xref: /openbmc/linux/tools/perf/util/annotate.c (revision 82e6fdd6)
1 /*
2  * Copyright (C) 2011, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
3  *
4  * Parts came from builtin-annotate.c, see those files for further
5  * copyright notes.
6  *
7  * Released under the GPL v2. (and only v2, not any later version)
8  */
9 
10 #include <errno.h>
11 #include <inttypes.h>
12 #include "util.h"
13 #include "ui/ui.h"
14 #include "sort.h"
15 #include "build-id.h"
16 #include "color.h"
17 #include "cache.h"
18 #include "symbol.h"
19 #include "debug.h"
20 #include "annotate.h"
21 #include "evsel.h"
22 #include "block-range.h"
23 #include "string2.h"
24 #include "arch/common.h"
25 #include <regex.h>
26 #include <pthread.h>
27 #include <linux/bitops.h>
28 #include <linux/kernel.h>
29 
30 #include "sane_ctype.h"
31 
32 const char 	*disassembler_style;
33 const char	*objdump_path;
34 static regex_t	 file_lineno;
35 
36 static struct ins_ops *ins__find(struct arch *arch, const char *name);
37 static void ins__sort(struct arch *arch);
38 static int disasm_line__parse(char *line, const char **namep, char **rawp);
39 
40 struct arch {
41 	const char	*name;
42 	struct ins	*instructions;
43 	size_t		nr_instructions;
44 	size_t		nr_instructions_allocated;
45 	struct ins_ops  *(*associate_instruction_ops)(struct arch *arch, const char *name);
46 	bool		sorted_instructions;
47 	bool		initialized;
48 	void		*priv;
49 	unsigned int	model;
50 	unsigned int	family;
51 	int		(*init)(struct arch *arch, char *cpuid);
52 	bool		(*ins_is_fused)(struct arch *arch, const char *ins1,
53 					const char *ins2);
54 	struct		{
55 		char comment_char;
56 		char skip_functions_char;
57 	} objdump;
58 };
59 
60 static struct ins_ops call_ops;
61 static struct ins_ops dec_ops;
62 static struct ins_ops jump_ops;
63 static struct ins_ops mov_ops;
64 static struct ins_ops nop_ops;
65 static struct ins_ops lock_ops;
66 static struct ins_ops ret_ops;
67 
68 static int arch__grow_instructions(struct arch *arch)
69 {
70 	struct ins *new_instructions;
71 	size_t new_nr_allocated;
72 
73 	if (arch->nr_instructions_allocated == 0 && arch->instructions)
74 		goto grow_from_non_allocated_table;
75 
76 	new_nr_allocated = arch->nr_instructions_allocated + 128;
77 	new_instructions = realloc(arch->instructions, new_nr_allocated * sizeof(struct ins));
78 	if (new_instructions == NULL)
79 		return -1;
80 
81 out_update_instructions:
82 	arch->instructions = new_instructions;
83 	arch->nr_instructions_allocated = new_nr_allocated;
84 	return 0;
85 
86 grow_from_non_allocated_table:
87 	new_nr_allocated = arch->nr_instructions + 128;
88 	new_instructions = calloc(new_nr_allocated, sizeof(struct ins));
89 	if (new_instructions == NULL)
90 		return -1;
91 
92 	memcpy(new_instructions, arch->instructions, arch->nr_instructions);
93 	goto out_update_instructions;
94 }
95 
96 static int arch__associate_ins_ops(struct arch* arch, const char *name, struct ins_ops *ops)
97 {
98 	struct ins *ins;
99 
100 	if (arch->nr_instructions == arch->nr_instructions_allocated &&
101 	    arch__grow_instructions(arch))
102 		return -1;
103 
104 	ins = &arch->instructions[arch->nr_instructions];
105 	ins->name = strdup(name);
106 	if (!ins->name)
107 		return -1;
108 
109 	ins->ops  = ops;
110 	arch->nr_instructions++;
111 
112 	ins__sort(arch);
113 	return 0;
114 }
115 
116 #include "arch/arm/annotate/instructions.c"
117 #include "arch/arm64/annotate/instructions.c"
118 #include "arch/x86/annotate/instructions.c"
119 #include "arch/powerpc/annotate/instructions.c"
120 #include "arch/s390/annotate/instructions.c"
121 
122 static struct arch architectures[] = {
123 	{
124 		.name = "arm",
125 		.init = arm__annotate_init,
126 	},
127 	{
128 		.name = "arm64",
129 		.init = arm64__annotate_init,
130 	},
131 	{
132 		.name = "x86",
133 		.init = x86__annotate_init,
134 		.instructions = x86__instructions,
135 		.nr_instructions = ARRAY_SIZE(x86__instructions),
136 		.ins_is_fused = x86__ins_is_fused,
137 		.objdump =  {
138 			.comment_char = '#',
139 		},
140 	},
141 	{
142 		.name = "powerpc",
143 		.init = powerpc__annotate_init,
144 	},
145 	{
146 		.name = "s390",
147 		.init = s390__annotate_init,
148 		.objdump =  {
149 			.comment_char = '#',
150 		},
151 	},
152 };
153 
154 static void ins__delete(struct ins_operands *ops)
155 {
156 	if (ops == NULL)
157 		return;
158 	zfree(&ops->source.raw);
159 	zfree(&ops->source.name);
160 	zfree(&ops->target.raw);
161 	zfree(&ops->target.name);
162 }
163 
164 static int ins__raw_scnprintf(struct ins *ins, char *bf, size_t size,
165 			      struct ins_operands *ops)
166 {
167 	return scnprintf(bf, size, "%-6s %s", ins->name, ops->raw);
168 }
169 
170 int ins__scnprintf(struct ins *ins, char *bf, size_t size,
171 		  struct ins_operands *ops)
172 {
173 	if (ins->ops->scnprintf)
174 		return ins->ops->scnprintf(ins, bf, size, ops);
175 
176 	return ins__raw_scnprintf(ins, bf, size, ops);
177 }
178 
179 bool ins__is_fused(struct arch *arch, const char *ins1, const char *ins2)
180 {
181 	if (!arch || !arch->ins_is_fused)
182 		return false;
183 
184 	return arch->ins_is_fused(arch, ins1, ins2);
185 }
186 
187 static int call__parse(struct arch *arch, struct ins_operands *ops, struct map *map)
188 {
189 	char *endptr, *tok, *name;
190 	struct addr_map_symbol target = {
191 		.map = map,
192 	};
193 
194 	ops->target.addr = strtoull(ops->raw, &endptr, 16);
195 
196 	name = strchr(endptr, '<');
197 	if (name == NULL)
198 		goto indirect_call;
199 
200 	name++;
201 
202 	if (arch->objdump.skip_functions_char &&
203 	    strchr(name, arch->objdump.skip_functions_char))
204 		return -1;
205 
206 	tok = strchr(name, '>');
207 	if (tok == NULL)
208 		return -1;
209 
210 	*tok = '\0';
211 	ops->target.name = strdup(name);
212 	*tok = '>';
213 
214 	if (ops->target.name == NULL)
215 		return -1;
216 find_target:
217 	target.addr = map__objdump_2mem(map, ops->target.addr);
218 
219 	if (map_groups__find_ams(&target) == 0 &&
220 	    map__rip_2objdump(target.map, map->map_ip(target.map, target.addr)) == ops->target.addr)
221 		ops->target.sym = target.sym;
222 
223 	return 0;
224 
225 indirect_call:
226 	tok = strchr(endptr, '*');
227 	if (tok != NULL)
228 		ops->target.addr = strtoull(tok + 1, NULL, 16);
229 	goto find_target;
230 }
231 
232 static int call__scnprintf(struct ins *ins, char *bf, size_t size,
233 			   struct ins_operands *ops)
234 {
235 	if (ops->target.sym)
236 		return scnprintf(bf, size, "%-6s %s", ins->name, ops->target.sym->name);
237 
238 	if (ops->target.addr == 0)
239 		return ins__raw_scnprintf(ins, bf, size, ops);
240 
241 	return scnprintf(bf, size, "%-6s *%" PRIx64, ins->name, ops->target.addr);
242 }
243 
244 static struct ins_ops call_ops = {
245 	.parse	   = call__parse,
246 	.scnprintf = call__scnprintf,
247 };
248 
249 bool ins__is_call(const struct ins *ins)
250 {
251 	return ins->ops == &call_ops || ins->ops == &s390_call_ops;
252 }
253 
254 static int jump__parse(struct arch *arch __maybe_unused, struct ins_operands *ops, struct map *map __maybe_unused)
255 {
256 	const char *s = strchr(ops->raw, '+');
257 	const char *c = strchr(ops->raw, ',');
258 
259 	/*
260 	 * skip over possible up to 2 operands to get to address, e.g.:
261 	 * tbnz	 w0, #26, ffff0000083cd190 <security_file_permission+0xd0>
262 	 */
263 	if (c++ != NULL) {
264 		ops->target.addr = strtoull(c, NULL, 16);
265 		if (!ops->target.addr) {
266 			c = strchr(c, ',');
267 			if (c++ != NULL)
268 				ops->target.addr = strtoull(c, NULL, 16);
269 		}
270 	} else {
271 		ops->target.addr = strtoull(ops->raw, NULL, 16);
272 	}
273 
274 	if (s++ != NULL) {
275 		ops->target.offset = strtoull(s, NULL, 16);
276 		ops->target.offset_avail = true;
277 	} else {
278 		ops->target.offset_avail = false;
279 	}
280 
281 	return 0;
282 }
283 
284 static int jump__scnprintf(struct ins *ins, char *bf, size_t size,
285 			   struct ins_operands *ops)
286 {
287 	const char *c = strchr(ops->raw, ',');
288 
289 	if (!ops->target.addr || ops->target.offset < 0)
290 		return ins__raw_scnprintf(ins, bf, size, ops);
