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;
52 int (*init)(struct arch *arch, char *cpuid);
53 bool (*ins_is_fused)(struct arch *arch, const char *ins1,
57 char skip_functions_char;
61 static struct ins_ops call_ops;
62 static struct ins_ops dec_ops;
63 static struct ins_ops jump_ops;
64 static struct ins_ops mov_ops;
65 static struct ins_ops nop_ops;
66 static struct ins_ops lock_ops;
67 static struct ins_ops ret_ops;
69 static int arch__grow_instructions(struct arch *arch)
71 struct ins *new_instructions;
72 size_t new_nr_allocated;
74 if (arch->nr_instructions_allocated == 0 && arch->instructions)
75 goto grow_from_non_allocated_table;
77 new_nr_allocated = arch->nr_instructions_allocated + 128;
78 new_instructions = realloc(arch->instructions, new_nr_allocated * sizeof(struct ins));
79 if (new_instructions == NULL)
82 out_update_instructions:
83 arch->instructions = new_instructions;
84 arch->nr_instructions_allocated = new_nr_allocated;
87 grow_from_non_allocated_table:
88 new_nr_allocated = arch->nr_instructions + 128;
89 new_instructions = calloc(new_nr_allocated, sizeof(struct ins));
90 if (new_instructions == NULL)
93 memcpy(new_instructions, arch->instructions, arch->nr_instructions);
94 goto out_update_instructions;
97 static int arch__associate_ins_ops(struct arch* arch, const char *name, struct ins_ops *ops)
101 if (arch->nr_instructions == arch->nr_instructions_allocated &&
102 arch__grow_instructions(arch))
105 ins = &arch->instructions[arch->nr_instructions];
106 ins->name = strdup(name);
111 arch->nr_instructions++;
117 #include "arch/arm/annotate/instructions.c"
118 #include "arch/arm64/annotate/instructions.c"
119 #include "arch/x86/annotate/instructions.c"
120 #include "arch/powerpc/annotate/instructions.c"
121 #include "arch/s390/annotate/instructions.c"
123 static struct arch architectures[] = {
126 .init = arm__annotate_init,
130 .init = arm64__annotate_init,
134 .init = x86__annotate_init,
135 .instructions = x86__instructions,
136 .nr_instructions = ARRAY_SIZE(x86__instructions),
137 .ins_is_fused = x86__ins_is_fused,
144 .init = powerpc__annotate_init,
148 .init = s390__annotate_init,
155 static void ins__delete(struct ins_operands *ops)
159 zfree(&ops->source.raw);
160 zfree(&ops->source.name);
161 zfree(&ops->target.raw);
162 zfree(&ops->target.name);
165 static int ins__raw_scnprintf(struct ins *ins, char *bf, size_t size,
166 struct ins_operands *ops)
168 return scnprintf(bf, size, "%-6.6s %s", ins->name, ops->raw);
171 int ins__scnprintf(struct ins *ins, char *bf, size_t size,
172 struct ins_operands *ops)
174 if (ins->ops->scnprintf)
175 return ins->ops->scnprintf(ins, bf, size, ops);
177 return ins__raw_scnprintf(ins, bf, size, ops);
180 bool ins__is_fused(struct arch *arch, const char *ins1, const char *ins2)
182 if (!arch || !arch->ins_is_fused)
185 return arch->ins_is_fused(arch, ins1, ins2);
188 static int call__parse(struct arch *arch, struct ins_operands *ops, struct map *map)
190 char *endptr, *tok, *name;
192 ops->target.addr = strtoull(ops->raw, &endptr, 16);
194 name = strchr(endptr, '<');
200 if (arch->objdump.skip_functions_char &&
201 strchr(name, arch->objdump.skip_functions_char))
204 tok = strchr(name, '>');
209 ops->target.name = strdup(name);
212 return ops->target.name == NULL ? -1 : 0;
215 tok = strchr(endptr, '*');
217 struct symbol *sym = map__find_symbol(map, map->map_ip(map, ops->target.addr));
219 ops->target.name = strdup(sym->name);
221 ops->target.addr = 0;
225 ops->target.addr = strtoull(tok + 1, NULL, 16);
229 static int call__scnprintf(struct ins *ins, char *bf, size_t size,
230 struct ins_operands *ops)
232 if (ops->target.name)
233 return scnprintf(bf, size, "%-6.6s %s", ins->name, ops->target.name);
235 if (ops->target.addr == 0)
236 return ins__raw_scnprintf(ins, bf, size, ops);
238 return scnprintf(bf, size, "%-6.6s *%" PRIx64, ins->name, ops->target.addr);
241 static struct ins_ops call_ops = {
242 .parse = call__parse,
243 .scnprintf = call__scnprintf,
246 bool ins__is_call(const struct ins *ins)
248 return ins->ops == &call_ops;
251 static int jump__parse(struct arch *arch __maybe_unused, struct ins_operands *ops, struct map *map __maybe_unused)
253 const char *s = strchr(ops->raw, '+');
254 const char *c = strchr(ops->raw, ',');
257 * skip over possible up to 2 operands to get to address, e.g.:
258 * tbnz w0, #26, ffff0000083cd190 <security_file_permission+0xd0>
261 ops->target.addr = strtoull(c, NULL, 16);
262 if (!ops->target.addr) {
265 ops->target.addr = strtoull(c, NULL, 16);
268 ops->target.addr = strtoull(ops->raw, NULL, 16);
272 ops->target.offset = strtoull(s, NULL, 16);
273 ops->target.offset_avail = true;
275 ops->target.offset_avail = false;
281 static int jump__scnprintf(struct ins *ins, char *bf, size_t size,
282 struct ins_operands *ops)
284 const char *c = strchr(ops->raw, ',');
286 if (!ops->target.addr || ops->target.offset < 0)
287 return ins__raw_scnprintf(ins, bf, size, ops);
290 const char *c2 = strchr(c + 1, ',');
292 /* check for 3-op insn */
297 /* mirror arch objdump's space-after-comma style */
302 return scnprintf(bf, size, "%-6.6s %.*s%" PRIx64,
303 ins->name, c ? c - ops->raw : 0, ops->raw,
307 static struct ins_ops jump_ops = {
308 .parse = jump__parse,
309 .scnprintf = jump__scnprintf,
312 bool ins__is_jump(const struct ins *ins)
314 return ins->ops == &jump_ops;
317 static int comment__symbol(char *raw, char *comment, u64 *addrp, char **namep)
319 char *endptr, *name, *t;
