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 <linux/bitops.h>
24 const char *disassembler_style
;
25 const char *objdump_path
;
26 static regex_t file_lineno
;
28 static struct ins
*ins__find(const char *name
);
29 static int disasm_line__parse(char *line
, char **namep
, char **rawp
);
31 static void ins__delete(struct ins_operands
*ops
)
35 zfree(&ops
->source
.raw
);
36 zfree(&ops
->source
.name
);
37 zfree(&ops
->target
.raw
);
38 zfree(&ops
->target
.name
);
41 static int ins__raw_scnprintf(struct ins
*ins
, char *bf
, size_t size
,
42 struct ins_operands
*ops
)
44 return scnprintf(bf
, size
, "%-6.6s %s", ins
->name
, ops
->raw
);
47 int ins__scnprintf(struct ins
*ins
, char *bf
, size_t size
,
48 struct ins_operands
*ops
)
50 if (ins
->ops
->scnprintf
)
51 return ins
->ops
->scnprintf(ins
, bf
, size
, ops
);
53 return ins__raw_scnprintf(ins
, bf
, size
, ops
);
56 static int call__parse(struct ins_operands
*ops
)
58 char *endptr
, *tok
, *name
;
60 ops
->target
.addr
= strtoull(ops
->raw
, &endptr
, 16);
62 name
= strchr(endptr
, '<');
69 if (strchr(name
, '+'))
73 tok
= strchr(name
, '>');
78 ops
->target
.name
= strdup(name
);
81 return ops
->target
.name
== NULL
? -1 : 0;
84 tok
= strchr(endptr
, '(');
90 tok
= strchr(endptr
, '*');
94 ops
->target
.addr
= strtoull(tok
+ 1, NULL
, 16);
98 static int call__scnprintf(struct ins
*ins
, char *bf
, size_t size
,
99 struct ins_operands
*ops
)
101 if (ops
->target
.name
)
102 return scnprintf(bf
, size
, "%-6.6s %s", ins
->name
, ops
->target
.name
);
104 if (ops
->target
.addr
== 0)
105 return ins__raw_scnprintf(ins
, bf
, size
, ops
);
107 return scnprintf(bf
, size
, "%-6.6s *%" PRIx64
, ins
->name
, ops
->target
.addr
);
110 static struct ins_ops call_ops
= {
111 .parse
= call__parse
,
112 .scnprintf
= call__scnprintf
,
115 bool ins__is_call(const struct ins
*ins
)
117 return ins
->ops
== &call_ops
;
120 static int jump__parse(struct ins_operands
*ops
)
122 const char *s
= strchr(ops
->raw
, '+');
124 ops
->target
.addr
= strtoull(ops
->raw
, NULL
, 16);
127 ops
->target
.offset
= strtoull(s
, NULL
, 16);
129 ops
->target
.offset
= UINT64_MAX
;
134 static int jump__scnprintf(struct ins
*ins
, char *bf
, size_t size
,
135 struct ins_operands
*ops
)
137 return scnprintf(bf
, size
, "%-6.6s %" PRIx64
, ins
->name
, ops
->target
.offset
);
140 static struct ins_ops jump_ops
= {
141 .parse
= jump__parse
,
142 .scnprintf
= jump__scnprintf
,
145 bool ins__is_jump(const struct ins
*ins
)
147 return ins
->ops
== &jump_ops
;
150 static int comment__symbol(char *raw
, char *comment
, u64
*addrp
, char **namep
)
152 char *endptr
, *name
, *t
;
154 if (strstr(raw
, "(%rip)") == NULL
)
157 *addrp
= strtoull(comment
, &endptr
, 16);
158 name
= strchr(endptr
, '<');
164 t
= strchr(name
, '>');
169 *namep
= strdup(name
);
175 static int lock__parse(struct ins_operands
*ops
)
179 ops
->locked
.ops
= zalloc(sizeof(*ops
->locked
.ops
));
180 if (ops
->locked
.ops
== NULL
)
183 if (disasm_line__parse(ops
->raw
, &name
, &ops
->locked
.ops
->raw
) < 0)
186 ops
->locked
.ins
= ins__find(name
);
189 if (ops
->locked
.ins
== NULL
)
192 if (!ops
->locked
.ins
->ops
)
195 if (ops
->locked
.ins
->ops
->parse
&&
196 ops
->locked
.ins
->ops
->parse(ops
->locked
.ops
) < 0)
202 zfree(&ops
->locked
.ops
);
206 static int lock__scnprintf(struct ins
*ins
, char *bf
, size_t size
,
207 struct ins_operands
*ops
)
211 if (ops
->locked
.ins
== NULL
)
212 return ins__raw_scnprintf(ins
, bf
, size
, ops
);
214 printed
= scnprintf(bf
, size
, "%-6.6s ", ins
->name
);
215 return printed
+ ins__scnprintf(ops
->locked
.ins
, bf
+ printed
,
216 size
- printed
, ops
->locked
.ops
);
219 static void lock__delete(struct ins_operands
*ops
)
221 struct ins
*ins
= ops
->locked
.ins
;
223 if (ins
&& ins
->ops
->free
)
224 ins
->ops
->free(ops
->locked
.ops
);
226 ins__delete(ops
->locked
.ops
);
228 zfree(&ops
->locked
.ops
);
229 zfree(&ops
->target
.raw
);
230 zfree(&ops
->target
.name
);
233 static struct ins_ops lock_ops
= {
234 .free
= lock__delete
,
235 .parse
= lock__parse
,
236 .scnprintf
= lock__scnprintf
,
239 static int mov__parse(struct ins_operands
*ops
)
241 char *s
= strchr(ops
->raw
, ','), *target
, *comment
, prev
;
247 ops
->source
.raw
= strdup(ops
->raw
);
250 if (ops
->source
.raw
== NULL
)
255 comment
