2 * Copyright (C) 2011, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
4 * Parts came from builtin-annotate.c, see those files for further
7 * Released under the GPL v2. (and only v2, not any later version)
20 #include "block-range.h"
21 #include "arch/common.h"
24 #include <linux/bitops.h>
25 #include <linux/kernel.h>
26 #include <sys/utsname.h>
28 const char *disassembler_style;
29 const char *objdump_path;
30 static regex_t file_lineno;
32 static struct ins_ops *ins__find(struct arch *arch, const char *name);
33 static void ins__sort(struct arch *arch);
34 static int disasm_line__parse(char *line, const char **namep, char **rawp);
38 struct ins *instructions;
39 size_t nr_instructions;
40 size_t nr_instructions_allocated;
41 struct ins_ops *(*associate_instruction_ops)(struct arch *arch, const char *name);
42 bool sorted_instructions;
45 int (*init)(struct arch *arch);
48 char skip_functions_char;
52 static struct ins_ops call_ops;
53 static struct ins_ops dec_ops;
54 static struct ins_ops jump_ops;
55 static struct ins_ops mov_ops;
56 static struct ins_ops nop_ops;
57 static struct ins_ops lock_ops;
58 static struct ins_ops ret_ops;
60 static int arch__grow_instructions(struct arch *arch)
62 struct ins *new_instructions;
63 size_t new_nr_allocated;
65 if (arch->nr_instructions_allocated == 0 && arch->instructions)
66 goto grow_from_non_allocated_table;
68 new_nr_allocated = arch->nr_instructions_allocated + 128;
69 new_instructions = realloc(arch->instructions, new_nr_allocated * sizeof(struct ins));
70 if (new_instructions == NULL)
73 out_update_instructions:
74 arch->instructions = new_instructions;
75 arch->nr_instructions_allocated = new_nr_allocated;
78 grow_from_non_allocated_table:
79 new_nr_allocated = arch->nr_instructions + 128;
80 new_instructions = calloc(new_nr_allocated, sizeof(struct ins));
81 if (new_instructions == NULL)
84 memcpy(new_instructions, arch->instructions, arch->nr_instructions);
85 goto out_update_instructions;
88 static int arch__associate_ins_ops(struct arch* arch, const char *name, struct ins_ops *ops)
92 if (arch->nr_instructions == arch->nr_instructions_allocated &&
93 arch__grow_instructions(arch))
96 ins = &arch->instructions[arch->nr_instructions];
97 ins->name = strdup(name);
102 arch->nr_instructions++;
108 #include "arch/arm/annotate/instructions.c"
109 #include "arch/arm64/annotate/instructions.c"
110 #include "arch/x86/annotate/instructions.c"
111 #include "arch/powerpc/annotate/instructions.c"
112 #include "arch/s390/annotate/instructions.c"
114 static struct arch architectures[] = {
117 .init = arm__annotate_init,
121 .init = arm64__annotate_init,
125 .instructions = x86__instructions,
126 .nr_instructions = ARRAY_SIZE(x86__instructions),
133 .init = powerpc__annotate_init,
137 .init = s390__annotate_init,
144 static void ins__delete(struct ins_operands *ops)
148 zfree(&ops->source.raw);
149 zfree(&ops->source.name);
150 zfree(&ops->target.raw);
151 zfree(&ops->target.name);
154 static int ins__raw_scnprintf(struct ins *ins, char *bf, size_t size,
155 struct ins_operands *ops)
157 return scnprintf(bf, size, "%-6.6s %s", ins->name, ops->raw);
160 int ins__scnprintf(struct ins *ins, char *bf, size_t size,
161 struct ins_operands *ops)
163 if (ins->ops->scnprintf)
164 return ins->ops->scnprintf(ins, bf, size, ops);
166 return ins__raw_scnprintf(ins, bf, size, ops);
169 static int call__parse(struct arch *arch, struct ins_operands *ops, struct map *map)
171 char *endptr, *tok, *name;
173 ops->target.addr = strtoull(ops->raw, &endptr, 16);
175 name = strchr(endptr, '<');
181 if (arch->objdump.skip_functions_char &&
182 strchr(name, arch->objdump.skip_functions_char))
185 tok = strchr(name, '>');
190 ops->target.name = strdup(name);
193 return ops->target.name == NULL ? -1 : 0;
196 tok = strchr(endptr, '*');
198 struct symbol *sym = map__find_symbol(map, map->map_ip(map, ops->target.addr));
200 ops->target.name = strdup(sym->name);
202 ops->target.addr = 0;
206 ops->target.addr = strtoull(tok + 1, NULL, 16);
210 static int call__scnprintf(struct ins *ins, char *bf, size_t size,
211 struct ins_operands *ops)
213 if (ops->target.name)
214 return scnprintf(bf, size, "%-6.6s %s", ins->name, ops->target.name);
216 if (ops->target.addr == 0)
217 return ins__raw_scnprintf(ins, bf, size, ops);
219 return scnprintf(bf, size, "%-6.6s *%" PRIx64, ins->name, ops->target.addr);
222 static struct ins_ops call_ops = {
223 .parse = call__parse,
224 .scnprintf = call__scnprintf,
227 bool ins__is_call(const struct ins *ins)
229 return ins->ops == &call_ops;
232 static int jump__parse(struct arch *arch __maybe_unused, struct ins_operands *ops, struct map *map __maybe_unused)
234 const char *s = strchr(ops->raw, '+');
235 const char *c = strchr(ops->raw, ',');
238 ops->target.addr = strtoull(c, NULL, 16);
240 ops->target.addr = strtoull(ops->raw, NULL, 16);
243 ops->target.offset = strtoull(s, NULL, 16);
244 ops->target.offset_avail = true;
246 ops->target.offset_avail = false;
252 static int jump__scnprintf(struct ins *ins, char *bf, size_t size,