291 
292 	if (c != NULL) {
293 		const char *c2 = strchr(c + 1, ',');
294 
295 		/* check for 3-op insn */
296 		if (c2 != NULL)
297 			c = c2;
298 		c++;
299 
300 		/* mirror arch objdump's space-after-comma style */
301 		if (*c == ' ')
302 			c++;
303 	}
304 
305 	return scnprintf(bf, size, "%-6s %.*s%" PRIx64,
306 			 ins->name, c ? c - ops->raw : 0, ops->raw,
307 			 ops->target.offset);
308 }
309 
310 static struct ins_ops jump_ops = {
311 	.parse	   = jump__parse,
312 	.scnprintf = jump__scnprintf,
313 };
314 
315 bool ins__is_jump(const struct ins *ins)
316 {
317 	return ins->ops == &jump_ops;
318 }
319 
320 static int comment__symbol(char *raw, char *comment, u64 *addrp, char **namep)
321 {
322 	char *endptr, *name, *t;
323 
324 	if (strstr(raw, "(%rip)") == NULL)
325 		return 0;
326 
327 	*addrp = strtoull(comment, &endptr, 16);
328 	if (endptr == comment)
329 		return 0;
330 	name = strchr(endptr, '<');
331 	if (name == NULL)
332 		return -1;
333 
334 	name++;
335 
336 	t = strchr(name, '>');
337 	if (t == NULL)
338 		return 0;
339 
340 	*t = '\0';
341 	*namep = strdup(name);
342 	*t = '>';
343 
344 	return 0;
345 }
346 
347 static int lock__parse(struct arch *arch, struct ins_operands *ops, struct map *map)
348 {
349 	ops->locked.ops = zalloc(sizeof(*ops->locked.ops));
350 	if (ops->locked.ops == NULL)
351 		return 0;
352 
353 	if (disasm_line__parse(ops->raw, &ops->locked.ins.name, &ops->locked.ops->raw) < 0)
354 		goto out_free_ops;
355 
356 	ops->locked.ins.ops = ins__find(arch, ops->locked.ins.name);
357 
358 	if (ops->locked.ins.ops == NULL)
359 		goto out_free_ops;
360 
361 	if (ops->locked.ins.ops->parse &&
362 	    ops->locked.ins.ops->parse(arch, ops->locked.ops, map) < 0)
363 		goto out_free_ops;
364 
365 	return 0;
366 
367 out_free_ops:
368 	zfree(&ops->locked.ops);
369 	return 0;
370 }
371 
372 static int lock__scnprintf(struct ins *ins, char *bf, size_t size,
373 			   struct ins_operands *ops)
374 {
375 	int printed;
376 
377 	if (ops->locked.ins.ops == NULL)
378 		return ins__raw_scnprintf(ins, bf, size, ops);
379 
380 	printed = scnprintf(bf, size, "%-6s ", ins->name);
381 	return printed + ins__scnprintf(&ops->locked.ins, bf + printed,
382 					size - printed, ops->locked.ops);
383 }
384 
385 static void lock__delete(struct ins_operands *ops)
386 {
387 	struct ins *ins = &ops->locked.ins;
388 
389 	if (ins->ops && ins->ops->free)
390 		ins->ops->free(ops->locked.ops);
391 	else
392 		ins__delete(ops->locked.ops);
393 
394 	zfree(&ops->locked.ops);
395 	zfree(&ops->target.raw);
396 	zfree(&ops->target.name);
397 }
398 
399 static struct ins_ops lock_ops = {
400 	.free	   = lock__delete,
401 	.parse	   = lock__parse,
402 	.scnprintf = lock__scnprintf,
403 };
404 
405 static int mov__parse(struct arch *arch, struct ins_operands *ops, struct map *map __maybe_unused)
406 {
407 	char *s = strchr(ops->raw, ','), *target, *comment, prev;
408 
409 	if (s == NULL)
410 		return -1;
411 
412 	*s = '\0';
413 	ops->source.raw = strdup(ops->raw);
414 	*s = ',';
415 
416 	if (ops->source.raw == NULL)
417 		return -1;
418 
419 	target = ++s;
420 	comment = strchr(s, arch->objdump.comment_char);
421 
422 	if (comment != NULL)
423 		s = comment - 1;
424 	else
425 		s = strchr(s, '\0') - 1;
426 
427 	while (s > target && isspace(s[0]))
428 		--s;
429 	s++;
430 	prev = *s;
431 	*s = '\0';
432 
433 	ops->target.raw = strdup(target);
434 	*s = prev;
435 
436 	if (ops->target.raw == NULL)
437 		goto out_free_source;
438 
439 	if (comment == NULL)
440 		return 0;
441 
442 	comment = ltrim(comment);
443 	comment__symbol(ops->source.raw, comment + 1, &ops->source.addr, &ops->source.name);
444 	comment__symbol(ops->target.raw, comment + 1, &ops->target.addr, &ops->target.name);
445 
446 	return 0;
447 
448 out_free_source:
449 	zfree(&ops->source.raw);
450 	return -1;
451 }
452 
453 static int mov__scnprintf(struct ins *ins, char *bf, size_t size,
454 			   struct ins_operands *ops)
455 {
456 	return scnprintf(bf, size, "%-6s %s,%s", ins->name,
457 			 ops->source.name ?: ops->source.raw,
458 			 ops->target.name ?: ops->target.raw);
459 }
460 
461 static struct ins_ops mov_ops = {
462 	.parse	   = mov__parse,
463 	.scnprintf = mov__scnprintf,
464 };
465 
466 static int dec__parse(struct arch *arch __maybe_unused, struct ins_operands *ops, struct map *map __maybe_unused)
467 {
468 	char *target, *comment, *s, prev;
469 
470 	target = s = ops->raw;
471 
472 	while (s[0] != '\0' && !isspace(s[0]))
473 		++s;
474 	prev = *s;
475 	*s = '\0';
476 
477 	ops->target.raw = strdup(target);
478 	*s = prev;
479 
480 	if (ops->target.raw == NULL)
481 		return -1;
482 
483 	comment = strchr(s, arch->objdump.comment_char);
484 	if (comment == NULL)
485 		return 0;
486 
487 	comment = ltrim(comment);
488 	comment__symbol(ops->target.raw, comment + 1, &ops->target.addr, &ops->target.name);
489 
490 	return 0;
491 }
492 
493 static int dec__scnprintf(struct ins *ins, char *bf, size_t size,
494 			   struct ins_operands *ops)
495 {
496 	return scnprintf(bf, size, "%-6s %s", ins->name,
497 			 ops->target.name ?: ops->target.raw);
498 }
499 
500 static struct ins_ops dec_ops = {
501 	.parse	   = dec__parse,
502 	.scnprintf = dec__scnprintf,
503 };
504 
505 static int nop__scnprintf(struct ins *ins __maybe_unused, char *bf, size_t size,
506 			  struct ins_operands *ops __maybe_unused)
507 {
508 	return scnprintf(bf, size, "%-6s", "nop");
509 }
510 
511 static struct ins_ops nop_ops = {
512 	.scnprintf = nop__scnprintf,
513 };
514 
515 static struct ins_ops ret_ops = {
516 	.scnprintf = ins__raw_scnprintf,
517 };
518 
519 bool ins__is_ret(const struct ins *ins)
520 {
521 	return ins->ops == &ret_ops;
522 }
523 
524 bool ins__is_lock(const struct ins *ins)
525 {
526 	return ins->ops == &lock_ops;
527 }
528 
529 static int ins__key_cmp(const void *name, const void *insp)
530 {
531 	const struct ins *ins = insp;
532 
533 	return strcmp(name, ins->name);
534 }
535 
536 static int ins__cmp(const void *a, const void *b)
537 {
538 	const struct ins *ia = a;
539 	const struct ins *ib = b;
540 
541 	return strcmp(ia->name, ib->name);
542 }
543 
544 static void ins__sort(struct arch *arch)
545 {
546 	const int nmemb = arch->nr_instructions;
547 
548 	qsort(arch->instructions, nmemb, sizeof(struct ins), ins__cmp);
549 }
550 
551 static struct ins_ops *__ins__find(struct arch *arch, const char *name)
552 {
553 	struct ins *ins;
554 	const int nmemb = arch->nr_instructions;
555 
556 	if (!arch->sorted_instructions) {
557 		ins__sort(arch);
558 		arch->sorted_instructions = true;
559 	}
560 
561 	ins = bsearch(name, arch->instructions, nmemb, sizeof(struct ins), ins__key_cmp);
562 	return ins ? ins->ops : NULL;
563 }
564 
565 static struct ins_ops *ins__find(struct arch *arch, const char *name)
566 {
567 	struct ins_ops *ops = __ins__find(arch, name);
568 
569 	if (!ops && arch->associate_instruction_ops)
570 		ops = arch->associate_instruction_ops(arch, name);
571 
572 	return ops;
573 }
574 
575 static int arch__key_cmp(const void *name, const void *archp)
576 {
577 	const struct arch *arch = archp;
578 
579 	return strcmp(name, arch->name);
580 }
581 
582 static int arch__cmp(const void *a, const void *b)
583 {
584 	const struct arch *aa = a;
585 	const struct arch *ab = b;
586 
587 	return strcmp(aa->name, ab->name);
588 }
589 
590 static void arch__sort(void)
591 {
592 	const int nmemb = ARRAY_SIZE(architectures);
593 
594 	qsort(architectures, nmemb, sizeof(struct arch), arch__cmp);
595 }
596 
597 static struct arch *arch__find(const char *name)
598 {
599 	const int nmemb = ARRAY_SIZE(architectures);
600 	static bool sorted;
601 
602 	if (!sorted) {
603 		arch__sort();
604 		sorted = true;
605 	}
606 
607 	return bsearch(name, architectures, nmemb, sizeof(struct arch), arch__key_cmp);
608 }
609 
610 int symbol__alloc_hist(struct symbol *sym)
611 {
612 	struct annotation *notes = symbol__annotation(sym);
613 	size_t size = symbol__size(sym);
614 	size_t sizeof_sym_hist;
615 
616 	/*
617 	 * Add buffer of one element for zero length symbol.