321 if (strstr(raw, "(%rip)") == NULL)
324 *addrp = strtoull(comment, &endptr, 16);
325 name = strchr(endptr, '<');
331 t = strchr(name, '>');
336 *namep = strdup(name);
342 static int lock__parse(struct arch *arch, struct ins_operands *ops, struct map *map)
344 ops->locked.ops = zalloc(sizeof(*ops->locked.ops));
345 if (ops->locked.ops == NULL)
348 if (disasm_line__parse(ops->raw, &ops->locked.ins.name, &ops->locked.ops->raw) < 0)
351 ops->locked.ins.ops = ins__find(arch, ops->locked.ins.name);
353 if (ops->locked.ins.ops == NULL)
356 if (ops->locked.ins.ops->parse &&
357 ops->locked.ins.ops->parse(arch, ops->locked.ops, map) < 0)
363 zfree(&ops->locked.ops);
367 static int lock__scnprintf(struct ins *ins, char *bf, size_t size,
368 struct ins_operands *ops)
372 if (ops->locked.ins.ops == NULL)
373 return ins__raw_scnprintf(ins, bf, size, ops);
375 printed = scnprintf(bf, size, "%-6.6s ", ins->name);
376 return printed + ins__scnprintf(&ops->locked.ins, bf + printed,
377 size - printed, ops->locked.ops);
380 static void lock__delete(struct ins_operands *ops)
382 struct ins *ins = &ops->locked.ins;
384 if (ins->ops && ins->ops->free)
385 ins->ops->free(ops->locked.ops);
387 ins__delete(ops->locked.ops);
389 zfree(&ops->locked.ops);
390 zfree(&ops->target.raw);
391 zfree(&ops->target.name);
394 static struct ins_ops lock_ops = {
395 .free = lock__delete,
396 .parse = lock__parse,
397 .scnprintf = lock__scnprintf,
400 static int mov__parse(struct arch *arch, struct ins_operands *ops, struct map *map __maybe_unused)
402 char *s = strchr(ops->raw, ','), *target, *comment, prev;
408 ops->source.raw = strdup(ops->raw);
411 if (ops->source.raw == NULL)
415 comment = strchr(s, arch->objdump.comment_char);
420 s = strchr(s, '\0') - 1;
422 while (s > target && isspace(s[0]))
428 ops->target.raw = strdup(target);
431 if (ops->target.raw == NULL)
432 goto out_free_source;
437 comment = ltrim(comment);
438 comment__symbol(ops->source.raw, comment, &ops->source.addr, &ops->source.name);
439 comment__symbol(ops->target.raw, comment, &ops->target.addr, &ops->target.name);
444 zfree(&ops->source.raw);
448 static int mov__scnprintf(struct ins *ins, char *bf, size_t size,
449 struct ins_operands *ops)
451 return scnprintf(bf, size, "%-6.6s %s,%s", ins->name,
452 ops->source.name ?: ops->source.raw,
453 ops->target.name ?: ops->target.raw);
456 static struct ins_ops mov_ops = {
458 .scnprintf = mov__scnprintf,
461 static int dec__parse(struct arch *arch __maybe_unused, struct ins_operands *ops, struct map *map __maybe_unused)
463 char *target, *comment, *s, prev;
465 target = s = ops->raw;
467 while (s[0] != '\0' && !isspace(s[0]))
472 ops->target.raw = strdup(target);
475 if (ops->target.raw == NULL)
478 comment = strchr(s, arch->objdump.comment_char);
482 comment = ltrim(comment);
483 comment__symbol(ops->target.raw, comment, &ops->target.addr, &ops->target.name);
488 static int dec__scnprintf(struct ins *ins, char *bf, size_t size,
489 struct ins_operands *ops)
491 return scnprintf(bf, size, "%-6.6s %s", ins->name,
492 ops->target.name ?: ops->target.raw);
495 static struct ins_ops dec_ops = {
497 .scnprintf = dec__scnprintf,
500 static int nop__scnprintf(struct ins *ins __maybe_unused, char *bf, size_t size,
501 struct ins_operands *ops __maybe_unused)
503 return scnprintf(bf, size, "%-6.6s", "nop");
506 static struct ins_ops nop_ops = {
507 .scnprintf = nop__scnprintf,
510 static struct ins_ops ret_ops = {
511 .scnprintf = ins__raw_scnprintf,
514 bool ins__is_ret(const struct ins *ins)
516 return ins->ops == &ret_ops;
519 bool ins__is_lock(const struct ins *ins)
521 return ins->ops == &lock_ops;
524 static int ins__key_cmp(const void *name, const void *insp)
526 const struct ins *ins = insp;
528 return strcmp(name, ins->name);
531 static int ins__cmp(const void *a, const void *b)
533 const struct ins *ia = a;
534 const struct ins *ib = b;
536 return strcmp(ia->name, ib->name);
539 static void ins__sort(struct arch *arch)
541 const int nmemb = arch->nr_instructions;
543 qsort(arch->instructions, nmemb, sizeof(struct ins), ins__cmp);
546 static struct ins_ops *__ins__find(struct arch *arch, const char *name)
549 const int nmemb = arch->nr_instructions;
551 if (!arch->sorted_instructions) {
553 arch->sorted_instructions = true;
556 ins = bsearch(name, arch->instructions, nmemb, sizeof(struct ins), ins__key_cmp);
557 return ins ? ins->ops : NULL;
560 static struct ins_ops *ins__find(struct arch *arch, const char *name)
562 struct ins_ops *ops = __ins__find(arch, name);
564 if (!ops && arch->associate_instruction_ops)
565 ops = arch->associate_instruction_ops(arch, name);
570 static int arch__key_cmp(const void *name, const void *archp)
572 const struct arch *arch = archp;
574 return strcmp(name, arch->name);
577 static int arch__cmp(const void *a, const void *b)
579 const struct arch *aa = a;
580 const struct arch *ab = b;
582 return strcmp(aa->name, ab->name);
585 static void arch__sort(void)
587 const int nmemb = ARRAY_SIZE(architectures);
589 qsort(architectures, nmemb, sizeof(struct arch), arch__cmp);
592 static struct arch *arch__find(const char *name)
594 const int nmemb = ARRAY_SIZE(architectures);
602 return bsearch(name, architectures, nmemb, sizeof(struct arch), arch__key_cmp);
605 int symbol__alloc_hist(struct symbol *sym)
607 struct annotation *notes = symbol__annotation(sym);
608 size_t size = symbol__size(sym);
609 size_t sizeof_sym_hist;
612 * Add buffer of one element for zero length symbol.