253 struct ins_operands *ops)
255 if (!ops->target.addr || ops->target.offset < 0)
256 return ins__raw_scnprintf(ins, bf, size, ops);
258 return scnprintf(bf, size, "%-6.6s %" PRIx64, ins->name, ops->target.offset);
261 static struct ins_ops jump_ops = {
262 .parse = jump__parse,
263 .scnprintf = jump__scnprintf,
266 bool ins__is_jump(const struct ins *ins)
268 return ins->ops == &jump_ops;
271 static int comment__symbol(char *raw, char *comment, u64 *addrp, char **namep)
273 char *endptr, *name, *t;
275 if (strstr(raw, "(%rip)") == NULL)
278 *addrp = strtoull(comment, &endptr, 16);
279 name = strchr(endptr, '<');
285 t = strchr(name, '>');
290 *namep = strdup(name);
296 static int lock__parse(struct arch *arch, struct ins_operands *ops, struct map *map)
298 ops->locked.ops = zalloc(sizeof(*ops->locked.ops));
299 if (ops->locked.ops == NULL)
302 if (disasm_line__parse(ops->raw, &ops->locked.ins.name, &ops->locked.ops->raw) < 0)
305 ops->locked.ins.ops = ins__find(arch, ops->locked.ins.name);
307 if (ops->locked.ins.ops == NULL)
310 if (ops->locked.ins.ops->parse &&
311 ops->locked.ins.ops->parse(arch, ops->locked.ops, map) < 0)
317 zfree(&ops->locked.ops);
321 static int lock__scnprintf(struct ins *ins, char *bf, size_t size,
322 struct ins_operands *ops)
326 if (ops->locked.ins.ops == NULL)
327 return ins__raw_scnprintf(ins, bf, size, ops);
329 printed = scnprintf(bf, size, "%-6.6s ", ins->name);
330 return printed + ins__scnprintf(&ops->locked.ins, bf + printed,
331 size - printed, ops->locked.ops);
334 static void lock__delete(struct ins_operands *ops)
336 struct ins *ins = &ops->locked.ins;
338 if (ins->ops && ins->ops->free)
339 ins->ops->free(ops->locked.ops);
341 ins__delete(ops->locked.ops);
343 zfree(&ops->locked.ops);
344 zfree(&ops->target.raw);
345 zfree(&ops->target.name);
348 static struct ins_ops lock_ops = {
349 .free = lock__delete,
350 .parse = lock__parse,
351 .scnprintf = lock__scnprintf,
354 static int mov__parse(struct arch *arch, struct ins_operands *ops, struct map *map __maybe_unused)
356 char *s = strchr(ops->raw, ','), *target, *comment, prev;
362 ops->source.raw = strdup(ops->raw);
365 if (ops->source.raw == NULL)
369 comment = strchr(s, arch->objdump.comment_char);
374 s = strchr(s, '\0') - 1;
376 while (s > target && isspace(s[0]))
382 ops->target.raw = strdup(target);
385 if (ops->target.raw == NULL)
386 goto out_free_source;
391 comment = ltrim(comment);
392 comment__symbol(ops->source.raw, comment, &ops->source.addr, &ops->source.name);
393 comment__symbol(ops->target.raw, comment, &ops->target.addr, &ops->target.name);
398 zfree(&ops->source.raw);
402 static int mov__scnprintf(struct ins *ins, char *bf, size_t size,
403 struct ins_operands *ops)
405 return scnprintf(bf, size, "%-6.6s %s,%s", ins->name,
406 ops->source.name ?: ops->source.raw,
407 ops->target.name ?: ops->target.raw);
410 static struct ins_ops mov_ops = {
412 .scnprintf = mov__scnprintf,
415 static int dec__parse(struct arch *arch __maybe_unused, struct ins_operands *ops, struct map *map __maybe_unused)
417 char *target, *comment, *s, prev;
419 target = s = ops->raw;
421 while (s[0] != '\0' && !isspace(s[0]))
426 ops->target.raw = strdup(target);
429 if (ops->target.raw == NULL)
432 comment = strchr(s, arch->objdump.comment_char);
436 comment = ltrim(comment);
437 comment__symbol(ops->target.raw, comment, &ops->target.addr, &ops->target.name);
442 static int dec__scnprintf(struct ins *ins, char *bf, size_t size,
443 struct ins_operands *ops)
445 return scnprintf(bf, size, "%-6.6s %s", ins->name,
446 ops->target.name ?: ops->target.raw);
449 static struct ins_ops dec_ops = {
451 .scnprintf = dec__scnprintf,
454 static int nop__scnprintf(struct ins *ins __maybe_unused, char *bf, size_t size,
455 struct ins_operands *ops __maybe_unused)
457 return scnprintf(bf, size, "%-6.6s", "nop");
460 static struct ins_ops nop_ops = {
461 .scnprintf = nop__scnprintf,
464 static struct ins_ops ret_ops = {
465 .scnprintf = ins__raw_scnprintf,
468 bool ins__is_ret(const struct ins *ins)
470 return ins->ops == &ret_ops;
473 static int ins__key_cmp(const void *name, const void *insp)
475 const struct ins *ins = insp;
477 return strcmp(name, ins->name);
480 static int ins__cmp(const void *a, const void *b)
482 const struct ins *ia = a;
483 const struct ins *ib = b;
485 return strcmp(ia->name, ib->name);
488 static void ins__sort(struct arch *arch)
490 const int nmemb = arch->nr_instructions;
492 qsort(arch->instructions, nmemb, sizeof(struct ins), ins__cmp);
495 static struct ins_ops *__ins__find(struct arch *arch, const char *name)
498 const int nmemb = arch->nr_instructions;
500 if (!arch->sorted_instructions) {
502 arch->sorted_instructions = true;
505 ins = bsearch(name, arch->instructions, nmemb, sizeof(struct ins), ins__key_cmp);
506 return ins ? ins->ops : NULL;
509 static struct ins_ops *ins__find(struct arch *arch, const char *name)
511 struct ins_ops *ops = __ins__find(arch, name);
513 if (!ops && arch->associate_instruction_ops)
