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)
22 #include "block-range.h"
24 #include "arch/common.h"
27 #include <linux/bitops.h>
28 #include <linux/kernel.h>
29 #include <sys/utsname.h>
31 #include "sane_ctype.h"
33 const char *disassembler_style;
34 const char *objdump_path;
35 static regex_t file_lineno;
37 static struct ins_ops *ins__find(struct arch *arch, const char *name);
38 static void ins__sort(struct arch *arch);
39 static int disasm_line__parse(char *line, const char **namep, char **rawp);
43 struct ins *instructions;
44 size_t nr_instructions;
45 size_t nr_instructions_allocated;
46 struct ins_ops *(*associate_instruction_ops)(struct arch *arch, const char *name);
47 bool sorted_instructions;
50 int (*init)(struct arch *arch);
53 char skip_functions_char;
57 static struct ins_ops call_ops;
58 static struct ins_ops dec_ops;
59 static struct ins_ops jump_ops;
60 static struct ins_ops mov_ops;
61 static struct ins_ops nop_ops;
62 static struct ins_ops lock_ops;
63 static struct ins_ops ret_ops;
65 static int arch__grow_instructions(struct arch *arch)
67 struct ins *new_instructions;
68 size_t new_nr_allocated;
70 if (arch->nr_instructions_allocated == 0 && arch->instructions)
71 goto grow_from_non_allocated_table;
73 new_nr_allocated = arch->nr_instructions_allocated + 128;
74 new_instructions = realloc(arch->instructions, new_nr_allocated * sizeof(struct ins));
75 if (new_instructions == NULL)
78 out_update_instructions:
79 arch->instructions = new_instructions;
80 arch->nr_instructions_allocated = new_nr_allocated;
83 grow_from_non_allocated_table:
84 new_nr_allocated = arch->nr_instructions + 128;
85 new_instructions = calloc(new_nr_allocated, sizeof(struct ins));
86 if (new_instructions == NULL)
89 memcpy(new_instructions, arch->instructions, arch->nr_instructions);
90 goto out_update_instructions;
93 static int arch__associate_ins_ops(struct arch* arch, const char *name, struct ins_ops *ops)
97 if (arch->nr_instructions == arch->nr_instructions_allocated &&
98 arch__grow_instructions(arch))
101 ins = &arch->instructions[arch->nr_instructions];
102 ins->name = strdup(name);
107 arch->nr_instructions++;
113 #include "arch/arm/annotate/instructions.c"
114 #include "arch/arm64/annotate/instructions.c"
115 #include "arch/x86/annotate/instructions.c"
116 #include "arch/powerpc/annotate/instructions.c"
117 #include "arch/s390/annotate/instructions.c"
119 static struct arch architectures[] = {
122 .init = arm__annotate_init,
126 .init = arm64__annotate_init,
130 .instructions = x86__instructions,
131 .nr_instructions = ARRAY_SIZE(x86__instructions),
138 .init = powerpc__annotate_init,
142 .init = s390__annotate_init,
149 static void ins__delete(struct ins_operands *ops)
153 zfree(&ops->source.raw);
154 zfree(&ops->source.name);
155 zfree(&ops->target.raw);
156 zfree(&ops->target.name);
159 static int ins__raw_scnprintf(struct ins *ins, char *bf, size_t size,
160 struct ins_operands *ops)
162 return scnprintf(bf, size, "%-6.6s %s", ins->name, ops->raw);
165 int ins__scnprintf(struct ins *ins, char *bf, size_t size,
166 struct ins_operands *ops)
168 if (ins->ops->scnprintf)
169 return ins->ops->scnprintf(ins, bf, size, ops);
171 return ins__raw_scnprintf(ins, bf, size, ops);
174 static int call__parse(struct arch *arch, struct ins_operands *ops, struct map *map)
176 char *endptr, *tok, *name;
178 ops->target.addr = strtoull(ops->raw, &endptr, 16);
180 name = strchr(endptr, '<');
186 if (arch->objdump.skip_functions_char &&
187 strchr(name, arch->objdump.skip_functions_char))
190 tok = strchr(name, '>');
195 ops->target.name = strdup(name);
198 return ops->target.name == NULL ? -1 : 0;
201 tok = strchr(endptr, '*');
203 struct symbol *sym = map__find_symbol(map, map->map_ip(map, ops->target.addr));
205 ops->target.name = strdup(sym->name);
207 ops->target.addr = 0;
211 ops->target.addr = strtoull(tok + 1, NULL, 16);
215 static int call__scnprintf(struct ins *ins, char *bf, size_t size,
216 struct ins_operands *ops)
218 if (ops->target.name)
219 return scnprintf(bf, size, "%-6.6s %s", ins->name, ops->target.name);
221 if (ops->target.addr == 0)
222 return ins__raw_scnprintf(ins, bf, size, ops);
224 return scnprintf(bf, size, "%-6.6s *%" PRIx64, ins->name, ops->target.addr);
227 static struct ins_ops call_ops = {
228 .parse = call__parse,
229 .scnprintf = call__scnprintf,
232 bool ins__is_call(const struct ins *ins)
234 return ins->ops == &call_ops;
237 static int jump__parse(struct arch *arch __maybe_unused, struct ins_operands *ops, struct map *map __maybe_unused)
239 const char *s = strchr(ops->raw, '+');
240 const char *c = strchr(ops->raw, ',');
243 * skip over possible up to 2 operands to get to address, e.g.:
244 * tbnz w0, #26, ffff0000083cd190 <security_file_permission+0xd0>
247 ops->target.addr = strtoull(c, NULL, 16);
248 if (!ops->target.addr) {
251 ops->target.addr = strtoull(c, NULL, 16);
254 ops->target.addr = strtoull(ops->raw, NULL, 16);
258 ops->target.offset = strtoull(s, NULL, 16);
259 ops->target.offset_avail = true;
261 ops->target.offset_avail = false;
267 static int jump__scnprintf(struct ins *ins, char *bf, size_t size,