618 	 * When sample is taken from first instruction of
619 	 * zero length symbol, perf still resolves it and
620 	 * shows symbol name in perf report and allows to
621 	 * annotate it.
622 	 */
623 	if (size == 0)
624 		size = 1;
625 
626 	/* Check for overflow when calculating sizeof_sym_hist */
627 	if (size > (SIZE_MAX - sizeof(struct sym_hist)) / sizeof(struct sym_hist_entry))
628 		return -1;
629 
630 	sizeof_sym_hist = (sizeof(struct sym_hist) + size * sizeof(struct sym_hist_entry));
631 
632 	/* Check for overflow in zalloc argument */
633 	if (sizeof_sym_hist > (SIZE_MAX - sizeof(*notes->src))
634 				/ symbol_conf.nr_events)
635 		return -1;
636 
637 	notes->src = zalloc(sizeof(*notes->src) + symbol_conf.nr_events * sizeof_sym_hist);
638 	if (notes->src == NULL)
639 		return -1;
640 	notes->src->sizeof_sym_hist = sizeof_sym_hist;
641 	notes->src->nr_histograms   = symbol_conf.nr_events;
642 	INIT_LIST_HEAD(&notes->src->source);
643 	return 0;
644 }
645 
646 /* The cycles histogram is lazily allocated. */
647 static int symbol__alloc_hist_cycles(struct symbol *sym)
648 {
649 	struct annotation *notes = symbol__annotation(sym);
650 	const size_t size = symbol__size(sym);
651 
652 	notes->src->cycles_hist = calloc(size, sizeof(struct cyc_hist));
653 	if (notes->src->cycles_hist == NULL)
654 		return -1;
655 	return 0;
656 }
657 
658 void symbol__annotate_zero_histograms(struct symbol *sym)
659 {
660 	struct annotation *notes = symbol__annotation(sym);
661 
662 	pthread_mutex_lock(&notes->lock);
663 	if (notes->src != NULL) {
664 		memset(notes->src->histograms, 0,
665 		       notes->src->nr_histograms * notes->src->sizeof_sym_hist);
666 		if (notes->src->cycles_hist)
667 			memset(notes->src->cycles_hist, 0,
668 				symbol__size(sym) * sizeof(struct cyc_hist));
669 	}
670 	pthread_mutex_unlock(&notes->lock);
671 }
672 
673 static int __symbol__account_cycles(struct annotation *notes,
674 				    u64 start,
675 				    unsigned offset, unsigned cycles,
676 				    unsigned have_start)
677 {
678 	struct cyc_hist *ch;
679 
680 	ch = notes->src->cycles_hist;
681 	/*
682 	 * For now we can only account one basic block per
683 	 * final jump. But multiple could be overlapping.
684 	 * Always account the longest one. So when
685 	 * a shorter one has been already seen throw it away.
686 	 *
687 	 * We separately always account the full cycles.
688 	 */
689 	ch[offset].num_aggr++;
690 	ch[offset].cycles_aggr += cycles;
691 
692 	if (!have_start && ch[offset].have_start)
693 		return 0;
694 	if (ch[offset].num) {
695 		if (have_start && (!ch[offset].have_start ||
696 				   ch[offset].start > start)) {
697 			ch[offset].have_start = 0;
698 			ch[offset].cycles = 0;
699 			ch[offset].num = 0;
700 			if (ch[offset].reset < 0xffff)
701 				ch[offset].reset++;
702 		} else if (have_start &&
703 			   ch[offset].start < start)
704 			return 0;
705 	}
706 	ch[offset].have_start = have_start;
707 	ch[offset].start = start;
708 	ch[offset].cycles += cycles;
709 	ch[offset].num++;
710 	return 0;
711 }
712 
713 static int __symbol__inc_addr_samples(struct symbol *sym, struct map *map,
714 				      struct annotation *notes, int evidx, u64 addr,
715 				      struct perf_sample *sample)
716 {
717 	unsigned offset;
718 	struct sym_hist *h;
719 
720 	pr_debug3("%s: addr=%#" PRIx64 "\n", __func__, map->unmap_ip(map, addr));
721 
722 	if ((addr < sym->start || addr >= sym->end) &&
723 	    (addr != sym->end || sym->start != sym->end)) {
724 		pr_debug("%s(%d): ERANGE! sym->name=%s, start=%#" PRIx64 ", addr=%#" PRIx64 ", end=%#" PRIx64 "\n",
725 		       __func__, __LINE__, sym->name, sym->start, addr, sym->end);
726 		return -ERANGE;
727 	}
728 
729 	offset = addr - sym->start;
730 	h = annotation__histogram(notes, evidx);
731 	h->nr_samples++;
732 	h->addr[offset].nr_samples++;
733 	h->period += sample->period;
734 	h->addr[offset].period += sample->period;
735 
736 	pr_debug3("%#" PRIx64 " %s: period++ [addr: %#" PRIx64 ", %#" PRIx64
737 		  ", evidx=%d] => nr_samples: %" PRIu64 ", period: %" PRIu64 "\n",
738 		  sym->start, sym->name, addr, addr - sym->start, evidx,
739 		  h->addr[offset].nr_samples, h->addr[offset].period);
740 	return 0;
741 }
742 
743 static struct annotation *symbol__get_annotation(struct symbol *sym, bool cycles)
744 {
745 	struct annotation *notes = symbol__annotation(sym);
746 
747 	if (notes->src == NULL) {
748 		if (symbol__alloc_hist(sym) < 0)
749 			return NULL;
750 	}
751 	if (!notes->src->cycles_hist && cycles) {
752 		if (symbol__alloc_hist_cycles(sym) < 0)
753 			return NULL;
754 	}
755 	return notes;
756 }
757 
758 static int symbol__inc_addr_samples(struct symbol *sym, struct map *map,
759 				    int evidx, u64 addr,
760 				    struct perf_sample *sample)
761 {
762 	struct annotation *notes;
763 
764 	if (sym == NULL)
765 		return 0;
766 	notes = symbol__get_annotation(sym, false);
767 	if (notes == NULL)
768 		return -ENOMEM;
769 	return __symbol__inc_addr_samples(sym, map, notes, evidx, addr, sample);
770 }
771 
772 static int symbol__account_cycles(u64 addr, u64 start,
773 				  struct symbol *sym, unsigned cycles)
774 {
775 	struct annotation *notes;
776 	unsigned offset;
777 
778 	if (sym == NULL)
779 		return 0;
780 	notes = symbol__get_annotation(sym, true);
781 	if (notes == NULL)
782 		return -ENOMEM;
783 	if (addr < sym->start || addr >= sym->end)
784 		return -ERANGE;
785 
786 	if (start) {
787 		if (start < sym->start || start >= sym->end)
788 			return -ERANGE;
789 		if (start >= addr)
790 			start = 0;
791 	}
792 	offset = addr - sym->start;
793 	return __symbol__account_cycles(notes,
794 					start ? start - sym->start : 0,
795 					offset, cycles,
796 					!!start);
797 }
798 
799 int addr_map_symbol__account_cycles(struct addr_map_symbol *ams,
800 				    struct addr_map_symbol *start,
801 				    unsigned cycles)
802 {
803 	u64 saddr = 0;
804 	int err;
805 
806 	if (!cycles)
807 		return 0;
808 
809 	/*
810 	 * Only set start when IPC can be computed. We can only
811 	 * compute it when the basic block is completely in a single
812 	 * function.