514 ops = arch->associate_instruction_ops(arch, name);
519 static int arch__key_cmp(const void *name, const void *archp)
521 const struct arch *arch = archp;
523 return strcmp(name, arch->name);
526 static int arch__cmp(const void *a, const void *b)
528 const struct arch *aa = a;
529 const struct arch *ab = b;
531 return strcmp(aa->name, ab->name);
534 static void arch__sort(void)
536 const int nmemb = ARRAY_SIZE(architectures);
538 qsort(architectures, nmemb, sizeof(struct arch), arch__cmp);
541 static struct arch *arch__find(const char *name)
543 const int nmemb = ARRAY_SIZE(architectures);
551 return bsearch(name, architectures, nmemb, sizeof(struct arch), arch__key_cmp);
554 int symbol__alloc_hist(struct symbol *sym)
556 struct annotation *notes = symbol__annotation(sym);
557 const size_t size = symbol__size(sym);
558 size_t sizeof_sym_hist;
560 /* Check for overflow when calculating sizeof_sym_hist */
561 if (size > (SIZE_MAX - sizeof(struct sym_hist)) / sizeof(u64))
564 sizeof_sym_hist = (sizeof(struct sym_hist) + size * sizeof(u64));
566 /* Check for overflow in zalloc argument */
567 if (sizeof_sym_hist > (SIZE_MAX - sizeof(*notes->src))
568 / symbol_conf.nr_events)
571 notes->src = zalloc(sizeof(*notes->src) + symbol_conf.nr_events * sizeof_sym_hist);
572 if (notes->src == NULL)
574 notes->src->sizeof_sym_hist = sizeof_sym_hist;
575 notes->src->nr_histograms = symbol_conf.nr_events;
576 INIT_LIST_HEAD(¬es->src->source);
580 /* The cycles histogram is lazily allocated. */
581 static int symbol__alloc_hist_cycles(struct symbol *sym)
583 struct annotation *notes = symbol__annotation(sym);
584 const size_t size = symbol__size(sym);
586 notes->src->cycles_hist = calloc(size, sizeof(struct cyc_hist));
587 if (notes->src->cycles_hist == NULL)
592 void symbol__annotate_zero_histograms(struct symbol *sym)
594 struct annotation *notes = symbol__annotation(sym);
596 pthread_mutex_lock(¬es->lock);
597 if (notes->src != NULL) {
598 memset(notes->src->histograms, 0,
599 notes->src->nr_histograms * notes->src->sizeof_sym_hist);
600 if (notes->src->cycles_hist)
601 memset(notes->src->cycles_hist, 0,
602 symbol__size(sym) * sizeof(struct cyc_hist));
604 pthread_mutex_unlock(¬es->lock);
607 static int __symbol__account_cycles(struct annotation *notes,
609 unsigned offset, unsigned cycles,
614 ch = notes->src->cycles_hist;
616 * For now we can only account one basic block per
617 * final jump. But multiple could be overlapping.
618 * Always account the longest one. So when
619 * a shorter one has been already seen throw it away.
621 * We separately always account the full cycles.
623 ch[offset].num_aggr++;
624 ch[offset].cycles_aggr += cycles;
626 if (!have_start && ch[offset].have_start)
628 if (ch[offset].num) {
629 if (have_start && (!ch[offset].have_start ||
630 ch[offset].start > start)) {
631 ch[offset].have_start = 0;
632 ch[offset].cycles = 0;
634 if (ch[offset].reset < 0xffff)
636 } else if (have_start &&
637 ch[offset].start < start)
640 ch[offset].have_start = have_start;
641 ch[offset].start = start;
642 ch[offset].cycles += cycles;
647 static int __symbol__inc_addr_samples(struct symbol *sym, struct map *map,
648 struct annotation *notes, int evidx, u64 addr)
653 pr_debug3("%s: addr=%#" PRIx64 "\n", __func__, map->unmap_ip(map, addr));
655 if ((addr < sym->start || addr >= sym->end) &&
656 (addr != sym->end || sym->start != sym->end)) {
657 pr_debug("%s(%d): ERANGE! sym->name=%s, start=%#" PRIx64 ", addr=%#" PRIx64 ", end=%#" PRIx64 "\n",
658 __func__, __LINE__, sym->name, sym->start, addr, sym->end);
662 offset = addr - sym->start;
663 h = annotation__histogram(notes, evidx);
667 pr_debug3("%#" PRIx64 " %s: period++ [addr: %#" PRIx64 ", %#" PRIx64
668 ", evidx=%d] => %" PRIu64 "\n", sym->start, sym->name,
669 addr, addr - sym->start, evidx, h->addr[offset]);
673 static struct annotation *symbol__get_annotation(struct symbol *sym, bool cycles)
675 struct annotation *notes = symbol__annotation(sym);
677 if (notes->src == NULL) {
678 if (symbol__alloc_hist(sym) < 0)
681 if (!notes->src->cycles_hist && cycles) {
682 if (symbol__alloc_hist_cycles(sym) < 0)
688 static int symbol__inc_addr_samples(struct symbol *sym, struct map *map,
691 struct annotation *notes;
695 notes = symbol__get_annotation(sym, false);
698 return __symbol__inc_addr_samples(sym, map, notes, evidx, addr);
701 static int symbol__account_cycles(u64 addr, u64 start,
702 struct symbol *sym, unsigned cycles)
704 struct annotation *notes;
709 notes = symbol__get_annotation(sym, true);
712 if (addr < sym->start || addr >= sym->end)
716 if (start < sym->start || start >= sym->end)
721 offset = addr - sym->start;
722 return __symbol__account_cycles(notes,
723 start ? start - sym->start : 0,
728 int addr_map_symbol__account_cycles(struct addr_map_symbol *ams,
729 struct addr_map_symbol *start,
739 * Only set start when IPC can be computed. We can only
740 * compute it when the basic block is completely in a single
742 * Special case the case when the jump is elsewhere, but
743 * it starts on the function start.
