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);
53 bool (*ins_is_fused)(struct arch *arch, const char *ins1,
55 int (*cpuid_parse)(struct arch *arch, char *cpuid);
58 char skip_functions_char;
62 static struct ins_ops call_ops;
63 static struct ins_ops dec_ops;
64 static struct ins_ops jump_ops;
65 static struct ins_ops mov_ops;
66 static struct ins_ops nop_ops;
67 static struct ins_ops lock_ops;
68 static struct ins_ops ret_ops;
70 static int arch__grow_instructions(struct arch *arch)
72 struct ins *new_instructions;
73 size_t new_nr_allocated;
75 if (arch->nr_instructions_allocated == 0 && arch->instructions)
76 goto grow_from_non_allocated_table;
78 new_nr_allocated = arch->nr_instructions_allocated + 128;
79 new_instructions = realloc(arch->instructions, new_nr_allocated * sizeof(struct ins));
80 if (new_instructions == NULL)
83 out_update_instructions:
84 arch->instructions = new_instructions;
85 arch->nr_instructions_allocated = new_nr_allocated;
88 grow_from_non_allocated_table:
89 new_nr_allocated = arch->nr_instructions + 128;
90 new_instructions = calloc(new_nr_allocated, sizeof(struct ins));
91 if (new_instructions == NULL)
94 memcpy(new_instructions, arch->instructions, arch->nr_instructions);
95 goto out_update_instructions;
98 static int arch__associate_ins_ops(struct arch* arch, const char *name, struct ins_ops *ops)
102 if (arch->nr_instructions == arch->nr_instructions_allocated &&
103 arch__grow_instructions(arch))
106 ins = &arch->instructions[arch->nr_instructions];
107 ins->name = strdup(name);
112 arch->nr_instructions++;
118 #include "arch/arm/annotate/instructions.c"
119 #include "arch/arm64/annotate/instructions.c"
120 #include "arch/x86/annotate/instructions.c"
121 #include "arch/powerpc/annotate/instructions.c"
122 #include "arch/s390/annotate/instructions.c"
124 static struct arch architectures[] = {
127 .init = arm__annotate_init,
131 .init = arm64__annotate_init,
135 .instructions = x86__instructions,
136 .nr_instructions = ARRAY_SIZE(x86__instructions),
137 .ins_is_fused = x86__ins_is_fused,
138 .cpuid_parse = x86__cpuid_parse,
145 .init = powerpc__annotate_init,
149 .init = s390__annotate_init,
156 static void ins__delete(struct ins_operands *ops)
160 zfree(&ops->source.raw);
161 zfree(&ops->source.name);
162 zfree(&ops->target.raw);
163 zfree(&ops->target.name);
166 static int ins__raw_scnprintf(struct ins *ins, char *bf, size_t size,
167 struct ins_operands *ops)
169 return scnprintf(bf, size, "%-6.6s %s", ins->name, ops->raw);
172 int ins__scnprintf(struct ins *ins, char *bf, size_t size,
173 struct ins_operands *ops)
175 if (ins->ops->scnprintf)
176 return ins->ops->scnprintf(ins, bf, size, ops);
178 return ins__raw_scnprintf(ins, bf, size, ops);
181 bool ins__is_fused(struct arch *arch, const char *ins1, const char *ins2)
183 if (!arch || !arch->ins_is_fused)
186 return arch->ins_is_fused(arch, ins1, ins2);
189 static int call__parse(struct arch *arch, struct ins_operands *ops, struct map *map)
191 char *endptr, *tok, *name;
193 ops->target.addr = strtoull(ops->raw, &endptr, 16);
195 name = strchr(endptr, '<');
201 if (arch->objdump.skip_functions_char &&
202 strchr(name, arch->objdump.skip_functions_char))
205 tok = strchr(name, '>');
210 ops->target.name = strdup(name);
213 return ops->target.name == NULL ? -1 : 0;
216 tok = strchr(endptr, '*');
218 struct symbol *sym = map__find_symbol(map, map->map_ip(map, ops->target.addr));
220 ops->target.name = strdup(sym->name);
222 ops->target.addr = 0;
226 ops->target.addr = strtoull(tok + 1, NULL, 16);
230 static int call__scnprintf(struct ins *ins, char *bf, size_t size,
231 struct ins_operands *ops)
233 if (ops->target.name)
234 return scnprintf(bf, size, "%-6.6s %s", ins->name, ops->target.name);
236 if (ops->target.addr == 0)
237 return ins__raw_scnprintf(ins, bf, size, ops);
239 return scnprintf(bf, size, "%-6.6s *%" PRIx64, ins->name, ops->target.addr);
242 static struct ins_ops call_ops = {
243 .parse = call__parse,
244 .scnprintf = call__scnprintf,
247 bool ins__is_call(const struct ins *ins)
249 return ins->ops == &call_ops;
252 static int jump__parse(struct arch *arch __maybe_unused, struct ins_operands *ops, struct map *map __maybe_unused)
254 const char *s = strchr(ops->raw, '+');
255 const char *c = strchr(ops->raw, ',');
258 * skip over possible up to 2 operands to get to address, e.g.:
259 * tbnz w0, #26, ffff0000083cd190 <security_file_permission+0xd0>
262 ops->target.addr = strtoull(c, NULL, 16);
263 if (!ops->target.addr) {
266 ops->target.addr = strtoull(c, NULL, 16);
269 ops->target.addr = strtoull(ops->raw, NULL, 16);
273 ops->target.offset = strtoull(s, NULL, 16);
274 ops->target.offset_avail = true;
276 ops->target.offset_avail = false;
282 static int jump__scnprintf(struct ins *ins, char *bf, size_t size,
283 struct ins_operands *ops)
285 const char *c = strchr(ops->raw, ',');
287 if (!ops->target.addr || ops->target.offset < 0)
288 return ins__raw_scnprintf(ins, bf, size, ops);
291 const char *c2 = strchr(c + 1, ',');
293 /* check for 3-op insn */
298 /* mirror arch objdump's space-after-comma style */
303 return scnprintf(bf, size, "%-6.6s %.*s%" PRIx64,
304 ins->name, c ? c - ops->raw : 0, ops->raw,
308 static struct ins_ops jump_ops = {
309 .parse = jump__parse,
310 .scnprintf = jump__scnprintf,
313 bool ins__is_jump(const struct ins *ins)
315 return ins->ops == &jump_ops;
318 static int comment__symbol(char *raw, char *comment, u64 *addrp, char **namep)
320 char *endptr, *name, *t;