= strchr(s
, ';');
257 comment
= strchr(s
, '#');
263 s
= strchr(s
, '\0') - 1;
265 while (s
> target
&& isspace(s
[0]))
271 ops
->target
.raw
= strdup(target
);
274 if (ops
->target
.raw
== NULL
)
275 goto out_free_source
;
280 while (comment
[0] != '\0' && isspace(comment
[0]))
283 comment__symbol(ops
->source
.raw
, comment
, &ops
->source
.addr
, &ops
->source
.name
);
284 comment__symbol(ops
->target
.raw
, comment
, &ops
->target
.addr
, &ops
->target
.name
);
289 zfree(&ops
->source
.raw
);
293 static int mov__scnprintf(struct ins
*ins
, char *bf
, size_t size
,
294 struct ins_operands
*ops
)
296 return scnprintf(bf
, size
, "%-6.6s %s,%s", ins
->name
,
297 ops
->source
.name
?: ops
->source
.raw
,
298 ops
->target
.name
?: ops
->target
.raw
);
301 static struct ins_ops mov_ops
= {
303 .scnprintf
= mov__scnprintf
,
306 static int dec__parse(struct ins_operands
*ops
)
308 char *target
, *comment
, *s
, prev
;
310 target
= s
= ops
->raw
;
312 while (s
[0] != '\0' && !isspace(s
[0]))
317 ops
->target
.raw
= strdup(target
);
320 if (ops
->target
.raw
== NULL
)
323 comment
= strchr(s
, '#');
327 while (comment
[0] != '\0' && isspace(comment
[0]))
330 comment__symbol(ops
->target
.raw
, comment
, &ops
->target
.addr
, &ops
->target
.name
);
335 static int dec__scnprintf(struct ins
*ins
, char *bf
, size_t size
,
336 struct ins_operands
*ops
)
338 return scnprintf(bf
, size
, "%-6.6s %s", ins
->name
,
339 ops
->target
.name
?: ops
->target
.raw
);
342 static struct ins_ops dec_ops
= {
344 .scnprintf
= dec__scnprintf
,
347 static int nop__scnprintf(struct ins
*ins __maybe_unused
, char *bf
, size_t size
,
348 struct ins_operands
*ops __maybe_unused
)
350 return scnprintf(bf
, size
, "%-6.6s", "nop");
353 static struct ins_ops nop_ops
= {
354 .scnprintf
= nop__scnprintf
,
357 static struct ins_ops ret_ops
= {
358 .scnprintf
= ins__raw_scnprintf
,
361 bool ins__is_ret(const struct ins
*ins
)
363 return ins
->ops
== &ret_ops
;
366 static struct ins instructions
[] = {
367 { .name
= "add", .ops
= &mov_ops
, },
368 { .name
= "addl", .ops
= &mov_ops
, },
369 { .name
= "addq", .ops
= &mov_ops
, },
370 { .name
= "addw", .ops
= &mov_ops
, },
371 { .name
= "and", .ops
= &mov_ops
, },
373 { .name
= "b", .ops
= &jump_ops
, }, // might also be a call
374 { .name
= "bcc", .ops
= &jump_ops
, },
375 { .name
= "bcs", .ops
= &jump_ops
, },
376 { .name
= "beq", .ops
= &jump_ops
, },
377 { .name
= "bge", .ops
= &jump_ops
, },
378 { .name
= "bgt", .ops
= &jump_ops
, },
379 { .name
= "bhi", .ops
= &jump_ops
, },
380 { .name
= "bl", .ops
= &call_ops
, },
381 { .name
= "bls", .ops
= &jump_ops
, },
382 { .name
= "blt", .ops
= &jump_ops
, },
383 { .name
= "blx", .ops
= &call_ops
, },
384 { .name
= "bne", .ops
= &jump_ops
, },
386 { .name
= "bts", .ops
= &mov_ops
, },
387 { .name
= "call", .ops
= &call_ops
, },
388 { .name
= "callq", .ops
= &call_ops
, },
389 { .name
= "cmp", .ops
= &mov_ops
, },
390 { .name
= "cmpb", .ops
= &mov_ops
, },
391 { .name
= "cmpl", .ops
= &mov_ops
, },
392 { .name
= "cmpq", .ops
= &mov_ops
, },
393 { .name
= "cmpw", .ops
= &mov_ops
, },
394 { .name
= "cmpxch", .ops
= &mov_ops
, },
395 { .name
= "dec", .ops
= &dec_ops
, },
396 { .name
= "decl", .ops
= &dec_ops
, },
397 { .name
= "imul", .ops
= &mov_ops
, },
398 { .name
= "inc", .ops
= &dec_ops
, },
399 { .name
= "incl", .ops
= &dec_ops
, },
400 { .name
= "ja", .ops
= &jump_ops
, },
401 { .name
= "jae", .ops
= &jump_ops
, },
402 { .name
= "jb", .ops
= &jump_ops
, },
403 { .name
= "jbe", .ops
= &jump_ops
, },
404 { .name
= "jc", .ops
= &jump_ops
, },
405 { .name
= "jcxz", .ops
= &jump_ops
, },
406 { .name
= "je", .ops
= &jump_ops
, },
407 { .name
= "jecxz", .ops
= &jump_ops
, },
408 { .name
= "jg", .ops
= &jump_ops
, },
409 { .name
= "jge", .ops
= &jump_ops
, },
410 { .name
= "jl", .ops
= &jump_ops
, },
411 { .name
= "jle", .ops
= &jump_ops
, },
412 { .name
= "jmp", .ops
= &jump_ops
, },
413 { .name
= "jmpq", .ops
= &jump_ops
, },
414 { .name
= "jna", .ops
= &jump_ops
, },
415 { .name
= "jnae", .ops
= &jump_ops
, },
416 { .name
= "jnb", .ops
= &jump_ops
, },
417 { .name
= "jnbe", .ops
= &jump_ops
, },
418 { .name
= "jnc", .ops
= &jump_ops
, },
419 { .name
= "jne", .ops
= &jump_ops
, },
420 { .name
= "jng", .ops
= &jump_ops
, },
421 { .name
= "jnge", .ops
= &jump_ops
, },
422 { .name
