perf annotate: Add a comment on the symbol__parse_objdump_line()
[linux-3.10.git] / tools / perf / util / annotate.c
1 /*
2  * Copyright (C) 2011, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
3  *
4  * Parts came from builtin-annotate.c, see those files for further
5  * copyright notes.
6  *
7  * Released under the GPL v2. (and only v2, not any later version)
8  */
9
10 #include "util.h"
11 #include "build-id.h"
12 #include "color.h"
13 #include "cache.h"
14 #include "symbol.h"
15 #include "debug.h"
16 #include "annotate.h"
17 #include "evsel.h"
18 #include <pthread.h>
19 #include <linux/bitops.h>
20
21 const char      *disassembler_style;
22 const char      *objdump_path;
23
24 static struct ins *ins__find(const char *name);
25 static int disasm_line__parse(char *line, char **namep, char **rawp);
26
27 static void ins__delete(struct ins_operands *ops)
28 {
29         free(ops->source.raw);
30         free(ops->source.name);
31         free(ops->target.raw);
32         free(ops->target.name);
33 }
34
35 static int ins__raw_scnprintf(struct ins *ins, char *bf, size_t size,
36                               struct ins_operands *ops)
37 {
38         return scnprintf(bf, size, "%-6.6s %s", ins->name, ops->raw);
39 }
40
41 int ins__scnprintf(struct ins *ins, char *bf, size_t size,
42                   struct ins_operands *ops)
43 {
44         if (ins->ops->scnprintf)
45                 return ins->ops->scnprintf(ins, bf, size, ops);
46
47         return ins__raw_scnprintf(ins, bf, size, ops);
48 }
49
50 static int call__parse(struct ins_operands *ops)
51 {
52         char *endptr, *tok, *name;
53
54         ops->target.addr = strtoull(ops->raw, &endptr, 16);
55
56         name = strchr(endptr, '<');
57         if (name == NULL)
58                 goto indirect_call;
59
60         name++;
61
62         tok = strchr(name, '>');
63         if (tok == NULL)
64                 return -1;
65
66         *tok = '\0';
67         ops->target.name = strdup(name);
68         *tok = '>';
69
70         return ops->target.name == NULL ? -1 : 0;
71
72 indirect_call:
73         tok = strchr(endptr, '(');
74         if (tok != NULL) {
75                 ops->target.addr = 0;
76                 return 0;
77         }
78
79         tok = strchr(endptr, '*');
80         if (tok == NULL)
81                 return -1;
82
83         ops->target.addr = strtoull(tok + 1, NULL, 16);
84         return 0;
85 }
86
87 static int call__scnprintf(struct ins *ins, char *bf, size_t size,
88                            struct ins_operands *ops)
89 {
90         if (ops->target.name)
91                 return scnprintf(bf, size, "%-6.6s %s", ins->name, ops->target.name);
92
93         if (ops->target.addr == 0)
94                 return ins__raw_scnprintf(ins, bf, size, ops);
95
96         return scnprintf(bf, size, "%-6.6s *%" PRIx64, ins->name, ops->target.addr);
97 }
98
99 static struct ins_ops call_ops = {
100         .parse     = call__parse,
101         .scnprintf = call__scnprintf,
102 };
103
104 bool ins__is_call(const struct ins *ins)
105 {
106         return ins->ops == &call_ops;
107 }
108
109 static int jump__parse(struct ins_operands *ops)
110 {
111         const char *s = strchr(ops->raw, '+');
112
113         ops->target.addr = strtoll(ops->raw, NULL, 16);
114
115         if (s++ != NULL)
116                 ops->target.offset = strtoll(s, NULL, 16);
117         else
118                 ops->target.offset = UINT64_MAX;
119
120         return 0;
121 }
122
123 static int jump__scnprintf(struct ins *ins, char *bf, size_t size,
124                            struct ins_operands *ops)
125 {
126         return scnprintf(bf, size, "%-6.6s %" PRIx64, ins->name, ops->target.offset);
127 }
128
129 static struct ins_ops jump_ops = {
130         .parse     = jump__parse,
131         .scnprintf = jump__scnprintf,
132 };
133
134 bool ins__is_jump(const struct ins *ins)
135 {
136         return ins->ops == &jump_ops;
137 }
138
139 static int comment__symbol(char *raw, char *comment, u64 *addrp, char **namep)
140 {
141         char *endptr, *name, *t;
142
143         if (strstr(raw, "(%rip)") == NULL)
144                 return 0;
145
146         *addrp = strtoull(comment, &endptr, 16);
147         name = strchr(endptr, '<');
148         if (name == NULL)
149                 return -1;
150
151         name++;
152
153         t = strchr(name, '>');
154         if (t == NULL)
155                 return 0;
156
157         *t = '\0';
158         *namep = strdup(name);
159         *t = '>';
160
161         return 0;
162 }
163
164 static int lock__parse(struct ins_operands *ops)
165 {
166         char *name;
167
168         ops->locked.ops = zalloc(sizeof(*ops->locked.ops));