268 struct ins_operands *ops)
270 const char *c = strchr(ops->raw, ',');
272 if (!ops->target.addr || ops->target.offset < 0)
273 return ins__raw_scnprintf(ins, bf, size, ops);
276 const char *c2 = strchr(c + 1, ',');
278 /* check for 3-op insn */
283 /* mirror arch objdump's space-after-comma style */
288 return scnprintf(bf, size, "%-6.6s %.*s%" PRIx64,
289 ins->name, c ? c - ops->raw : 0, ops->raw,
293 static struct ins_ops jump_ops = {
294 .parse = jump__parse,
295 .scnprintf = jump__scnprintf,
298 bool ins__is_jump(const struct ins *ins)
300 return ins->ops == &jump_ops;
303 static int comment__symbol(char *raw, char *comment, u64 *addrp, char **namep)
305 char *endptr, *name, *t;
307 if (strstr(raw, "(%rip)") == NULL)
310 *addrp = strtoull(comment, &endptr, 16);
311 name = strchr(endptr, '<');
317 t = strchr(name, '>');
322 *namep = strdup(name);
328 static int lock__parse(struct arch *arch, struct ins_operands *ops, struct map *map)
330 ops->locked.ops = zalloc(sizeof(*ops->locked.ops));
331 if (ops->locked.ops == NULL)
334 if (disasm_line__parse(ops->raw, &ops->locked.ins.name, &ops->locked.ops->raw) < 0)
337 ops->locked.ins.ops = ins__find(arch, ops->locked.ins.name);
339 if (ops->locked.ins.ops == NULL)
342 if (ops->locked.ins.ops->parse &&
343 ops->locked.ins.ops->parse(arch, ops->locked.ops, map) < 0)
349 zfree(&ops->locked.ops);
353 static int lock__scnprintf(struct ins *ins, char *bf, size_t size,
354 struct ins_operands *ops)
358 if (ops->locked.ins.ops == NULL)
359 return ins__raw_scnprintf(ins, bf, size, ops);
361 printed = scnprintf(bf, size, "%-6.6s ", ins->name);
362 return printed + ins__scnprintf(&ops->locked.ins, bf + printed,
363 size - printed, ops->locked.ops);
366 static void lock__delete(struct ins_operands *ops)
368 struct ins *ins = &ops->locked.ins;
370 if (ins->ops && ins->ops->free)
371 ins->ops->free(ops->locked.ops);
373 ins__delete(ops->locked.ops);
375 zfree(&ops->locked.ops);
376 zfree(&ops->target.raw);
377 zfree(&ops->target.name);
380 static struct ins_ops lock_ops = {
381 .free = lock__delete,
382 .parse = lock__parse,
383 .scnprintf = lock__scnprintf,
386 static int mov__parse(struct arch *arch, struct ins_operands *ops, struct map *map __maybe_unused)
388 char *s = strchr(ops->raw, ','), *target, *comment, prev;
394 ops->source.raw = strdup(ops->raw);
397 if (ops->source.raw == NULL)
401 comment = strchr(s, arch->objdump.comment_char);
406 s = strchr(s, '\0') - 1;
408 while (s > target && isspace(s[0]))
414 ops->target.raw = strdup(target);
417 if (ops->target.raw == NULL)
418 goto out_free_source;
423 comment = ltrim(comment);
424 comment__symbol(ops->source.raw, comment, &ops->source.addr, &ops->source.name);
425 comment__symbol(ops->target.raw, comment, &ops->target.addr, &ops->target.name);
430 zfree(&ops->source.raw);
434 static int mov__scnprintf(struct ins *ins, char *bf, size_t size,
435 struct ins_operands *ops)
437 return scnprintf(bf, size, "%-6.6s %s,%s", ins->name,
438 ops->source.name ?: ops->source.raw,
439 ops->target.name ?: ops->target.raw);
442 static struct ins_ops mov_ops = {
444 .scnprintf = mov__scnprintf,
447 static int dec__parse(struct arch *arch __maybe_unused, struct ins_operands *ops, struct map *map __maybe_unused)
449 char *target, *comment, *s, prev;
451 target = s = ops->raw;
453 while (s[0] != '\0' && !isspace(s[0]))
458 ops->target.raw = strdup(target);
461 if (ops->target.raw == NULL)
464 comment = strchr(s, arch->objdump.comment_char);
468 comment = ltrim(comment);
469 comment__symbol(ops->target.raw, comment, &ops->target.addr, &ops->target.name);
474 static int dec__scnprintf(struct ins *ins, char *bf, size_t size,
475 struct ins_operands *ops)
477 return scnprintf(bf, size, "%-6.6s %s", ins->name,
478 ops->target.name ?: ops->target.raw);
481 static struct ins_ops dec_ops = {
483 .scnprintf = dec__scnprintf,
486 static int nop__scnprintf(struct ins *ins __maybe_unused, char *bf, size_t size,
487 struct ins_operands *ops __maybe_unused)
489 return scnprintf(bf, size, "%-6.6s", "nop");
492 static struct ins_ops nop_ops = {
493 .scnprintf = nop__scnprintf,
496 static struct ins_ops ret_ops = {
497 .scnprintf = ins__raw_scnprintf,
500 bool ins__is_ret(const struct ins *ins)
502 return ins->ops == &ret_ops;
505 static int ins__key_cmp(const void *name, const void *insp)
507 const struct ins *ins = insp;
509 return strcmp(name, ins->name);
512 static int ins__cmp(const void *a, const void *b)
514 const struct ins *ia = a;
515 const struct ins *ib = b;
517 return strcmp(ia->name, ib->name);
520 static void ins__sort(struct arch *arch)
522 const int nmemb = arch->nr_instructions;
524 qsort(arch->instructions, nmemb, sizeof(struct ins), ins__cmp);
527 static struct ins_ops *__ins__find(struct arch *arch, const char *name)
530 const int nmemb = arch->nr_instructions;
532 if (!arch->sorted_instructions) {
534 arch->sorted_instructions = true;
537 ins = bsearch(name, arch->instructions, nmemb, sizeof(struct ins), ins__key_cmp);
538 return ins ? ins->ops : NULL;
541 static struct ins_ops *ins__find(struct arch *arch, const char *name)
543 struct ins_ops *ops = __ins__find(arch, name);