= "jnl", .ops
= &jump_ops
, },
423 { .name
= "jnle", .ops
= &jump_ops
, },
424 { .name
= "jno", .ops
= &jump_ops
, },
425 { .name
= "jnp", .ops
= &jump_ops
, },
426 { .name
= "jns", .ops
= &jump_ops
, },
427 { .name
= "jnz", .ops
= &jump_ops
, },
428 { .name
= "jo", .ops
= &jump_ops
, },
429 { .name
= "jp", .ops
= &jump_ops
, },
430 { .name
= "jpe", .ops
= &jump_ops
, },
431 { .name
= "jpo", .ops
= &jump_ops
, },
432 { .name
= "jrcxz", .ops
= &jump_ops
, },
433 { .name
= "js", .ops
= &jump_ops
, },
434 { .name
= "jz", .ops
= &jump_ops
, },
435 { .name
= "lea", .ops
= &mov_ops
, },
436 { .name
= "lock", .ops
= &lock_ops
, },
437 { .name
= "mov", .ops
= &mov_ops
, },
438 { .name
= "movb", .ops
= &mov_ops
, },
439 { .name
= "movdqa",.ops
= &mov_ops
, },
440 { .name
= "movl", .ops
= &mov_ops
, },
441 { .name
= "movq", .ops
= &mov_ops
, },
442 { .name
= "movslq", .ops
= &mov_ops
, },
443 { .name
= "movzbl", .ops
= &mov_ops
, },
444 { .name
= "movzwl", .ops
= &mov_ops
, },
445 { .name
= "nop", .ops
= &nop_ops
, },
446 { .name
= "nopl", .ops
= &nop_ops
, },
447 { .name
= "nopw", .ops
= &nop_ops
, },
448 { .name
= "or", .ops
= &mov_ops
, },
449 { .name
= "orl", .ops
= &mov_ops
, },
450 { .name
= "test", .ops
= &mov_ops
, },
451 { .name
= "testb", .ops
= &mov_ops
, },
452 { .name
= "testl", .ops
= &mov_ops
, },
453 { .name
= "xadd", .ops
= &mov_ops
, },
454 { .name
= "xbeginl", .ops
= &jump_ops
, },
455 { .name
= "xbeginq", .ops
= &jump_ops
, },
456 { .name
= "retq", .ops
= &ret_ops
, },
459 static int ins__key_cmp(const void *name
, const void *insp
)
461 const struct ins
*ins
= insp
;
463 return strcmp(name
, ins
->name
);
466 static int ins__cmp(const void *a
, const void *b
)
468 const struct ins
*ia
= a
;
469 const struct ins
*ib
= b
;
471 return strcmp(ia
->name
, ib
->name
);
474 static void ins__sort(void)
476 const int nmemb
= ARRAY_SIZE(instructions
);
478 qsort(instructions
, nmemb
, sizeof(struct ins
), ins__cmp
);
481 static struct ins
*ins__find(const char *name
)
483 const int nmemb
= ARRAY_SIZE(instructions
);
491 return bsearch(name
, instructions
, nmemb
, sizeof(struct ins
), ins__key_cmp
);
494 int symbol__annotate_init(struct map
*map __maybe_unused
, struct symbol
*sym
)
496 struct annotation
*notes
= symbol__annotation(sym
);
497 pthread_mutex_init(¬es
->lock
, NULL
);
501 int symbol__alloc_hist(struct symbol
*sym
)
503 struct annotation
*notes
= symbol__annotation(sym
);
504 const size_t size
= symbol__size(sym
);
505 size_t sizeof_sym_hist
;
507 /* Check for overflow when calculating sizeof_sym_hist */
508 if (size
> (SIZE_MAX
- sizeof(struct sym_hist
)) / sizeof(u64
))
511 sizeof_sym_hist
= (sizeof(struct sym_hist
) + size
* sizeof(u64
));
513 /* Check for overflow in zalloc argument */
514 if (sizeof_sym_hist
> (SIZE_MAX
- sizeof(*notes
->src
))
515 / symbol_conf
.nr_events
)
518 notes
->src
= zalloc(sizeof(*notes
->src
) + symbol_conf
.nr_events
* sizeof_sym_hist
);
519 if (notes
->src
== NULL
)
521 notes
->src
->sizeof_sym_hist
= sizeof_sym_hist
;
522 notes
->src
->nr_histograms
= symbol_conf
.nr_events
;
523 INIT_LIST_HEAD(¬es
->src
->source
);
527 /* The cycles histogram is lazily allocated. */
528 static int symbol__alloc_hist_cycles(struct symbol
*sym
)
530 struct annotation
*notes
= symbol__annotation(sym
);
531 const size_t size
= symbol__size(sym
);
533 notes
->src
->cycles_hist
= calloc(size
, sizeof(struct cyc_hist
));
534 if (notes
->src
->cycles_hist
== NULL
)
539 void symbol__annotate_zero_histograms(struct symbol
*sym
)
541 struct annotation
*notes
= symbol__annotation(sym
);
543 pthread_mutex_lock(¬es
->lock
);
544 if (notes
->src
!= NULL
) {
545 memset(notes
->src
->histograms
, 0,
546 notes
->src
->nr_histograms
* notes
->src
->sizeof_sym_hist
);
547 if (notes
->src
->cycles_hist
)
548 memset(notes
->src
->cycles_hist
, 0,
549 symbol__size(sym
) * sizeof(struct cyc_hist
));
551 pthread_mutex_unlock(¬es
->lock
);
554 static int __symbol__account_cycles(struct annotation
*notes
,
556 unsigned offset
, unsigned cycles
,
561 ch
= notes
->src
->cycles_hist
;
563 * For now we can only account one basic block per
564 * final jump. But multiple could be overlapping.
565 * Always account the longest one. So when
566 * a shorter one has been already seen throw it away.
568 * We separately always account the full cycles.