322 if (strstr(raw, "(%rip)") == NULL)
325 *addrp = strtoull(comment, &endptr, 16);
326 name = strchr(endptr, '<');
332 t = strchr(name, '>');
337 *namep = strdup(name);
343 static int lock__parse(struct arch *arch, struct ins_operands *ops, struct map *map)
345 ops->locked.ops = zalloc(sizeof(*ops->locked.ops));
346 if (ops->locked.ops == NULL)
349 if (disasm_line__parse(ops->raw, &ops->locked.ins.name, &ops->locked.ops->raw) < 0)
352 ops->locked.ins.ops = ins__find(arch, ops->locked.ins.name);
354 if (ops->locked.ins.ops == NULL)
357 if (ops->locked.ins.ops->parse &&
358 ops->locked.ins.ops->parse(arch, ops->locked.ops, map) < 0)
364 zfree(&ops->locked.ops);
368 static int lock__scnprintf(struct ins *ins, char *bf, size_t size,
369 struct ins_operands *ops)
373 if (ops->locked.ins.ops == NULL)
374 return ins__raw_scnprintf(ins, bf, size, ops);
376 printed = scnprintf(bf, size, "%-6.6s ", ins->name);
377 return printed + ins__scnprintf(&ops->locked.ins, bf + printed,
378 size - printed, ops->locked.ops);
381 static void lock__delete(struct ins_operands *ops)
383 struct ins *ins = &ops->locked.ins;
385 if (ins->ops && ins->ops->free)
386 ins->ops->free(ops->locked.ops);
388 ins__delete(ops->locked.ops);
390 zfree(&ops->locked.ops);
391 zfree(&ops->target.raw);
392 zfree(&ops->target.name);
395 static struct ins_ops lock_ops = {
396 .free = lock__delete,
397 .parse = lock__parse,
398 .scnprintf = lock__scnprintf,
401 static int mov__parse(struct arch *arch, struct ins_operands *ops, struct map *map __maybe_unused)
403 char *s = strchr(ops->raw, ','), *target, *comment, prev;
409 ops->source.raw = strdup(ops->raw);
412 if (ops->source.raw == NULL)
416 comment = strchr(s, arch->objdump.comment_char);
421 s = strchr(s, '\0') - 1;
423 while (s > target && isspace(s[0]))
429 ops->target.raw = strdup(target);
432 if (ops->target.raw == NULL)
433 goto out_free_source;
438 comment = ltrim(comment);
439 comment__symbol(ops->source.raw, comment, &ops->source.addr, &ops->source.name);
440 comment__symbol(ops->target.raw, comment, &ops->target.addr, &ops->target.name);
445 zfree(&ops->source.raw);
449 static int mov__scnprintf(struct ins *ins, char *bf, size_t size,
450 struct ins_operands *ops)
452 return scnprintf(bf, size, "%-6.6s %s,%s", ins->name,
453 ops->source.name ?: ops->source.raw,
454 ops->target.name ?: ops->target.raw);
457 static struct ins_ops mov_ops = {
459 .scnprintf = mov__scnprintf,
462 static int dec__parse(struct arch *arch __maybe_unused, struct ins_operands *ops, struct map *map __maybe_unused)
464 char *target, *comment, *s, prev;
466 target = s = ops->raw;
468 while (s[0] != '\0' && !isspace(s[0]))
473 ops->target.raw = strdup(target);
476 if (ops->target.raw == NULL)
479 comment = strchr(s, arch->objdump.comment_char);
483 comment = ltrim(comment);
484 comment__symbol(ops->target.raw, comment, &ops->target.addr, &ops->target.name);
489 static int dec__scnprintf(struct ins *ins, char *bf, size_t size,
490 struct ins_operands *ops)
492 return scnprintf(bf, size, "%-6.6s %s", ins->name,
493 ops->target.name ?: ops->target.raw);
496 static struct ins_ops dec_ops = {
498 .scnprintf = dec__scnprintf,
501 static int nop__scnprintf(struct ins *ins __maybe_unused, char *bf, size_t size,
502 struct ins_operands *ops __maybe_unused)
504 return scnprintf(bf, size, "%-6.6s", "nop");
507 static struct ins_ops nop_ops = {
508 .scnprintf = nop__scnprintf,
511 static struct ins_ops ret_ops = {
512 .scnprintf = ins__raw_scnprintf,
515 bool ins__is_ret(const struct ins *ins)
517 return ins->ops == &ret_ops;
520 bool ins__is_lock(const struct ins *ins)
522 return ins->ops == &lock_ops;
525 static int ins__key_cmp(const void *name, const void *insp)
527 const struct ins *ins = insp;
529 return strcmp(name, ins->name);
532 static int ins__cmp(const void *a, const void *b)
534 const struct ins *ia = a;
535 const struct ins *ib = b;
537 return strcmp(ia->name, ib->name);
540 static void ins__sort(struct arch *arch)
542 const int nmemb = arch->nr_instructions;
544 qsort(arch->instructions, nmemb, sizeof(struct ins), ins__cmp);
547 static struct ins_ops *__ins__find(struct arch *arch, const char *name)
550 const int nmemb = arch->nr_instructions;
552 if (!arch->sorted_instructions) {
554 arch->sorted_instructions = true;
557 ins = bsearch(name, arch->instructions, nmemb, sizeof(struct ins), ins__key_cmp);
558 return ins ? ins->ops : NULL;
561 static struct ins_ops *ins__find(struct arch *arch, const char *name)
563 struct ins_ops *ops = __ins__find(arch, name);
565 if (!ops && arch->associate_instruction_ops)
566 ops = arch->associate_instruction_ops(arch, name);
571 static int arch__key_cmp(const void *name, const void *archp)
573 const struct arch *arch = archp;
575 return strcmp(name, arch->name);
578 static int arch__cmp(const void *a, const void *b)
580 const struct arch *aa = a;
581 const struct arch *ab = b;
583 return strcmp(aa->name, ab->name);
586 static void arch__sort(void)
588 const int nmemb = ARRAY_SIZE(architectures);
590 qsort(architectures, nmemb, sizeof(struct arch), arch__cmp);
593 static struct arch *arch__find(const char *name)
595 const int nmemb = ARRAY_SIZE(architectures);
603 return bsearch(name, architectures, nmemb, sizeof(struct arch), arch__key_cmp);
606 int symbol__alloc_hist(struct symbol *sym)
608 struct annotation *notes = symbol__annotation(sym);
609 size_t size = symbol__size(sym);
610 size_t sizeof_sym_hist;
613 * Add buffer of one element for zero length symbol.
