xref: /openbmc/linux/tools/perf/util/symbol.c (revision 95c96174)
1 #include <dirent.h>
2 #include <errno.h>
3 #include <stdlib.h>
4 #include <stdio.h>
5 #include <string.h>
6 #include <sys/types.h>
7 #include <sys/stat.h>
8 #include <sys/param.h>
9 #include <fcntl.h>
10 #include <unistd.h>
11 #include <inttypes.h>
12 #include "build-id.h"
13 #include "util.h"
14 #include "debug.h"
15 #include "symbol.h"
16 #include "strlist.h"
17 
18 #include <libelf.h>
19 #include <gelf.h>
20 #include <elf.h>
21 #include <limits.h>
22 #include <sys/utsname.h>
23 
24 #ifndef KSYM_NAME_LEN
25 #define KSYM_NAME_LEN 256
26 #endif
27 
28 #ifndef NT_GNU_BUILD_ID
29 #define NT_GNU_BUILD_ID 3
30 #endif
31 
32 static bool dso__build_id_equal(const struct dso *dso, u8 *build_id);
33 static int elf_read_build_id(Elf *elf, void *bf, size_t size);
34 static void dsos__add(struct list_head *head, struct dso *dso);
35 static struct map *map__new2(u64 start, struct dso *dso, enum map_type type);
36 static int dso__load_kernel_sym(struct dso *dso, struct map *map,
37 				symbol_filter_t filter);
38 static int dso__load_guest_kernel_sym(struct dso *dso, struct map *map,
39 			symbol_filter_t filter);
40 static int vmlinux_path__nr_entries;
41 static char **vmlinux_path;
42 
43 struct symbol_conf symbol_conf = {
44 	.exclude_other	  = true,
45 	.use_modules	  = true,
46 	.try_vmlinux_path = true,
47 	.annotate_src	  = true,
48 	.symfs            = "",
49 };
50 
51 int dso__name_len(const struct dso *dso)
52 {
53 	if (!dso)
54 		return strlen("[unknown]");
55 	if (verbose)
56 		return dso->long_name_len;
57 
58 	return dso->short_name_len;
59 }
60 
61 bool dso__loaded(const struct dso *dso, enum map_type type)
62 {
63 	return dso->loaded & (1 << type);
64 }
65 
66 bool dso__sorted_by_name(const struct dso *dso, enum map_type type)
67 {
68 	return dso->sorted_by_name & (1 << type);
69 }
70 
71 static void dso__set_sorted_by_name(struct dso *dso, enum map_type type)
72 {
73 	dso->sorted_by_name |= (1 << type);
74 }
75 
76 bool symbol_type__is_a(char symbol_type, enum map_type map_type)
77 {
78 	symbol_type = toupper(symbol_type);
79 
80 	switch (map_type) {
81 	case MAP__FUNCTION:
82 		return symbol_type == 'T' || symbol_type == 'W';
83 	case MAP__VARIABLE:
84 		return symbol_type == 'D';
85 	default:
86 		return false;
87 	}
88 }
89 
90 static int prefix_underscores_count(const char *str)
91 {
92 	const char *tail = str;
93 
94 	while (*tail == '_')
95 		tail++;
96 
97 	return tail - str;
98 }
99 
100 #define SYMBOL_A 0
101 #define SYMBOL_B 1
102 
103 static int choose_best_symbol(struct symbol *syma, struct symbol *symb)
104 {
105 	s64 a;
106 	s64 b;
107 
108 	/* Prefer a symbol with non zero length */
109 	a = syma->end - syma->start;
110 	b = symb->end - symb->start;
111 	if ((b == 0) && (a > 0))
112 		return SYMBOL_A;
113 	else if ((a == 0) && (b > 0))
114 		return SYMBOL_B;
115 
116 	/* Prefer a non weak symbol over a weak one */
117 	a = syma->binding == STB_WEAK;
118 	b = symb->binding == STB_WEAK;
119 	if (b && !a)
120 		return SYMBOL_A;
121 	if (a && !b)
122 		return SYMBOL_B;
123 
124 	/* Prefer a global symbol over a non global one */
125 	a = syma->binding == STB_GLOBAL;
126 	b = symb->binding == STB_GLOBAL;
127 	if (a && !b)
