cycles",
1349                                 iter_cycles / iter_count,
1350                                 bf + printed, bfsize - printed);
1351         }
1352
1353         if (i)
1354                 printed += scnprintf(bf + printed, bfsize - printed, ")");
1355
1356         return printed;
1357 }
1358
1359 static int counts_str_build(char *bf, int bfsize,
1360                              u64 branch_count, u64 predicted_count,
1361                              u64 abort_count, u64 cycles_count,
1362                              u64 iter_count, u64 iter_cycles,
1363                              struct branch_type_stat *brtype_stat)
1364 {
1365         int printed;
1366
1367         if (branch_count == 0)
1368                 return scnprintf(bf, bfsize, " (calltrace)");
1369
1370         if (brtype_stat->branch_to) {
1371                 printed = branch_to_str(bf, bfsize, branch_count,
1372                                 predicted_count, abort_count, brtype_stat);
1373         } else {
1374                 printed = branch_from_str(bf, bfsize, branch_count,
1375                                 cycles_count, iter_count, iter_cycles);
1376         }
1377
1378         if (!printed)
1379                 bf[0] = 0;
1380
1381         return printed;
1382 }
1383
1384 static int callchain_counts_printf(FILE *fp, char *bf, int bfsize,
1385                                    u64 branch_count, u64 predicted_count,
1386                                    u64 abort_count, u64 cycles_count,
1387                                    u64 iter_count, u64 iter_cycles,
1388                                    struct branch_type_stat *brtype_stat)
1389 {
1390         char str[256];
1391
1392         counts_str_build(str, sizeof(str), branch_count,
1393                          predicted_count, abort_count, cycles_count,
1394                          iter_count, iter_cycles, brtype_stat);
1395
1396         if (fp)
1397                 return fprintf(fp, "%s", str);
1398
1399         return scnprintf(bf, bfsize, "%s", str);
1400 }
1401
1402 int callchain_list_counts__printf_value(struct callchain_list *clist,
1403                                         FILE *fp, char *bf, int bfsize)
1404 {
1405         u64 branch_count, predicted_count;
1406         u64 abort_count, cycles_count;
1407         u64 iter_count, iter_cycles;
1408
1409         branch_count = clist->branch_count;
1410         predicted_count = clist->predicted_count;
1411         abort_count = clist->abort_count;
1412         cycles_count = clist->cycles_count;
1413         iter_count = clist->iter_count;
1414         iter_cycles = clist->iter_cycles;
1415
1416         return callchain_counts_printf(fp, bf, bfsize, branch_count,
1417                                        predicted_count, abort_count,
1418                                        cycles_count, iter_count, iter_cycles,
1419                                        &clist->brtype_stat);
1420 }
1421
1422 static void free_callchain_node(struct callchain_node *node)
1423 {
1424         struct callchain_list *list, *tmp;
1425         struct callchain_node *child;
1426         struct rb_node *n;
1427
1428         list_for_each_entry_safe(list, tmp, &node->parent_val, list) {
1429                 list_del(&list->list);
1430                 map__zput(list->ms.map);
1431                 free(list);
1432         }
1433
1434         list_for_each_entry_safe(list, tmp, &node->val, list) {
1435                 list_del(&list->list);
1436                 map__zput(list->ms.map);
1437                 free(list);
1438         }
1439
1440         n = rb_first(&node->rb_root_in);
1441         while (n) {
1442                 child = container_of(n, struct callchain_node, rb_node_in);
1443                 n = rb_next(n);
1444                 rb_erase(&child->rb_node_in, &node->rb_root_in);
1445
1446                 free_callchain_node(child);
1447                 free(child);
1448         }
1449 }
1450
1451 void free_callchain(struct callchain_root *root)
1452 {
1453         if (!symbol_conf.use_callchain)
1454                 return;
1455
1456         free_callchain_node(&root->node);
1457 }
1458
1459 static u64 decay_callchain_node(struct callchain_node *node)
1460 {
1461         struct callchain_node *child;
1462         struct rb_node *n;
1463         u64 child_hits = 0;
1464
1465         n = rb_first(&node->rb_root_in);
1466         while (n) {
1467                 child = container_of(n, struct callchain_node, rb_node_in);
1468
1469                 child_hits += decay_callchain_node(child);
1470                 n = rb_next(n);
1471         }
1472
1473         node->hit = (node->hit * 7) / 8;
1474         node->children_hit = child_hits;
1475
1476         return node->hit;
1477 }
1478
1479 void decay_callchain(struct callchain_root *root)
1480 {
1481         if (!symbol_conf.use_callchain)
1482                 return;
1483
1484         decay_callchain_node(&root->node);
1485 }
1486
1487 int callchain_node__make_parent_list(struct callchain_node *node)
1488 {
1489         struct callchain_node *parent = node->parent;
1490         struct callchain_list *chain, *new;
1491         LIST_HEAD(head);
1492
1493         while (parent) {
1494                 list_for_each_entry_reverse(chain, &parent->val, list) {
1495                         new = malloc(sizeof(*new));
1496                         if (new == NULL)
1497                                 goto out;
1498                         *new = *chain;
1499                         new->has_children = false;
1500                         map__get(new->ms.map);
1501                         list_add_tail(&new->list, &head);
1502                 }
1503                 parent = parent->parent;
1504         }
1505
1506         list_for_each_entry_safe_reverse(chain, new, &head, list)
1507                 list_move_tail(&chain->list, &node->parent_val);
1508
1509         if (!list_empty(&node->parent_val)) {
1510                 chain = list_first_entry(&node->parent_val, struct callchain_list, list);
1511                 chain->has_children = rb_prev(&node->rb_node) || rb_next(&node->rb_node);
1512
1513                 chain = list_first_entry(&node->val, struct callchain_list, list);
1514                 chain->has_children = false;
1515         }
1516         return 0;
1517
1518 out:
1519         list_for_each_entry_safe(chain, new, &head, list) {
1520                 list_del(&chain->list);
1521                 map__zput(chain->ms.map);
1522                 free(chain);
1523         }
1524         return -ENOMEM;
1525 }
1526
1527 int callchain_cursor__copy(struct callchain_cursor *dst,
1528                            struct callchain_cursor *src)
1529 {
1530         int rc = 0;
1531
1532         callchain_cursor_reset(dst);
1533         callchain_cursor_commit(src);
1534
1535         while (true) {
1536                 struct callchain_cursor_node *node;
1537
1538                 node = callchain_cursor_current(src);
1539                 if (node == NULL)
1540                         break;
1541
1542                 rc = callchain_cursor_append(dst, node->ip, node->map, node->sym,
1543                                              node->branch, &node->branch_flags,
1544                                              node->nr_loop_iter,
1545                                              node->iter_cycles,
1546                                              node->branch_from);
1547                 if (rc)
1548                         break;
1549
1550                 callchain_cursor_advance(src);
1551         }
1552
1553         return rc;
1554 }