614 * When sample is taken from first instruction of
615 * zero length symbol, perf still resolves it and
616 * shows symbol name in perf report and allows to
622 /* Check for overflow when calculating sizeof_sym_hist */
623 if (size > (SIZE_MAX - sizeof(struct sym_hist)) / sizeof(struct sym_hist_entry))
626 sizeof_sym_hist = (sizeof(struct sym_hist) + size * sizeof(struct sym_hist_entry));
628 /* Check for overflow in zalloc argument */
629 if (sizeof_sym_hist > (SIZE_MAX - sizeof(*notes->src))
630 / symbol_conf.nr_events)
633 notes->src = zalloc(sizeof(*notes->src) + symbol_conf.nr_events * sizeof_sym_hist);
634 if (notes->src == NULL)
636 notes->src->sizeof_sym_hist = sizeof_sym_hist;
637 notes->src->nr_histograms = symbol_conf.nr_events;
638 INIT_LIST_HEAD(¬es->src->source);
642 /* The cycles histogram is lazily allocated. */
643 static int symbol__alloc_hist_cycles(struct symbol *sym)
645 struct annotation *notes = symbol__annotation(sym);
646 const size_t size = symbol__size(sym);
648 notes->src->cycles_hist = calloc(size, sizeof(struct cyc_hist));
649 if (notes->src->cycles_hist == NULL)
654 void symbol__annotate_zero_histograms(struct symbol *sym)
656 struct annotation *notes = symbol__annotation(sym);
658 pthread_mutex_lock(¬es->lock);
659 if (notes->src != NULL) {
660 memset(notes->src->histograms, 0,
661 notes->src->nr_histograms * notes->src->sizeof_sym_hist);
662 if (notes->src->cycles_hist)
663 memset(notes->src->cycles_hist, 0,
664 symbol__size(sym) * sizeof(struct cyc_hist));
666 pthread_mutex_unlock(¬es->lock);
669 static int __symbol__account_cycles(struct annotation *notes,
671 unsigned offset, unsigned cycles,
676 ch = notes->src->cycles_hist;
678 * For now we can only account one basic block per
679 * final jump. But multiple could be overlapping.
680 * Always account the longest one. So when
681 * a shorter one has been already seen throw it away.
683 * We separately always account the full cycles.
685 ch[offset].num_aggr++;
686 ch[offset].cycles_aggr += cycles;
688 if (!have_start && ch[offset].have_start)
690 if (ch[offset].num) {
691 if (have_start && (!ch[offset].have_start ||
692 ch[offset].start > start)) {
693 ch[offset].have_start = 0;
694 ch[offset].cycles = 0;
696 if (ch[offset].reset < 0xffff)
698 } else if (have_start &&
699 ch[offset].start < start)
702 ch[offset].have_start = have_start;
703 ch[offset].start = start;
704 ch[offset].cycles += cycles;
709 static int __symbol__inc_addr_samples(struct symbol *sym, struct map *map,
710 struct annotation *notes, int evidx, u64 addr,
711 struct perf_sample *sample)
716 pr_debug3("%s: addr=%#" PRIx64 "\n", __func__, map->unmap_ip(map, addr));
718 if ((addr < sym->start || addr >= sym->end) &&
719 (addr != sym->end || sym->start != sym->end)) {
720 pr_debug("%s(%d): ERANGE! sym->name=%s, start=%#" PRIx64 ", addr=%#" PRIx64 ", end=%#" PRIx64 "\n",
721 __func__, __LINE__, sym->name, sym->start, addr, sym->end);
725 offset = addr - sym->start;
726 h = annotation__histogram(notes, evidx);
728 h->addr[offset].nr_samples++;
729 h->period += sample->period;
730 h->addr[offset].period += sample->period;
732 pr_debug3("%#" PRIx64 " %s: period++ [addr: %#" PRIx64 ", %#" PRIx64
733 ", evidx=%d] => nr_samples: %" PRIu64 ", period: %" PRIu64 "\n",
734 sym->start, sym->name, addr, addr - sym->start, evidx,
735 h->addr[offset].nr_samples, h->addr[offset].period);
739 static struct annotation *symbol__get_annotation(struct symbol *sym, bool cycles)
741 struct annotation *notes = symbol__annotation(sym);
743 if (notes->src == NULL) {
744 if (symbol__alloc_hist(sym) < 0)
747 if (!notes->src->cycles_hist && cycles) {
748 if (symbol__alloc_hist_cycles(sym) < 0)
754 static int symbol__inc_addr_samples(struct symbol *sym, struct map *map,
756 struct perf_sample *sample)
758 struct annotation *notes;
762 notes = symbol__get_annotation(sym, false);
765 return __symbol__inc_addr_samples(sym, map, notes, evidx, addr, sample);
768 static int symbol__account_cycles(u64 addr, u64 start,
769 struct symbol *sym, unsigned cycles)
771 struct annotation *notes;
776 notes = symbol__get_annotation(sym, true);
779 if (addr < sym->start || addr >= sym->end)
783 if (start < sym->start || start >= sym->end)
788 offset = addr - sym->start;
789 return __symbol__account_cycles(notes,
790 start ? start - sym->start : 0,
795 int addr_map_symbol__account_cycles(struct addr_map_symbol *ams,
796 struct addr_map_symbol *start,
806 * Only set start when IPC can be computed. We can only
807 * compute it when the basic block is completely in a single
809 * Special case the case when the jump is elsewhere, but
810 * it starts on the function start.