813 	 * Special case the case when the jump is elsewhere, but
814 	 * it starts on the function start.
815 	 */
816 	if (start &&
817 		(start->sym == ams->sym ||
818 		 (ams->sym &&
819 		   start->addr == ams->sym->start + ams->map->start)))
820 		saddr = start->al_addr;
821 	if (saddr == 0)
822 		pr_debug2("BB with bad start: addr %"PRIx64" start %"PRIx64" sym %"PRIx64" saddr %"PRIx64"\n",
823 			ams->addr,
824 			start ? start->addr : 0,
825 			ams->sym ? ams->sym->start + ams->map->start : 0,
826 			saddr);
827 	err = symbol__account_cycles(ams->al_addr, saddr, ams->sym, cycles);
828 	if (err)
829 		pr_debug2("account_cycles failed %d\n", err);
830 	return err;
831 }
832 
833 int addr_map_symbol__inc_samples(struct addr_map_symbol *ams, struct perf_sample *sample,
834 				 int evidx)
835 {
836 	return symbol__inc_addr_samples(ams->sym, ams->map, evidx, ams->al_addr, sample);
837 }
838 
839 int hist_entry__inc_addr_samples(struct hist_entry *he, struct perf_sample *sample,
840 				 int evidx, u64 ip)
841 {
842 	return symbol__inc_addr_samples(he->ms.sym, he->ms.map, evidx, ip, sample);
843 }
844 
845 static void disasm_line__init_ins(struct disasm_line *dl, struct arch *arch, struct map *map)
846 {
847 	dl->ins.ops = ins__find(arch, dl->ins.name);
848 
849 	if (!dl->ins.ops)
850 		return;
851 
852 	if (dl->ins.ops->parse && dl->ins.ops->parse(arch, &dl->ops, map) < 0)
853 		dl->ins.ops = NULL;
854 }
855 
856 static int disasm_line__parse(char *line, const char **namep, char **rawp)
857 {
858 	char tmp, *name = ltrim(line);
859 
860 	if (name[0] == '\0')
861 		return -1;
862 
863 	*rawp = name + 1;
864 
865 	while ((*rawp)[0] != '\0' && !isspace((*rawp)[0]))
866 		++*rawp;
867 
868 	tmp = (*rawp)[0];
869 	(*rawp)[0] = '\0';
870 	*namep = strdup(name);
871 
872 	if (*namep == NULL)
873 		goto out_free_name;
874 
875 	(*rawp)[0] = tmp;
876 	*rawp = ltrim(*rawp);
877 
878 	return 0;
879 
880 out_free_name:
881 	free((void *)namep);
882 	*namep = NULL;
883 	return -1;
884 }
885 
886 struct annotate_args {
887 	size_t			 privsize;
888 	struct arch		*arch;
889 	struct map		*map;
890 	struct perf_evsel	*evsel;
891 	s64			 offset;
892 	char			*line;
893 	int			 line_nr;
894 };
895 
896 static void annotation_line__delete(struct annotation_line *al)
897 {
898 	void *ptr = (void *) al - al->privsize;
899 
900 	free_srcline(al->path);
901 	zfree(&al->line);
902 	free(ptr);
903 }
904 
905 /*
906  * Allocating the annotation line data with following
907  * structure:
908  *
909  *    --------------------------------------
910  *    private space | struct annotation_line
911  *    --------------------------------------
912  *
913  * Size of the private space is stored in 'struct annotation_line'.
914  *
915  */
916 static struct annotation_line *
917 annotation_line__new(struct annotate_args *args, size_t privsize)
918 {
919 	struct annotation_line *al;
920 	struct perf_evsel *evsel = args->evsel;
921 	size_t size = privsize + sizeof(*al);
922 	int nr = 1;
923 
924 	if (perf_evsel__is_group_event(evsel))
925 		nr = evsel->nr_members;
926 
927 	size += sizeof(al->samples[0]) * nr;
928 
929 	al = zalloc(size);
930 	if (al) {
931 		al = (void *) al + privsize;
932 		al->privsize   = privsize;
933 		al->offset     = args->offset;
934 		al->line       = strdup(args->line);
935 		al->line_nr    = args->line_nr;
936 		al->samples_nr = nr;
937 	}
938 
939 	return al;
940 }
941 
942 /*
943  * Allocating the disasm annotation line data with
944  * following structure:
945  *
946  *    ------------------------------------------------------------
947  *    privsize space | struct disasm_line | struct annotation_line
948  *    ------------------------------------------------------------
949  *
950  * We have 'struct annotation_line' member as last member
951  * of 'struct disasm_line' to have an easy access.
952  *
953  */
954 static struct disasm_line *disasm_line__new(struct annotate_args *args)
955 {
956 	struct disasm_line *dl = NULL;
957 	struct annotation_line *al;
958 	size_t privsize = args->privsize + offsetof(struct disasm_line, al);
959 
960 	al = annotation_line__new(args, privsize);
961 	if (al != NULL) {
962 		dl = disasm_line(al);
963 
964 		if (dl->al.line == NULL)
965 			goto out_delete;
966 
967 		if (args->offset != -1) {
968 			if (disasm_line__parse(dl->al.line, &dl->ins.name, &dl->ops.raw) < 0)
969 				goto out_free_line;
970 
971 			disasm_line__init_ins(dl, args->arch, args->map);
972 		}
973 	}
974 
975 	return dl;
976 
977 out_free_line:
978 	zfree(&dl->al.line);
979 out_delete:
980 	free(dl);
981 	return NULL;
982 }
983 
984 void disasm_line__free(struct disasm_line *dl)
985 {
986 	if (dl->ins.ops && dl->ins.ops->free)
987 		dl->ins.ops->free(&dl->ops);
988 	else
989 		ins__delete(&dl->ops);
990 	free((void *)dl->ins.name);
991 	dl->ins.name = NULL;
992 	annotation_line__delete(&dl->al);
993 }
994 
995 int disasm_line__scnprintf(struct disasm_line *dl, char *bf, size_t size, bool raw)
996 {
997 	if (raw || !dl->ins.ops)
998 		return scnprintf(bf, size, "%-6s %s", dl->ins.name, dl->ops.raw);
999 
1000 	return ins__scnprintf(&dl->ins, bf, size, &dl->ops);
1001 }
1002 
1003 static void annotation_line__add(struct annotation_line *al, struct list_head *head)
1004 {
1005 	list_add_tail(&al->node, head);
1006 }
1007 
1008 struct annotation_line *
1009 annotation_line__next(struct annotation_line *pos, struct list_head *head)
1010 {
1011 	list_for_each_entry_continue(pos, head, node)
1012 		if (pos->offset >= 0)
1013 			return pos;
1014 
1015 	return NULL;
1016 }
1017 
1018 static const char *annotate__address_color(struct block_range *br)
1019 {
1020 	double cov = block_range__coverage(br);
1021 
1022 	if (cov >= 0) {
1023 		/* mark red for >75% coverage */
1024 		if (cov > 0.75)
1025 			return PERF_COLOR_RED;
1026 
1027 		/* mark dull for <1% coverage */
1028 		if (cov < 0.01)
1029 			return PERF_COLOR_NORMAL;
1030 	}
1031 
1032 	return PERF_COLOR_MAGENTA;
1033 }
1034 
1035 static const char *annotate__asm_color(struct block_range *br)
1036 {
1037 	double cov = block_range__coverage(br);
1038 
1039 	if (cov >= 0) {
1040 		/* mark dull for <1% coverage */
1041 		if (cov < 0.01)
1042 			return PERF_COLOR_NORMAL;
1043 	}
1044 
1045 	return PERF_COLOR_BLUE;
1046 }
1047 
1048 static void annotate__branch_printf(struct block_range *br, u64 addr)
1049 {
1050 	bool emit_comment = true;
1051 
1052 	if (!br)
1053 		return;
1054 
1055 #if 1
1056 	if (br->is_target && br->start == addr) {
1057 		struct block_range *branch = br;
1058 		double p;
1059 
1060 		/*
1061 		 * Find matching branch to our target.