169         if (ops->locked.ops == NULL)
170                 return 0;
171
172         if (disasm_line__parse(ops->raw, &name, &ops->locked.ops->raw) < 0)
173                 goto out_free_ops;
174
175         ops->locked.ins = ins__find(name);
176         if (ops->locked.ins == NULL)
177                 goto out_free_ops;
178
179         if (!ops->locked.ins->ops)
180                 return 0;
181
182         if (ops->locked.ins->ops->parse)
183                 ops->locked.ins->ops->parse(ops->locked.ops);
184
185         return 0;
186
187 out_free_ops:
188         free(ops->locked.ops);
189         ops->locked.ops = NULL;
190         return 0;
191 }
192
193 static int lock__scnprintf(struct ins *ins, char *bf, size_t size,
194                            struct ins_operands *ops)
195 {
196         int printed;
197
198         if (ops->locked.ins == NULL)
199                 return ins__raw_scnprintf(ins, bf, size, ops);
200
201         printed = scnprintf(bf, size, "%-6.6s ", ins->name);
202         return printed + ins__scnprintf(ops->locked.ins, bf + printed,
203                                         size - printed, ops->locked.ops);
204 }
205
206 static void lock__delete(struct ins_operands *ops)
207 {
208         free(ops->locked.ops);
209         free(ops->target.raw);
210         free(ops->target.name);
211 }
212
213 static struct ins_ops lock_ops = {
214         .free      = lock__delete,
215         .parse     = lock__parse,
216         .scnprintf = lock__scnprintf,
217 };
218
219 static int mov__parse(struct ins_operands *ops)
220 {
221         char *s = strchr(ops->raw, ','), *target, *comment, prev;
222
223         if (s == NULL)
224                 return -1;
225
226         *s = '\0';
227         ops->source.raw = strdup(ops->raw);
228         *s = ',';
229         
230         if (ops->source.raw == NULL)
231                 return -1;
232
233         target = ++s;
234
235         while (s[0] != '\0' && !isspace(s[0]))
236                 ++s;
237         prev = *s;
238         *s = '\0';
239
240         ops->target.raw = strdup(target);
241         *s = prev;
242
243         if (ops->target.raw == NULL)
244                 goto out_free_source;
245
246         comment = strchr(s, '#');
247         if (comment == NULL)
248                 return 0;
249
250         while (comment[0] != '\0' && isspace(comment[0]))
251                 ++comment;
252
253         comment__symbol(ops->source.raw, comment, &ops->source.addr, &ops->source.name);
254         comment__symbol(ops->target.raw, comment, &ops->target.addr, &ops->target.name);
255
256         return 0;
257
258 out_free_source:
259         free(ops->source.raw);
260         ops->source.raw = NULL;
261         return -1;
262 }
263
264 static int mov__scnprintf(struct ins *ins, char *bf, size_t size,
265                            struct ins_operands *ops)
266 {
267         return scnprintf(bf, size, "%-6.6s %s,%s", ins->name,
268                          ops->source.name ?: ops->source.raw,
269                          ops->target.name ?: ops->target.raw);
270 }
271
272 static struct ins_ops mov_ops = {
273         .parse     = mov__parse,
274         .scnprintf = mov__scnprintf,
275 };
276
277 static int dec__parse(struct ins_operands *ops)
278 {
279         char *target, *comment, *s, prev;
280
281         target = s = ops->raw;
282
283         while (s[0] != '\0' && !isspace(s[0]))
284                 ++s;
285         prev = *s;
286         *s = '\0';
287
288         ops->target.raw = strdup(target);
289         *s = prev;
290
291         if (ops->target.raw == NULL)
292                 return -1;
293
294         comment = strchr(s, '#');
295         if (comment == NULL)
296                 return 0;
297
298         while (comment[0] != '\0' && isspace(comment[0]))
299                 ++comment;
300
301         comment__symbol(ops->target.raw, comment, &ops->target.addr, &ops->target.name);
302
303         return 0;
304 }
305
306 static int dec__scnprintf(struct ins *ins, char *bf, size_t size,
307                            struct ins_operands *ops)
308 {
309         return scnprintf(bf, size, "%-6.6s %s", ins->name,
310                          ops->target.name ?: ops->target.raw);
311 }
312
313 static struct ins_ops dec_ops = {
314         .parse     = dec__parse,
315         .scnprintf = dec__scnprintf,
316 };
317
318 static int nop__scnprintf(struct ins *ins __maybe_unused, char *bf, size_t size,
319                           struct ins_operands *ops __maybe_unused)
320 {
321         return scnprintf(bf, size, "%-6.6s", "nop");
322 }
323
324 static struct ins_ops nop_ops = {
325         .scnprintf = nop__scnprintf,
326 };
327
328 /*
329  * Must be sorted by name!