813 (start->sym == ams->sym ||
815 start->addr == ams->sym->start + ams->map->start)))
816 saddr = start->al_addr;
818 pr_debug2("BB with bad start: addr %"PRIx64" start %"PRIx64" sym %"PRIx64" saddr %"PRIx64"\n",
820 start ? start->addr : 0,
821 ams->sym ? ams->sym->start + ams->map->start : 0,
823 err = symbol__account_cycles(ams->al_addr, saddr, ams->sym, cycles);
825 pr_debug2("account_cycles failed %d\n", err);
829 int addr_map_symbol__inc_samples(struct addr_map_symbol *ams, struct perf_sample *sample,
832 return symbol__inc_addr_samples(ams->sym, ams->map, evidx, ams->al_addr, sample);
835 int hist_entry__inc_addr_samples(struct hist_entry *he, struct perf_sample *sample,
838 return symbol__inc_addr_samples(he->ms.sym, he->ms.map, evidx, ip, sample);
841 static void disasm_line__init_ins(struct disasm_line *dl, struct arch *arch, struct map *map)
843 dl->ins.ops = ins__find(arch, dl->ins.name);
848 if (dl->ins.ops->parse && dl->ins.ops->parse(arch, &dl->ops, map) < 0)
852 static int disasm_line__parse(char *line, const char **namep, char **rawp)
854 char tmp, *name = ltrim(line);
861 while ((*rawp)[0] != '\0' && !isspace((*rawp)[0]))
866 *namep = strdup(name);
872 *rawp = ltrim(*rawp);
882 static struct disasm_line *disasm_line__new(s64 offset, char *line,
883 size_t privsize, int line_nr,
887 struct disasm_line *dl = zalloc(sizeof(*dl) + privsize);
891 dl->line = strdup(line);
892 dl->line_nr = line_nr;
893 if (dl->line == NULL)
897 if (disasm_line__parse(dl->line, &dl->ins.name, &dl->ops.raw) < 0)
900 disasm_line__init_ins(dl, arch, map);
913 void disasm_line__free(struct disasm_line *dl)
916 if (dl->ins.ops && dl->ins.ops->free)
917 dl->ins.ops->free(&dl->ops);
919 ins__delete(&dl->ops);
920 free((void *)dl->ins.name);
925 int disasm_line__scnprintf(struct disasm_line *dl, char *bf, size_t size, bool raw)
927 if (raw || !dl->ins.ops)
928 return scnprintf(bf, size, "%-6.6s %s", dl->ins.name, dl->ops.raw);
930 return ins__scnprintf(&dl->ins, bf, size, &dl->ops);
933 static void disasm__add(struct list_head *head, struct disasm_line *line)
935 list_add_tail(&line->node, head);
938 struct disasm_line *disasm__get_next_ip_line(struct list_head *head, struct disasm_line *pos)
940 list_for_each_entry_continue(pos, head, node)
941 if (pos->offset >= 0)
947 double disasm__calc_percent(struct annotation *notes, int evidx, s64 offset,
948 s64 end, const char **path, struct sym_hist_entry *sample)
950 struct source_line *src_line = notes->src->lines;
951 double percent = 0.0;
953 sample->nr_samples = sample->period = 0;
956 size_t sizeof_src_line = sizeof(*src_line) +
957 sizeof(src_line->samples) * (src_line->nr_pcnt - 1);
959 while (offset < end) {
960 src_line = (void *)notes->src->lines +
961 (sizeof_src_line * offset);
964 *path = src_line->path;
966 percent += src_line->samples[evidx].percent;
967 sample->nr_samples += src_line->samples[evidx].nr;
971 struct sym_hist *h = annotation__histogram(notes, evidx);
972 unsigned int hits = 0;
975 while (offset < end) {
976 hits += h->addr[offset].nr_samples;
977 period += h->addr[offset].period;
982 sample->period = period;
983 sample->nr_samples = hits;
984 percent = 100.0 * hits / h->nr_samples;
991 static const char *annotate__address_color(struct block_range *br)
993 double cov = block_range__coverage(br);
996 /* mark red for >75% coverage */
998 return PERF_COLOR_RED;
1000 /* mark dull for <1% coverage */
1002 return PERF_COLOR_NORMAL;
1005 return PERF_COLOR_MAGENTA;
1008 static const char *annotate__asm_color(struct block_range *br)
1010 double cov = block_range__coverage(br);
1013 /* mark dull for <1% coverage */
1015 return PERF_COLOR_NORMAL;
1018 return PERF_COLOR_BLUE;
1021 static void annotate__branch_printf(struct block_range *br, u64 addr)
1023 bool emit_comment = true;
1029 if (br->is_target && br->start == addr) {
1030 struct block_range *branch = br;
1034 * Find matching branch to our target.
1036 while (!branch->is_branch)
1037 branch = block_range__next(branch);
1039 p = 100 *(double)br->entry / branch->coverage;
1043 emit_comment = false;
1048 * The percentage of coverage joined at this target in relation
1049 * to the next branch.
1051 printf(" +%.2f%%", p);
1055 if (br->is_branch && br->end == addr) {
1056 double p = 100*(double)br->taken / br->coverage;
1060 emit_comment = false;
1065 * The percentage of coverage leaving at this branch, and
1066 * its prediction ratio.