746 (start->sym == ams->sym ||
748 start->addr == ams->sym->start + ams->map->start)))
749 saddr = start->al_addr;
751 pr_debug2("BB with bad start: addr %"PRIx64" start %"PRIx64" sym %"PRIx64" saddr %"PRIx64"\n",
753 start ? start->addr : 0,
754 ams->sym ? ams->sym->start + ams->map->start : 0,
756 err = symbol__account_cycles(ams->al_addr, saddr, ams->sym, cycles);
758 pr_debug2("account_cycles failed %d\n", err);
762 int addr_map_symbol__inc_samples(struct addr_map_symbol *ams, int evidx)
764 return symbol__inc_addr_samples(ams->sym, ams->map, evidx, ams->al_addr);
767 int hist_entry__inc_addr_samples(struct hist_entry *he, int evidx, u64 ip)
769 return symbol__inc_addr_samples(he->ms.sym, he->ms.map, evidx, ip);
772 static void disasm_line__init_ins(struct disasm_line *dl, struct arch *arch, struct map *map)
774 dl->ins.ops = ins__find(arch, dl->ins.name);
779 if (dl->ins.ops->parse && dl->ins.ops->parse(arch, &dl->ops, map) < 0)
783 static int disasm_line__parse(char *line, const char **namep, char **rawp)
785 char tmp, *name = ltrim(line);
792 while ((*rawp)[0] != '\0' && !isspace((*rawp)[0]))
797 *namep = strdup(name);
803 *rawp = ltrim(*rawp);
813 static struct disasm_line *disasm_line__new(s64 offset, char *line,
814 size_t privsize, int line_nr,
818 struct disasm_line *dl = zalloc(sizeof(*dl) + privsize);
822 dl->line = strdup(line);
823 dl->line_nr = line_nr;
824 if (dl->line == NULL)
828 if (disasm_line__parse(dl->line, &dl->ins.name, &dl->ops.raw) < 0)
831 disasm_line__init_ins(dl, arch, map);
844 void disasm_line__free(struct disasm_line *dl)
847 if (dl->ins.ops && dl->ins.ops->free)
848 dl->ins.ops->free(&dl->ops);
850 ins__delete(&dl->ops);
851 free((void *)dl->ins.name);
856 int disasm_line__scnprintf(struct disasm_line *dl, char *bf, size_t size, bool raw)
858 if (raw || !dl->ins.ops)
859 return scnprintf(bf, size, "%-6.6s %s", dl->ins.name, dl->ops.raw);
861 return ins__scnprintf(&dl->ins, bf, size, &dl->ops);
864 static void disasm__add(struct list_head *head, struct disasm_line *line)
866 list_add_tail(&line->node, head);
869 struct disasm_line *disasm__get_next_ip_line(struct list_head *head, struct disasm_line *pos)
871 list_for_each_entry_continue(pos, head, node)
872 if (pos->offset >= 0)
878 double disasm__calc_percent(struct annotation *notes, int evidx, s64 offset,
879 s64 end, const char **path, u64 *nr_samples)
881 struct source_line *src_line = notes->src->lines;
882 double percent = 0.0;
886 size_t sizeof_src_line = sizeof(*src_line) +
887 sizeof(src_line->samples) * (src_line->nr_pcnt - 1);
889 while (offset < end) {
890 src_line = (void *)notes->src->lines +
891 (sizeof_src_line * offset);
894 *path = src_line->path;
896 percent += src_line->samples[evidx].percent;
897 *nr_samples += src_line->samples[evidx].nr;
901 struct sym_hist *h = annotation__histogram(notes, evidx);
902 unsigned int hits = 0;
905 hits += h->addr[offset++];
909 percent = 100.0 * hits / h->sum;
916 static const char *annotate__address_color(struct block_range *br)
918 double cov = block_range__coverage(br);
921 /* mark red for >75% coverage */
923 return PERF_COLOR_RED;
925 /* mark dull for <1% coverage */
927 return PERF_COLOR_NORMAL;
930 return PERF_COLOR_MAGENTA;
933 static const char *annotate__asm_color(struct block_range *br)
935 double cov = block_range__coverage(br);
938 /* mark dull for <1% coverage */
940 return PERF_COLOR_NORMAL;
943 return PERF_COLOR_BLUE;
946 static void annotate__branch_printf(struct block_range *br, u64 addr)
948 bool emit_comment = true;
954 if (br->is_target && br->start == addr) {
955 struct block_range *branch = br;
959 * Find matching branch to our target.
961 while (!branch->is_branch)
962 branch = block_range__next(branch);
964 p = 100 *(double)br->entry / branch->coverage;
968 emit_comment = false;
973 * The percentage of coverage joined at this target in relation
974 * to the next branch.
976 printf(" +%.2f%%", p);
980 if (br->is_branch && br->end == addr) {
981 double p = 100*(double)br->taken / br->coverage;
985 emit_comment = false;
990 * The percentage of coverage leaving at this branch, and
991 * its prediction ratio.