330  */
331 static struct ins instructions[] = {
332         { .name = "add",   .ops  = &mov_ops, },
333         { .name = "addl",  .ops  = &mov_ops, },
334         { .name = "addq",  .ops  = &mov_ops, },
335         { .name = "addw",  .ops  = &mov_ops, },
336         { .name = "and",   .ops  = &mov_ops, },
337         { .name = "bts",   .ops  = &mov_ops, },
338         { .name = "call",  .ops  = &call_ops, },
339         { .name = "callq", .ops  = &call_ops, },
340         { .name = "cmp",   .ops  = &mov_ops, },
341         { .name = "cmpb",  .ops  = &mov_ops, },
342         { .name = "cmpl",  .ops  = &mov_ops, },
343         { .name = "cmpq",  .ops  = &mov_ops, },
344         { .name = "cmpw",  .ops  = &mov_ops, },
345         { .name = "cmpxch", .ops  = &mov_ops, },
346         { .name = "dec",   .ops  = &dec_ops, },
347         { .name = "decl",  .ops  = &dec_ops, },
348         { .name = "imul",  .ops  = &mov_ops, },
349         { .name = "inc",   .ops  = &dec_ops, },
350         { .name = "incl",  .ops  = &dec_ops, },
351         { .name = "ja",    .ops  = &jump_ops, },
352         { .name = "jae",   .ops  = &jump_ops, },
353         { .name = "jb",    .ops  = &jump_ops, },
354         { .name = "jbe",   .ops  = &jump_ops, },
355         { .name = "jc",    .ops  = &jump_ops, },
356         { .name = "jcxz",  .ops  = &jump_ops, },
357         { .name = "je",    .ops  = &jump_ops, },
358         { .name = "jecxz", .ops  = &jump_ops, },
359         { .name = "jg",    .ops  = &jump_ops, },
360         { .name = "jge",   .ops  = &jump_ops, },
361         { .name = "jl",    .ops  = &jump_ops, },
362         { .name = "jle",   .ops  = &jump_ops, },
363         { .name = "jmp",   .ops  = &jump_ops, },
364         { .name = "jmpq",  .ops  = &jump_ops, },
365         { .name = "jna",   .ops  = &jump_ops, },
366         { .name = "jnae",  .ops  = &jump_ops, },
367         { .name = "jnb",   .ops  = &jump_ops, },
368         { .name = "jnbe",  .ops  = &jump_ops, },
369         { .name = "jnc",   .ops  = &jump_ops, },
370         { .name = "jne",   .ops  = &jump_ops, },
371         { .name = "jng",   .ops  = &jump_ops, },
372         { .name = "jnge",  .ops  = &jump_ops, },
373         { .name = "jnl",   .ops  = &jump_ops, },
374         { .name = "jnle",  .ops  = &jump_ops, },
375         { .name = "jno",   .ops  = &jump_ops, },
376         { .name = "jnp",   .ops  = &jump_ops, },
377         { .name = "jns",   .ops  = &jump_ops, },
378         { .name = "jnz",   .ops  = &jump_ops, },
379         { .name = "jo",    .ops  = &jump_ops, },
380         { .name = "jp",    .ops  = &jump_ops, },
381         { .name = "jpe",   .ops  = &jump_ops, },
382         { .name = "jpo",   .ops  = &jump_ops, },
383         { .name = "jrcxz", .ops  = &jump_ops, },
384         { .name = "js",    .ops  = &jump_ops, },
385         { .name = "jz",    .ops  = &jump_ops, },
386         { .name = "lea",   .ops  = &mov_ops, },
387         { .name = "lock",  .ops  = &lock_ops, },
388         { .name = "mov",   .ops  = &mov_ops, },
389         { .name = "movb",  .ops  = &mov_ops, },
390         { .name = "movdqa",.ops  = &mov_ops, },
391         { .name = "movl",  .ops  = &mov_ops, },
392         { .name = "movq",  .ops  = &mov_ops, },
393         { .name = "movslq", .ops  = &mov_ops, },
394         { .name = "movzbl", .ops  = &mov_ops, },
395         { .name = "movzwl", .ops  = &mov_ops, },
396         { .name = "nop",   .ops  = &nop_ops, },
397         { .name = "nopl",  .ops  = &nop_ops, },
398         { .name = "nopw",  .ops  = &nop_ops, },
399         { .name = "or",    .ops  = &mov_ops, },
400         { .name = "orl",   .ops  = &mov_ops, },
401         { .name = "test",  .ops  = &mov_ops, },
402         { .name = "testb", .ops  = &mov_ops, },
403         { .name = "testl", .ops  = &mov_ops, },
404         { .name = "xadd",  .ops  = &mov_ops, },
405         { .name = "xbeginl", .ops  = &jump_ops, },
406         { .name = "xbeginq", .ops  = &jump_ops, },
407 };
408
409 static int ins__cmp(const void *name, const void *insp)
410 {
411         const struct ins *ins = insp;
412
413         return strcmp(name, ins->name);
414 }
415
416 static struct ins *ins__find(const char *name)
417 {
418         const int nmemb = ARRAY_SIZE(instructions);
419
420         return bsearch(name, instructions, nmemb, sizeof(struct ins), ins__cmp);
421 }
422
423 int symbol__annotate_init(struct map *map __maybe_unused, struct symbol *sym)
424 {
425         struct annotation *notes = symbol__annotation(sym);
426         pthread_mutex_init(&notes->lock, NULL);
427         return 0;
428 }
429
430 int symbol__alloc_hist(struct symbol *sym)
431 {
432         struct annotation *notes = symbol__annotation(sym);
433         const size_t size = symbol__size(sym);
434         size_t sizeof_sym_hist;
435
436         /* Check for overflow when calculating sizeof_sym_hist */