1068 printf(" -%.2f%% (p:%.2f%%)", p, 100*(double)br->pred / br->taken);
1074 static int disasm_line__print(struct disasm_line *dl, struct symbol *sym, u64 start,
1075 struct perf_evsel *evsel, u64 len, int min_pcnt, int printed,
1076 int max_lines, struct disasm_line *queue)
1078 static const char *prev_line;
1079 static const char *prev_color;
1081 if (dl->offset != -1) {
1082 const char *path = NULL;
1083 double percent, max_percent = 0.0;
1084 double *ppercents = &percent;
1085 struct sym_hist_entry sample;
1086 struct sym_hist_entry *psamples = &sample;
1087 int i, nr_percent = 1;
1089 struct annotation *notes = symbol__annotation(sym);
1090 s64 offset = dl->offset;
1091 const u64 addr = start + offset;
1092 struct disasm_line *next;
1093 struct block_range *br;
1095 next = disasm__get_next_ip_line(¬es->src->source, dl);
1097 if (perf_evsel__is_group_event(evsel)) {
1098 nr_percent = evsel->nr_members;
1099 ppercents = calloc(nr_percent, sizeof(double));
1100 psamples = calloc(nr_percent, sizeof(struct sym_hist_entry));
1101 if (ppercents == NULL || psamples == NULL) {
1106 for (i = 0; i < nr_percent; i++) {
1107 percent = disasm__calc_percent(notes,
1108 notes->src->lines ? i : evsel->idx + i,
1110 next ? next->offset : (s64) len,
1113 ppercents[i] = percent;
1114 psamples[i] = sample;
1115 if (percent > max_percent)
1116 max_percent = percent;
1119 if (max_percent < min_pcnt)
1122 if (max_lines && printed >= max_lines)
1125 if (queue != NULL) {
1126 list_for_each_entry_from(queue, ¬es->src->source, node) {
1129 disasm_line__print(queue, sym, start, evsel, len,
1134 color = get_percent_color(max_percent);
1137 * Also color the filename and line if needed, with
1138 * the same color than the percentage. Don't print it
1139 * twice for close colored addr with the same filename:line
1142 if (!prev_line || strcmp(prev_line, path)
1143 || color != prev_color) {
1144 color_fprintf(stdout, color, " %s", path);
1150 for (i = 0; i < nr_percent; i++) {
1151 percent = ppercents[i];
1152 sample = psamples[i];
1153 color = get_percent_color(percent);
1155 if (symbol_conf.show_total_period)
1156 color_fprintf(stdout, color, " %11" PRIu64,
1158 else if (symbol_conf.show_nr_samples)
1159 color_fprintf(stdout, color, " %7" PRIu64,
1162 color_fprintf(stdout, color, " %7.2f", percent);
1167 br = block_range__find(addr);
1168 color_fprintf(stdout, annotate__address_color(br), " %" PRIx64 ":", addr);
1169 color_fprintf(stdout, annotate__asm_color(br), "%s", dl->line);
1170 annotate__branch_printf(br, addr);
1173 if (ppercents != &percent)
1176 if (psamples != &sample)
1179 } else if (max_lines && printed >= max_lines)
1182 int width = symbol_conf.show_total_period ? 12 : 8;
1187 if (perf_evsel__is_group_event(evsel))
1188 width *= evsel->nr_members;
1191 printf(" %*s:\n", width, " ");
1193 printf(" %*s: %s\n", width, " ", dl->line);
1200 * symbol__parse_objdump_line() parses objdump output (with -d --no-show-raw)
1201 * which looks like following
1203 * 0000000000415500 <_init>:
1204 * 415500: sub $0x8,%rsp
1205 * 415504: mov 0x2f5ad5(%rip),%rax # 70afe0 <_DYNAMIC+0x2f8>
1206 * 41550b: test %rax,%rax
1207 * 41550e: je 415515 <_init+0x15>
1208 * 415510: callq 416e70 <__gmon_start__@plt>
1209 * 415515: add $0x8,%rsp
1212 * it will be parsed and saved into struct disasm_line as
1213 * <offset> <name> <ops.raw>
1215 * The offset will be a relative offset from the start of the symbol and -1
1216 * means that it's not a disassembly line so should be treated differently.
1217 * The ops.raw part will be parsed further according to type of the instruction.
1219 static int symbol__parse_objdump_line(struct symbol *sym, struct map *map,
1221 FILE *file, size_t privsize,
1224 struct annotation *notes = symbol__annotation(sym);
1225 struct disasm_line *dl;
1226 char *line = NULL, *parsed_line, *tmp, *tmp2;
1228 s64 line_ip, offset = -1;
1229 regmatch_t match[2];
1231 if (getline(&line, &line_len, file) < 0)
1238 parsed_line = rtrim(line);
1240 /* /filename:linenr ? Save line number and ignore. */
1241 if (regexec(&file_lineno, parsed_line, 2, match, 0) == 0) {
1242 *line_nr = atoi(parsed_line + match[1].rm_so);
1246 tmp = ltrim(parsed_line);
1249 * Parse hexa addresses followed by ':'
1251 line_ip = strtoull(tmp, &tmp2, 16);
1252 if (*tmp2 != ':' || tmp == tmp2 || tmp2[1] == '\0')
1256 if (line_ip != -1) {
1257 u64 start = map__rip_2objdump(map, sym->start),
1258 end = map__rip_2objdump(map, sym->end);
1260 offset = line_ip - start;
1261 if ((u64)line_ip < start || (u64)line_ip >= end)
1264 parsed_line = tmp2 + 1;
1267 dl = disasm_line__new(offset, parsed_line, privsize, *line_nr, arch, map);
1274 if (!disasm_line__has_offset(dl)) {
1275 dl->ops.target.offset = dl->ops.target.addr -
1276 map__rip_2objdump(map, sym->start);
1277 dl->ops.target.offset_avail = true;
1280 /* kcore has no symbols, so add the call target name */
1281 if (dl->ins.ops && ins__is_call(&dl->ins) && !dl->ops.target.name) {
1282 struct addr_map_symbol target = {
1284 .addr = dl->ops.target.addr,
1287 if (!map_groups__find_ams(&target) &&
1288 target.sym->start == target.al_addr)
1289 dl->ops.target.name = strdup(target.sym->name);
1292 disasm__add(¬es->src->source, dl);
1297 static __attribute__((constructor)) void symbol__init_regexpr(void)
1299 regcomp(&file_lineno, "^/[^:]+:([0-9]+)", REG_EXTENDED);
1302 static void delete_last_nop(struct symbol *sym)
1304 struct annotation *notes = symbol__annotation(sym);
1305 struct list_head *list = ¬es->src->source;
1306 struct disasm_line *dl;