993 printf(" -%.2f%% (p:%.2f%%)", p, 100*(double)br->pred / br->taken);
999 static int disasm_line__print(struct disasm_line *dl, struct symbol *sym, u64 start,
1000 struct perf_evsel *evsel, u64 len, int min_pcnt, int printed,
1001 int max_lines, struct disasm_line *queue)
1003 static const char *prev_line;
1004 static const char *prev_color;
1006 if (dl->offset != -1) {
1007 const char *path = NULL;
1009 double percent, max_percent = 0.0;
1010 double *ppercents = &percent;
1011 u64 *psamples = &nr_samples;
1012 int i, nr_percent = 1;
1014 struct annotation *notes = symbol__annotation(sym);
1015 s64 offset = dl->offset;
1016 const u64 addr = start + offset;
1017 struct disasm_line *next;
1018 struct block_range *br;
1020 next = disasm__get_next_ip_line(¬es->src->source, dl);
1022 if (perf_evsel__is_group_event(evsel)) {
1023 nr_percent = evsel->nr_members;
1024 ppercents = calloc(nr_percent, sizeof(double));
1025 psamples = calloc(nr_percent, sizeof(u64));
1026 if (ppercents == NULL || psamples == NULL) {
1031 for (i = 0; i < nr_percent; i++) {
1032 percent = disasm__calc_percent(notes,
1033 notes->src->lines ? i : evsel->idx + i,
1035 next ? next->offset : (s64) len,
1036 &path, &nr_samples);
1038 ppercents[i] = percent;
1039 psamples[i] = nr_samples;
1040 if (percent > max_percent)
1041 max_percent = percent;
1044 if (max_percent < min_pcnt)
1047 if (max_lines && printed >= max_lines)
1050 if (queue != NULL) {
1051 list_for_each_entry_from(queue, ¬es->src->source, node) {
1054 disasm_line__print(queue, sym, start, evsel, len,
1059 color = get_percent_color(max_percent);
1062 * Also color the filename and line if needed, with
1063 * the same color than the percentage. Don't print it
1064 * twice for close colored addr with the same filename:line
1067 if (!prev_line || strcmp(prev_line, path)
1068 || color != prev_color) {
1069 color_fprintf(stdout, color, " %s", path);
1075 for (i = 0; i < nr_percent; i++) {
1076 percent = ppercents[i];
1077 nr_samples = psamples[i];
1078 color = get_percent_color(percent);
1080 if (symbol_conf.show_total_period)
1081 color_fprintf(stdout, color, " %7" PRIu64,
1084 color_fprintf(stdout, color, " %7.2f", percent);
1089 br = block_range__find(addr);
1090 color_fprintf(stdout, annotate__address_color(br), " %" PRIx64 ":", addr);
1091 color_fprintf(stdout, annotate__asm_color(br), "%s", dl->line);
1092 annotate__branch_printf(br, addr);
1095 if (ppercents != &percent)
1098 if (psamples != &nr_samples)
1101 } else if (max_lines && printed >= max_lines)
1109 if (perf_evsel__is_group_event(evsel))
1110 width *= evsel->nr_members;
1113 printf(" %*s:\n", width, " ");
1115 printf(" %*s: %s\n", width, " ", dl->line);
1122 * symbol__parse_objdump_line() parses objdump output (with -d --no-show-raw)
1123 * which looks like following
1125 * 0000000000415500 <_init>:
1126 * 415500: sub $0x8,%rsp
1127 * 415504: mov 0x2f5ad5(%rip),%rax # 70afe0 <_DYNAMIC+0x2f8>
1128 * 41550b: test %rax,%rax
1129 * 41550e: je 415515 <_init+0x15>
1130 * 415510: callq 416e70 <__gmon_start__@plt>
1131 * 415515: add $0x8,%rsp
1134 * it will be parsed and saved into struct disasm_line as
1135 * <offset> <name> <ops.raw>
1137 * The offset will be a relative offset from the start of the symbol and -1
1138 * means that it's not a disassembly line so should be treated differently.
1139 * The ops.raw part will be parsed further according to type of the instruction.
1141 static int symbol__parse_objdump_line(struct symbol *sym, struct map *map,
1143 FILE *file, size_t privsize,
1146 struct annotation *notes = symbol__annotation(sym);
1147 struct disasm_line *dl;
1148 char *line = NULL, *parsed_line, *tmp, *tmp2;
1150 s64 line_ip, offset = -1;
1151 regmatch_t match[2];
1153 if (getline(&line, &line_len, file) < 0)
1160 parsed_line = rtrim(line);
1162 /* /filename:linenr ? Save line number and ignore. */
1163 if (regexec(&file_lineno, parsed_line, 2, match, 0) == 0) {
1164 *line_nr = atoi(parsed_line + match[1].rm_so);
1168 tmp = ltrim(parsed_line);
1171 * Parse hexa addresses followed by ':'
1173 line_ip = strtoull(tmp, &tmp2, 16);
1174 if (*tmp2 != ':' || tmp == tmp2 || tmp2[1] == '\0')
1178 if (line_ip != -1) {
1179 u64 start = map__rip_2objdump(map, sym->start),
1180 end = map__rip_2objdump(map, sym->end);
1182 offset = line_ip - start;
1183 if ((u64)line_ip < start || (u64)line_ip >= end)
1186 parsed_line = tmp2 + 1;
1189 dl = disasm_line__new(offset, parsed_line, privsize, *line_nr, arch, map);
1196 if (!disasm_line__has_offset(dl)) {
1197 dl->ops.target.offset = dl->ops.target.addr -
1198 map__rip_2objdump(map, sym->start);
1199 dl->ops.target.offset_avail = true;
1202 /* kcore has no symbols, so add the call target name */
1203 if (dl->ins.ops && ins__is_call(&dl->ins) && !dl->ops.target.name) {
1204 struct addr_map_symbol target = {
1206 .addr = dl->ops.target.addr,
1209 if (!map_groups__find_ams(&target) &&
1210 target.sym->start == target.al_addr)
1211 dl->ops.target.name = strdup(target.sym->name);
1214 disasm__add(¬es->src->source, dl);
1219 static __attribute__((constructor)) void symbol__init_regexpr(void)
1221 regcomp(&file_lineno, "^/[^:]+:([0-9]+)", REG_EXTENDED);
1224 static void delete_last_nop(struct symbol *sym)
1226 struct annotation *notes = symbol__annotation(sym);
1227 struct list_head *list = ¬es->src->source;