545 if (!ops && arch->associate_instruction_ops)
546 ops = arch->associate_instruction_ops(arch, name);
551 static int arch__key_cmp(const void *name, const void *archp)
553 const struct arch *arch = archp;
555 return strcmp(name, arch->name);
558 static int arch__cmp(const void *a, const void *b)
560 const struct arch *aa = a;
561 const struct arch *ab = b;
563 return strcmp(aa->name, ab->name);
566 static void arch__sort(void)
568 const int nmemb = ARRAY_SIZE(architectures);
570 qsort(architectures, nmemb, sizeof(struct arch), arch__cmp);
573 static struct arch *arch__find(const char *name)
575 const int nmemb = ARRAY_SIZE(architectures);
583 return bsearch(name, architectures, nmemb, sizeof(struct arch), arch__key_cmp);
586 int symbol__alloc_hist(struct symbol *sym)
588 struct annotation *notes = symbol__annotation(sym);
589 const size_t size = symbol__size(sym);
590 size_t sizeof_sym_hist;
592 /* Check for overflow when calculating sizeof_sym_hist */
593 if (size > (SIZE_MAX - sizeof(struct sym_hist)) / sizeof(u64))
596 sizeof_sym_hist = (sizeof(struct sym_hist) + size * sizeof(u64));
598 /* Check for overflow in zalloc argument */
599 if (sizeof_sym_hist > (SIZE_MAX - sizeof(*notes->src))
600 / symbol_conf.nr_events)
603 notes->src = zalloc(sizeof(*notes->src) + symbol_conf.nr_events * sizeof_sym_hist);
604 if (notes->src == NULL)
606 notes->src->sizeof_sym_hist = sizeof_sym_hist;
607 notes->src->nr_histograms = symbol_conf.nr_events;
608 INIT_LIST_HEAD(¬es->src->source);
612 /* The cycles histogram is lazily allocated. */
613 static int symbol__alloc_hist_cycles(struct symbol *sym)
615 struct annotation *notes = symbol__annotation(sym);
616 const size_t size = symbol__size(sym);
618 notes->src->cycles_hist = calloc(size, sizeof(struct cyc_hist));
619 if (notes->src->cycles_hist == NULL)
624 void symbol__annotate_zero_histograms(struct symbol *sym)
626 struct annotation *notes = symbol__annotation(sym);
628 pthread_mutex_lock(¬es->lock);
629 if (notes->src != NULL) {
630 memset(notes->src->histograms, 0,
631 notes->src->nr_histograms * notes->src->sizeof_sym_hist);
632 if (notes->src->cycles_hist)
633 memset(notes->src->cycles_hist, 0,
634 symbol__size(sym) * sizeof(struct cyc_hist));
636 pthread_mutex_unlock(¬es->lock);
639 static int __symbol__account_cycles(struct annotation *notes,
641 unsigned offset, unsigned cycles,
646 ch = notes->src->cycles_hist;
648 * For now we can only account one basic block per
649 * final jump. But multiple could be overlapping.
650 * Always account the longest one. So when
651 * a shorter one has been already seen throw it away.
653 * We separately always account the full cycles.
655 ch[offset].num_aggr++;
656 ch[offset].cycles_aggr += cycles;
658 if (!have_start && ch[offset].have_start)
660 if (ch[offset].num) {
661 if (have_start && (!ch[offset].have_start ||
662 ch[offset].start > start)) {
663 ch[offset].have_start = 0;
664 ch[offset].cycles = 0;
666 if (ch[offset].reset < 0xffff)
668 } else if (have_start &&
669 ch[offset].start < start)
672 ch[offset].have_start = have_start;
673 ch[offset].start = start;
674 ch[offset].cycles += cycles;
679 static int __symbol__inc_addr_samples(struct symbol *sym, struct map *map,
680 struct annotation *notes, int evidx, u64 addr)
685 pr_debug3("%s: addr=%#" PRIx64 "\n", __func__, map->unmap_ip(map, addr));
687 if ((addr < sym->start || addr >= sym->end) &&
688 (addr != sym->end || sym->start != sym->end)) {
689 pr_debug("%s(%d): ERANGE! sym->name=%s, start=%#" PRIx64 ", addr=%#" PRIx64 ", end=%#" PRIx64 "\n",
690 __func__, __LINE__, sym->name, sym->start, addr, sym->end);
694 offset = addr - sym->start;
695 h = annotation__histogram(notes, evidx);
699 pr_debug3("%#" PRIx64 " %s: period++ [addr: %#" PRIx64 ", %#" PRIx64
700 ", evidx=%d] => %" PRIu64 "\n", sym->start, sym->name,
701 addr, addr - sym->start, evidx, h->addr[offset]);
705 static struct annotation *symbol__get_annotation(struct symbol *sym, bool cycles)
707 struct annotation *notes = symbol__annotation(sym);
709 if (notes->src == NULL) {
710 if (symbol__alloc_hist(sym) < 0)
713 if (!notes->src->cycles_hist && cycles) {
714 if (symbol__alloc_hist_cycles(sym) < 0)
720 static int symbol__inc_addr_samples(struct symbol *sym, struct map *map,
723 struct annotation *notes;
727 notes = symbol__get_annotation(sym, false);
730 return __symbol__inc_addr_samples(sym, map, notes, evidx, addr);
733 static int symbol__account_cycles(u64 addr, u64 start,
734 struct symbol *sym, unsigned cycles)
736 struct annotation *notes;
741 notes = symbol__get_annotation(sym, true);
744 if (addr < sym->start || addr >= sym->end)
748 if (start < sym->start || start >= sym->end)
753 offset = addr - sym->start;
754 return __symbol__account_cycles(notes,
755 start ? start - sym->start : 0,
760 int addr_map_symbol__account_cycles(struct addr_map_symbol *ams,
761 struct addr_map_symbol *start,
771 * Only set start when IPC can be computed. We can only
772 * compute it when the basic block is completely in a single
774 * Special case the case when the jump is elsewhere, but
775 * it starts on the function start.