1308 while (!list_empty(list)) {
1309 dl = list_entry(list->prev, struct disasm_line, node);
1312 if (dl->ins.ops != &nop_ops)
1315 if (!strstr(dl->line, " nop ") &&
1316 !strstr(dl->line, " nopl ") &&
1317 !strstr(dl->line, " nopw "))
1321 list_del(&dl->node);
1322 disasm_line__free(dl);
1326 int symbol__strerror_disassemble(struct symbol *sym __maybe_unused, struct map *map,
1327 int errnum, char *buf, size_t buflen)
1329 struct dso *dso = map->dso;
1331 BUG_ON(buflen == 0);
1334 str_error_r(errnum, buf, buflen);
1339 case SYMBOL_ANNOTATE_ERRNO__NO_VMLINUX: {
1340 char bf[SBUILD_ID_SIZE + 15] = " with build id ";
1341 char *build_id_msg = NULL;
1343 if (dso->has_build_id) {
1344 build_id__sprintf(dso->build_id,
1345 sizeof(dso->build_id), bf + 15);
1348 scnprintf(buf, buflen,
1349 "No vmlinux file%s\nwas found in the path.\n\n"
1350 "Note that annotation using /proc/kcore requires CAP_SYS_RAWIO capability.\n\n"
1352 " perf buildid-cache -vu vmlinux\n\n"
1354 " --vmlinux vmlinux\n", build_id_msg ?: "");
1358 scnprintf(buf, buflen, "Internal error: Invalid %d error code\n", errnum);
1365 static int dso__disassemble_filename(struct dso *dso, char *filename, size_t filename_size)
1367 char linkname[PATH_MAX];
1368 char *build_id_filename;
1369 char *build_id_path = NULL;
1372 if (dso->symtab_type == DSO_BINARY_TYPE__KALLSYMS &&
1373 !dso__is_kcore(dso))
1374 return SYMBOL_ANNOTATE_ERRNO__NO_VMLINUX;
1376 build_id_filename = dso__build_id_filename(dso, NULL, 0, false);
1377 if (build_id_filename) {
1378 __symbol__join_symfs(filename, filename_size, build_id_filename);
1379 free(build_id_filename);
1381 if (dso->has_build_id)
1386 build_id_path = strdup(filename);
1391 * old style build-id cache has name of XX/XXXXXXX.. while
1392 * new style has XX/XXXXXXX../{elf,kallsyms,vdso}.
1393 * extract the build-id part of dirname in the new style only.
1395 pos = strrchr(build_id_path, '/');
1396 if (pos && strlen(pos) < SBUILD_ID_SIZE - 2)
1397 dirname(build_id_path);
1399 if (dso__is_kcore(dso) ||
1400 readlink(build_id_path, linkname, sizeof(linkname)) < 0 ||
1401 strstr(linkname, DSO__NAME_KALLSYMS) ||
1402 access(filename, R_OK)) {
1405 * If we don't have build-ids or the build-id file isn't in the
1406 * cache, or is just a kallsyms file, well, lets hope that this
1407 * DSO is the same as when 'perf record' ran.
1409 __symbol__join_symfs(filename, filename_size, dso->long_name);
1412 free(build_id_path);
1416 static const char *annotate__norm_arch(const char *arch_name)
1420 if (!arch_name) { /* Assume we are annotating locally. */
1421 if (uname(&uts) < 0)
1423 arch_name = uts.machine;
1425 return normalize_arch((char *)arch_name);
1428 int symbol__disassemble(struct symbol *sym, struct map *map,
1429 const char *arch_name, size_t privsize,
1430 struct arch **parch, char *cpuid)
1432 struct dso *dso = map->dso;
1433 char command[PATH_MAX * 2];
1434 struct arch *arch = NULL;
1436 char symfs_filename[PATH_MAX];
1437 struct kcore_extract kce;
1438 bool delete_extract = false;
1443 int err = dso__disassemble_filename(dso, symfs_filename, sizeof(symfs_filename));
1448 arch_name = annotate__norm_arch(arch_name);
1452 arch = arch__find(arch_name);
1460 err = arch->init(arch);
1462 pr_err("%s: failed to initialize %s arch priv area\n", __func__, arch->name);
1467 if (arch->cpuid_parse && cpuid)
1468 arch->cpuid_parse(arch, cpuid);
1470 pr_debug("%s: filename=%s, sym=%s, start=%#" PRIx64 ", end=%#" PRIx64 "\n", __func__,
1471 symfs_filename, sym->name, map->unmap_ip(map, sym->start),
1472 map->unmap_ip(map, sym->end));
1474 pr_debug("annotating [%p] %30s : [%p] %30s\n",
1475 dso, dso->long_name, sym, sym->name);
1477 if (dso__is_kcore(dso)) {
1478 kce.kcore_filename = symfs_filename;
1479 kce.addr = map__rip_2objdump(map, sym->start);
1480 kce.offs = sym->start;
1481 kce.len = sym->end - sym->start;
1482 if (!kcore_extract__create(&kce)) {
1483 delete_extract = true;
1484 strlcpy(symfs_filename, kce.extract_filename,
1485 sizeof(symfs_filename));
1487 } else if (dso__needs_decompress(dso)) {
1488 char tmp[KMOD_DECOMP_LEN];
1490 if (dso__decompress_kmodule_path(dso, symfs_filename,
1491 tmp, sizeof(tmp)) < 0)
1494 strcpy(symfs_filename, tmp);
1497 snprintf(command, sizeof(command),
1498 "%s %s%s --start-address=0x%016" PRIx64
1499 " --stop-address=0x%016" PRIx64
1500 " -l -d %s %s -C \"%s\" 2>/dev/null|grep -v \"%s:\"|expand",
1501 objdump_path ? objdump_path : "objdump",
1502 disassembler_style ? "-M " : "",
1503 disassembler_style ? disassembler_style : "",
1504 map__rip_2objdump(map, sym->start),
1505 map__rip_2objdump(map, sym->end),
1506 symbol_conf.annotate_asm_raw ? "" : "--no-show-raw",
1507 symbol_conf.annotate_src ? "-S" : "",
1508 symfs_filename, symfs_filename);
1510 pr_debug("Executing: %s\n", command);
1513 if (pipe(stdout_fd) < 0) {
1514 pr_err("Failure creating the pipe to run %s\n", command);
1515 goto out_remove_tmp;
1520 pr_err("Failure forking to run %s\n", command);
1521 goto out_close_stdout;
1525 close(stdout_fd[0]);
1526 dup2(stdout_fd[1], 1);
1527 close(stdout_fd[1]);
1528 execl("/bin/sh", "sh", "-c", command, NULL);
1533 close(stdout_fd[1]);
1535 file = fdopen(stdout_fd[0], "r");
1537 pr_err("Failure creating FILE stream for %s\n", command);
1539 * If we were using debug info should retry with
1542 goto out_remove_tmp;
1546 while (!feof(file)) {
1548 * The source code line number (lineno) needs to be kept in
1549 * accross calls to symbol__parse_objdump_line(), so that it
1550 * can associate it with the instructions till the next one.