613 * When sample is taken from first instruction of
614 * zero length symbol, perf still resolves it and
615 * shows symbol name in perf report and allows to
621 /* Check for overflow when calculating sizeof_sym_hist */
622 if (size > (SIZE_MAX - sizeof(struct sym_hist)) / sizeof(struct sym_hist_entry))
625 sizeof_sym_hist = (sizeof(struct sym_hist) + size * sizeof(struct sym_hist_entry));
627 /* Check for overflow in zalloc argument */
628 if (sizeof_sym_hist > (SIZE_MAX - sizeof(*notes->src))
629 / symbol_conf.nr_events)
632 notes->src = zalloc(sizeof(*notes->src) + symbol_conf.nr_events * sizeof_sym_hist);
633 if (notes->src == NULL)
635 notes->src->sizeof_sym_hist = sizeof_sym_hist;
636 notes->src->nr_histograms = symbol_conf.nr_events;
637 INIT_LIST_HEAD(¬es->src->source);
641 /* The cycles histogram is lazily allocated. */
642 static int symbol__alloc_hist_cycles(struct symbol *sym)
644 struct annotation *notes = symbol__annotation(sym);
645 const size_t size = symbol__size(sym);
647 notes->src->cycles_hist = calloc(size, sizeof(struct cyc_hist));
648 if (notes->src->cycles_hist == NULL)
653 void symbol__annotate_zero_histograms(struct symbol *sym)
655 struct annotation *notes = symbol__annotation(sym);
657 pthread_mutex_lock(¬es->lock);
658 if (notes->src != NULL) {
659 memset(notes->src->histograms, 0,
660 notes->src->nr_histograms * notes->src->sizeof_sym_hist);
661 if (notes->src->cycles_hist)
662 memset(notes->src->cycles_hist, 0,
663 symbol__size(sym) * sizeof(struct cyc_hist));
665 pthread_mutex_unlock(¬es->lock);
668 static int __symbol__account_cycles(struct annotation *notes,
670 unsigned offset, unsigned cycles,
675 ch = notes->src->cycles_hist;
677 * For now we can only account one basic block per
678 * final jump. But multiple could be overlapping.
679 * Always account the longest one. So when
680 * a shorter one has been already seen throw it away.
682 * We separately always account the full cycles.
684 ch[offset].num_aggr++;
685 ch[offset].cycles_aggr += cycles;
687 if (!have_start && ch[offset].have_start)
689 if (ch[offset].num) {
690 if (have_start && (!ch[offset].have_start ||
691 ch[offset].start > start)) {
692 ch[offset].have_start = 0;
693 ch[offset].cycles = 0;
695 if (ch[offset].reset < 0xffff)
697 } else if (have_start &&
698 ch[offset].start < start)
701 ch[offset].have_start = have_start;
702 ch[offset].start = start;
703 ch[offset].cycles += cycles;
708 static int __symbol__inc_addr_samples(struct symbol *sym, struct map *map,
709 struct annotation *notes, int evidx, u64 addr,
710 struct perf_sample *sample)
715 pr_debug3("%s: addr=%#" PRIx64 "\n", __func__, map->unmap_ip(map, addr));
717 if ((addr < sym->start || addr >= sym->end) &&
718 (addr != sym->end || sym->start != sym->end)) {
719 pr_debug("%s(%d): ERANGE! sym->name=%s, start=%#" PRIx64 ", addr=%#" PRIx64 ", end=%#" PRIx64 "\n",
720 __func__, __LINE__, sym->name, sym->start, addr, sym->end);
724 offset = addr - sym->start;
725 h = annotation__histogram(notes, evidx);
727 h->addr[offset].nr_samples++;
728 h->period += sample->period;
729 h->addr[offset].period += sample->period;
731 pr_debug3("%#" PRIx64 " %s: period++ [addr: %#" PRIx64 ", %#" PRIx64
732 ", evidx=%d] => nr_samples: %" PRIu64 ", period: %" PRIu64 "\n",
733 sym->start, sym->name, addr, addr - sym->start, evidx,
734 h->addr[offset].nr_samples, h->addr[offset].period);
738 static struct annotation *symbol__get_annotation(struct symbol *sym, bool cycles)
740 struct annotation *notes = symbol__annotation(sym);
742 if (notes->src == NULL) {
743 if (symbol__alloc_hist(sym) < 0)
746 if (!notes->src->cycles_hist && cycles) {
747 if (symbol__alloc_hist_cycles(sym) < 0)
753 static int symbol__inc_addr_samples(struct symbol *sym, struct map *map,
755 struct perf_sample *sample)
757 struct annotation *notes;
761 notes = symbol__get_annotation(sym, false);
764 return __symbol__inc_addr_samples(sym, map, notes, evidx, addr, sample);
767 static int symbol__account_cycles(u64 addr, u64 start,
768 struct symbol *sym, unsigned cycles)
770 struct annotation *notes;
775 notes = symbol__get_annotation(sym, true);
778 if (addr < sym->start || addr >= sym->end)
782 if (start < sym->start || start >= sym->end)
787 offset = addr - sym->start;
788 return __symbol__account_cycles(notes,
789 start ? start - sym->start : 0,
794 int addr_map_symbol__account_cycles(struct addr_map_symbol *ams,
795 struct addr_map_symbol *start,
805 * Only set start when IPC can be computed. We can only
806 * compute it when the basic block is completely in a single
808 * Special case the case when the jump is elsewhere, but
809 * it starts on the function start.