1228 struct disasm_line *dl;
1230 while (!list_empty(list)) {
1231 dl = list_entry(list->prev, struct disasm_line, node);
1234 if (dl->ins.ops != &nop_ops)
1237 if (!strstr(dl->line, " nop ") &&
1238 !strstr(dl->line, " nopl ") &&
1239 !strstr(dl->line, " nopw "))
1243 list_del(&dl->node);
1244 disasm_line__free(dl);
1248 int symbol__strerror_disassemble(struct symbol *sym __maybe_unused, struct map *map,
1249 int errnum, char *buf, size_t buflen)
1251 struct dso *dso = map->dso;
1253 BUG_ON(buflen == 0);
1256 str_error_r(errnum, buf, buflen);
1261 case SYMBOL_ANNOTATE_ERRNO__NO_VMLINUX: {
1262 char bf[SBUILD_ID_SIZE + 15] = " with build id ";
1263 char *build_id_msg = NULL;
1265 if (dso->has_build_id) {
1266 build_id__sprintf(dso->build_id,
1267 sizeof(dso->build_id), bf + 15);
1270 scnprintf(buf, buflen,
1271 "No vmlinux file%s\nwas found in the path.\n\n"
1272 "Note that annotation using /proc/kcore requires CAP_SYS_RAWIO capability.\n\n"
1274 " perf buildid-cache -vu vmlinux\n\n"
1276 " --vmlinux vmlinux\n", build_id_msg ?: "");
1280 scnprintf(buf, buflen, "Internal error: Invalid %d error code\n", errnum);
1287 static int dso__disassemble_filename(struct dso *dso, char *filename, size_t filename_size)
1289 char linkname[PATH_MAX];
1290 char *build_id_filename;
1291 char *build_id_path = NULL;
1293 if (dso->symtab_type == DSO_BINARY_TYPE__KALLSYMS &&
1294 !dso__is_kcore(dso))
1295 return SYMBOL_ANNOTATE_ERRNO__NO_VMLINUX;
1297 build_id_filename = dso__build_id_filename(dso, NULL, 0);
1298 if (build_id_filename) {
1299 __symbol__join_symfs(filename, filename_size, build_id_filename);
1300 free(build_id_filename);
1302 if (dso->has_build_id)
1307 build_id_path = strdup(filename);
1311 dirname(build_id_path);
1313 if (dso__is_kcore(dso) ||
1314 readlink(build_id_path, linkname, sizeof(linkname)) < 0 ||
1315 strstr(linkname, DSO__NAME_KALLSYMS) ||
1316 access(filename, R_OK)) {
1319 * If we don't have build-ids or the build-id file isn't in the
1320 * cache, or is just a kallsyms file, well, lets hope that this
1321 * DSO is the same as when 'perf record' ran.
1323 __symbol__join_symfs(filename, filename_size, dso->long_name);
1326 free(build_id_path);
1330 static const char *annotate__norm_arch(const char *arch_name)
1334 if (!arch_name) { /* Assume we are annotating locally. */
1335 if (uname(&uts) < 0)
1337 arch_name = uts.machine;
1339 return normalize_arch((char *)arch_name);
1342 int symbol__disassemble(struct symbol *sym, struct map *map, const char *arch_name, size_t privsize)
1344 struct dso *dso = map->dso;
1345 char command[PATH_MAX * 2];
1346 struct arch *arch = NULL;
1348 char symfs_filename[PATH_MAX];
1349 struct kcore_extract kce;
1350 bool delete_extract = false;
1355 int err = dso__disassemble_filename(dso, symfs_filename, sizeof(symfs_filename));
1360 arch_name = annotate__norm_arch(arch_name);
1364 arch = arch__find(arch_name);
1369 err = arch->init(arch);
1371 pr_err("%s: failed to initialize %s arch priv area\n", __func__, arch->name);
1376 pr_debug("%s: filename=%s, sym=%s, start=%#" PRIx64 ", end=%#" PRIx64 "\n", __func__,
1377 symfs_filename, sym->name, map->unmap_ip(map, sym->start),
1378 map->unmap_ip(map, sym->end));
1380 pr_debug("annotating [%p] %30s : [%p] %30s\n",
1381 dso, dso->long_name, sym, sym->name);
1383 if (dso__is_kcore(dso)) {
1384 kce.kcore_filename = symfs_filename;
1385 kce.addr = map__rip_2objdump(map, sym->start);
1386 kce.offs = sym->start;
1387 kce.len = sym->end - sym->start;
1388 if (!kcore_extract__create(&kce)) {
1389 delete_extract = true;
1390 strlcpy(symfs_filename, kce.extract_filename,
1391 sizeof(symfs_filename));
1393 } else if (dso__needs_decompress(dso)) {
1399 if (kmod_path__parse_ext(&m, symfs_filename))
1402 snprintf(tmp, PATH_MAX, "/tmp/perf-kmod-XXXXXX");
1410 ret = decompress_to_file(m.ext, symfs_filename, fd);
1413 pr_err("Cannot decompress %s %s\n", m.ext, symfs_filename);
1421 strcpy(symfs_filename, tmp);
1424 snprintf(command, sizeof(command),
1425 "%s %s%s --start-address=0x%016" PRIx64
1426 " --stop-address=0x%016" PRIx64
1427 " -l -d %s %s -C %s 2>/dev/null|grep -v %s:|expand",
1428 objdump_path ? objdump_path : "objdump",
1429 disassembler_style ? "-M " : "",
1430 disassembler_style ? disassembler_style : "",
1431 map__rip_2objdump(map, sym->start),
1432 map__rip_2objdump(map, sym->end),
1433 symbol_conf.annotate_asm_raw ? "" : "--no-show-raw",
1434 symbol_conf.annotate_src ? "-S" : "",
1435 symfs_filename, symfs_filename);
1437 pr_debug("Executing: %s\n", command);
1440 if (pipe(stdout_fd) < 0) {
1441 pr_err("Failure creating the pipe to run %s\n", command);
1442 goto out_remove_tmp;
1447 pr_err("Failure forking to run %s\n", command);
1448 goto out_close_stdout;
1452 close(stdout_fd[0]);
1453 dup2(stdout_fd[1], 1);
1454 close(stdout_fd[1]);
1455 execl("/bin/sh", "sh", "-c", command, NULL);
1460 close(stdout_fd[1]);
1462 file = fdopen(stdout_fd[0], "r");
1464 pr_err("Failure creating FILE stream for %s\n", command);
1466 * If we were using debug info should retry with
1469 goto out_remove_tmp;
1473 while (!feof(file)) {
1475 * The source code line number (lineno) needs to be kept in
1476 * accross calls to symbol__parse_objdump_line(), so that it
1477 * can associate it with the instructions till the next one.