570 ch
[offset
].num_aggr
++;
571 ch
[offset
].cycles_aggr
+= cycles
;
573 if (!have_start
&& ch
[offset
].have_start
)
575 if (ch
[offset
].num
) {
576 if (have_start
&& (!ch
[offset
].have_start
||
577 ch
[offset
].start
> start
)) {
578 ch
[offset
].have_start
= 0;
579 ch
[offset
].cycles
= 0;
581 if (ch
[offset
].reset
< 0xffff)
583 } else if (have_start
&&
584 ch
[offset
].start
< start
)
587 ch
[offset
].have_start
= have_start
;
588 ch
[offset
].start
= start
;
589 ch
[offset
].cycles
+= cycles
;
594 static int __symbol__inc_addr_samples(struct symbol
*sym
, struct map
*map
,
595 struct annotation
*notes
, int evidx
, u64 addr
)
600 pr_debug3("%s: addr=%#" PRIx64
"\n", __func__
, map
->unmap_ip(map
, addr
));
602 if (addr
< sym
->start
|| addr
>= sym
->end
) {
603 pr_debug("%s(%d): ERANGE! sym->name=%s, start=%#" PRIx64
", addr=%#" PRIx64
", end=%#" PRIx64
"\n",
604 __func__
, __LINE__
, sym
->name
, sym
->start
, addr
, sym
->end
);
608 offset
= addr
- sym
->start
;
609 h
= annotation__histogram(notes
, evidx
);
613 pr_debug3("%#" PRIx64
" %s: period++ [addr: %#" PRIx64
", %#" PRIx64
614 ", evidx=%d] => %" PRIu64
"\n", sym
->start
, sym
->name
,
615 addr
, addr
- sym
->start
, evidx
, h
->addr
[offset
]);
619 static struct annotation
*symbol__get_annotation(struct symbol
*sym
, bool cycles
)
621 struct annotation
*notes
= symbol__annotation(sym
);
623 if (notes
->src
== NULL
) {
624 if (symbol__alloc_hist(sym
) < 0)
627 if (!notes
->src
->cycles_hist
&& cycles
) {
628 if (symbol__alloc_hist_cycles(sym
) < 0)
634 static int symbol__inc_addr_samples(struct symbol
*sym
, struct map
*map
,
637 struct annotation
*notes
;
641 notes
= symbol__get_annotation(sym
, false);
644 return __symbol__inc_addr_samples(sym
, map
, notes
, evidx
, addr
);
647 static int symbol__account_cycles(u64 addr
, u64 start
,
648 struct symbol
*sym
, unsigned cycles
)
650 struct annotation
*notes
;
655 notes
= symbol__get_annotation(sym
, true);
658 if (addr
< sym
->start
|| addr
>= sym
->end
)
662 if (start
< sym
->start
|| start
>= sym
->end
)
667 offset
= addr
- sym
->start
;
668 return __symbol__account_cycles(notes
,
669 start
? start
- sym
->start
: 0,
674 int addr_map_symbol__account_cycles(struct addr_map_symbol
*ams
,
675 struct addr_map_symbol
*start
,
685 * Only set start when IPC can be computed. We can only
686 * compute it when the basic block is completely in a single
688 * Special case the case when the jump is elsewhere, but
689 * it starts on the function start.
692 (start
->sym
== ams
->sym
||
694 start
->addr
== ams
->sym
->start
+ ams
->map
->start
)))
695 saddr
= start
->al_addr
;
697 pr_debug2("BB with bad start: addr %"PRIx64
" start %"PRIx64
" sym %"PRIx64
" saddr %"PRIx64
"\n",
699 start
? start
->addr
: 0,
700 ams
->sym
? ams
->sym
->start
+ ams
->map
->start
: 0,
702 err
= symbol__account_cycles(ams
->al_addr
, saddr
, ams
->sym
, cycles
);
704 pr_debug2("account_cycles failed %d\n", err
);
708 int addr_map_symbol__inc_samples(struct addr_map_symbol
*ams
, int evidx
)
710 return symbol__inc_addr_samples(ams
->sym
, ams
->map
, evidx
, ams
->al_addr
);
713 int hist_entry__inc_addr_samples(struct hist_entry
*he
, int evidx
, u64 ip
)
715 return symbol__inc_addr_samples(he
->ms
.sym
, he
->ms
.map
, evidx
, ip
);
718 static void disasm_line__init_ins(struct disasm_line
*dl
)
720 dl
->ins
= ins__find(dl
->name
);
728 if (dl
->ins
->ops
->parse
&& dl
->ins
->ops
->parse(&dl
->ops
) < 0)
732 static int disasm_line__parse(char *line
, char **namep
, char **rawp
)
734 char *name
= line
, tmp
;
736 while (isspace(name
[0]))
744 while ((*rawp
)[0] != '\0' && !isspace((*rawp
)[0]))
749 *namep
= strdup(name
);
756 if ((*rawp
)[0] != '\0') {
758 while (isspace((*rawp
)[0]))
769 static struct disasm_line
*disasm_line__new(s64 offset
, char *line
,
770 size_t privsize
, int line_nr
)
772 struct disasm_line
*dl
= zalloc(sizeof(*dl
) + privsize
);
776 dl
->line
= strdup(line
);
777 dl
->line_nr
= line_nr
;
778 if (dl
->line
== NULL
)
782 if (disasm_line__parse(dl
->line
, &dl
->name
, &dl
->ops
.raw
) < 0)
785 disasm_line__init_ins(dl
);
798 void disasm_line__free(struct disasm_line
*dl
)
802 if (dl
->ins
&& dl
->ins
->ops
->free
)
803 dl
->ins
->ops
->free(&dl
->ops
);
805 ins__delete(&dl
->ops
);
809 int disasm_line__scnprintf(struct disasm_line
*dl
, char *bf
, size_t size
, bool raw
)
812 return scnprintf(bf
, size
, "%-6.6s %s", dl
->name
, dl
->ops
.raw
);
814 return ins__scnprintf(dl
->ins
, bf
, size
, &dl
->ops
);
817 static void disasm__add(struct list_head
*head
, struct disasm_line
*line
)
819 list_add_tail(&line
->node
, head
);
822 struct disasm_line
*disasm__get_next_ip_line(struct list_head
*head
, struct disasm_line
*pos
)
824 list_for_each_entry_continue(pos
, head
, node
)
825 if (pos
->offset
>= 0)
831 double disasm__calc_percent(struct annotation
*notes
, int evidx
, s64 offset
,
832 s64 end
, const char **path
, u64
*nr_samples
)
834 struct source_line
*src_line
= notes
->src
->lines
;