812 (start->sym == ams->sym ||
814 start->addr == ams->sym->start + ams->map->start)))
815 saddr = start->al_addr;
817 pr_debug2("BB with bad start: addr %"PRIx64" start %"PRIx64" sym %"PRIx64" saddr %"PRIx64"\n",
819 start ? start->addr : 0,
820 ams->sym ? ams->sym->start + ams->map->start : 0,
822 err = symbol__account_cycles(ams->al_addr, saddr, ams->sym, cycles);
824 pr_debug2("account_cycles failed %d\n", err);
828 int addr_map_symbol__inc_samples(struct addr_map_symbol *ams, struct perf_sample *sample,
831 return symbol__inc_addr_samples(ams->sym, ams->map, evidx, ams->al_addr, sample);
834 int hist_entry__inc_addr_samples(struct hist_entry *he, struct perf_sample *sample,
837 return symbol__inc_addr_samples(he->ms.sym, he->ms.map, evidx, ip, sample);
840 static void disasm_line__init_ins(struct disasm_line *dl, struct arch *arch, struct map *map)
842 dl->ins.ops = ins__find(arch, dl->ins.name);
847 if (dl->ins.ops->parse && dl->ins.ops->parse(arch, &dl->ops, map) < 0)
851 static int disasm_line__parse(char *line, const char **namep, char **rawp)
853 char tmp, *name = ltrim(line);
860 while ((*rawp)[0] != '\0' && !isspace((*rawp)[0]))
865 *namep = strdup(name);
871 *rawp = ltrim(*rawp);
881 static struct disasm_line *disasm_line__new(s64 offset, char *line,
882 size_t privsize, int line_nr,
886 struct disasm_line *dl = zalloc(sizeof(*dl) + privsize);
890 dl->line = strdup(line);
891 dl->line_nr = line_nr;
892 if (dl->line == NULL)
896 if (disasm_line__parse(dl->line, &dl->ins.name, &dl->ops.raw) < 0)
899 disasm_line__init_ins(dl, arch, map);
912 void disasm_line__free(struct disasm_line *dl)
915 if (dl->ins.ops && dl->ins.ops->free)
916 dl->ins.ops->free(&dl->ops);
918 ins__delete(&dl->ops);
919 free((void *)dl->ins.name);
924 int disasm_line__scnprintf(struct disasm_line *dl, char *bf, size_t size, bool raw)
926 if (raw || !dl->ins.ops)
927 return scnprintf(bf, size, "%-6.6s %s", dl->ins.name, dl->ops.raw);
929 return ins__scnprintf(&dl->ins, bf, size, &dl->ops);
932 static void disasm__add(struct list_head *head, struct disasm_line *line)
934 list_add_tail(&line->node, head);
937 struct disasm_line *disasm__get_next_ip_line(struct list_head *head, struct disasm_line *pos)
939 list_for_each_entry_continue(pos, head, node)
940 if (pos->offset >= 0)
946 double disasm__calc_percent(struct annotation *notes, int evidx, s64 offset,
947 s64 end, const char **path, struct sym_hist_entry *sample)
949 struct source_line *src_line = notes->src->lines;
950 double percent = 0.0;
952 sample->nr_samples = sample->period = 0;
955 size_t sizeof_src_line = sizeof(*src_line) +
956 sizeof(src_line->samples) * (src_line->nr_pcnt - 1);
958 while (offset < end) {
959 src_line = (void *)notes->src->lines +
960 (sizeof_src_line * offset);
963 *path = src_line->path;
965 percent += src_line->samples[evidx].percent;
966 sample->nr_samples += src_line->samples[evidx].nr;
970 struct sym_hist *h = annotation__histogram(notes, evidx);
971 unsigned int hits = 0;
974 while (offset < end) {
975 hits += h->addr[offset].nr_samples;
976 period += h->addr[offset].period;
981 sample->period = period;
982 sample->nr_samples = hits;
983 percent = 100.0 * hits / h->nr_samples;
990 static const char *annotate__address_color(struct block_range *br)
992 double cov = block_range__coverage(br);
995 /* mark red for >75% coverage */
997 return PERF_COLOR_RED;
999 /* mark dull for <1% coverage */
1001 return PERF_COLOR_NORMAL;
1004 return PERF_COLOR_MAGENTA;
1007 static const char *annotate__asm_color(struct block_range *br)
1009 double cov = block_range__coverage(br);
1012 /* mark dull for <1% coverage */
1014 return PERF_COLOR_NORMAL;
1017 return PERF_COLOR_BLUE;
1020 static void annotate__branch_printf(struct block_range *br, u64 addr)
1022 bool emit_comment = true;
1028 if (br->is_target && br->start == addr) {
1029 struct block_range *branch = br;
1033 * Find matching branch to our target.
1035 while (!branch->is_branch)
1036 branch = block_range__next(branch);
1038 p = 100 *(double)br->entry / branch->coverage;
1042 emit_comment = false;
1047 * The percentage of coverage joined at this target in relation
1048 * to the next branch.
1050 printf(" +%.2f%%", p);
1054 if (br->is_branch && br->end == addr) {
1055 double p = 100*(double)br->taken / br->coverage;
1059 emit_comment = false;
1064 * The percentage of coverage leaving at this branch, and
1065 * its prediction ratio.