1551 * See disasm_line__new() and struct disasm_line::line_nr.
1553 if (symbol__parse_objdump_line(sym, map, arch, file, privsize,
1560 pr_err("No output from %s\n", command);
1563 * kallsyms does not have symbol sizes so there may a nop at the end.
1566 if (dso__is_kcore(dso))
1567 delete_last_nop(sym);
1572 close(stdout_fd[0]);
1574 if (dso__needs_decompress(dso))
1575 unlink(symfs_filename);
1578 kcore_extract__delete(&kce);
1583 close(stdout_fd[1]);
1584 goto out_remove_tmp;
1587 static void insert_source_line(struct rb_root *root, struct source_line *src_line)
1589 struct source_line *iter;
1590 struct rb_node **p = &root->rb_node;
1591 struct rb_node *parent = NULL;
1594 while (*p != NULL) {
1596 iter = rb_entry(parent, struct source_line, node);
1598 ret = strcmp(iter->path, src_line->path);
1600 for (i = 0; i < src_line->nr_pcnt; i++)
1601 iter->samples[i].percent_sum += src_line->samples[i].percent;
1608 p = &(*p)->rb_right;
1611 for (i = 0; i < src_line->nr_pcnt; i++)
1612 src_line->samples[i].percent_sum = src_line->samples[i].percent;
1614 rb_link_node(&src_line->node, parent, p);
1615 rb_insert_color(&src_line->node, root);
1618 static int cmp_source_line(struct source_line *a, struct source_line *b)
1622 for (i = 0; i < a->nr_pcnt; i++) {
1623 if (a->samples[i].percent_sum == b->samples[i].percent_sum)
1625 return a->samples[i].percent_sum > b->samples[i].percent_sum;
1631 static void __resort_source_line(struct rb_root *root, struct source_line *src_line)
1633 struct source_line *iter;
1634 struct rb_node **p = &root->rb_node;
1635 struct rb_node *parent = NULL;
1637 while (*p != NULL) {
1639 iter = rb_entry(parent, struct source_line, node);
1641 if (cmp_source_line(src_line, iter))
1644 p = &(*p)->rb_right;
1647 rb_link_node(&src_line->node, parent, p);
1648 rb_insert_color(&src_line->node, root);
1651 static void resort_source_line(struct rb_root *dest_root, struct rb_root *src_root)
1653 struct source_line *src_line;
1654 struct rb_node *node;
1656 node = rb_first(src_root);
1658 struct rb_node *next;
1660 src_line = rb_entry(node, struct source_line, node);
1661 next = rb_next(node);
1662 rb_erase(node, src_root);
1664 __resort_source_line(dest_root, src_line);
1669 static void symbol__free_source_line(struct symbol *sym, int len)
1671 struct annotation *notes = symbol__annotation(sym);
1672 struct source_line *src_line = notes->src->lines;
1673 size_t sizeof_src_line;
1676 sizeof_src_line = sizeof(*src_line) +
1677 (sizeof(src_line->samples) * (src_line->nr_pcnt - 1));
1679 for (i = 0; i < len; i++) {
1680 free_srcline(src_line->path);
1681 src_line = (void *)src_line + sizeof_src_line;
1684 zfree(¬es->src->lines);
1687 /* Get the filename:line for the colored entries */
1688 static int symbol__get_source_line(struct symbol *sym, struct map *map,
1689 struct perf_evsel *evsel,
1690 struct rb_root *root, int len)
1694 int evidx = evsel->idx;
1695 struct source_line *src_line;
1696 struct annotation *notes = symbol__annotation(sym);
1697 struct sym_hist *h = annotation__histogram(notes, evidx);
1698 struct rb_root tmp_root = RB_ROOT;
1700 u64 nr_samples = h->nr_samples;
1701 size_t sizeof_src_line = sizeof(struct source_line);
1703 if (perf_evsel__is_group_event(evsel)) {
1704 for (i = 1; i < evsel->nr_members; i++) {
1705 h = annotation__histogram(notes, evidx + i);
1706 nr_samples += h->nr_samples;
1708 nr_pcnt = evsel->nr_members;
1709 sizeof_src_line += (nr_pcnt - 1) * sizeof(src_line->samples);
1715 src_line = notes->src->lines = calloc(len, sizeof_src_line);
1716 if (!notes->src->lines)
1719 start = map__rip_2objdump(map, sym->start);
1721 for (i = 0; i < len; i++) {
1723 double percent_max = 0.0;
1725 src_line->nr_pcnt = nr_pcnt;
1727 for (k = 0; k < nr_pcnt; k++) {
1728 double percent = 0.0;
1730 h = annotation__histogram(notes, evidx + k);
1731 nr_samples = h->addr[i].nr_samples;
1733 percent = 100.0 * nr_samples / h->nr_samples;
1735 if (percent > percent_max)
1736 percent_max = percent;
1737 src_line->samples[k].percent = percent;
1738 src_line->samples[k].nr = nr_samples;
1741 if (percent_max <= 0.5)
1745 src_line->path = get_srcline(map->dso, offset, NULL,
1747 insert_source_line(&tmp_root, src_line);
1750 src_line = (void *)src_line + sizeof_src_line;
1753 resort_source_line(root, &tmp_root);
1757 static void print_summary(struct rb_root *root, const char *filename)
1759 struct source_line *src_line;
1760 struct rb_node *node;
1762 printf("\nSorted summary for file %s\n", filename);
1763 printf("----------------------------------------------\n\n");
1765 if (RB_EMPTY_ROOT(root)) {
1766 printf(" Nothing higher than %1.1f%%\n", MIN_GREEN);
1770 node = rb_first(root);
1772 double percent, percent_max = 0.0;