778 (start->sym == ams->sym ||
780 start->addr == ams->sym->start + ams->map->start)))
781 saddr = start->al_addr;
783 pr_debug2("BB with bad start: addr %"PRIx64" start %"PRIx64" sym %"PRIx64" saddr %"PRIx64"\n",
785 start ? start->addr : 0,
786 ams->sym ? ams->sym->start + ams->map->start : 0,
788 err = symbol__account_cycles(ams->al_addr, saddr, ams->sym, cycles);
790 pr_debug2("account_cycles failed %d\n", err);
794 int addr_map_symbol__inc_samples(struct addr_map_symbol *ams, int evidx)
796 return symbol__inc_addr_samples(ams->sym, ams->map, evidx, ams->al_addr);
799 int hist_entry__inc_addr_samples(struct hist_entry *he, int evidx, u64 ip)
801 return symbol__inc_addr_samples(he->ms.sym, he->ms.map, evidx, ip);
804 static void disasm_line__init_ins(struct disasm_line *dl, struct arch *arch, struct map *map)
806 dl->ins.ops = ins__find(arch, dl->ins.name);
811 if (dl->ins.ops->parse && dl->ins.ops->parse(arch, &dl->ops, map) < 0)
815 static int disasm_line__parse(char *line, const char **namep, char **rawp)
817 char tmp, *name = ltrim(line);
824 while ((*rawp)[0] != '\0' && !isspace((*rawp)[0]))
829 *namep = strdup(name);
835 *rawp = ltrim(*rawp);
845 static struct disasm_line *disasm_line__new(s64 offset, char *line,
846 size_t privsize, int line_nr,
850 struct disasm_line *dl = zalloc(sizeof(*dl) + privsize);
854 dl->line = strdup(line);
855 dl->line_nr = line_nr;
856 if (dl->line == NULL)
860 if (disasm_line__parse(dl->line, &dl->ins.name, &dl->ops.raw) < 0)
863 disasm_line__init_ins(dl, arch, map);
876 void disasm_line__free(struct disasm_line *dl)
879 if (dl->ins.ops && dl->ins.ops->free)
880 dl->ins.ops->free(&dl->ops);
882 ins__delete(&dl->ops);
883 free((void *)dl->ins.name);
888 int disasm_line__scnprintf(struct disasm_line *dl, char *bf, size_t size, bool raw)
890 if (raw || !dl->ins.ops)
891 return scnprintf(bf, size, "%-6.6s %s", dl->ins.name, dl->ops.raw);
893 return ins__scnprintf(&dl->ins, bf, size, &dl->ops);
896 static void disasm__add(struct list_head *head, struct disasm_line *line)
898 list_add_tail(&line->node, head);
901 struct disasm_line *disasm__get_next_ip_line(struct list_head *head, struct disasm_line *pos)
903 list_for_each_entry_continue(pos, head, node)
904 if (pos->offset >= 0)
910 double disasm__calc_percent(struct annotation *notes, int evidx, s64 offset,
911 s64 end, const char **path, u64 *nr_samples)
913 struct source_line *src_line = notes->src->lines;
914 double percent = 0.0;
918 size_t sizeof_src_line = sizeof(*src_line) +
919 sizeof(src_line->samples) * (src_line->nr_pcnt - 1);
921 while (offset < end) {
922 src_line = (void *)notes->src->lines +
923 (sizeof_src_line * offset);
926 *path = src_line->path;
928 percent += src_line->samples[evidx].percent;
929 *nr_samples += src_line->samples[evidx].nr;
933 struct sym_hist *h = annotation__histogram(notes, evidx);
934 unsigned int hits = 0;
937 hits += h->addr[offset++];
941 percent = 100.0 * hits / h->sum;
948 static const char *annotate__address_color(struct block_range *br)
950 double cov = block_range__coverage(br);
953 /* mark red for >75% coverage */
955 return PERF_COLOR_RED;
957 /* mark dull for <1% coverage */
959 return PERF_COLOR_NORMAL;
962 return PERF_COLOR_MAGENTA;
965 static const char *annotate__asm_color(struct block_range *br)
967 double cov = block_range__coverage(br);
970 /* mark dull for <1% coverage */
972 return PERF_COLOR_NORMAL;
975 return PERF_COLOR_BLUE;
978 static void annotate__branch_printf(struct block_range *br, u64 addr)
980 bool emit_comment = true;
986 if (br->is_target && br->start == addr) {
987 struct block_range *branch = br;
991 * Find matching branch to our target.
993 while (!branch->is_branch)
994 branch = block_range__next(branch);
996 p = 100 *(double)br->entry / branch->coverage;
1000 emit_comment = false;
1005 * The percentage of coverage joined at this target in relation
1006 * to the next branch.
1008 printf(" +%.2f%%", p);
1012 if (br->is_branch && br->end == addr) {
1013 double p = 100*(double)br->taken / br->coverage;
1017 emit_comment = false;
1022 * The percentage of coverage leaving at this branch, and
1023 * its prediction ratio.