1067 printf(" -%.2f%% (p:%.2f%%)", p, 100*(double)br->pred / br->taken);
1073 static int disasm_line__print(struct disasm_line *dl, struct symbol *sym, u64 start,
1074 struct perf_evsel *evsel, u64 len, int min_pcnt, int printed,
1075 int max_lines, struct disasm_line *queue)
1077 static const char *prev_line;
1078 static const char *prev_color;
1080 if (dl->offset != -1) {
1081 const char *path = NULL;
1082 double percent, max_percent = 0.0;
1083 double *ppercents = &percent;
1084 struct sym_hist_entry sample;
1085 struct sym_hist_entry *psamples = &sample;
1086 int i, nr_percent = 1;
1088 struct annotation *notes = symbol__annotation(sym);
1089 s64 offset = dl->offset;
1090 const u64 addr = start + offset;
1091 struct disasm_line *next;
1092 struct block_range *br;
1094 next = disasm__get_next_ip_line(¬es->src->source, dl);
1096 if (perf_evsel__is_group_event(evsel)) {
1097 nr_percent = evsel->nr_members;
1098 ppercents = calloc(nr_percent, sizeof(double));
1099 psamples = calloc(nr_percent, sizeof(struct sym_hist_entry));
1100 if (ppercents == NULL || psamples == NULL) {
1105 for (i = 0; i < nr_percent; i++) {
1106 percent = disasm__calc_percent(notes,
1107 notes->src->lines ? i : evsel->idx + i,
1109 next ? next->offset : (s64) len,
1112 ppercents[i] = percent;
1113 psamples[i] = sample;
1114 if (percent > max_percent)
1115 max_percent = percent;
1118 if (max_percent < min_pcnt)
1121 if (max_lines && printed >= max_lines)
1124 if (queue != NULL) {
1125 list_for_each_entry_from(queue, ¬es->src->source, node) {
1128 disasm_line__print(queue, sym, start, evsel, len,
1133 color = get_percent_color(max_percent);
1136 * Also color the filename and line if needed, with
1137 * the same color than the percentage. Don't print it
1138 * twice for close colored addr with the same filename:line
1141 if (!prev_line || strcmp(prev_line, path)
1142 || color != prev_color) {
1143 color_fprintf(stdout, color, " %s", path);
1149 for (i = 0; i < nr_percent; i++) {
1150 percent = ppercents[i];
1151 sample = psamples[i];
1152 color = get_percent_color(percent);
1154 if (symbol_conf.show_total_period)
1155 color_fprintf(stdout, color, " %11" PRIu64,
1157 else if (symbol_conf.show_nr_samples)
1158 color_fprintf(stdout, color, " %7" PRIu64,
1161 color_fprintf(stdout, color, " %7.2f", percent);
1166 br = block_range__find(addr);
1167 color_fprintf(stdout, annotate__address_color(br), " %" PRIx64 ":", addr);
1168 color_fprintf(stdout, annotate__asm_color(br), "%s", dl->line);
1169 annotate__branch_printf(br, addr);
1172 if (ppercents != &percent)
1175 if (psamples != &sample)
1178 } else if (max_lines && printed >= max_lines)
1181 int width = symbol_conf.show_total_period ? 12 : 8;
1186 if (perf_evsel__is_group_event(evsel))
1187 width *= evsel->nr_members;
1190 printf(" %*s:\n", width, " ");
1192 printf(" %*s: %s\n", width, " ", dl->line);
1199 * symbol__parse_objdump_line() parses objdump output (with -d --no-show-raw)
1200 * which looks like following
1202 * 0000000000415500 <_init>:
1203 * 415500: sub $0x8,%rsp
1204 * 415504: mov 0x2f5ad5(%rip),%rax # 70afe0 <_DYNAMIC+0x2f8>
1205 * 41550b: test %rax,%rax
1206 * 41550e: je 415515 <_init+0x15>
1207 * 415510: callq 416e70 <__gmon_start__@plt>
1208 * 415515: add $0x8,%rsp
1211 * it will be parsed and saved into struct disasm_line as
1212 * <offset> <name> <ops.raw>
1214 * The offset will be a relative offset from the start of the symbol and -1
1215 * means that it's not a disassembly line so should be treated differently.
1216 * The ops.raw part will be parsed further according to type of the instruction.
1218 static int symbol__parse_objdump_line(struct symbol *sym, struct map *map,
1220 FILE *file, size_t privsize,
1223 struct annotation *notes = symbol__annotation(sym);
1224 struct disasm_line *dl;
1225 char *line = NULL, *parsed_line, *tmp, *tmp2;
1227 s64 line_ip, offset = -1;
1228 regmatch_t match[2];
1230 if (getline(&line, &line_len, file) < 0)
1237 parsed_line = rtrim(line);
1239 /* /filename:linenr ? Save line number and ignore. */
1240 if (regexec(&file_lineno, parsed_line, 2, match, 0) == 0) {
1241 *line_nr = atoi(parsed_line + match[1].rm_so);
1245 tmp = ltrim(parsed_line);
1248 * Parse hexa addresses followed by ':'
1250 line_ip = strtoull(tmp, &tmp2, 16);
1251 if (*tmp2 != ':' || tmp == tmp2 || tmp2[1] == '\0')
1255 if (line_ip != -1) {
1256 u64 start = map__rip_2objdump(map, sym->start),
1257 end = map__rip_2objdump(map, sym->end);
1259 offset = line_ip - start;
1260 if ((u64)line_ip < start || (u64)line_ip >= end)
1263 parsed_line = tmp2 + 1;
1266 dl = disasm_line__new(offset, parsed_line, privsize, *line_nr, arch, map);
1273 if (!disasm_line__has_offset(dl)) {
1274 dl->ops.target.offset = dl->ops.target.addr -
1275 map__rip_2objdump(map, sym->start);
1276 dl->ops.target.offset_avail = true;
1279 /* kcore has no symbols, so add the call target name */
1280 if (dl->ins.ops && ins__is_call(&dl->ins) && !dl->ops.target.name) {
1281 struct addr_map_symbol target = {
1283 .addr = dl->ops.target.addr,
1286 if (!map_groups__find_ams(&target) &&
1287 target.sym->start == target.al_addr)
1288 dl->ops.target.name = strdup(target.sym->name);
1291 disasm__add(¬es->src->source, dl);
1296 static __attribute__((constructor)) void symbol__init_regexpr(void)
1298 regcomp(&file_lineno, "^/[^:]+:([0-9]+)", REG_EXTENDED);
1301 static void delete_last_nop(struct symbol *sym)
1303 struct annotation *notes = symbol__annotation(sym);
1304 struct list_head *list = ¬es->src->source;