437         if (size > (SIZE_MAX - sizeof(struct sym_hist)) / sizeof(u64))
438                 return -1;
439
440         sizeof_sym_hist = (sizeof(struct sym_hist) + size * sizeof(u64));
441
442         /* Check for overflow in zalloc argument */
443         if (sizeof_sym_hist > (SIZE_MAX - sizeof(*notes->src))
444                                 / symbol_conf.nr_events)
445                 return -1;
446
447         notes->src = zalloc(sizeof(*notes->src) + symbol_conf.nr_events * sizeof_sym_hist);
448         if (notes->src == NULL)
449                 return -1;
450         notes->src->sizeof_sym_hist = sizeof_sym_hist;
451         notes->src->nr_histograms   = symbol_conf.nr_events;
452         INIT_LIST_HEAD(&notes->src->source);
453         return 0;
454 }
455
456 void symbol__annotate_zero_histograms(struct symbol *sym)
457 {
458         struct annotation *notes = symbol__annotation(sym);
459
460         pthread_mutex_lock(&notes->lock);
461         if (notes->src != NULL)
462                 memset(notes->src->histograms, 0,
463                        notes->src->nr_histograms * notes->src->sizeof_sym_hist);
464         pthread_mutex_unlock(&notes->lock);
465 }
466
467 int symbol__inc_addr_samples(struct symbol *sym, struct map *map,
468                              int evidx, u64 addr)
469 {
470         unsigned offset;
471         struct annotation *notes;
472         struct sym_hist *h;
473
474         notes = symbol__annotation(sym);
475         if (notes->src == NULL)
476                 return -ENOMEM;
477
478         pr_debug3("%s: addr=%#" PRIx64 "\n", __func__, map->unmap_ip(map, addr));
479
480         if (addr < sym->start || addr > sym->end)
481                 return -ERANGE;
482
483         offset = addr - sym->start;
484         h = annotation__histogram(notes, evidx);
485         h->sum++;
486         h->addr[offset]++;
487
488         pr_debug3("%#" PRIx64 " %s: period++ [addr: %#" PRIx64 ", %#" PRIx64
489                   ", evidx=%d] => %" PRIu64 "\n", sym->start, sym->name,
490                   addr, addr - sym->start, evidx, h->addr[offset]);
491         return 0;
492 }
493
494 static void disasm_line__init_ins(struct disasm_line *dl)
495 {
496         dl->ins = ins__find(dl->name);
497
498         if (dl->ins == NULL)
499                 return;
500
501         if (!dl->ins->ops)
502                 return;
503
504         if (dl->ins->ops->parse)
505                 dl->ins->ops->parse(&dl->ops);
506 }
507
508 static int disasm_line__parse(char *line, char **namep, char **rawp)
509 {
510         char *name = line, tmp;
511
512         while (isspace(name[0]))
513                 ++name;
514
515         if (name[0] == '\0')
516                 return -1;
517
518         *rawp = name + 1;
519
520         while ((*rawp)[0] != '\0' && !isspace((*rawp)[0]))
521                 ++*rawp;
522
523         tmp = (*rawp)[0];
524         (*rawp)[0] = '\0';
525         *namep = strdup(name);
526
527         if (*namep == NULL)
528                 goto out_free_name;
529
530         (*rawp)[0] = tmp;
531
532         if ((*rawp)[0] != '\0') {
533                 (*rawp)++;
534                 while (isspace((*rawp)[0]))
535                         ++(*rawp);
536         }
537
538         return 0;
539
540 out_free_name:
541         free(*namep);
542         *namep = NULL;
543         return -1;
544 }
545
546 static struct disasm_line *disasm_line__new(s64 offset, char *line, size_t privsize)
547 {
548         struct disasm_line *dl = zalloc(sizeof(*dl) + privsize);
549
550         if (dl != NULL) {
551                 dl->offset = offset;
552                 dl->line = strdup(line);
553                 if (dl->line == NULL)
554                         goto out_delete;
555
556                 if (offset != -1) {
557                         if (disasm_line__parse(dl->line, &dl->name, &dl->ops.raw) < 0)
558                                 goto out_free_line;
559
560                         disasm_line__init_ins(dl);
561                 }
562         }
563
564         return dl;
565
566 out_free_line:
567         free(dl->line);
568 out_delete:
569         free(dl);
570         return NULL;
571 }
572
573 void disasm_line__free(struct disasm_line *dl)
574 {
575         free(dl->line);
576         free(dl->name);
577         if (dl->ins && dl->ins->ops->free)
578                 dl->ins->ops->free(&dl->ops);
579         else
580                 ins__delete(&dl->ops);
581         free(dl);
582 }
583
584 int disasm_line__scnprintf(struct disasm_line *dl, char *bf, size_t size, bool raw)
585 {
586         if (raw || !dl->ins)
587                 return scnprintf(bf, size, "%-6.6s %s", dl->name, dl->ops.raw);
588
589         return ins__scnprintf(dl->ins, bf, size, &dl->ops);
590 }
591
592 static void disasm__add(struct list_head *head, struct disasm_line *line)
593 {
594         list_add_tail(&line->node, head);