1025 printf(" -%.2f%% (p:%.2f%%)", p, 100*(double)br->pred / br->taken);
1031 static int disasm_line__print(struct disasm_line *dl, struct symbol *sym, u64 start,
1032 struct perf_evsel *evsel, u64 len, int min_pcnt, int printed,
1033 int max_lines, struct disasm_line *queue)
1035 static const char *prev_line;
1036 static const char *prev_color;
1038 if (dl->offset != -1) {
1039 const char *path = NULL;
1041 double percent, max_percent = 0.0;
1042 double *ppercents = &percent;
1043 u64 *psamples = &nr_samples;
1044 int i, nr_percent = 1;
1046 struct annotation *notes = symbol__annotation(sym);
1047 s64 offset = dl->offset;
1048 const u64 addr = start + offset;
1049 struct disasm_line *next;
1050 struct block_range *br;
1052 next = disasm__get_next_ip_line(¬es->src->source, dl);
1054 if (perf_evsel__is_group_event(evsel)) {
1055 nr_percent = evsel->nr_members;
1056 ppercents = calloc(nr_percent, sizeof(double));
1057 psamples = calloc(nr_percent, sizeof(u64));
1058 if (ppercents == NULL || psamples == NULL) {
1063 for (i = 0; i < nr_percent; i++) {
1064 percent = disasm__calc_percent(notes,
1065 notes->src->lines ? i : evsel->idx + i,
1067 next ? next->offset : (s64) len,
1068 &path, &nr_samples);
1070 ppercents[i] = percent;
1071 psamples[i] = nr_samples;
1072 if (percent > max_percent)
1073 max_percent = percent;
1076 if (max_percent < min_pcnt)
1079 if (max_lines && printed >= max_lines)
1082 if (queue != NULL) {
1083 list_for_each_entry_from(queue, ¬es->src->source, node) {
1086 disasm_line__print(queue, sym, start, evsel, len,
1091 color = get_percent_color(max_percent);
1094 * Also color the filename and line if needed, with
1095 * the same color than the percentage. Don't print it
1096 * twice for close colored addr with the same filename:line
1099 if (!prev_line || strcmp(prev_line, path)
1100 || color != prev_color) {
1101 color_fprintf(stdout, color, " %s", path);
1107 for (i = 0; i < nr_percent; i++) {
1108 percent = ppercents[i];
1109 nr_samples = psamples[i];
1110 color = get_percent_color(percent);
1112 if (symbol_conf.show_total_period)
1113 color_fprintf(stdout, color, " %7" PRIu64,
1116 color_fprintf(stdout, color, " %7.2f", percent);
1121 br = block_range__find(addr);
1122 color_fprintf(stdout, annotate__address_color(br), " %" PRIx64 ":", addr);
1123 color_fprintf(stdout, annotate__asm_color(br), "%s", dl->line);
1124 annotate__branch_printf(br, addr);
1127 if (ppercents != &percent)
1130 if (psamples != &nr_samples)
1133 } else if (max_lines && printed >= max_lines)
1141 if (perf_evsel__is_group_event(evsel))
1142 width *= evsel->nr_members;
1145 printf(" %*s:\n", width, " ");
1147 printf(" %*s: %s\n", width, " ", dl->line);
1154 * symbol__parse_objdump_line() parses objdump output (with -d --no-show-raw)
1155 * which looks like following
1157 * 0000000000415500 <_init>:
1158 * 415500: sub $0x8,%rsp
1159 * 415504: mov 0x2f5ad5(%rip),%rax # 70afe0 <_DYNAMIC+0x2f8>
1160 * 41550b: test %rax,%rax
1161 * 41550e: je 415515 <_init+0x15>
1162 * 415510: callq 416e70 <__gmon_start__@plt>
1163 * 415515: add $0x8,%rsp
1166 * it will be parsed and saved into struct disasm_line as
1167 * <offset> <name> <ops.raw>
1169 * The offset will be a relative offset from the start of the symbol and -1
1170 * means that it's not a disassembly line so should be treated differently.
1171 * The ops.raw part will be parsed further according to type of the instruction.
1173 static int symbol__parse_objdump_line(struct symbol *sym, struct map *map,
1175 FILE *file, size_t privsize,
1178 struct annotation *notes = symbol__annotation(sym);
1179 struct disasm_line *dl;
1180 char *line = NULL, *parsed_line, *tmp, *tmp2;
1182 s64 line_ip, offset = -1;
1183 regmatch_t match[2];
1185 if (getline(&line, &line_len, file) < 0)
1192 parsed_line = rtrim(line);
1194 /* /filename:linenr ? Save line number and ignore. */
1195 if (regexec(&file_lineno, parsed_line, 2, match, 0) == 0) {
1196 *line_nr = atoi(parsed_line + match[1].rm_so);
1200 tmp = ltrim(parsed_line);
1203 * Parse hexa addresses followed by ':'
1205 line_ip = strtoull(tmp, &tmp2, 16);
1206 if (*tmp2 != ':' || tmp == tmp2 || tmp2[1] == '\0')
1210 if (line_ip != -1) {
1211 u64 start = map__rip_2objdump(map, sym->start),
1212 end = map__rip_2objdump(map, sym->end);
1214 offset = line_ip - start;
1215 if ((u64)line_ip < start || (u64)line_ip >= end)
1218 parsed_line = tmp2 + 1;
1221 dl = disasm_line__new(offset, parsed_line, privsize, *line_nr, arch, map);
1228 if (!disasm_line__has_offset(dl)) {
1229 dl->ops.target.offset = dl->ops.target.addr -
1230 map__rip_2objdump(map, sym->start);
1231 dl->ops.target.offset_avail = true;
1234 /* kcore has no symbols, so add the call target name */
1235 if (dl->ins.ops && ins__is_call(&dl->ins) && !dl->ops.target.name) {
1236 struct addr_map_symbol target = {
1238 .addr = dl->ops.target.addr,
1241 if (!map_groups__find_ams(&target) &&
1242 target.sym->start == target.al_addr)
1243 dl->ops.target.name = strdup(target.sym->name);
1246 disasm__add(¬es->src->source, dl);
1251 static __attribute__((constructor)) void symbol__init_regexpr(void)
1253 regcomp(&file_lineno, "^/[^:]+:([0-9]+)", REG_EXTENDED);
1256 static void delete_last_nop(struct symbol *sym)
1258 struct annotation *notes = symbol__annotation(sym);
1259 struct list_head *list = ¬es->src->source;
1260 struct disasm_line *dl;
1262 while (!list_empty(list)) {