1305 struct disasm_line *dl;
1307 while (!list_empty(list)) {
1308 dl = list_entry(list->prev, struct disasm_line, node);
1311 if (dl->ins.ops != &nop_ops)
1314 if (!strstr(dl->line, " nop ") &&
1315 !strstr(dl->line, " nopl ") &&
1316 !strstr(dl->line, " nopw "))
1320 list_del(&dl->node);
1321 disasm_line__free(dl);
1325 int symbol__strerror_disassemble(struct symbol *sym __maybe_unused, struct map *map,
1326 int errnum, char *buf, size_t buflen)
1328 struct dso *dso = map->dso;
1330 BUG_ON(buflen == 0);
1333 str_error_r(errnum, buf, buflen);
1338 case SYMBOL_ANNOTATE_ERRNO__NO_VMLINUX: {
1339 char bf[SBUILD_ID_SIZE + 15] = " with build id ";
1340 char *build_id_msg = NULL;
1342 if (dso->has_build_id) {
1343 build_id__sprintf(dso->build_id,
1344 sizeof(dso->build_id), bf + 15);
1347 scnprintf(buf, buflen,
1348 "No vmlinux file%s\nwas found in the path.\n\n"
1349 "Note that annotation using /proc/kcore requires CAP_SYS_RAWIO capability.\n\n"
1351 " perf buildid-cache -vu vmlinux\n\n"
1353 " --vmlinux vmlinux\n", build_id_msg ?: "");
1357 scnprintf(buf, buflen, "Internal error: Invalid %d error code\n", errnum);
1364 static int dso__disassemble_filename(struct dso *dso, char *filename, size_t filename_size)
1366 char linkname[PATH_MAX];
1367 char *build_id_filename;
1368 char *build_id_path = NULL;
1371 if (dso->symtab_type == DSO_BINARY_TYPE__KALLSYMS &&
1372 !dso__is_kcore(dso))
1373 return SYMBOL_ANNOTATE_ERRNO__NO_VMLINUX;
1375 build_id_filename = dso__build_id_filename(dso, NULL, 0, false);
1376 if (build_id_filename) {
1377 __symbol__join_symfs(filename, filename_size, build_id_filename);
1378 free(build_id_filename);
1380 if (dso->has_build_id)
1385 build_id_path = strdup(filename);
1390 * old style build-id cache has name of XX/XXXXXXX.. while
1391 * new style has XX/XXXXXXX../{elf,kallsyms,vdso}.
1392 * extract the build-id part of dirname in the new style only.
1394 pos = strrchr(build_id_path, '/');
1395 if (pos && strlen(pos) < SBUILD_ID_SIZE - 2)
1396 dirname(build_id_path);
1398 if (dso__is_kcore(dso) ||
1399 readlink(build_id_path, linkname, sizeof(linkname)) < 0 ||
1400 strstr(linkname, DSO__NAME_KALLSYMS) ||
1401 access(filename, R_OK)) {
1404 * If we don't have build-ids or the build-id file isn't in the
1405 * cache, or is just a kallsyms file, well, lets hope that this
1406 * DSO is the same as when 'perf record' ran.
1408 __symbol__join_symfs(filename, filename_size, dso->long_name);
1411 free(build_id_path);
1415 static const char *annotate__norm_arch(const char *arch_name)
1419 if (!arch_name) { /* Assume we are annotating locally. */
1420 if (uname(&uts) < 0)
1422 arch_name = uts.machine;
1424 return normalize_arch((char *)arch_name);
1427 int symbol__disassemble(struct symbol *sym, struct map *map,
1428 const char *arch_name, size_t privsize,
1429 struct arch **parch, char *cpuid)
1431 struct dso *dso = map->dso;
1432 char command[PATH_MAX * 2];
1433 struct arch *arch = NULL;
1435 char symfs_filename[PATH_MAX];
1436 struct kcore_extract kce;
1437 bool delete_extract = false;
1442 int err = dso__disassemble_filename(dso, symfs_filename, sizeof(symfs_filename));
1447 arch_name = annotate__norm_arch(arch_name);
1451 arch = arch__find(arch_name);
1459 err = arch->init(arch, cpuid);
1461 pr_err("%s: failed to initialize %s arch priv area\n", __func__, arch->name);
1466 pr_debug("%s: filename=%s, sym=%s, start=%#" PRIx64 ", end=%#" PRIx64 "\n", __func__,
1467 symfs_filename, sym->name, map->unmap_ip(map, sym->start),
1468 map->unmap_ip(map, sym->end));
1470 pr_debug("annotating [%p] %30s : [%p] %30s\n",
1471 dso, dso->long_name, sym, sym->name);
1473 if (dso__is_kcore(dso)) {
1474 kce.kcore_filename = symfs_filename;
1475 kce.addr = map__rip_2objdump(map, sym->start);
1476 kce.offs = sym->start;
1477 kce.len = sym->end - sym->start;
1478 if (!kcore_extract__create(&kce)) {
1479 delete_extract = true;
1480 strlcpy(symfs_filename, kce.extract_filename,
1481 sizeof(symfs_filename));
1483 } else if (dso__needs_decompress(dso)) {
1484 char tmp[KMOD_DECOMP_LEN];
1486 if (dso__decompress_kmodule_path(dso, symfs_filename,
1487 tmp, sizeof(tmp)) < 0)
1490 strcpy(symfs_filename, tmp);
1493 snprintf(command, sizeof(command),
1494 "%s %s%s --start-address=0x%016" PRIx64
1495 " --stop-address=0x%016" PRIx64
1496 " -l -d %s %s -C \"%s\" 2>/dev/null|grep -v \"%s:\"|expand",
1497 objdump_path ? objdump_path : "objdump",
1498 disassembler_style ? "-M " : "",
1499 disassembler_style ? disassembler_style : "",
1500 map__rip_2objdump(map, sym->start),
1501 map__rip_2objdump(map, sym->end),
1502 symbol_conf.annotate_asm_raw ? "" : "--no-show-raw",
1503 symbol_conf.annotate_src ? "-S" : "",
1504 symfs_filename, symfs_filename);
1506 pr_debug("Executing: %s\n", command);
1509 if (pipe(stdout_fd) < 0) {
1510 pr_err("Failure creating the pipe to run %s\n", command);
1511 goto out_remove_tmp;
1516 pr_err("Failure forking to run %s\n", command);
1517 goto out_close_stdout;
1521 close(stdout_fd[0]);
1522 dup2(stdout_fd[1], 1);
1523 close(stdout_fd[1]);
1524 execl("/bin/sh", "sh", "-c", command, NULL);
1529 close(stdout_fd[1]);
1531 file = fdopen(stdout_fd[0], "r");
1533 pr_err("Failure creating FILE stream for %s\n", command);
1535 * If we were using debug info should retry with
1538 goto out_remove_tmp;
1542 while (!feof(file)) {
1544 * The source code line number (lineno) needs to be kept in
1545 * accross calls to symbol__parse_objdump_line(), so that it
1546 * can associate it with the instructions till the next one.