1062 		 */
1063 		while (!branch->is_branch)
1064 			branch = block_range__next(branch);
1065 
1066 		p = 100 *(double)br->entry / branch->coverage;
1067 
1068 		if (p > 0.1) {
1069 			if (emit_comment) {
1070 				emit_comment = false;
1071 				printf("\t#");
1072 			}
1073 
1074 			/*
1075 			 * The percentage of coverage joined at this target in relation
1076 			 * to the next branch.
1077 			 */
1078 			printf(" +%.2f%%", p);
1079 		}
1080 	}
1081 #endif
1082 	if (br->is_branch && br->end == addr) {
1083 		double p = 100*(double)br->taken / br->coverage;
1084 
1085 		if (p > 0.1) {
1086 			if (emit_comment) {
1087 				emit_comment = false;
1088 				printf("\t#");
1089 			}
1090 
1091 			/*
1092 			 * The percentage of coverage leaving at this branch, and
1093 			 * its prediction ratio.
1094 			 */
1095 			printf(" -%.2f%% (p:%.2f%%)", p, 100*(double)br->pred  / br->taken);
1096 		}
1097 	}
1098 }
1099 
1100 static int disasm_line__print(struct disasm_line *dl, u64 start, int addr_fmt_width)
1101 {
1102 	s64 offset = dl->al.offset;
1103 	const u64 addr = start + offset;
1104 	struct block_range *br;
1105 
1106 	br = block_range__find(addr);
1107 	color_fprintf(stdout, annotate__address_color(br), "  %*" PRIx64 ":", addr_fmt_width, addr);
1108 	color_fprintf(stdout, annotate__asm_color(br), "%s", dl->al.line);
1109 	annotate__branch_printf(br, addr);
1110 	return 0;
1111 }
1112 
1113 static int
1114 annotation_line__print(struct annotation_line *al, struct symbol *sym, u64 start,
1115 		       struct perf_evsel *evsel, u64 len, int min_pcnt, int printed,
1116 		       int max_lines, struct annotation_line *queue, int addr_fmt_width)
1117 {
1118 	struct disasm_line *dl = container_of(al, struct disasm_line, al);
1119 	static const char *prev_line;
1120 	static const char *prev_color;
1121 
1122 	if (al->offset != -1) {
1123 		double max_percent = 0.0;
1124 		int i, nr_percent = 1;
1125 		const char *color;
1126 		struct annotation *notes = symbol__annotation(sym);
1127 
1128 		for (i = 0; i < al->samples_nr; i++) {
1129 			struct annotation_data *sample = &al->samples[i];
1130 
1131 			if (sample->percent > max_percent)
1132 				max_percent = sample->percent;
1133 		}
1134 
1135 		if (max_percent < min_pcnt)
1136 			return -1;
1137 
1138 		if (max_lines && printed >= max_lines)
1139 			return 1;
1140 
1141 		if (queue != NULL) {
1142 			list_for_each_entry_from(queue, &notes->src->source, node) {
1143 				if (queue == al)
1144 					break;
1145 				annotation_line__print(queue, sym, start, evsel, len,
1146 						       0, 0, 1, NULL, addr_fmt_width);
1147 			}
1148 		}
1149 
1150 		color = get_percent_color(max_percent);
1151 
1152 		/*
1153 		 * Also color the filename and line if needed, with
1154 		 * the same color than the percentage. Don't print it
1155 		 * twice for close colored addr with the same filename:line
1156 		 */
1157 		if (al->path) {
1158 			if (!prev_line || strcmp(prev_line, al->path)
1159 				       || color != prev_color) {
1160 				color_fprintf(stdout, color, " %s", al->path);
1161 				prev_line = al->path;
1162 				prev_color = color;
1163 			}
1164 		}
1165 
1166 		for (i = 0; i < nr_percent; i++) {
1167 			struct annotation_data *sample = &al->samples[i];
1168 
1169 			color = get_percent_color(sample->percent);
1170 
1171 			if (symbol_conf.show_total_period)
1172 				color_fprintf(stdout, color, " %11" PRIu64,
1173 					      sample->he.period);
1174 			else if (symbol_conf.show_nr_samples)
1175 				color_fprintf(stdout, color, " %7" PRIu64,
1176 					      sample->he.nr_samples);
1177 			else
1178 				color_fprintf(stdout, color, " %7.2f", sample->percent);
1179 		}
1180 
1181 		printf(" : ");
1182 
1183 		disasm_line__print(dl, start, addr_fmt_width);
1184 		printf("\n");
1185 	} else if (max_lines && printed >= max_lines)
1186 		return 1;
1187 	else {
1188 		int width = symbol_conf.show_total_period ? 12 : 8;
1189 
1190 		if (queue)
1191 			return -1;
1192 
1193 		if (perf_evsel__is_group_event(evsel))
1194 			width *= evsel->nr_members;
1195 
1196 		if (!*al->line)
1197 			printf(" %*s:\n", width, " ");
1198 		else
1199 			printf(" %*s:     %*s %s\n", width, " ", addr_fmt_width, " ", al->line);
1200 	}
1201 
1202 	return 0;
1203 }
1204 
1205 /*
1206  * symbol__parse_objdump_line() parses objdump output (with -d --no-show-raw)
1207  * which looks like following
1208  *
1209  *  0000000000415500 <_init>:
1210  *    415500:       sub    $0x8,%rsp
1211  *    415504:       mov    0x2f5ad5(%rip),%rax        # 70afe0 <_DYNAMIC+0x2f8>
1212  *    41550b:       test   %rax,%rax
1213  *    41550e:       je     415515 <_init+0x15>
1214  *    415510:       callq  416e70 <__gmon_start__@plt>
1215  *    415515:       add    $0x8,%rsp
1216  *    415519:       retq
1217  *
1218  * it will be parsed and saved into struct disasm_line as
1219  *  <offset>       <name>  <ops.raw>
1220  *
1221  * The offset will be a relative offset from the start of the symbol and -1
1222  * means that it's not a disassembly line so should be treated differently.
1223  * The ops.raw part will be parsed further according to type of the instruction.
1224  */
1225 static int symbol__parse_objdump_line(struct symbol *sym, FILE *file,
1226 				      struct annotate_args *args,
1227 				      int *line_nr)
1228 {
1229 	struct map *map = args->map;
1230 	struct annotation *notes = symbol__annotation(sym);
1231 	struct disasm_line *dl;
1232 	char *line = NULL, *parsed_line, *tmp, *tmp2;
1233 	size_t line_len;
1234 	s64 line_ip, offset = -1;
1235 	regmatch_t match[2];
1236 
1237 	if (getline(&line, &line_len, file) < 0)
1238 		return -1;
1239 
1240 	if (!line)
1241 		return -1;
1242 
1243 	line_ip = -1;
1244 	parsed_line = rtrim(line);
1245 
1246 	/* /filename:linenr ? Save line number and ignore. */
1247 	if (regexec(&file_lineno, parsed_line, 2, match, 0) == 0) {
1248 		*line_nr = atoi(parsed_line + match[1].rm_so);
1249 		return 0;
1250 	}
1251 
1252 	tmp = ltrim(parsed_line);
1253 	if (*tmp) {
1254 		/*
1255 		 * Parse hexa addresses followed by ':'
1256 		 */
1257 		line_ip = strtoull(tmp, &tmp2, 16);
1258 		if (*tmp2 != ':' || tmp == tmp2 || tmp2[1] == '\0')
1259 			line_ip = -1;
1260 	}
1261 
1262 	if (line_ip != -1) {
1263 		u64 start = map__rip_2objdump(map, sym->start),
1264 		    end = map__rip_2objdump(map, sym->end);
1265 
1266 		offset = line_ip - start;
1267 		if ((u64)line_ip < start || (u64)line_ip >= end)
1268 			offset = -1;
1269 		else
1270 			parsed_line = tmp2 + 1;
1271 	}
1272 
1273 	args->offset  = offset;
1274 	args->line    = parsed_line;
1275 	args->line_nr = *line_nr;
1276 
1277 	dl = disasm_line__new(args);
1278 	free(line);
1279 	(*line_nr)++;
1280 
1281 	if (dl == NULL)
1282 		return -1;
1283 
1284 	if (!disasm_line__has_offset(dl)) {
1285 		dl->ops.target.offset = dl->ops.target.addr -
1286 					map__rip_2objdump(map, sym->start);
1287 		dl->ops.target.offset_avail = true;
1288 	}
1289 