835 double percent
= 0.0;
839 size_t sizeof_src_line
= sizeof(*src_line
) +
840 sizeof(src_line
->samples
) * (src_line
->nr_pcnt
- 1);
842 while (offset
< end
) {
843 src_line
= (void *)notes
->src
->lines
+
844 (sizeof_src_line
* offset
);
847 *path
= src_line
->path
;
849 percent
+= src_line
->samples
[evidx
].percent
;
850 *nr_samples
+= src_line
->samples
[evidx
].nr
;
854 struct sym_hist
*h
= annotation__histogram(notes
, evidx
);
855 unsigned int hits
= 0;
858 hits
+= h
->addr
[offset
++];
862 percent
= 100.0 * hits
/ h
->sum
;
869 static int disasm_line__print(struct disasm_line
*dl
, struct symbol
*sym
, u64 start
,
870 struct perf_evsel
*evsel
, u64 len
, int min_pcnt
, int printed
,
871 int max_lines
, struct disasm_line
*queue
)
873 static const char *prev_line
;
874 static const char *prev_color
;
876 if (dl
->offset
!= -1) {
877 const char *path
= NULL
;
879 double percent
, max_percent
= 0.0;
880 double *ppercents
= &percent
;
881 u64
*psamples
= &nr_samples
;
882 int i
, nr_percent
= 1;
884 struct annotation
*notes
= symbol__annotation(sym
);
885 s64 offset
= dl
->offset
;
886 const u64 addr
= start
+ offset
;
887 struct disasm_line
*next
;
889 next
= disasm__get_next_ip_line(¬es
->src
->source
, dl
);
891 if (perf_evsel__is_group_event(evsel
)) {
892 nr_percent
= evsel
->nr_members
;
893 ppercents
= calloc(nr_percent
, sizeof(double));
894 psamples
= calloc(nr_percent
, sizeof(u64
));
895 if (ppercents
== NULL
|| psamples
== NULL
) {
900 for (i
= 0; i
< nr_percent
; i
++) {
901 percent
= disasm__calc_percent(notes
,
902 notes
->src
->lines
? i
: evsel
->idx
+ i
,
904 next
? next
->offset
: (s64
) len
,
907 ppercents
[i
] = percent
;
908 psamples
[i
] = nr_samples
;
909 if (percent
> max_percent
)
910 max_percent
= percent
;
913 if (max_percent
< min_pcnt
)
916 if (max_lines
&& printed
>= max_lines
)
920 list_for_each_entry_from(queue
, ¬es
->src
->source
, node
) {
923 disasm_line__print(queue
, sym
, start
, evsel
, len
,
928 color
= get_percent_color(max_percent
);
931 * Also color the filename and line if needed, with
932 * the same color than the percentage. Don't print it
933 * twice for close colored addr with the same filename:line
936 if (!prev_line
|| strcmp(prev_line
, path
)
937 || color
!= prev_color
) {
938 color_fprintf(stdout
, color
, " %s", path
);
944 for (i
= 0; i
< nr_percent
; i
++) {
945 percent
= ppercents
[i
];
946 nr_samples
= psamples
[i
];
947 color
= get_percent_color(percent
);
949 if (symbol_conf
.show_total_period
)
950 color_fprintf(stdout
, color
, " %7" PRIu64
,
953 color_fprintf(stdout
, color
, " %7.2f", percent
);
957 color_fprintf(stdout
, PERF_COLOR_MAGENTA
, " %" PRIx64
":", addr
);
958 color_fprintf(stdout
, PERF_COLOR_BLUE
, "%s\n", dl
->line
);
960 if (ppercents
!= &percent
)
963 if (psamples
!= &nr_samples
)
966 } else if (max_lines
&& printed
>= max_lines
)
974 if (perf_evsel__is_group_event(evsel
))
975 width
*= evsel
->nr_members
;
978 printf(" %*s:\n", width
, " ");
980 printf(" %*s: %s\n", width
, " ", dl
->line
);
987 * symbol__parse_objdump_line() parses objdump output (with -d --no-show-raw)
988 * which looks like following
990 * 0000000000415500 <_init>:
991 * 415500: sub $0x8,%rsp
992 * 415504: mov 0x2f5ad5(%rip),%rax # 70afe0 <_DYNAMIC+0x2f8>
993 * 41550b: test %rax,%rax
994 * 41550e: je 415515 <_init+0x15>
995 * 415510: callq 416e70 <__gmon_start__@plt>
996 * 415515: add $0x8,%rsp
999 * it will be parsed and saved into struct disasm_line as
1000 * <offset> <name> <ops.raw>
1002 * The offset will be a relative offset from the start of the symbol and -1
1003 * means that it's not a disassembly line so should be treated differently.
1004 * The ops.raw part will be parsed further according to type of the instruction.
1006 static int symbol__parse_objdump_line(struct symbol
*sym
, struct map
*map
,
1007 FILE *file
, size_t privsize
,
1010 struct annotation
*notes
= symbol__annotation(sym
);
1011 struct disasm_line
*dl
;
1012 char *line
= NULL
, *parsed_line
, *tmp
, *tmp2
, *c
;
1014 s64 line_ip
, offset
= -1;
1015 regmatch_t match
[2];
1017 if (getline(&line
, &line_len
, file
) < 0)
1023 while (line_len
!= 0 && isspace(line
[line_len
- 1]))
1024 line
[--line_len
] = '\0';
1026 c
= strchr(line
, '\n');
1033 /* /filename:linenr ? Save line number and ignore. */
1034 if (regexec(&file_lineno
, line
, 2, match
, 0) == 0) {
1035 *line_nr
= atoi(line
+ match
[1].rm_so
);
1040 * Strip leading spaces:
1051 * Parse hexa addresses followed by ':'
1053 line_ip
= strtoull(tmp
, &tmp2
, 16);
1054 if (*tmp2
!= ':' || tmp
== tmp2
|| tmp2
[1] == '\0')
1058 if (line_ip
!= -1) {
1059 u64 start
= map__rip_2objdump(map
, sym
->start
),
1060 end
= map__rip_2objdump(map
, sym
->end
);
1062 offset
= line_ip
- start
;
1063 if ((u64
)line_ip
< start
|| (u64
)line_ip
>= end
)
1066 parsed_line
= tmp2
+ 1;
1069 dl
= disasm_line__new(offset
, parsed_line
, privsize
, *line_nr
);
1076 if (dl
->ops
.target
.offset
== UINT64_MAX
)
1077 dl
->ops
.target
.offset
= dl
->ops