1777 src_line = rb_entry(node, struct source_line, node);
1778 for (i = 0; i < src_line->nr_pcnt; i++) {
1779 percent = src_line->samples[i].percent_sum;
1780 color = get_percent_color(percent);
1781 color_fprintf(stdout, color, " %7.2f", percent);
1783 if (percent > percent_max)
1784 percent_max = percent;
1787 path = src_line->path;
1788 color = get_percent_color(percent_max);
1789 color_fprintf(stdout, color, " %s\n", path);
1791 node = rb_next(node);
1795 static void symbol__annotate_hits(struct symbol *sym, struct perf_evsel *evsel)
1797 struct annotation *notes = symbol__annotation(sym);
1798 struct sym_hist *h = annotation__histogram(notes, evsel->idx);
1799 u64 len = symbol__size(sym), offset;
1801 for (offset = 0; offset < len; ++offset)
1802 if (h->addr[offset].nr_samples != 0)
1803 printf("%*" PRIx64 ": %" PRIu64 "\n", BITS_PER_LONG / 2,
1804 sym->start + offset, h->addr[offset].nr_samples);
1805 printf("%*s: %" PRIu64 "\n", BITS_PER_LONG / 2, "h->nr_samples", h->nr_samples);
1808 int symbol__annotate_printf(struct symbol *sym, struct map *map,
1809 struct perf_evsel *evsel, bool full_paths,
1810 int min_pcnt, int max_lines, int context)
1812 struct dso *dso = map->dso;
1814 const char *d_filename;
1815 const char *evsel_name = perf_evsel__name(evsel);
1816 struct annotation *notes = symbol__annotation(sym);
1817 struct sym_hist *h = annotation__histogram(notes, evsel->idx);
1818 struct disasm_line *pos, *queue = NULL;
1819 u64 start = map__rip_2objdump(map, sym->start);
1820 int printed = 2, queue_len = 0;
1823 int width = symbol_conf.show_total_period ? 12 : 8;
1824 int graph_dotted_len;
1826 filename = strdup(dso->long_name);
1831 d_filename = filename;
1833 d_filename = basename(filename);
1835 len = symbol__size(sym);
1837 if (perf_evsel__is_group_event(evsel))
1838 width *= evsel->nr_members;
1840 graph_dotted_len = printf(" %-*.*s| Source code & Disassembly of %s for %s (%" PRIu64 " samples)\n",
1841 width, width, symbol_conf.show_total_period ? "Period" :
1842 symbol_conf.show_nr_samples ? "Samples" : "Percent",
1843 d_filename, evsel_name, h->nr_samples);
1845 printf("%-*.*s----\n",
1846 graph_dotted_len, graph_dotted_len, graph_dotted_line);
1849 symbol__annotate_hits(sym, evsel);
1851 list_for_each_entry(pos, ¬es->src->source, node) {
1852 if (context && queue == NULL) {
1857 switch (disasm_line__print(pos, sym, start, evsel, len,
1858 min_pcnt, printed, max_lines,
1863 printed += queue_len;
1869 /* filtered by max_lines */
1875 * Filtered by min_pcnt or non IP lines when
1880 if (queue_len == context)
1881 queue = list_entry(queue->node.next, typeof(*queue), node);
1893 void symbol__annotate_zero_histogram(struct symbol *sym, int evidx)
1895 struct annotation *notes = symbol__annotation(sym);
1896 struct sym_hist *h = annotation__histogram(notes, evidx);
1898 memset(h, 0, notes->src->sizeof_sym_hist);
1901 void symbol__annotate_decay_histogram(struct symbol *sym, int evidx)
1903 struct annotation *notes = symbol__annotation(sym);
1904 struct sym_hist *h = annotation__histogram(notes, evidx);
1905 int len = symbol__size(sym), offset;
1908 for (offset = 0; offset < len; ++offset) {
1909 h->addr[offset].nr_samples = h->addr[offset].nr_samples * 7 / 8;
1910 h->nr_samples += h->addr[offset].nr_samples;
1914 void disasm__purge(struct list_head *head)
1916 struct disasm_line *pos, *n;
1918 list_for_each_entry_safe(pos, n, head, node) {
1919 list_del(&pos->node);
1920 disasm_line__free(pos);
1924 static size_t disasm_line__fprintf(struct disasm_line *dl, FILE *fp)
1928 if (dl->offset == -1)
1929 return fprintf(fp, "%s\n", dl->line);
1931 printed = fprintf(fp, "%#" PRIx64 " %s", dl->offset, dl->ins.name);
1933 if (dl->ops.raw[0] != '\0') {
1934 printed += fprintf(fp, "%.*s %s\n", 6 - (int)printed, " ",
1938 return printed + fprintf(fp, "\n");
1941 size_t disasm__fprintf(struct list_head *head, FILE *fp)
1943 struct disasm_line *pos;
1946 list_for_each_entry(pos, head, node)
1947 printed += disasm_line__fprintf(pos, fp);
1952 int symbol__tty_annotate(struct symbol *sym, struct map *map,
1953 struct perf_evsel *evsel, bool print_lines,
1954 bool full_paths, int min_pcnt, int max_lines)
1956 struct dso *dso = map->dso;
1957 struct rb_root source_line = RB_ROOT;
1960 if (symbol__disassemble(sym, map, perf_evsel__env_arch(evsel),
1964 len = symbol__size(sym);
1967 srcline_full_filename = full_paths;
1968 symbol__get_source_line(sym, map, evsel, &source_line, len);
1969 print_summary(&source_line, dso->long_name);
1972 symbol__annotate_printf(sym, map, evsel, full_paths,
1973 min_pcnt, max_lines, 0);
1975 symbol__free_source_line(sym, len);
1977 disasm__purge(&symbol__annotation(sym)->src->source);
1982 bool ui__has_annotation(void)
1984 return use_browser == 1 && perf_hpp_list.sym;