595 }
596
597 struct disasm_line *disasm__get_next_ip_line(struct list_head *head, struct disasm_line *pos)
598 {
599         list_for_each_entry_continue(pos, head, node)
600                 if (pos->offset >= 0)
601                         return pos;
602
603         return NULL;
604 }
605
606 static int disasm_line__print(struct disasm_line *dl, struct symbol *sym, u64 start,
607                       struct perf_evsel *evsel, u64 len, int min_pcnt, int printed,
608                       int max_lines, struct disasm_line *queue)
609 {
610         static const char *prev_line;
611         static const char *prev_color;
612
613         if (dl->offset != -1) {
614                 const char *path = NULL;
615                 unsigned int hits = 0;
616                 double percent = 0.0;
617                 const char *color;
618                 struct annotation *notes = symbol__annotation(sym);
619                 struct source_line *src_line = notes->src->lines;
620                 struct sym_hist *h = annotation__histogram(notes, evsel->idx);
621                 s64 offset = dl->offset;
622                 const u64 addr = start + offset;
623                 struct disasm_line *next;
624
625                 next = disasm__get_next_ip_line(&notes->src->source, dl);
626
627                 while (offset < (s64)len &&
628                        (next == NULL || offset < next->offset)) {
629                         if (src_line) {
630                                 if (path == NULL)
631                                         path = src_line[offset].path;
632                                 percent += src_line[offset].percent;
633                         } else
634                                 hits += h->addr[offset];
635
636                         ++offset;
637                 }
638
639                 if (src_line == NULL && h->sum)
640                         percent = 100.0 * hits / h->sum;
641
642                 if (percent < min_pcnt)
643                         return -1;
644
645                 if (max_lines && printed >= max_lines)
646                         return 1;
647
648                 if (queue != NULL) {
649                         list_for_each_entry_from(queue, &notes->src->source, node) {
650                                 if (queue == dl)
651                                         break;
652                                 disasm_line__print(queue, sym, start, evsel, len,
653                                                     0, 0, 1, NULL);
654                         }
655                 }
656
657                 color = get_percent_color(percent);
658
659                 /*
660                  * Also color the filename and line if needed, with
661                  * the same color than the percentage. Don't print it
662                  * twice for close colored addr with the same filename:line
663                  */
664                 if (path) {
665                         if (!prev_line || strcmp(prev_line, path)
666                                        || color != prev_color) {
667                                 color_fprintf(stdout, color, " %s", path);
668                                 prev_line = path;
669                                 prev_color = color;
670                         }
671                 }
672
673                 color_fprintf(stdout, color, " %7.2f", percent);
674                 printf(" :      ");
675                 color_fprintf(stdout, PERF_COLOR_MAGENTA, "  %" PRIx64 ":", addr);
676                 color_fprintf(stdout, PERF_COLOR_BLUE, "%s\n", dl->line);
677         } else if (max_lines && printed >= max_lines)
678                 return 1;
679         else {
680                 if (queue)
681                         return -1;
682
683                 if (!*dl->line)
684                         printf("         :\n");
685                 else
686                         printf("         :      %s\n", dl->line);
687         }
688
689         return 0;
690 }
691
692 /*
693  * symbol__parse_objdump_line() parses objdump output (with -d --no-show-raw)
694  * which looks like following
695  *
696  *  0000000000415500 <_init>:
697  *    415500:       sub    $0x8,%rsp
698  *    415504:       mov    0x2f5ad5(%rip),%rax        # 70afe0 <_DYNAMIC+0x2f8>
699  *    41550b:       test   %rax,%rax
700  *    41550e:       je     415515 <_init+0x15>
701  *    415510:       callq  416e70 <__gmon_start__@plt>
702  *    415515:       add    $0x8,%rsp
703  *    415519:       retq
704  *
705  * it will be parsed and saved into struct disasm_line as
706  *  <offset>       <name>  <ops.raw>
707  *
708  * The offset will be a relative offset from the start of the symbol and -1
709  * means that it's not a disassembly line so should be treated differently.
710  * The ops.raw part will be parsed further according to type of the instruction.