.target
.addr
-
1078 map__rip_2objdump(map
, sym
->start
);
1080 /* kcore has no symbols, so add the call target name */
1081 if (dl
->ins
&& ins__is_call(dl
->ins
) && !dl
->ops
.target
.name
) {
1082 struct addr_map_symbol target
= {
1084 .addr
= dl
->ops
.target
.addr
,
1087 if (!map_groups__find_ams(&target
, NULL
) &&
1088 target
.sym
->start
== target
.al_addr
)
1089 dl
->ops
.target
.name
= strdup(target
.sym
->name
);
1092 disasm__add(¬es
->src
->source
, dl
);
1097 static __attribute__((constructor
)) void symbol__init_regexpr(void)
1099 regcomp(&file_lineno
, "^/[^:]+:([0-9]+)", REG_EXTENDED
);
1102 static void delete_last_nop(struct symbol
*sym
)
1104 struct annotation
*notes
= symbol__annotation(sym
);
1105 struct list_head
*list
= ¬es
->src
->source
;
1106 struct disasm_line
*dl
;
1108 while (!list_empty(list
)) {
1109 dl
= list_entry(list
->prev
, struct disasm_line
, node
);
1111 if (dl
->ins
&& dl
->ins
->ops
) {
1112 if (dl
->ins
->ops
!= &nop_ops
)
1115 if (!strstr(dl
->line
, " nop ") &&
1116 !strstr(dl
->line
, " nopl ") &&
1117 !strstr(dl
->line
, " nopw "))
1121 list_del(&dl
->node
);
1122 disasm_line__free(dl
);
1126 int symbol__annotate(struct symbol
*sym
, struct map
*map
, size_t privsize
)
1128 struct dso
*dso
= map
->dso
;
1129 char *filename
= dso__build_id_filename(dso
, NULL
, 0);
1130 bool free_filename
= true;
1131 char command
[PATH_MAX
* 2];
1134 char symfs_filename
[PATH_MAX
];
1135 struct kcore_extract kce
;
1136 bool delete_extract
= false;
1141 symbol__join_symfs(symfs_filename
, filename
);
1143 if (filename
== NULL
) {
1144 if (dso
->has_build_id
) {
1145 pr_err("Can't annotate %s: not enough memory\n",
1150 } else if (dso__is_kcore(dso
)) {
1152 } else if (readlink(symfs_filename
, command
, sizeof(command
)) < 0 ||
1153 strstr(command
, DSO__NAME_KALLSYMS
) ||
1154 access(symfs_filename
, R_OK
)) {
1158 * If we don't have build-ids or the build-id file isn't in the
1159 * cache, or is just a kallsyms file, well, lets hope that this
1160 * DSO is the same as when 'perf record' ran.
1162 filename
= (char *)dso
->long_name
;
1163 symbol__join_symfs(symfs_filename
, filename
);
1164 free_filename
= false;
1167 if (dso
->symtab_type
== DSO_BINARY_TYPE__KALLSYMS
&&
1168 !dso__is_kcore(dso
)) {
1169 char bf
[SBUILD_ID_SIZE
+ 15] = " with build id ";
1170 char *build_id_msg
= NULL
;
1172 if (dso
->annotate_warned
)
1173 goto out_free_filename
;
1175 if (dso
->has_build_id
) {
1176 build_id__sprintf(dso
->build_id
,
1177 sizeof(dso
->build_id
), bf
+ 15);
1181 dso
->annotate_warned
= 1;
1182 pr_err("Can't annotate %s:\n\n"
1183 "No vmlinux file%s\nwas found in the path.\n\n"
1184 "Note that annotation using /proc/kcore requires CAP_SYS_RAWIO capability.\n\n"
1186 " perf buildid-cache -vu vmlinux\n\n"
1188 " --vmlinux vmlinux\n",
1189 sym
->name
, build_id_msg
?: "");
1190 goto out_free_filename
;
1193 pr_debug("%s: filename=%s, sym=%s, start=%#" PRIx64
", end=%#" PRIx64
"\n", __func__
,
1194 filename
, sym
->name
, map
->unmap_ip(map
, sym
->start
),
1195 map
->unmap_ip(map
, sym
->end
));
1197 pr_debug("annotating [%p] %30s : [%p] %30s\n",
1198 dso
, dso
->long_name
, sym
, sym
->name
);
1200 if (dso__is_kcore(dso
)) {
1201 kce
.kcore_filename
= symfs_filename
;
1202 kce
.addr
= map__rip_2objdump(map
, sym
->start
);
1203 kce
.offs
= sym
->start
;
1204 kce
.len
= sym
->end
- sym
->start
;
1205 if (!kcore_extract__create(&kce
)) {
1206 delete_extract
= true;
1207 strlcpy(symfs_filename
, kce
.extract_filename
,
1208 sizeof(symfs_filename
));
1209 if (free_filename
) {
1211 free_filename
= false;
1213 filename
= symfs_filename
;
1215 } else if (dso__needs_decompress(dso
)) {
1221 if (kmod_path__parse_ext(&m
, symfs_filename
))
1222 goto out_free_filename
;
1224 snprintf(tmp
, PATH_MAX
, "/tmp/perf-kmod-XXXXXX");
1229 goto out_free_filename
;
1232 ret
= decompress_to_file(m
.ext
, symfs_filename
, fd
);
1235 pr_err("Cannot decompress %s %s\n", m
.ext
, symfs_filename
);
1241 goto out_free_filename
;
1243 strcpy(symfs_filename
, tmp
);
1246 snprintf(command
, sizeof(command
),
1247 "%s %s%s --start-address=0x%016" PRIx64
1248 " --stop-address=0x%016" PRIx64
1249 " -l -d %s %s -C %s 2>/dev/null|grep -v %s|expand",
1250 objdump_path
? objdump_path
: "objdump",
1251 disassembler_style
? "-M " : "",
1252 disassembler_style
? disassembler_style
: "",
1253 map__rip_2objdump(map
, sym
->start
),
1254 map__rip_2objdump(map
, sym
->end
),
1255 symbol_conf
.annotate_asm_raw
? "" : "--no-show-raw",
1256 symbol_conf
.annotate_src
? "-S" : "",
1257 symfs_filename
, filename
);
1259 pr_debug("Executing: %s\n", command
);
1261 file
= popen(command
, "r");
1263 pr_err("Failure running %s\n", command
);
1265 * If we were using debug info should retry with
1268 goto out_remove_tmp
;
1272 while (!feof(file
)) {
1273 if (symbol__parse_objdump_line(sym
, map
, file
, privsize
,
1280 pr_err("No output from %s\n", command
);
1283 * kallsyms does not have symbol sizes so there may a nop at the end.