1290 	/* kcore has no symbols, so add the call target symbol */
1291 	if (dl->ins.ops && ins__is_call(&dl->ins) && !dl->ops.target.sym) {
1292 		struct addr_map_symbol target = {
1293 			.map = map,
1294 			.addr = dl->ops.target.addr,
1295 		};
1296 
1297 		if (!map_groups__find_ams(&target) &&
1298 		    target.sym->start == target.al_addr)
1299 			dl->ops.target.sym = target.sym;
1300 	}
1301 
1302 	annotation_line__add(&dl->al, &notes->src->source);
1303 
1304 	return 0;
1305 }
1306 
1307 static __attribute__((constructor)) void symbol__init_regexpr(void)
1308 {
1309 	regcomp(&file_lineno, "^/[^:]+:([0-9]+)", REG_EXTENDED);
1310 }
1311 
1312 static void delete_last_nop(struct symbol *sym)
1313 {
1314 	struct annotation *notes = symbol__annotation(sym);
1315 	struct list_head *list = &notes->src->source;
1316 	struct disasm_line *dl;
1317 
1318 	while (!list_empty(list)) {
1319 		dl = list_entry(list->prev, struct disasm_line, al.node);
1320 
1321 		if (dl->ins.ops) {
1322 			if (dl->ins.ops != &nop_ops)
1323 				return;
1324 		} else {
1325 			if (!strstr(dl->al.line, " nop ") &&
1326 			    !strstr(dl->al.line, " nopl ") &&
1327 			    !strstr(dl->al.line, " nopw "))
1328 				return;
1329 		}
1330 
1331 		list_del(&dl->al.node);
1332 		disasm_line__free(dl);
1333 	}
1334 }
1335 
1336 int symbol__strerror_disassemble(struct symbol *sym __maybe_unused, struct map *map,
1337 			      int errnum, char *buf, size_t buflen)
1338 {
1339 	struct dso *dso = map->dso;
1340 
1341 	BUG_ON(buflen == 0);
1342 
1343 	if (errnum >= 0) {
1344 		str_error_r(errnum, buf, buflen);
1345 		return 0;
1346 	}
1347 
1348 	switch (errnum) {
1349 	case SYMBOL_ANNOTATE_ERRNO__NO_VMLINUX: {
1350 		char bf[SBUILD_ID_SIZE + 15] = " with build id ";
1351 		char *build_id_msg = NULL;
1352 
1353 		if (dso->has_build_id) {
1354 			build_id__sprintf(dso->build_id,
1355 					  sizeof(dso->build_id), bf + 15);
1356 			build_id_msg = bf;
1357 		}
1358 		scnprintf(buf, buflen,
1359 			  "No vmlinux file%s\nwas found in the path.\n\n"
1360 			  "Note that annotation using /proc/kcore requires CAP_SYS_RAWIO capability.\n\n"
1361 			  "Please use:\n\n"
1362 			  "  perf buildid-cache -vu vmlinux\n\n"
1363 			  "or:\n\n"
1364 			  "  --vmlinux vmlinux\n", build_id_msg ?: "");
1365 	}
1366 		break;
1367 	default:
1368 		scnprintf(buf, buflen, "Internal error: Invalid %d error code\n", errnum);
1369 		break;
1370 	}
1371 
1372 	return 0;
1373 }
1374 
1375 static int dso__disassemble_filename(struct dso *dso, char *filename, size_t filename_size)
1376 {
1377 	char linkname[PATH_MAX];
1378 	char *build_id_filename;
1379 	char *build_id_path = NULL;
1380 	char *pos;
1381 
1382 	if (dso->symtab_type == DSO_BINARY_TYPE__KALLSYMS &&
1383 	    !dso__is_kcore(dso))
1384 		return SYMBOL_ANNOTATE_ERRNO__NO_VMLINUX;
1385 
1386 	build_id_filename = dso__build_id_filename(dso, NULL, 0, false);
1387 	if (build_id_filename) {
1388 		__symbol__join_symfs(filename, filename_size, build_id_filename);
1389 		free(build_id_filename);
1390 	} else {
1391 		if (dso->has_build_id)
1392 			return ENOMEM;
1393 		goto fallback;
1394 	}
1395 
1396 	build_id_path = strdup(filename);
1397 	if (!build_id_path)
1398 		return -1;
1399 
1400 	/*
1401 	 * old style build-id cache has name of XX/XXXXXXX.. while
1402 	 * new style has XX/XXXXXXX../{elf,kallsyms,vdso}.
1403 	 * extract the build-id part of dirname in the new style only.
1404 	 */
1405 	pos = strrchr(build_id_path, '/');
1406 	if (pos && strlen(pos) < SBUILD_ID_SIZE - 2)
1407 		dirname(build_id_path);
1408 
1409 	if (dso__is_kcore(dso) ||
1410 	    readlink(build_id_path, linkname, sizeof(linkname)) < 0 ||
1411 	    strstr(linkname, DSO__NAME_KALLSYMS) ||
1412 	    access(filename, R_OK)) {
1413 fallback:
1414 		/*
1415 		 * If we don't have build-ids or the build-id file isn't in the
1416 		 * cache, or is just a kallsyms file, well, lets hope that this
1417 		 * DSO is the same as when 'perf record' ran.
1418 		 */
1419 		__symbol__join_symfs(filename, filename_size, dso->long_name);
1420 	}
1421 
1422 	free(build_id_path);
1423 	return 0;
1424 }
1425 
1426 static int symbol__disassemble(struct symbol *sym, struct annotate_args *args)
1427 {
1428 	struct map *map = args->map;
1429 	struct dso *dso = map->dso;
1430 	char command[PATH_MAX * 2];
1431 	FILE *file;
1432 	char symfs_filename[PATH_MAX];
1433 	struct kcore_extract kce;
1434 	bool delete_extract = false;
1435 	int stdout_fd[2];
1436 	int lineno = 0;
1437 	int nline;
1438 	pid_t pid;
1439 	int err = dso__disassemble_filename(dso, symfs_filename, sizeof(symfs_filename));
1440 
1441 	if (err)
1442 		return err;
1443 
1444 	pr_debug("%s: filename=%s, sym=%s, start=%#" PRIx64 ", end=%#" PRIx64 "\n", __func__,
1445 		 symfs_filename, sym->name, map->unmap_ip(map, sym->start),
1446 		 map->unmap_ip(map, sym->end));
1447 
1448 	pr_debug("annotating [%p] %30s : [%p] %30s\n",
1449 		 dso, dso->long_name, sym, sym->name);
1450 
1451 	if (dso__is_kcore(dso)) {
1452 		kce.kcore_filename = symfs_filename;
1453 		kce.addr = map__rip_2objdump(map, sym->start);
1454 		kce.offs = sym->start;
1455 		kce.len = sym->end - sym->start;
1456 		if (!kcore_extract__create(&kce)) {
1457 			delete_extract = true;
1458 			strlcpy(symfs_filename, kce.extract_filename,
1459 				sizeof(symfs_filename));
1460 		}
1461 	} else if (dso__needs_decompress(dso)) {
1462 		char tmp[KMOD_DECOMP_LEN];
1463 
1464 		if (dso__decompress_kmodule_path(dso, symfs_filename,
1465 						 tmp, sizeof(tmp)) < 0)
1466 			goto out;
1467 
1468 		strcpy(symfs_filename, tmp);
1469 	}
1470 
1471 	snprintf(command, sizeof(command),
1472 		 "%s %s%s --start-address=0x%016" PRIx64
1473 		 " --stop-address=0x%016" PRIx64
1474 		 " -l -d %s %s -C \"%s\" 2>/dev/null|grep -v \"%s:\"|expand",
1475 		 objdump_path ? objdump_path : "objdump",
1476 		 disassembler_style ? "-M " : "",
1477 		 disassembler_style ? disassembler_style : "",
1478 		 map__rip_2objdump(map, sym->start),
1479 		 map__rip_2objdump(map, sym->end),
1480 		 symbol_conf.annotate_asm_raw ? "" : "--no-show-raw",
1481 		 symbol_conf.annotate_src ? "-S" : "",
1482 		 symfs_filename, symfs_filename);
1483 
1484 	pr_debug("Executing: %s\n", command);
1485 
1486 	err = -1;
1487 	if (pipe(stdout_fd) < 0) {
1488 		pr_err("Failure creating the pipe to run %s\n", command);
1489 		goto out_remove_tmp;
1490 	}
1491 
1492 	pid = fork();
1493 	if (pid < 0) {
1494 		pr_err("Failure forking to run %s\n", command);
1495 		goto out_close_stdout;
1496 	}
1497 
1498 	if (pid == 0) {
1499 		close(stdout_fd[0]);
1500 		dup2(stdout_fd[1], 1);
1501 		close(stdout_fd[1]);
1502 		execl("/bin/sh", "sh", "-c", command, NULL);
1503 		perror(command);
1504 		exit(-1);
1505 	}
1506 
1507 	close(stdout_fd[1]);
1508 
1509 	file = fdopen(stdout_fd[0], "r");
1510 	if (!file) {
1511 		pr_err("Failure creating FILE stream for %s\n", command);
1512 		/*
1513 		 * If we were using debug info should retry with
1514 		 * original binary.