1263 dl = list_entry(list->prev, struct disasm_line, node);
1266 if (dl->ins.ops != &nop_ops)
1269 if (!strstr(dl->line, " nop ") &&
1270 !strstr(dl->line, " nopl ") &&
1271 !strstr(dl->line, " nopw "))
1275 list_del(&dl->node);
1276 disasm_line__free(dl);
1280 int symbol__strerror_disassemble(struct symbol *sym __maybe_unused, struct map *map,
1281 int errnum, char *buf, size_t buflen)
1283 struct dso *dso = map->dso;
1285 BUG_ON(buflen == 0);
1288 str_error_r(errnum, buf, buflen);
1293 case SYMBOL_ANNOTATE_ERRNO__NO_VMLINUX: {
1294 char bf[SBUILD_ID_SIZE + 15] = " with build id ";
1295 char *build_id_msg = NULL;
1297 if (dso->has_build_id) {
1298 build_id__sprintf(dso->build_id,
1299 sizeof(dso->build_id), bf + 15);
1302 scnprintf(buf, buflen,
1303 "No vmlinux file%s\nwas found in the path.\n\n"
1304 "Note that annotation using /proc/kcore requires CAP_SYS_RAWIO capability.\n\n"
1306 " perf buildid-cache -vu vmlinux\n\n"
1308 " --vmlinux vmlinux\n", build_id_msg ?: "");
1312 scnprintf(buf, buflen, "Internal error: Invalid %d error code\n", errnum);
1319 static int dso__disassemble_filename(struct dso *dso, char *filename, size_t filename_size)
1321 char linkname[PATH_MAX];
1322 char *build_id_filename;
1323 char *build_id_path = NULL;
1326 if (dso->symtab_type == DSO_BINARY_TYPE__KALLSYMS &&
1327 !dso__is_kcore(dso))
1328 return SYMBOL_ANNOTATE_ERRNO__NO_VMLINUX;
1330 build_id_filename = dso__build_id_filename(dso, NULL, 0);
1331 if (build_id_filename) {
1332 __symbol__join_symfs(filename, filename_size, build_id_filename);
1333 free(build_id_filename);
1335 if (dso->has_build_id)
1340 build_id_path = strdup(filename);
1345 * old style build-id cache has name of XX/XXXXXXX.. while
1346 * new style has XX/XXXXXXX../{elf,kallsyms,vdso}.
1347 * extract the build-id part of dirname in the new style only.
1349 pos = strrchr(build_id_path, '/');
1350 if (pos && strlen(pos) < SBUILD_ID_SIZE - 2)
1351 dirname(build_id_path);
1353 if (dso__is_kcore(dso) ||
1354 readlink(build_id_path, linkname, sizeof(linkname)) < 0 ||
1355 strstr(linkname, DSO__NAME_KALLSYMS) ||
1356 access(filename, R_OK)) {
1359 * If we don't have build-ids or the build-id file isn't in the
1360 * cache, or is just a kallsyms file, well, lets hope that this
1361 * DSO is the same as when 'perf record' ran.
1363 __symbol__join_symfs(filename, filename_size, dso->long_name);
1366 free(build_id_path);
1370 static const char *annotate__norm_arch(const char *arch_name)
1374 if (!arch_name) { /* Assume we are annotating locally. */
1375 if (uname(&uts) < 0)
1377 arch_name = uts.machine;
1379 return normalize_arch((char *)arch_name);
1382 int symbol__disassemble(struct symbol *sym, struct map *map,
1383 const char *arch_name, size_t privsize,
1384 struct arch **parch)
1386 struct dso *dso = map->dso;
1387 char command[PATH_MAX * 2];
1388 struct arch *arch = NULL;
1390 char symfs_filename[PATH_MAX];
1391 struct kcore_extract kce;
1392 bool delete_extract = false;
1397 int err = dso__disassemble_filename(dso, symfs_filename, sizeof(symfs_filename));
1402 arch_name = annotate__norm_arch(arch_name);
1406 arch = arch__find(arch_name);
1414 err = arch->init(arch);
1416 pr_err("%s: failed to initialize %s arch priv area\n", __func__, arch->name);
1421 pr_debug("%s: filename=%s, sym=%s, start=%#" PRIx64 ", end=%#" PRIx64 "\n", __func__,
1422 symfs_filename, sym->name, map->unmap_ip(map, sym->start),
1423 map->unmap_ip(map, sym->end));
1425 pr_debug("annotating [%p] %30s : [%p] %30s\n",
1426 dso, dso->long_name, sym, sym->name);
1428 if (dso__is_kcore(dso)) {
1429 kce.kcore_filename = symfs_filename;
1430 kce.addr = map__rip_2objdump(map, sym->start);
1431 kce.offs = sym->start;
1432 kce.len = sym->end - sym->start;
1433 if (!kcore_extract__create(&kce)) {
1434 delete_extract = true;
1435 strlcpy(symfs_filename, kce.extract_filename,
1436 sizeof(symfs_filename));
1438 } else if (dso__needs_decompress(dso)) {
1439 char tmp[KMOD_DECOMP_LEN];
1441 if (dso__decompress_kmodule_path(dso, symfs_filename,
1442 tmp, sizeof(tmp)) < 0)
1445 strcpy(symfs_filename, tmp);
1448 snprintf(command, sizeof(command),
1449 "%s %s%s --start-address=0x%016" PRIx64
1450 " --stop-address=0x%016" PRIx64
1451 " -l -d %s %s -C \"%s\" 2>/dev/null|grep -v \"%s:\"|expand",
1452 objdump_path ? objdump_path : "objdump",
1453 disassembler_style ? "-M " : "",
1454 disassembler_style ? disassembler_style : "",
1455 map__rip_2objdump(map, sym->start),
1456 map__rip_2objdump(map, sym->end),
1457 symbol_conf.annotate_asm_raw ? "" : "--no-show-raw",
1458 symbol_conf.annotate_src ? "-S" : "",
1459 symfs_filename, symfs_filename);
1461 pr_debug("Executing: %s\n", command);
1464 if (pipe(stdout_fd) < 0) {
1465 pr_err("Failure creating the pipe to run %s\n", command);
1466 goto out_remove_tmp;
1471 pr_err("Failure forking to run %s\n", command);
1472 goto out_close_stdout;
1476 close(stdout_fd[0]);
1477 dup2(stdout_fd[1], 1);
1478 close(stdout_fd[1]);
1479 execl("/bin/sh", "sh", "-c", command, NULL);
1484 close(stdout_fd[1]);
1486 file = fdopen(stdout_fd[0], "r");
1488 pr_err("Failure creating FILE stream for %s\n", command);
1490 * If we were using debug info should retry with
1493 goto out_remove_tmp;
1497 while (!feof(file)) {
1499 * The source code line number (lineno) needs to be kept in
1500 * accross calls to symbol__parse_objdump_line(), so that it
1501 * can associate it with the instructions till the next one.