1478 * See disasm_line__new() and struct disasm_line::line_nr.
1480 if (symbol__parse_objdump_line(sym, map, arch, file, privsize,
1487 pr_err("No output from %s\n", command);
1490 * kallsyms does not have symbol sizes so there may a nop at the end.
1493 if (dso__is_kcore(dso))
1494 delete_last_nop(sym);
1499 close(stdout_fd[0]);
1501 if (dso__needs_decompress(dso))
1502 unlink(symfs_filename);
1505 kcore_extract__delete(&kce);
1510 close(stdout_fd[1]);
1511 goto out_remove_tmp;
1514 static void insert_source_line(struct rb_root *root, struct source_line *src_line)
1516 struct source_line *iter;
1517 struct rb_node **p = &root->rb_node;
1518 struct rb_node *parent = NULL;
1521 while (*p != NULL) {
1523 iter = rb_entry(parent, struct source_line, node);
1525 ret = strcmp(iter->path, src_line->path);
1527 for (i = 0; i < src_line->nr_pcnt; i++)
1528 iter->samples[i].percent_sum += src_line->samples[i].percent;
1535 p = &(*p)->rb_right;
1538 for (i = 0; i < src_line->nr_pcnt; i++)
1539 src_line->samples[i].percent_sum = src_line->samples[i].percent;
1541 rb_link_node(&src_line->node, parent, p);
1542 rb_insert_color(&src_line->node, root);
1545 static int cmp_source_line(struct source_line *a, struct source_line *b)
1549 for (i = 0; i < a->nr_pcnt; i++) {
1550 if (a->samples[i].percent_sum == b->samples[i].percent_sum)
1552 return a->samples[i].percent_sum > b->samples[i].percent_sum;
1558 static void __resort_source_line(struct rb_root *root, struct source_line *src_line)
1560 struct source_line *iter;
1561 struct rb_node **p = &root->rb_node;
1562 struct rb_node *parent = NULL;
1564 while (*p != NULL) {
1566 iter = rb_entry(parent, struct source_line, node);
1568 if (cmp_source_line(src_line, iter))
1571 p = &(*p)->rb_right;
1574 rb_link_node(&src_line->node, parent, p);
1575 rb_insert_color(&src_line->node, root);
1578 static void resort_source_line(struct rb_root *dest_root, struct rb_root *src_root)
1580 struct source_line *src_line;
1581 struct rb_node *node;
1583 node = rb_first(src_root);
1585 struct rb_node *next;
1587 src_line = rb_entry(node, struct source_line, node);
1588 next = rb_next(node);
1589 rb_erase(node, src_root);
1591 __resort_source_line(dest_root, src_line);
1596 static void symbol__free_source_line(struct symbol *sym, int len)
1598 struct annotation *notes = symbol__annotation(sym);
1599 struct source_line *src_line = notes->src->lines;
1600 size_t sizeof_src_line;
1603 sizeof_src_line = sizeof(*src_line) +
1604 (sizeof(src_line->samples) * (src_line->nr_pcnt - 1));
1606 for (i = 0; i < len; i++) {
1607 free_srcline(src_line->path);
1608 src_line = (void *)src_line + sizeof_src_line;
1611 zfree(¬es->src->lines);
1614 /* Get the filename:line for the colored entries */
1615 static int symbol__get_source_line(struct symbol *sym, struct map *map,
1616 struct perf_evsel *evsel,
1617 struct rb_root *root, int len)
1621 int evidx = evsel->idx;
1622 struct source_line *src_line;
1623 struct annotation *notes = symbol__annotation(sym);
1624 struct sym_hist *h = annotation__histogram(notes, evidx);
1625 struct rb_root tmp_root = RB_ROOT;
1628 size_t sizeof_src_line = sizeof(struct source_line);
1630 if (perf_evsel__is_group_event(evsel)) {
1631 for (i = 1; i < evsel->nr_members; i++) {
1632 h = annotation__histogram(notes, evidx + i);
1635 nr_pcnt = evsel->nr_members;
1636 sizeof_src_line += (nr_pcnt - 1) * sizeof(src_line->samples);
1642 src_line = notes->src->lines = calloc(len, sizeof_src_line);
1643 if (!notes->src->lines)
1646 start = map__rip_2objdump(map, sym->start);
1648 for (i = 0; i < len; i++) {
1649 u64 offset, nr_samples;
1650 double percent_max = 0.0;
1652 src_line->nr_pcnt = nr_pcnt;
1654 for (k = 0; k < nr_pcnt; k++) {
1655 double percent = 0.0;
1657 h = annotation__histogram(notes, evidx + k);
1658 nr_samples = h->addr[i];
1660 percent = 100.0 * nr_samples / h->sum;
1662 if (percent > percent_max)
1663 percent_max = percent;
1664 src_line->samples[k].percent = percent;
1665 src_line->samples[k].nr = nr_samples;
1668 if (percent_max <= 0.5)
1672 src_line->path = get_srcline(map->dso, offset, NULL,
1674 insert_source_line(&tmp_root, src_line);
1677 src_line = (void *)src_line + sizeof_src_line;
1680 resort_source_line(root, &tmp_root);
1684 static void print_summary(struct rb_root *root, const char *filename)
1686 struct source_line *src_line;
1687 struct rb_node *node;
1689 printf("\nSorted summary for file %s\n", filename);
1690 printf("----------------------------------------------\n\n");
1692 if (RB_EMPTY_ROOT(root)) {
1693 printf(" Nothing higher than %1.1f%%\n", MIN_GREEN);