1286 if (dso__is_kcore(dso
))
1287 delete_last_nop(sym
);
1292 if (dso__needs_decompress(dso
))
1293 unlink(symfs_filename
);
1296 kcore_extract__delete(&kce
);
1302 static void insert_source_line(struct rb_root
*root
, struct source_line
*src_line
)
1304 struct source_line
*iter
;
1305 struct rb_node
**p
= &root
->rb_node
;
1306 struct rb_node
*parent
= NULL
;
1309 while (*p
!= NULL
) {
1311 iter
= rb_entry(parent
, struct source_line
, node
);
1313 ret
= strcmp(iter
->path
, src_line
->path
);
1315 for (i
= 0; i
< src_line
->nr_pcnt
; i
++)
1316 iter
->samples
[i
].percent_sum
+= src_line
->samples
[i
].percent
;
1323 p
= &(*p
)->rb_right
;
1326 for (i
= 0; i
< src_line
->nr_pcnt
; i
++)
1327 src_line
->samples
[i
].percent_sum
= src_line
->samples
[i
].percent
;
1329 rb_link_node(&src_line
->node
, parent
, p
);
1330 rb_insert_color(&src_line
->node
, root
);
1333 static int cmp_source_line(struct source_line
*a
, struct source_line
*b
)
1337 for (i
= 0; i
< a
->nr_pcnt
; i
++) {
1338 if (a
->samples
[i
].percent_sum
== b
->samples
[i
].percent_sum
)
1340 return a
->samples
[i
].percent_sum
> b
->samples
[i
].percent_sum
;
1346 static void __resort_source_line(struct rb_root
*root
, struct source_line
*src_line
)
1348 struct source_line
*iter
;
1349 struct rb_node
**p
= &root
->rb_node
;
1350 struct rb_node
*parent
= NULL
;
1352 while (*p
!= NULL
) {
1354 iter
= rb_entry(parent
, struct source_line
, node
);
1356 if (cmp_source_line(src_line
, iter
))
1359 p
= &(*p
)->rb_right
;
1362 rb_link_node(&src_line
->node
, parent
, p
);
1363 rb_insert_color(&src_line
->node
, root
);
1366 static void resort_source_line(struct rb_root
*dest_root
, struct rb_root
*src_root
)
1368 struct source_line
*src_line
;
1369 struct rb_node
*node
;
1371 node
= rb_first(src_root
);
1373 struct rb_node
*next
;
1375 src_line
= rb_entry(node
, struct source_line
, node
);
1376 next
= rb_next(node
);
1377 rb_erase(node
, src_root
);
1379 __resort_source_line(dest_root
, src_line
);
1384 static void symbol__free_source_line(struct symbol
*sym
, int len
)
1386 struct annotation
*notes
= symbol__annotation(sym
);
1387 struct source_line
*src_line
= notes
->src
->lines
;
1388 size_t sizeof_src_line
;
1391 sizeof_src_line
= sizeof(*src_line
) +
1392 (sizeof(src_line
->samples
) * (src_line
->nr_pcnt
- 1));
1394 for (i
= 0; i
< len
; i
++) {
1395 free_srcline(src_line
->path
);
1396 src_line
= (void *)src_line
+ sizeof_src_line
;
1399 zfree(¬es
->src
->lines
);
1402 /* Get the filename:line for the colored entries */
1403 static int symbol__get_source_line(struct symbol
*sym
, struct map
*map
,
1404 struct perf_evsel
*evsel
,
1405 struct rb_root
*root
, int len
)
1409 int evidx
= evsel
->idx
;
1410 struct source_line
*src_line
;
1411 struct annotation
*notes
= symbol__annotation(sym
);
1412 struct sym_hist
*h
= annotation__histogram(notes
, evidx
);
1413 struct rb_root tmp_root
= RB_ROOT
;
1416 size_t sizeof_src_line
= sizeof(struct source_line
);
1418 if (perf_evsel__is_group_event(evsel
)) {
1419 for (i
= 1; i
< evsel
->nr_members
; i
++) {
1420 h
= annotation__histogram(notes
, evidx
+ i
);
1423 nr_pcnt
= evsel
->nr_members
;
1424 sizeof_src_line
+= (nr_pcnt
- 1) * sizeof(src_line
->samples
);
1430 src_line
= notes
->src
->lines
= calloc(len
, sizeof_src_line
);
1431 if (!notes
->src
->lines
)
1434 start
= map__rip_2objdump(map
, sym
->start
);
1436 for (i
= 0; i
< len
; i
++) {
1438 double percent_max
= 0.0;
1440 src_line
->nr_pcnt
= nr_pcnt
;
1442 for (k
= 0; k
< nr_pcnt
; k
++) {
1443 h
= annotation__histogram(notes
, evidx
+ k
);
1444 src_line
->samples
[k
].percent
= 100.0 * h
->addr
[i
] / h
->sum
;
1446 if (src_line
->samples
[k
].percent
> percent_max
)
1447 percent_max
= src_line
->samples
[k
].percent
;
1450 if (percent_max
<= 0.5)
1454 src_line
->path
= get_srcline(map
->dso
, offset
, NULL
, false);
1455 insert_source_line(&tmp_root
, src_line
);
1458 src_line
= (void *)src_line
+ sizeof_src_line
;
1461 resort_source_line(root
, &tmp_root
);
1465 static void print_summary(struct rb_root
*root
, const char *filename
)
1467 struct source_line
*src_line
;
1468 struct rb_node
*node
;
1470 printf("\nSorted summary for file %s\n", filename
);
1471 printf("----------------------------------------------\n\n");
1473 if (RB_EMPTY_ROOT(root
)) {
1474 printf(" Nothing higher than %1.1f%%\n", MIN_GREEN
);
1478 node
= rb_first(root
);
1480 double percent
, percent_max
= 0.0;
1485 src_line
= rb_entry(node
, struct source_line
, node
);
1486 for (i
= 0; i
< src_line
->nr_pcnt
; i
++) {
1487 percent
= src_line
->samples