1502 * See disasm_line__new() and struct disasm_line::line_nr.
1504 if (symbol__parse_objdump_line(sym, map, arch, file, privsize,
1511 pr_err("No output from %s\n", command);
1514 * kallsyms does not have symbol sizes so there may a nop at the end.
1517 if (dso__is_kcore(dso))
1518 delete_last_nop(sym);
1523 close(stdout_fd[0]);
1525 if (dso__needs_decompress(dso))
1526 unlink(symfs_filename);
1529 kcore_extract__delete(&kce);
1534 close(stdout_fd[1]);
1535 goto out_remove_tmp;
1538 static void insert_source_line(struct rb_root *root, struct source_line *src_line)
1540 struct source_line *iter;
1541 struct rb_node **p = &root->rb_node;
1542 struct rb_node *parent = NULL;
1545 while (*p != NULL) {
1547 iter = rb_entry(parent, struct source_line, node);
1549 ret = strcmp(iter->path, src_line->path);
1551 for (i = 0; i < src_line->nr_pcnt; i++)
1552 iter->samples[i].percent_sum += src_line->samples[i].percent;
1559 p = &(*p)->rb_right;
1562 for (i = 0; i < src_line->nr_pcnt; i++)
1563 src_line->samples[i].percent_sum = src_line->samples[i].percent;
1565 rb_link_node(&src_line->node, parent, p);
1566 rb_insert_color(&src_line->node, root);
1569 static int cmp_source_line(struct source_line *a, struct source_line *b)
1573 for (i = 0; i < a->nr_pcnt; i++) {
1574 if (a->samples[i].percent_sum == b->samples[i].percent_sum)
1576 return a->samples[i].percent_sum > b->samples[i].percent_sum;
1582 static void __resort_source_line(struct rb_root *root, struct source_line *src_line)
1584 struct source_line *iter;
1585 struct rb_node **p = &root->rb_node;
1586 struct rb_node *parent = NULL;
1588 while (*p != NULL) {
1590 iter = rb_entry(parent, struct source_line, node);
1592 if (cmp_source_line(src_line, iter))
1595 p = &(*p)->rb_right;
1598 rb_link_node(&src_line->node, parent, p);
1599 rb_insert_color(&src_line->node, root);
1602 static void resort_source_line(struct rb_root *dest_root, struct rb_root *src_root)
1604 struct source_line *src_line;
1605 struct rb_node *node;
1607 node = rb_first(src_root);
1609 struct rb_node *next;
1611 src_line = rb_entry(node, struct source_line, node);
1612 next = rb_next(node);
1613 rb_erase(node, src_root);
1615 __resort_source_line(dest_root, src_line);
1620 static void symbol__free_source_line(struct symbol *sym, int len)
1622 struct annotation *notes = symbol__annotation(sym);
1623 struct source_line *src_line = notes->src->lines;
1624 size_t sizeof_src_line;
1627 sizeof_src_line = sizeof(*src_line) +
1628 (sizeof(src_line->samples) * (src_line->nr_pcnt - 1));
1630 for (i = 0; i < len; i++) {
1631 free_srcline(src_line->path);
1632 src_line = (void *)src_line + sizeof_src_line;
1635 zfree(¬es->src->lines);
1638 /* Get the filename:line for the colored entries */
1639 static int symbol__get_source_line(struct symbol *sym, struct map *map,
1640 struct perf_evsel *evsel,
1641 struct rb_root *root, int len)
1645 int evidx = evsel->idx;
1646 struct source_line *src_line;
1647 struct annotation *notes = symbol__annotation(sym);
1648 struct sym_hist *h = annotation__histogram(notes, evidx);
1649 struct rb_root tmp_root = RB_ROOT;
1652 size_t sizeof_src_line = sizeof(struct source_line);
1654 if (perf_evsel__is_group_event(evsel)) {
1655 for (i = 1; i < evsel->nr_members; i++) {
1656 h = annotation__histogram(notes, evidx + i);
1659 nr_pcnt = evsel->nr_members;
1660 sizeof_src_line += (nr_pcnt - 1) * sizeof(src_line->samples);
1666 src_line = notes->src->lines = calloc(len, sizeof_src_line);
1667 if (!notes->src->lines)
1670 start = map__rip_2objdump(map, sym->start);
1672 for (i = 0; i < len; i++) {
1673 u64 offset, nr_samples;
1674 double percent_max = 0.0;
1676 src_line->nr_pcnt = nr_pcnt;
1678 for (k = 0; k < nr_pcnt; k++) {
1679 double percent = 0.0;
1681 h = annotation__histogram(notes, evidx + k);
1682 nr_samples = h->addr[i];
1684 percent = 100.0 * nr_samples / h->sum;
1686 if (percent > percent_max)
1687 percent_max = percent;
1688 src_line->samples[k].percent = percent;
1689 src_line->samples[k].nr = nr_samples;
1692 if (percent_max <= 0.5)
1696 src_line->path = get_srcline(map->dso, offset, NULL,
1698 insert_source_line(&tmp_root, src_line);
1701 src_line = (void *)src_line + sizeof_src_line;
1704 resort_source_line(root, &tmp_root);
1708 static void print_summary(struct rb_root *root, const char *filename)
1710 struct source_line *src_line;
1711 struct rb_node *node;
1713 printf("\nSorted summary for file %s\n", filename);
1714 printf("----------------------------------------------\n\n");
1716 if (RB_EMPTY_ROOT(root)) {
1717 printf(" Nothing higher than %1.1f%%\n", MIN_GREEN);