1697 node = rb_first(root);
1699 double percent, percent_max = 0.0;
1704 src_line = rb_entry(node, struct source_line, node);
1705 for (i = 0; i < src_line->nr_pcnt; i++) {
1706 percent = src_line->samples[i].percent_sum;
1707 color = get_percent_color(percent);
1708 color_fprintf(stdout, color, " %7.2f", percent);
1710 if (percent > percent_max)
1711 percent_max = percent;
1714 path = src_line->path;
1715 color = get_percent_color(percent_max);
1716 color_fprintf(stdout, color, " %s\n", path);
1718 node = rb_next(node);
1722 static void symbol__annotate_hits(struct symbol *sym, struct perf_evsel *evsel)
1724 struct annotation *notes = symbol__annotation(sym);
1725 struct sym_hist *h = annotation__histogram(notes, evsel->idx);
1726 u64 len = symbol__size(sym), offset;
1728 for (offset = 0; offset < len; ++offset)
1729 if (h->addr[offset] != 0)
1730 printf("%*" PRIx64 ": %" PRIu64 "\n", BITS_PER_LONG / 2,
1731 sym->start + offset, h->addr[offset]);
1732 printf("%*s: %" PRIu64 "\n", BITS_PER_LONG / 2, "h->sum", h->sum);
1735 int symbol__annotate_printf(struct symbol *sym, struct map *map,
1736 struct perf_evsel *evsel, bool full_paths,
1737 int min_pcnt, int max_lines, int context)
1739 struct dso *dso = map->dso;
1741 const char *d_filename;
1742 const char *evsel_name = perf_evsel__name(evsel);
1743 struct annotation *notes = symbol__annotation(sym);
1744 struct sym_hist *h = annotation__histogram(notes, evsel->idx);
1745 struct disasm_line *pos, *queue = NULL;
1746 u64 start = map__rip_2objdump(map, sym->start);
1747 int printed = 2, queue_len = 0;
1751 int graph_dotted_len;
1753 filename = strdup(dso->long_name);
1758 d_filename = filename;
1760 d_filename = basename(filename);
1762 len = symbol__size(sym);
1764 if (perf_evsel__is_group_event(evsel))
1765 width *= evsel->nr_members;
1767 graph_dotted_len = printf(" %-*.*s| Source code & Disassembly of %s for %s (%" PRIu64 " samples)\n",
1768 width, width, "Percent", d_filename, evsel_name, h->sum);
1770 printf("%-*.*s----\n",
1771 graph_dotted_len, graph_dotted_len, graph_dotted_line);
1774 symbol__annotate_hits(sym, evsel);
1776 list_for_each_entry(pos, ¬es->src->source, node) {
1777 if (context && queue == NULL) {
1782 switch (disasm_line__print(pos, sym, start, evsel, len,
1783 min_pcnt, printed, max_lines,
1788 printed += queue_len;
1794 /* filtered by max_lines */
1800 * Filtered by min_pcnt or non IP lines when
1805 if (queue_len == context)
1806 queue = list_entry(queue->node.next, typeof(*queue), node);
1818 void symbol__annotate_zero_histogram(struct symbol *sym, int evidx)
1820 struct annotation *notes = symbol__annotation(sym);
1821 struct sym_hist *h = annotation__histogram(notes, evidx);
1823 memset(h, 0, notes->src->sizeof_sym_hist);
1826 void symbol__annotate_decay_histogram(struct symbol *sym, int evidx)
1828 struct annotation *notes = symbol__annotation(sym);
1829 struct sym_hist *h = annotation__histogram(notes, evidx);
1830 int len = symbol__size(sym), offset;
1833 for (offset = 0; offset < len; ++offset) {
1834 h->addr[offset] = h->addr[offset] * 7 / 8;
1835 h->sum += h->addr[offset];
1839 void disasm__purge(struct list_head *head)
1841 struct disasm_line *pos, *n;
1843 list_for_each_entry_safe(pos, n, head, node) {
1844 list_del(&pos->node);
1845 disasm_line__free(pos);
1849 static size_t disasm_line__fprintf(struct disasm_line *dl, FILE *fp)
1853 if (dl->offset == -1)
1854 return fprintf(fp, "%s\n", dl->line);
1856 printed = fprintf(fp, "%#" PRIx64 " %s", dl->offset, dl->ins.name);
1858 if (dl->ops.raw[0] != '\0') {
1859 printed += fprintf(fp, "%.*s %s\n", 6 - (int)printed, " ",
1863 return printed + fprintf(fp, "\n");
1866 size_t disasm__fprintf(struct list_head *head, FILE *fp)
1868 struct disasm_line *pos;
1871 list_for_each_entry(pos, head, node)
1872 printed += disasm_line__fprintf(pos, fp);
1877 int symbol__tty_annotate(struct symbol *sym, struct map *map,
1878 struct perf_evsel *evsel, bool print_lines,
1879 bool full_paths, int min_pcnt, int max_lines)
1881 struct dso *dso = map->dso;
1882 struct rb_root source_line = RB_ROOT;
1885 if (symbol__disassemble(sym, map, perf_evsel__env_arch(evsel), 0) < 0)
1888 len = symbol__size(sym);
1891 srcline_full_filename = full_paths;
1892 symbol__get_source_line(sym, map, evsel, &source_line, len);
1893 print_summary(&source_line, dso->long_name);
1896 symbol__annotate_printf(sym, map, evsel, full_paths,
1897 min_pcnt, max_lines, 0);
1899 symbol__free_source_line(sym, len);
1901 disasm__purge(&symbol__annotation(sym)->src->source);
1906 bool ui__has_annotation(void)
1908 return use_browser == 1 && perf_hpp_list.sym;