1515 		 */
1516 		goto out_remove_tmp;
1517 	}
1518 
1519 	nline = 0;
1520 	while (!feof(file)) {
1521 		/*
1522 		 * The source code line number (lineno) needs to be kept in
1523 		 * accross calls to symbol__parse_objdump_line(), so that it
1524 		 * can associate it with the instructions till the next one.
1525 		 * See disasm_line__new() and struct disasm_line::line_nr.
1526 		 */
1527 		if (symbol__parse_objdump_line(sym, file, args, &lineno) < 0)
1528 			break;
1529 		nline++;
1530 	}
1531 
1532 	if (nline == 0)
1533 		pr_err("No output from %s\n", command);
1534 
1535 	/*
1536 	 * kallsyms does not have symbol sizes so there may a nop at the end.
1537 	 * Remove it.
1538 	 */
1539 	if (dso__is_kcore(dso))
1540 		delete_last_nop(sym);
1541 
1542 	fclose(file);
1543 	err = 0;
1544 out_remove_tmp:
1545 	close(stdout_fd[0]);
1546 
1547 	if (dso__needs_decompress(dso))
1548 		unlink(symfs_filename);
1549 
1550 	if (delete_extract)
1551 		kcore_extract__delete(&kce);
1552 out:
1553 	return err;
1554 
1555 out_close_stdout:
1556 	close(stdout_fd[1]);
1557 	goto out_remove_tmp;
1558 }
1559 
1560 static void calc_percent(struct sym_hist *hist,
1561 			 struct annotation_data *sample,
1562 			 s64 offset, s64 end)
1563 {
1564 	unsigned int hits = 0;
1565 	u64 period = 0;
1566 
1567 	while (offset < end) {
1568 		hits   += hist->addr[offset].nr_samples;
1569 		period += hist->addr[offset].period;
1570 		++offset;
1571 	}
1572 
1573 	if (hist->nr_samples) {
1574 		sample->he.period     = period;
1575 		sample->he.nr_samples = hits;
1576 		sample->percent = 100.0 * hits / hist->nr_samples;
1577 	}
1578 }
1579 
1580 static void annotation__calc_percent(struct annotation *notes,
1581 				     struct perf_evsel *evsel, s64 len)
1582 {
1583 	struct annotation_line *al, *next;
1584 
1585 	list_for_each_entry(al, &notes->src->source, node) {
1586 		s64 end;
1587 		int i;
1588 
1589 		if (al->offset == -1)
1590 			continue;
1591 
1592 		next = annotation_line__next(al, &notes->src->source);
1593 		end  = next ? next->offset : len;
1594 
1595 		for (i = 0; i < al->samples_nr; i++) {
1596 			struct annotation_data *sample;
1597 			struct sym_hist *hist;
1598 
1599 			hist   = annotation__histogram(notes, evsel->idx + i);
1600 			sample = &al->samples[i];
1601 
1602 			calc_percent(hist, sample, al->offset, end);
1603 		}
1604 	}
1605 }
1606 
1607 void symbol__calc_percent(struct symbol *sym, struct perf_evsel *evsel)
1608 {
1609 	struct annotation *notes = symbol__annotation(sym);
1610 
1611 	annotation__calc_percent(notes, evsel, symbol__size(sym));
1612 }
1613 
1614 int symbol__annotate(struct symbol *sym, struct map *map,
1615 		     struct perf_evsel *evsel, size_t privsize,
1616 		     struct arch **parch)
1617 {
1618 	struct annotate_args args = {
1619 		.privsize	= privsize,
1620 		.map		= map,
1621 		.evsel		= evsel,
1622 	};
1623 	struct perf_env *env = perf_evsel__env(evsel);
1624 	const char *arch_name = perf_env__arch(env);
1625 	struct arch *arch;
1626 	int err;
1627 
1628 	if (!arch_name)
1629 		return -1;
1630 
1631 	args.arch = arch = arch__find(arch_name);
1632 	if (arch == NULL)
1633 		return -ENOTSUP;
1634 
1635 	if (parch)
1636 		*parch = arch;
1637 
1638 	if (arch->init) {
1639 		err = arch->init(arch, env ? env->cpuid : NULL);
1640 		if (err) {
1641 			pr_err("%s: failed to initialize %s arch priv area\n", __func__, arch->name);
1642 			return err;
1643 		}
1644 	}
1645 
1646 	return symbol__disassemble(sym, &args);
1647 }
1648 
1649 static void insert_source_line(struct rb_root *root, struct annotation_line *al)
1650 {
1651 	struct annotation_line *iter;
1652 	struct rb_node **p = &root->rb_node;
1653 	struct rb_node *parent = NULL;
1654 	int i, ret;
1655 
1656 	while (*p != NULL) {
1657 		parent = *p;
1658 		iter = rb_entry(parent, struct annotation_line, rb_node);
1659 
1660 		ret = strcmp(iter->path, al->path);
1661 		if (ret == 0) {
1662 			for (i = 0; i < al->samples_nr; i++)
1663 				iter->samples[i].percent_sum += al->samples[i].percent;
1664 			return;
1665 		}
1666 
1667 		if (ret < 0)
1668 			p = &(*p)->rb_left;
1669 		else
1670 			p = &(*p)->rb_right;
1671 	}
1672 
1673 	for (i = 0; i < al->samples_nr; i++)
1674 		al->samples[i].percent_sum = al->samples[i].percent;
1675 
1676 	rb_link_node(&al->rb_node, parent, p);
1677 	rb_insert_color(&al->rb_node, root);
1678 }
1679 
1680 static int cmp_source_line(struct annotation_line *a, struct annotation_line *b)
1681 {
1682 	int i;
1683 
1684 	for (i = 0; i < a->samples_nr; i++) {
1685 		if (a->samples[i].percent_sum == b->samples[i].percent_sum)
1686 			continue;
1687 		return a->samples[i].percent_sum > b->samples[i].percent_sum;
1688 	}
1689 
1690 	return 0;
1691 }
1692 
1693 static void __resort_source_line(struct rb_root *root, struct annotation_line *al)
1694 {
1695 	struct annotation_line *iter;
1696 	struct rb_node **p = &root->rb_node;
1697 	struct rb_node *parent = NULL;
1698 
1699 	while (*p != NULL) {
1700 		parent = *p;
1701 		iter = rb_entry(parent, struct annotation_line, rb_node);
1702 
1703 		if (cmp_source_line(al, iter))
1704 			p = &(*p)->rb_left;
1705 		else
1706 			p = &(*p)->rb_right;
1707 	}
1708 
1709 	rb_link_node(&al->rb_node, parent, p);
1710 	rb_insert_color(&al->rb_node, root);
1711 }
1712 
1713 static void resort_source_line(struct rb_root *dest_root, struct rb_root *src_root)
1714 {
1715 	struct annotation_line *al;
1716 	struct rb_node *node;
1717 
1718 	node = rb_first(src_root);
1719 	while (node) {
1720 		struct rb_node *next;
1721 
1722 		al = rb_entry(node, struct annotation_line, rb_node);
1723 		next = rb_next(node);
1724 		rb_erase(node, src_root);
1725 
1726 		__resort_source_line(dest_root, al);
1727 		node = next;
1728 	}
1729 }
1730 
1731 static void print_summary(struct rb_root *root, const char *filename)
1732 {
1733 	struct annotation_line *al;
1734 	struct rb_node *node;
1735 
1736 	printf("\nSorted summary for file %s\n", filename);
1737 	printf("----------------------------------------------\n\n");
1738 
1739 	if (RB_EMPTY_ROOT(root)) {
1740 		printf(" Nothing higher than %1.1f%%\n", MIN_GREEN);
1741 		return;
1742 	}
1743 
1744 	node = rb_first(root);
1745 	while (node) {
1746 		double percent, percent_max = 0.0;
1747 		const char *color;
1748 		char *path;
1749 		int i;
1750 
1751 		al = rb_entry(node, struct annotation_line, rb_node);
1752 		for (i = 0; i < al->samples_nr; i++) {
1753 			percent = al->samples[i].percent_sum;
1754 			color = get_percent_color(percent);
1755 			color_fprintf(stdout, color, " %7.2f", percent);
1756 
1757 			if (percent > percent_max)
1758 				percent_max = percent;
1759 		}
1760 
1761 		path = al->path;
1762 		color = get_percent_color(percent_max);
1763 		color_fprintf(stdout, color, " %s\n", path);
1764 
1765 		node = rb_next(node);
1766 	}
1767 }
1768 
1769 static void symbol__annotate_hits(struct symbol *sym, struct perf_evsel *evsel)
1770 {
1771 	struct annotation *notes = symbol__annotation(sym);
1772 	struct sym_hist *h = annotation__histogram(notes, evsel->idx);
1773 	u64 len = symbol__size(sym), offset;