711  */
712 static int symbol__parse_objdump_line(struct symbol *sym, struct map *map,
713                                       FILE *file, size_t privsize)
714 {
715         struct annotation *notes = symbol__annotation(sym);
716         struct disasm_line *dl;
717         char *line = NULL, *parsed_line, *tmp, *tmp2, *c;
718         size_t line_len;
719         s64 line_ip, offset = -1;
720
721         if (getline(&line, &line_len, file) < 0)
722                 return -1;
723
724         if (!line)
725                 return -1;
726
727         while (line_len != 0 && isspace(line[line_len - 1]))
728                 line[--line_len] = '\0';
729
730         c = strchr(line, '\n');
731         if (c)
732                 *c = 0;
733
734         line_ip = -1;
735         parsed_line = line;
736
737         /*
738          * Strip leading spaces:
739          */
740         tmp = line;
741         while (*tmp) {
742                 if (*tmp != ' ')
743                         break;
744                 tmp++;
745         }
746
747         if (*tmp) {
748                 /*
749                  * Parse hexa addresses followed by ':'
750                  */
751                 line_ip = strtoull(tmp, &tmp2, 16);
752                 if (*tmp2 != ':' || tmp == tmp2 || tmp2[1] == '\0')
753                         line_ip = -1;
754         }
755
756         if (line_ip != -1) {
757                 u64 start = map__rip_2objdump(map, sym->start),
758                     end = map__rip_2objdump(map, sym->end);
759
760                 offset = line_ip - start;
761                 if (offset < 0 || (u64)line_ip > end)
762                         offset = -1;
763                 else
764                         parsed_line = tmp2 + 1;
765         }
766
767         dl = disasm_line__new(offset, parsed_line, privsize);
768         free(line);
769
770         if (dl == NULL)
771                 return -1;
772
773         disasm__add(&notes->src->source, dl);
774
775         return 0;
776 }
777
778 int symbol__annotate(struct symbol *sym, struct map *map, size_t privsize)
779 {
780         struct dso *dso = map->dso;
781         char *filename = dso__build_id_filename(dso, NULL, 0);
782         bool free_filename = true;
783         char command[PATH_MAX * 2];
784         FILE *file;
785         int err = 0;
786         char symfs_filename[PATH_MAX];
787
788         if (filename) {
789                 snprintf(symfs_filename, sizeof(symfs_filename), "%s%s",
790                          symbol_conf.symfs, filename);
791         }
792
793         if (filename == NULL) {
794                 if (dso->has_build_id) {
795                         pr_err("Can't annotate %s: not enough memory\n",
796                                sym->name);
797                         return -ENOMEM;
798                 }
799                 goto fallback;
800         } else if (readlink(symfs_filename, command, sizeof(command)) < 0 ||
801                    strstr(command, "[kernel.kallsyms]") ||
802                    access(symfs_filename, R_OK)) {
803                 free(filename);
804 fallback:
805                 /*
806                  * If we don't have build-ids or the build-id file isn't in the
807                  * cache, or is just a kallsyms file, well, lets hope that this
808                  * DSO is the same as when 'perf record' ran.
809                  */
810                 filename = dso->long_name;
811                 snprintf(symfs_filename, sizeof(symfs_filename), "%s%s",
812                          symbol_conf.symfs, filename);
813                 free_filename = false;
814         }
815
816         if (dso->symtab_type == DSO_BINARY_TYPE__KALLSYMS) {
817                 char bf[BUILD_ID_SIZE * 2 + 16] = " with build id ";
818                 char *build_id_msg = NULL;
819
820                 if (dso->annotate_warned)
821                         goto out_free_filename;
822
823                 if (dso->has_build_id) {
824                         build_id__sprintf(dso->build_id,
825                                           sizeof(dso->build_id), bf + 15);
826                         build_id_msg = bf;
827                 }
828                 err = -ENOENT;
829                 dso->annotate_warned = 1;
830                 pr_err("Can't annotate %s:\n\n"
831                        "No vmlinux file%s\nwas found in the path.\n\n"
832                        "Please use:\n\n"
833                        "  perf buildid-cache -vu vmlinux\n\n"
834                        "or:\n\n"
835                        "  --vmlinux vmlinux\n",
836                        sym->name, build_id_msg ?: "");
837                 goto out_free_filename;
838         }
839
840         pr_debug("%s: filename=%s, sym=%s, start=%#" PRIx64 ", end=%#" PRIx64 "\n", __func__,
841                  filename, sym->name, map->unmap_ip(map, sym->start),
842                  map->unmap_ip(map, sym->end));
843
844         pr_debug("annotating [%p] %30s : [%p] %30s\n",
845                  dso, dso->long_name, sym, sym->name);
846
847         snprintf(command, sizeof(command),
848                  "%s %s%s --start-address=0x%016" PRIx64
849                  " --stop-address=0x%016" PRIx64
850                  " -d %s %s -C %s|grep -v %s|expand",
851                  objdump_path ? objdump_path : "objdump",
852                  disassembler_style ? "-M " : "",
853                  disassembler_style ? disassembler_style : "",
854                  map__rip_2objdump(map, sym->start),
855                  map__rip_2objdump(map, sym->end+1),
856                  symbol_conf.annotate_asm_raw ? "" : "--no-show-raw",
857                  symbol_conf.annotate_src ? "-S" : "",
858                  symfs_filename, filename);
859
860         pr_debug("Executing: %s\n", command);
861
862         file = popen(command, "r");
863         if (!file)
864                 goto out_free_filename;
865
866         while (!feof(file))
867                 if (symbol__parse_objdump_line(sym, map, file, privsize) < 0)
868                         break;
869
870         pclose(file);
871 out_free_filename:
872         if (free_filename)
873                 free(filename);
874         return err;
875 }
876
877 static void insert_source_line(struct rb_root *root, struct source_line *src_line)
878 {
879         struct source_line *iter;
880         struct rb_node **p = &root->rb_node;
881         struct rb_node *parent = NULL;
882         int ret;
883
884         while (*p != NULL) {
885                 parent = *p;
886                 iter = rb_entry(parent, struct source_line, node);
887
888                 ret = strcmp(iter->path, src_line->path);
889                 if (ret == 0) {
890                         iter->percent_sum += src_line->percent;
891                         return;
892                 }
893
894                 if (ret < 0)
895                         p = &(*p)->rb_left;
896                 else
897                         p = &(*p)->rb_right;
898         }
899
900         src_line->percent_sum = src_line->percent;
901
902         rb_link_node(&src_line->node, parent, p);
903         rb_insert_color(&src_line->node, root);
904 }
905
906 static void __resort_source_line(struct rb_root *root, struct source_line *src_line)
907 {
908         struct source_line *iter;
909         struct rb_node **p = &root->rb_node;
910         struct rb_node *parent = NULL;
911
912         while (*p != NULL) {
913                 parent = *p;
914                 iter = rb_entry(parent, struct source_line, node);
915
916                 if (src_line->percent_sum > iter->percent_sum)
917                         p = &(*p)->rb_left;
918                 else
919                         p = &(*p)->rb_right;
920         }
921
922         rb_link_node(&src_line->node, parent, p);
923         rb_insert_color(&src_line->node, root);
924 }
925
926 static void resort_source_line(struct rb_root *dest_root, struct rb_root *src_root)
927 {
928         struct source_line *src_line;
929         struct rb_node *node;
930
931         node = rb_first(src_root);
932         while (node) {
933                 struct rb_node *next;
934
935                 src_line = rb_entry(node, struct source_line, node);
936                 next = rb_next(node);
937                 rb_erase(node, src_root);
938
939                 __resort_source_line(dest_root, src_line);
940                 node = next;
941         }
942 }
943
944 static void symbol__free_source_line(struct symbol *sym, int len)
945 {
946         struct annotation *notes = symbol__annotation(sym);
947         struct source_line *src_line = notes->src->lines;
948         int i;
949
950         for (i = 0; i < len; i++)
951                 free(src_line[i].path);
952
953         free(src_line);
954         notes->src->lines = NULL;
955 }
956
957 /* Get the filename:line for the colored entries */
958 static int symbol__get_source_line(struct symbol *sym, struct map *map,
959                                    struct perf_evsel *evsel,
960                                    struct rb_root *root, int len,
961                                    const char *filename)
962 {
963         u64 start;
964         int i;
965         char cmd[PATH_MAX * 2];
966         struct source_line *src_line;
967         struct annotation *notes = symbol__annotation(sym);
968         struct sym_hist *h = annotation__histogram(notes, evsel->idx);
969         struct rb_root tmp_root = RB_ROOT;
970
971         if (!h->sum)
972                 return 0;
973
974         src_line = notes->src->lines = calloc(len, sizeof(struct source_line));
975         if (!notes->src->lines)
976                 return -1;
977
978         start = map__rip_2objdump(map, sym->start);
979
980         for (i = 0; i < len; i++) {
981                 char *path = NULL;
982                 size_t line_len;
983                 u64 offset;
984                 FILE *fp;
985
986                 src_line[i].percent = 100.0 * h->addr[i] / h->sum;
987                 if (src_line[i].percent <= 0.5)
988                         continue;
989
990                 offset = start + i;
991                 sprintf(cmd, "addr2line -e %s %016" PRIx64, filename, offset);
992                 fp = popen(cmd, "r");
993                 if (!fp)
994                         continue;
995
996                 if (getline(&path, &line_len, fp) < 0 || !line_len)
997                         goto next;
998
999                 src_line[i].path = malloc(sizeof(char) * line_len + 1);
1000                 if (!src_line[i].path)
1001                         goto next;
1002
1003                 strcpy(src_line[i].path, path);
1004                 insert_source_line(&tmp_root, &src_line[i]);
1005
1006         next:
1007                 pclose(fp);
1008         }
1009
1010         resort_source_line(root, &tmp_root);
1011         return 0;
1012 }
1013
1014 static void print_summary(struct rb_root *root, const char *filename)
1015 {
1016         struct source_line *src_line;
1017         struct rb_node *node;
1018
1019         printf("\nSorted summary for file %s\n", filename);
1020         printf("----------------------------------------------\n\n");
1021
1022         if (RB_EMPTY_ROOT(root)) {
1023                 printf(" Nothing higher than %1.1f%%\n", MIN_GREEN);
1024                 return;
1025         }
1026
1027         node = rb_first(root);
1028         while (node) {
1029                 double percent;
1030                 const char *color;
1031                 char *path;
1032
1033                 src_line = rb_entry(node, struct source_line, node);
1034                 percent = src_line->percent_sum;
1035                 color = get_percent_color(percent);
1036                 path = src_line->path;
1037
1038                 color_fprintf(stdout, color, " %7.2f %s", percent, path);
1039                 node = rb_next(node);
1040         }
1041 }
1042
1043 static void symbol__annotate_hits(struct symbol *sym, struct perf_evsel *evsel)
1044 {
1045         struct annotation *notes = symbol__annotation(sym);
1046         struct sym_hist *h = annotation__histogram(notes, evsel->idx);
1047         u64 len = symbol__size(sym), offset;
1048
1049         for (offset = 0; offset < len; ++offset)