1547 * See disasm_line__new() and struct disasm_line::line_nr.
1549 if (symbol__parse_objdump_line(sym, map, arch, file, privsize,
1556 pr_err("No output from %s\n", command);
1559 * kallsyms does not have symbol sizes so there may a nop at the end.
1562 if (dso__is_kcore(dso))
1563 delete_last_nop(sym);
1568 close(stdout_fd[0]);
1570 if (dso__needs_decompress(dso))
1571 unlink(symfs_filename);
1574 kcore_extract__delete(&kce);
1579 close(stdout_fd[1]);
1580 goto out_remove_tmp;
1583 static void insert_source_line(struct rb_root *root, struct source_line *src_line)
1585 struct source_line *iter;
1586 struct rb_node **p = &root->rb_node;
1587 struct rb_node *parent = NULL;
1590 while (*p != NULL) {
1592 iter = rb_entry(parent, struct source_line, node);
1594 ret = strcmp(iter->path, src_line->path);
1596 for (i = 0; i < src_line->nr_pcnt; i++)
1597 iter->samples[i].percent_sum += src_line->samples[i].percent;
1604 p = &(*p)->rb_right;
1607 for (i = 0; i < src_line->nr_pcnt; i++)
1608 src_line->samples[i].percent_sum = src_line->samples[i].percent;
1610 rb_link_node(&src_line->node, parent, p);
1611 rb_insert_color(&src_line->node, root);
1614 static int cmp_source_line(struct source_line *a, struct source_line *b)
1618 for (i = 0; i < a->nr_pcnt; i++) {
1619 if (a->samples[i].percent_sum == b->samples[i].percent_sum)
1621 return a->samples[i].percent_sum > b->samples[i].percent_sum;
1627 static void __resort_source_line(struct rb_root *root, struct source_line *src_line)
1629 struct source_line *iter;
1630 struct rb_node **p = &root->rb_node;
1631 struct rb_node *parent = NULL;
1633 while (*p != NULL) {
1635 iter = rb_entry(parent, struct source_line, node);
1637 if (cmp_source_line(src_line, iter))
1640 p = &(*p)->rb_right;
1643 rb_link_node(&src_line->node, parent, p);
1644 rb_insert_color(&src_line->node, root);
1647 static void resort_source_line(struct rb_root *dest_root, struct rb_root *src_root)
1649 struct source_line *src_line;
1650 struct rb_node *node;
1652 node = rb_first(src_root);
1654 struct rb_node *next;
1656 src_line = rb_entry(node, struct source_line, node);
1657 next = rb_next(node);
1658 rb_erase(node, src_root);
1660 __resort_source_line(dest_root, src_line);
1665 static void symbol__free_source_line(struct symbol *sym, int len)
1667 struct annotation *notes = symbol__annotation(sym);
1668 struct source_line *src_line = notes->src->lines;
1669 size_t sizeof_src_line;
1672 sizeof_src_line = sizeof(*src_line) +
1673 (sizeof(src_line->samples) * (src_line->nr_pcnt - 1));
1675 for (i = 0; i < len; i++) {
1676 free_srcline(src_line->path);
1677 src_line = (void *)src_line + sizeof_src_line;
1680 zfree(¬es->src->lines);
1683 /* Get the filename:line for the colored entries */
1684 static int symbol__get_source_line(struct symbol *sym, struct map *map,
1685 struct perf_evsel *evsel,
1686 struct rb_root *root, int len)
1690 int evidx = evsel->idx;
1691 struct source_line *src_line;
1692 struct annotation *notes = symbol__annotation(sym);
1693 struct sym_hist *h = annotation__histogram(notes, evidx);
1694 struct rb_root tmp_root = RB_ROOT;
1696 u64 nr_samples = h->nr_samples;
1697 size_t sizeof_src_line = sizeof(struct source_line);
1699 if (perf_evsel__is_group_event(evsel)) {
1700 for (i = 1; i < evsel->nr_members; i++) {
1701 h = annotation__histogram(notes, evidx + i);
1702 nr_samples += h->nr_samples;
1704 nr_pcnt = evsel->nr_members;
1705 sizeof_src_line += (nr_pcnt - 1) * sizeof(src_line->samples);
1711 src_line = notes->src->lines = calloc(len, sizeof_src_line);
1712 if (!notes->src->lines)
1715 start = map__rip_2objdump(map, sym->start);
1717 for (i = 0; i < len; i++) {
1719 double percent_max = 0.0;
1721 src_line->nr_pcnt = nr_pcnt;
1723 for (k = 0; k < nr_pcnt; k++) {
1724 double percent = 0.0;
1726 h = annotation__histogram(notes, evidx + k);
1727 nr_samples = h->addr[i].nr_samples;
1729 percent = 100.0 * nr_samples / h->nr_samples;
1731 if (percent > percent_max)
1732 percent_max = percent;
1733 src_line->samples[k].percent = percent;
1734 src_line->samples[k].nr = nr_samples;
1737 if (percent_max <= 0.5)
1741 src_line->path = get_srcline(map->dso, offset, NULL,
1743 insert_source_line(&tmp_root, src_line);
1746 src_line = (void *)src_line + sizeof_src_line;
1749 resort_source_line(root, &tmp_root);
1753 static void print_summary(struct rb_root *root, const char *filename)
1755 struct source_line *src_line;
1756 struct rb_node *node;
1758 printf("\nSorted summary for file %s\n", filename);
1759 printf("----------------------------------------------\n\n");
1761 if (RB_EMPTY_ROOT(root)) {
1762 printf(" Nothing higher than %1.1f%%\n", MIN_GREEN);
1766 node = rb_first(root);
1768 double percent, percent_max = 0.0;