1774 
1775 	for (offset = 0; offset < len; ++offset)
1776 		if (h->addr[offset].nr_samples != 0)
1777 			printf("%*" PRIx64 ": %" PRIu64 "\n", BITS_PER_LONG / 2,
1778 			       sym->start + offset, h->addr[offset].nr_samples);
1779 	printf("%*s: %" PRIu64 "\n", BITS_PER_LONG / 2, "h->nr_samples", h->nr_samples);
1780 }
1781 
1782 static int annotated_source__addr_fmt_width(struct list_head *lines, u64 start)
1783 {
1784 	char bf[32];
1785 	struct annotation_line *line;
1786 
1787 	list_for_each_entry_reverse(line, lines, node) {
1788 		if (line->offset != -1)
1789 			return scnprintf(bf, sizeof(bf), "%" PRIx64, start + line->offset);
1790 	}
1791 
1792 	return 0;
1793 }
1794 
1795 int symbol__annotate_printf(struct symbol *sym, struct map *map,
1796 			    struct perf_evsel *evsel, bool full_paths,
1797 			    int min_pcnt, int max_lines, int context)
1798 {
1799 	struct dso *dso = map->dso;
1800 	char *filename;
1801 	const char *d_filename;
1802 	const char *evsel_name = perf_evsel__name(evsel);
1803 	struct annotation *notes = symbol__annotation(sym);
1804 	struct sym_hist *h = annotation__histogram(notes, evsel->idx);
1805 	struct annotation_line *pos, *queue = NULL;
1806 	u64 start = map__rip_2objdump(map, sym->start);
1807 	int printed = 2, queue_len = 0, addr_fmt_width;
1808 	int more = 0;
1809 	u64 len;
1810 	int width = symbol_conf.show_total_period ? 12 : 8;
1811 	int graph_dotted_len;
1812 
1813 	filename = strdup(dso->long_name);
1814 	if (!filename)
1815 		return -ENOMEM;
1816 
1817 	if (full_paths)
1818 		d_filename = filename;
1819 	else
1820 		d_filename = basename(filename);
1821 
1822 	len = symbol__size(sym);
1823 
1824 	if (perf_evsel__is_group_event(evsel))
1825 		width *= evsel->nr_members;
1826 
1827 	graph_dotted_len = printf(" %-*.*s|	Source code & Disassembly of %s for %s (%" PRIu64 " samples)\n",
1828 				  width, width, symbol_conf.show_total_period ? "Period" :
1829 				  symbol_conf.show_nr_samples ? "Samples" : "Percent",
1830 				  d_filename, evsel_name, h->nr_samples);
1831 
1832 	printf("%-*.*s----\n",
1833 	       graph_dotted_len, graph_dotted_len, graph_dotted_line);
1834 
1835 	if (verbose > 0)
1836 		symbol__annotate_hits(sym, evsel);
1837 
1838 	addr_fmt_width = annotated_source__addr_fmt_width(&notes->src->source, start);
1839 
1840 	list_for_each_entry(pos, &notes->src->source, node) {
1841 		int err;
1842 
1843 		if (context && queue == NULL) {
1844 			queue = pos;
1845 			queue_len = 0;
1846 		}
1847 
1848 		err = annotation_line__print(pos, sym, start, evsel, len,
1849 					     min_pcnt, printed, max_lines,
1850 					     queue, addr_fmt_width);
1851 
1852 		switch (err) {
1853 		case 0:
1854 			++printed;
1855 			if (context) {
1856 				printed += queue_len;
1857 				queue = NULL;
1858 				queue_len = 0;
1859 			}
1860 			break;
1861 		case 1:
1862 			/* filtered by max_lines */
1863 			++more;
1864 			break;
1865 		case -1:
1866 		default:
1867 			/*
1868 			 * Filtered by min_pcnt or non IP lines when
1869 			 * context != 0
1870 			 */
1871 			if (!context)
1872 				break;
1873 			if (queue_len == context)
1874 				queue = list_entry(queue->node.next, typeof(*queue), node);
1875 			else
1876 				++queue_len;
1877 			break;
1878 		}
1879 	}
1880 
1881 	free(filename);
1882 
1883 	return more;
1884 }
1885 
1886 void symbol__annotate_zero_histogram(struct symbol *sym, int evidx)
1887 {
1888 	struct annotation *notes = symbol__annotation(sym);
1889 	struct sym_hist *h = annotation__histogram(notes, evidx);
1890 
1891 	memset(h, 0, notes->src->sizeof_sym_hist);
1892 }
1893 
1894 void symbol__annotate_decay_histogram(struct symbol *sym, int evidx)
1895 {
1896 	struct annotation *notes = symbol__annotation(sym);
1897 	struct sym_hist *h = annotation__histogram(notes, evidx);
1898 	int len = symbol__size(sym), offset;
1899 
1900 	h->nr_samples = 0;
1901 	for (offset = 0; offset < len; ++offset) {
1902 		h->addr[offset].nr_samples = h->addr[offset].nr_samples * 7 / 8;
1903 		h->nr_samples += h->addr[offset].nr_samples;
1904 	}
1905 }
1906 
1907 void annotated_source__purge(struct annotated_source *as)
1908 {
1909 	struct annotation_line *al, *n;
1910 
1911 	list_for_each_entry_safe(al, n, &as->source, node) {
1912 		list_del(&al->node);
1913 		disasm_line__free(disasm_line(al));
1914 	}
1915 }
1916 
1917 static size_t disasm_line__fprintf(struct disasm_line *dl, FILE *fp)
1918 {
1919 	size_t printed;
1920 
1921 	if (dl->al.offset == -1)
1922 		return fprintf(fp, "%s\n", dl->al.line);
1923 
1924 	printed = fprintf(fp, "%#" PRIx64 " %s", dl->al.offset, dl->ins.name);
1925 
1926 	if (dl->ops.raw[0] != '\0') {
1927 		printed += fprintf(fp, "%.*s %s\n", 6 - (int)printed, " ",
1928 				   dl->ops.raw);
1929 	}
1930 
1931 	return printed + fprintf(fp, "\n");
1932 }
1933 
1934 size_t disasm__fprintf(struct list_head *head, FILE *fp)
1935 {
1936 	struct disasm_line *pos;
1937 	size_t printed = 0;
1938 
1939 	list_for_each_entry(pos, head, al.node)
1940 		printed += disasm_line__fprintf(pos, fp);
1941 
1942 	return printed;
1943 }
1944 
1945 static void annotation__calc_lines(struct annotation *notes, struct map *map,
1946 				  struct rb_root *root, u64 start)
1947 {
1948 	struct annotation_line *al;
1949 	struct rb_root tmp_root = RB_ROOT;
1950 
1951 	list_for_each_entry(al, &notes->src->source, node) {
1952 		double percent_max = 0.0;
1953 		int i;
1954 
1955 		for (i = 0; i < al->samples_nr; i++) {
1956 			struct annotation_data *sample;
1957 
1958 			sample = &al->samples[i];
1959 
1960 			if (sample->percent > percent_max)
1961 				percent_max = sample->percent;
1962 		}
1963 
1964 		if (percent_max <= 0.5)
1965 			continue;
1966 
1967 		al->path = get_srcline(map->dso, start + al->offset, NULL,
1968 				       false, true, start + al->offset);
1969 		insert_source_line(&tmp_root, al);
1970 	}
1971 
1972 	resort_source_line(root, &tmp_root);
1973 }
1974 
1975 static void symbol__calc_lines(struct symbol *sym, struct map *map,
1976 			      struct rb_root *root)
1977 {
1978 	struct annotation *notes = symbol__annotation(sym);
1979 	u64 start = map__rip_2objdump(map, sym->start);
1980 
1981 	annotation__calc_lines(notes, map, root, start);
1982 }
1983 
1984 int symbol__tty_annotate(struct symbol *sym, struct map *map,
1985 			 struct perf_evsel *evsel, bool print_lines,
1986 			 bool full_paths, int min_pcnt, int max_lines)
1987 {
1988 	struct dso *dso = map->dso;
1989 	struct rb_root source_line = RB_ROOT;
1990 
1991 	if (symbol__annotate(sym, map, evsel, 0, NULL) < 0)
1992 		return -1;
1993 
1994 	symbol__calc_percent(sym, evsel);
1995 
1996 	if (print_lines) {
1997 		srcline_full_filename = full_paths;
1998 		symbol__calc_lines(sym, map, &source_line);
1999 		print_summary(&source_line, dso->long_name);
2000 	}
2001 
2002 	symbol__annotate_printf(sym, map, evsel, full_paths,
2003 				min_pcnt, max_lines, 0);
2004 
2005 	annotated_source__purge(symbol__annotation(sym)->src);
2006 
2007 	return 0;
2008 }
2009 
2010 bool ui__has_annotation(void)
2011 {
2012 	return use_browser == 1 && perf_hpp_list.sym;
2013 }
2014