1050                 if (h->addr[offset] != 0)
1051                         printf("%*" PRIx64 ": %" PRIu64 "\n", BITS_PER_LONG / 2,
1052                                sym->start + offset, h->addr[offset]);
1053         printf("%*s: %" PRIu64 "\n", BITS_PER_LONG / 2, "h->sum", h->sum);
1054 }
1055
1056 int symbol__annotate_printf(struct symbol *sym, struct map *map,
1057                             struct perf_evsel *evsel, bool full_paths,
1058                             int min_pcnt, int max_lines, int context)
1059 {
1060         struct dso *dso = map->dso;
1061         char *filename;
1062         const char *d_filename;
1063         struct annotation *notes = symbol__annotation(sym);
1064         struct disasm_line *pos, *queue = NULL;
1065         u64 start = map__rip_2objdump(map, sym->start);
1066         int printed = 2, queue_len = 0;
1067         int more = 0;
1068         u64 len;
1069
1070         filename = strdup(dso->long_name);
1071         if (!filename)
1072                 return -ENOMEM;
1073
1074         if (full_paths)
1075                 d_filename = filename;
1076         else
1077                 d_filename = basename(filename);
1078
1079         len = symbol__size(sym);
1080
1081         printf(" Percent |      Source code & Disassembly of %s\n", d_filename);
1082         printf("------------------------------------------------\n");
1083
1084         if (verbose)
1085                 symbol__annotate_hits(sym, evsel);
1086
1087         list_for_each_entry(pos, &notes->src->source, node) {
1088                 if (context && queue == NULL) {
1089                         queue = pos;
1090                         queue_len = 0;
1091                 }
1092
1093                 switch (disasm_line__print(pos, sym, start, evsel, len,
1094                                             min_pcnt, printed, max_lines,
1095                                             queue)) {
1096                 case 0:
1097                         ++printed;
1098                         if (context) {
1099                                 printed += queue_len;
1100                                 queue = NULL;
1101                                 queue_len = 0;
1102                         }
1103                         break;
1104                 case 1:
1105                         /* filtered by max_lines */
1106                         ++more;
1107                         break;
1108                 case -1:
1109                 default:
1110                         /*
1111                          * Filtered by min_pcnt or non IP lines when
1112                          * context != 0
1113                          */
1114                         if (!context)
1115                                 break;
1116                         if (queue_len == context)
1117                                 queue = list_entry(queue->node.next, typeof(*queue), node);
1118                         else
1119                                 ++queue_len;
1120                         break;
1121                 }
1122         }
1123
1124         free(filename);
1125
1126         return more;
1127 }
1128
1129 void symbol__annotate_zero_histogram(struct symbol *sym, int evidx)
1130 {
1131         struct annotation *notes = symbol__annotation(sym);
1132         struct sym_hist *h = annotation__histogram(notes, evidx);
1133
1134         memset(h, 0, notes->src->sizeof_sym_hist);
1135 }
1136
1137 void symbol__annotate_decay_histogram(struct symbol *sym, int evidx)
1138 {
1139         struct annotation *notes = symbol__annotation(sym);
1140         struct sym_hist *h = annotation__histogram(notes, evidx);
1141         int len = symbol__size(sym), offset;
1142
1143         h->sum = 0;
1144         for (offset = 0; offset < len; ++offset) {
1145                 h->addr[offset] = h->addr[offset] * 7 / 8;
1146                 h->sum += h->addr[offset];
1147         }
1148 }
1149
1150 void disasm__purge(struct list_head *head)
1151 {
1152         struct disasm_line *pos, *n;
1153
1154         list_for_each_entry_safe(pos, n, head, node) {
1155                 list_del(&pos->node);
1156                 disasm_line__free(pos);
1157         }
1158 }
1159
1160 static size_t disasm_line__fprintf(struct disasm_line *dl, FILE *fp)
1161 {
1162         size_t printed;
1163
1164         if (dl->offset == -1)
1165                 return fprintf(fp, "%s\n", dl->line);
1166
1167         printed = fprintf(fp, "%#" PRIx64 " %s", dl->offset, dl->name);
1168
1169         if (dl->ops.raw[0] != '\0') {
1170                 printed += fprintf(fp, "%.*s %s\n", 6 - (int)printed, " ",
1171                                    dl->ops.raw);
1172         }
1173
1174         return printed + fprintf(fp, "\n");
1175 }
1176
1177 size_t disasm__fprintf(struct list_head *head, FILE *fp)
1178 {
1179         struct disasm_line *pos;
1180         size_t printed = 0;
1181
1182         list_for_each_entry(pos, head, node)
1183                 printed += disasm_line__fprintf(pos, fp);
1184
1185         return printed;
1186 }
1187
1188 int symbol__tty_annotate(struct symbol *sym, struct map *map,
1189                          struct perf_evsel *evsel, bool print_lines,
1190                          bool full_paths, int min_pcnt, int max_lines)
1191 {
1192         struct dso *dso = map->dso;
1193         const char *filename = dso->long_name;
1194         struct rb_root source_line = RB_ROOT;
1195         u64 len;
1196
1197         if (symbol__annotate(sym, map, 0) < 0)
1198                 return -1;
1199
1200         len = symbol__size(sym);
1201
1202         if (print_lines) {
1203                 symbol__get_source_line(sym, map, evsel, &source_line,
1204                                         len, filename);
1205                 print_summary(&source_line, filename);
1206         }
1207
1208         symbol__annotate_printf(sym, map, evsel, full_paths,
1209                                 min_pcnt, max_lines, 0);
1210         if (print_lines)
1211                 symbol__free_source_line(sym, len);
1212
1213         disasm__purge(&symbol__annotation(sym)->src->source);
1214
1215         return 0;
1216 }