1773 src_line = rb_entry(node, struct source_line, node);
1774 for (i = 0; i < src_line->nr_pcnt; i++) {
1775 percent = src_line->samples[i].percent_sum;
1776 color = get_percent_color(percent);
1777 color_fprintf(stdout, color, " %7.2f", percent);
1779 if (percent > percent_max)
1780 percent_max = percent;
1783 path = src_line->path;
1784 color = get_percent_color(percent_max);
1785 color_fprintf(stdout, color, " %s\n", path);
1787 node = rb_next(node);
1791 static void symbol__annotate_hits(struct symbol *sym, struct perf_evsel *evsel)
1793 struct annotation *notes = symbol__annotation(sym);
1794 struct sym_hist *h = annotation__histogram(notes, evsel->idx);
1795 u64 len = symbol__size(sym), offset;
1797 for (offset = 0; offset < len; ++offset)
1798 if (h->addr[offset].nr_samples != 0)
1799 printf("%*" PRIx64 ": %" PRIu64 "\n", BITS_PER_LONG / 2,
1800 sym->start + offset, h->addr[offset].nr_samples);
1801 printf("%*s: %" PRIu64 "\n", BITS_PER_LONG / 2, "h->nr_samples", h->nr_samples);
1804 int symbol__annotate_printf(struct symbol *sym, struct map *map,
1805 struct perf_evsel *evsel, bool full_paths,
1806 int min_pcnt, int max_lines, int context)
1808 struct dso *dso = map->dso;
1810 const char *d_filename;
1811 const char *evsel_name = perf_evsel__name(evsel);
1812 struct annotation *notes = symbol__annotation(sym);
1813 struct sym_hist *h = annotation__histogram(notes, evsel->idx);
1814 struct disasm_line *pos, *queue = NULL;
1815 u64 start = map__rip_2objdump(map, sym->start);
1816 int printed = 2, queue_len = 0;
1819 int width = symbol_conf.show_total_period ? 12 : 8;
1820 int graph_dotted_len;
1822 filename = strdup(dso->long_name);
1827 d_filename = filename;
1829 d_filename = basename(filename);
1831 len = symbol__size(sym);
1833 if (perf_evsel__is_group_event(evsel))
1834 width *= evsel->nr_members;
1836 graph_dotted_len = printf(" %-*.*s| Source code & Disassembly of %s for %s (%" PRIu64 " samples)\n",
1837 width, width, symbol_conf.show_total_period ? "Period" :
1838 symbol_conf.show_nr_samples ? "Samples" : "Percent",
1839 d_filename, evsel_name, h->nr_samples);
1841 printf("%-*.*s----\n",
1842 graph_dotted_len, graph_dotted_len, graph_dotted_line);
1845 symbol__annotate_hits(sym, evsel);
1847 list_for_each_entry(pos, ¬es->src->source, node) {
1848 if (context && queue == NULL) {
1853 switch (disasm_line__print(pos, sym, start, evsel, len,
1854 min_pcnt, printed, max_lines,
1859 printed += queue_len;
1865 /* filtered by max_lines */
1871 * Filtered by min_pcnt or non IP lines when
1876 if (queue_len == context)
1877 queue = list_entry(queue->node.next, typeof(*queue), node);
1889 void symbol__annotate_zero_histogram(struct symbol *sym, int evidx)
1891 struct annotation *notes = symbol__annotation(sym);
1892 struct sym_hist *h = annotation__histogram(notes, evidx);
1894 memset(h, 0, notes->src->sizeof_sym_hist);
1897 void symbol__annotate_decay_histogram(struct symbol *sym, int evidx)
1899 struct annotation *notes = symbol__annotation(sym);
1900 struct sym_hist *h = annotation__histogram(notes, evidx);
1901 int len = symbol__size(sym), offset;
1904 for (offset = 0; offset < len; ++offset) {
1905 h->addr[offset].nr_samples = h->addr[offset].nr_samples * 7 / 8;
1906 h->nr_samples += h->addr[offset].nr_samples;
1910 void disasm__purge(struct list_head *head)
1912 struct disasm_line *pos, *n;
1914 list_for_each_entry_safe(pos, n, head, node) {
1915 list_del(&pos->node);
1916 disasm_line__free(pos);
1920 static size_t disasm_line__fprintf(struct disasm_line *dl, FILE *fp)
1924 if (dl->offset == -1)
1925 return fprintf(fp, "%s\n", dl->line);
1927 printed = fprintf(fp, "%#" PRIx64 " %s", dl->offset, dl->ins.name);
1929 if (dl->ops.raw[0] != '\0') {
1930 printed += fprintf(fp, "%.*s %s\n", 6 - (int)printed, " ",
1934 return printed + fprintf(fp, "\n");
1937 size_t disasm__fprintf(struct list_head *head, FILE *fp)
1939 struct disasm_line *pos;
1942 list_for_each_entry(pos, head, node)
1943 printed += disasm_line__fprintf(pos, fp);
1948 int symbol__tty_annotate(struct symbol *sym, struct map *map,
1949 struct perf_evsel *evsel, bool print_lines,
1950 bool full_paths, int min_pcnt, int max_lines)
1952 struct dso *dso = map->dso;
1953 struct rb_root source_line = RB_ROOT;
1956 if (symbol__disassemble(sym, map, perf_evsel__env_arch(evsel),
1960 len = symbol__size(sym);
1963 srcline_full_filename = full_paths;
1964 symbol__get_source_line(sym, map, evsel, &source_line, len);
1965 print_summary(&source_line, dso->long_name);
1968 symbol__annotate_printf(sym, map, evsel, full_paths,
1969 min_pcnt, max_lines, 0);
1971 symbol__free_source_line(sym, len);
1973 disasm__purge(&symbol__annotation(sym)->src->source);
1978 bool ui__has_annotation(void)
1980 return use_browser == 1 && perf_hpp_list.sym;