[i
].percent_sum
;
1488 color
= get_percent_color(percent
);
1489 color_fprintf(stdout
, color
, " %7.2f", percent
);
1491 if (percent
> percent_max
)
1492 percent_max
= percent
;
1495 path
= src_line
->path
;
1496 color
= get_percent_color(percent_max
);
1497 color_fprintf(stdout
, color
, " %s\n", path
);
1499 node
= rb_next(node
);
1503 static void symbol__annotate_hits(struct symbol
*sym
, struct perf_evsel
*evsel
)
1505 struct annotation
*notes
= symbol__annotation(sym
);
1506 struct sym_hist
*h
= annotation__histogram(notes
, evsel
->idx
);
1507 u64 len
= symbol__size(sym
), offset
;
1509 for (offset
= 0; offset
< len
; ++offset
)
1510 if (h
->addr
[offset
] != 0)
1511 printf("%*" PRIx64
": %" PRIu64
"\n", BITS_PER_LONG
/ 2,
1512 sym
->start
+ offset
, h
->addr
[offset
]);
1513 printf("%*s: %" PRIu64
"\n", BITS_PER_LONG
/ 2, "h->sum", h
->sum
);
1516 int symbol__annotate_printf(struct symbol
*sym
, struct map
*map
,
1517 struct perf_evsel
*evsel
, bool full_paths
,
1518 int min_pcnt
, int max_lines
, int context
)
1520 struct dso
*dso
= map
->dso
;
1522 const char *d_filename
;
1523 const char *evsel_name
= perf_evsel__name(evsel
);
1524 struct annotation
*notes
= symbol__annotation(sym
);
1525 struct sym_hist
*h
= annotation__histogram(notes
, evsel
->idx
);
1526 struct disasm_line
*pos
, *queue
= NULL
;
1527 u64 start
= map__rip_2objdump(map
, sym
->start
);
1528 int printed
= 2, queue_len
= 0;
1532 int graph_dotted_len
;
1534 filename
= strdup(dso
->long_name
);
1539 d_filename
= filename
;
1541 d_filename
= basename(filename
);
1543 len
= symbol__size(sym
);
1545 if (perf_evsel__is_group_event(evsel
))
1546 width
*= evsel
->nr_members
;
1548 graph_dotted_len
= printf(" %-*.*s| Source code & Disassembly of %s for %s (%" PRIu64
" samples)\n",
1549 width
, width
, "Percent", d_filename
, evsel_name
, h
->sum
);
1551 printf("%-*.*s----\n",
1552 graph_dotted_len
, graph_dotted_len
, graph_dotted_line
);
1555 symbol__annotate_hits(sym
, evsel
);
1557 list_for_each_entry(pos
, ¬es
->src
->source
, node
) {
1558 if (context
&& queue
== NULL
) {
1563 switch (disasm_line__print(pos
, sym
, start
, evsel
, len
,
1564 min_pcnt
, printed
, max_lines
,
1569 printed
+= queue_len
;
1575 /* filtered by max_lines */
1581 * Filtered by min_pcnt or non IP lines when
1586 if (queue_len
== context
)
1587 queue
= list_entry(queue
->node
.next
, typeof(*queue
), node
);
1599 void symbol__annotate_zero_histogram(struct symbol
*sym
, int evidx
)
1601 struct annotation
*notes
= symbol__annotation(sym
);
1602 struct sym_hist
*h
= annotation__histogram(notes
, evidx
);
1604 memset(h
, 0, notes
->src
->sizeof_sym_hist
);
1607 void symbol__annotate_decay_histogram(struct symbol
*sym
, int evidx
)
1609 struct annotation
*notes
= symbol__annotation(sym
);
1610 struct sym_hist
*h
= annotation__histogram(notes
, evidx
);
1611 int len
= symbol__size(sym
), offset
;
1614 for (offset
= 0; offset
< len
; ++offset
) {
1615 h
->addr
[offset
] = h
->addr
[offset
] * 7 / 8;
1616 h
->sum
+= h
->addr
[offset
];
1620 void disasm__purge(struct list_head
*head
)
1622 struct disasm_line
*pos
, *n
;
1624 list_for_each_entry_safe(pos
, n
, head
, node
) {
1625 list_del(&pos
->node
);
1626 disasm_line__free(pos
);
1630 static size_t disasm_line__fprintf(struct disasm_line
*dl
, FILE *fp
)
1634 if (dl
->offset
== -1)
1635 return fprintf(fp
, "%s\n", dl
->line
);
1637 printed
= fprintf(fp
, "%#" PRIx64
" %s", dl
->offset
, dl
->name
);
1639 if (dl
->ops
.raw
[0] != '\0') {
1640 printed
+= fprintf(fp
, "%.*s %s\n", 6 - (int)printed
, " ",
1644 return printed
+ fprintf(fp
, "\n");
1647 size_t disasm__fprintf(struct list_head
*head
, FILE *fp
)
1649 struct disasm_line
*pos
;
1652 list_for_each_entry(pos
, head
, node
)
1653 printed
+= disasm_line__fprintf(pos
, fp
);
1658 int symbol__tty_annotate(struct symbol
*sym
, struct map
*map
,
1659 struct perf_evsel
*evsel
, bool print_lines
,
1660 bool full_paths
, int min_pcnt
, int max_lines
)
1662 struct dso
*dso
= map
->dso
;
1663 struct rb_root source_line
= RB_ROOT
;
1666 if (symbol__annotate(sym
, map
, 0) < 0)
1669 len
= symbol__size(sym
);
1672 srcline_full_filename
= full_paths
;
1673 symbol__get_source_line(sym
, map
, evsel
, &source_line
, len
);
1674 print_summary(&source_line
, dso
->long_name
);
1677 symbol__annotate_printf(sym
, map
, evsel
, full_paths
,
1678 min_pcnt
, max_lines
, 0);
1680 symbol__free_source_line(sym
, len
);
1682 disasm__purge(&symbol__annotation(sym
)->src
->source
);
1687 bool ui__has_annotation(void)
1689 return use_browser
== 1 && perf_hpp_list
.sym
;