1721 node = rb_first(root);
1723 double percent, percent_max = 0.0;
1728 src_line = rb_entry(node, struct source_line, node);
1729 for (i = 0; i < src_line->nr_pcnt; i++) {
1730 percent = src_line->samples[i].percent_sum;
1731 color = get_percent_color(percent);
1732 color_fprintf(stdout, color, " %7.2f", percent);
1734 if (percent > percent_max)
1735 percent_max = percent;
1738 path = src_line->path;
1739 color = get_percent_color(percent_max);
1740 color_fprintf(stdout, color, " %s\n", path);
1742 node = rb_next(node);
1746 static void symbol__annotate_hits(struct symbol *sym, struct perf_evsel *evsel)
1748 struct annotation *notes = symbol__annotation(sym);
1749 struct sym_hist *h = annotation__histogram(notes, evsel->idx);
1750 u64 len = symbol__size(sym), offset;
1752 for (offset = 0; offset < len; ++offset)
1753 if (h->addr[offset] != 0)
1754 printf("%*" PRIx64 ": %" PRIu64 "\n", BITS_PER_LONG / 2,
1755 sym->start + offset, h->addr[offset]);
1756 printf("%*s: %" PRIu64 "\n", BITS_PER_LONG / 2, "h->sum", h->sum);
1759 int symbol__annotate_printf(struct symbol *sym, struct map *map,
1760 struct perf_evsel *evsel, bool full_paths,
1761 int min_pcnt, int max_lines, int context)
1763 struct dso *dso = map->dso;
1765 const char *d_filename;
1766 const char *evsel_name = perf_evsel__name(evsel);
1767 struct annotation *notes = symbol__annotation(sym);
1768 struct sym_hist *h = annotation__histogram(notes, evsel->idx);
1769 struct disasm_line *pos, *queue = NULL;
1770 u64 start = map__rip_2objdump(map, sym->start);
1771 int printed = 2, queue_len = 0;
1775 int graph_dotted_len;
1777 filename = strdup(dso->long_name);
1782 d_filename = filename;
1784 d_filename = basename(filename);
1786 len = symbol__size(sym);
1788 if (perf_evsel__is_group_event(evsel))
1789 width *= evsel->nr_members;
1791 graph_dotted_len = printf(" %-*.*s| Source code & Disassembly of %s for %s (%" PRIu64 " samples)\n",
1792 width, width, "Percent", d_filename, evsel_name, h->sum);
1794 printf("%-*.*s----\n",
1795 graph_dotted_len, graph_dotted_len, graph_dotted_line);
1798 symbol__annotate_hits(sym, evsel);
1800 list_for_each_entry(pos, ¬es->src->source, node) {
1801 if (context && queue == NULL) {
1806 switch (disasm_line__print(pos, sym, start, evsel, len,
1807 min_pcnt, printed, max_lines,
1812 printed += queue_len;
1818 /* filtered by max_lines */
1824 * Filtered by min_pcnt or non IP lines when
1829 if (queue_len == context)
1830 queue = list_entry(queue->node.next, typeof(*queue), node);
1842 void symbol__annotate_zero_histogram(struct symbol *sym, int evidx)
1844 struct annotation *notes = symbol__annotation(sym);
1845 struct sym_hist *h = annotation__histogram(notes, evidx);
1847 memset(h, 0, notes->src->sizeof_sym_hist);
1850 void symbol__annotate_decay_histogram(struct symbol *sym, int evidx)
1852 struct annotation *notes = symbol__annotation(sym);
1853 struct sym_hist *h = annotation__histogram(notes, evidx);
1854 int len = symbol__size(sym), offset;
1857 for (offset = 0; offset < len; ++offset) {
1858 h->addr[offset] = h->addr[offset] * 7 / 8;
1859 h->sum += h->addr[offset];
1863 void disasm__purge(struct list_head *head)
1865 struct disasm_line *pos, *n;
1867 list_for_each_entry_safe(pos, n, head, node) {
1868 list_del(&pos->node);
1869 disasm_line__free(pos);
1873 static size_t disasm_line__fprintf(struct disasm_line *dl, FILE *fp)
1877 if (dl->offset == -1)
1878 return fprintf(fp, "%s\n", dl->line);
1880 printed = fprintf(fp, "%#" PRIx64 " %s", dl->offset, dl->ins.name);
1882 if (dl->ops.raw[0] != '\0') {
1883 printed += fprintf(fp, "%.*s %s\n", 6 - (int)printed, " ",
1887 return printed + fprintf(fp, "\n");
1890 size_t disasm__fprintf(struct list_head *head, FILE *fp)
1892 struct disasm_line *pos;
1895 list_for_each_entry(pos, head, node)
1896 printed += disasm_line__fprintf(pos, fp);
1901 int symbol__tty_annotate(struct symbol *sym, struct map *map,
1902 struct perf_evsel *evsel, bool print_lines,
1903 bool full_paths, int min_pcnt, int max_lines)
1905 struct dso *dso = map->dso;
1906 struct rb_root source_line = RB_ROOT;
1909 if (symbol__disassemble(sym, map, perf_evsel__env_arch(evsel),
1913 len = symbol__size(sym);
1916 srcline_full_filename = full_paths;
1917 symbol__get_source_line(sym, map, evsel, &source_line, len);
1918 print_summary(&source_line, dso->long_name);
1921 symbol__annotate_printf(sym, map, evsel, full_paths,
1922 min_pcnt, max_lines, 0);
1924 symbol__free_source_line(sym, len);
1926 disasm__purge(&symbol__annotation(sym)->src->source);
1931 bool ui__has_annotation(void)
1933 return use_browser == 1 && perf_hpp_list.sym;