128 		return SYMBOL_A;
129 	if (b && !a)
130 		return SYMBOL_B;
131 
132 	/* Prefer a symbol with less underscores */
133 	a = prefix_underscores_count(syma->name);
134 	b = prefix_underscores_count(symb->name);
135 	if (b > a)
136 		return SYMBOL_A;
137 	else if (a > b)
138 		return SYMBOL_B;
139 
140 	/* If all else fails, choose the symbol with the longest name */
141 	if (strlen(syma->name) >= strlen(symb->name))
142 		return SYMBOL_A;
143 	else
144 		return SYMBOL_B;
145 }
146 
147 static void symbols__fixup_duplicate(struct rb_root *symbols)
148 {
149 	struct rb_node *nd;
150 	struct symbol *curr, *next;
151 
152 	nd = rb_first(symbols);
153 
154 	while (nd) {
155 		curr = rb_entry(nd, struct symbol, rb_node);
156 again:
157 		nd = rb_next(&curr->rb_node);
158 		next = rb_entry(nd, struct symbol, rb_node);
159 
160 		if (!nd)
161 			break;
162 
163 		if (curr->start != next->start)
164 			continue;
165 
166 		if (choose_best_symbol(curr, next) == SYMBOL_A) {
167 			rb_erase(&next->rb_node, symbols);
168 			goto again;
169 		} else {
170 			nd = rb_next(&curr->rb_node);
171 			rb_erase(&curr->rb_node, symbols);
172 		}
173 	}
174 }
175 
176 static void symbols__fixup_end(struct rb_root *symbols)
177 {
178 	struct rb_node *nd, *prevnd = rb_first(symbols);
179 	struct symbol *curr, *prev;
180 
181 	if (prevnd == NULL)
182 		return;
183 
184 	curr = rb_entry(prevnd, struct symbol, rb_node);
185 
186 	for (nd = rb_next(prevnd); nd; nd = rb_next(nd)) {
187 		prev = curr;
188 		curr = rb_entry(nd, struct symbol, rb_node);
189 
190 		if (prev->end == prev->start && prev->end != curr->start)
191 			prev->end = curr->start - 1;
192 	}
193 
194 	/* Last entry */
195 	if (curr->end == curr->start)
196 		curr->end = roundup(curr->start, 4096);
197 }
198 
199 static void __map_groups__fixup_end(struct map_groups *mg, enum map_type type)
200 {
201 	struct map *prev, *curr;
202 	struct rb_node *nd, *prevnd = rb_first(&mg->maps[type]);
203 
204 	if (prevnd == NULL)
205 		return;
206 
207 	curr = rb_entry(prevnd, struct map, rb_node);
208 
209 	for (nd = rb_next(prevnd); nd; nd = rb_next(nd)) {
210 		prev = curr;
211 		curr = rb_entry(nd, struct map, rb_node);
212 		prev->end = curr->start - 1;
213 	}
214 
215 	/*
216 	 * We still haven't the actual symbols, so guess the
217 	 * last map final address.
218 	 */
219 	curr->end = ~0ULL;
220 }
221 
222 static void map_groups__fixup_end(struct map_groups *mg)
223 {
224 	int i;
225 	for (i = 0; i < MAP__NR_TYPES; ++i)
226 		__map_groups__fixup_end(mg, i);
227 }
228 
229 static struct symbol *symbol__new(u64 start, u64 len, u8 binding,
230 				  const char *name)
231 {
232 	size_t namelen = strlen(name) + 1;
233 	struct symbol *sym = calloc(1, (symbol_conf.priv_size +
234 					sizeof(*sym) + namelen));
235 	if (sym == NULL)
236 		return NULL;
237 
238 	if (symbol_conf.priv_size)
239 		sym = ((void *)sym) + symbol_conf.priv_size;
240 
241 	sym->start   = start;
242 	sym->end     = len ? start + len - 1 : start;
243 	sym->binding = binding;
244 	sym->namelen = namelen - 1;
245 
246 	pr_debug4("%s: %s %#" PRIx64 "-%#" PRIx64 "\n",
247 		  __func__, name, start, sym->end);
248 	memcpy(sym->name, name, namelen);
249 
250 	return sym;
251 }
252 
253 void symbol__delete(struct symbol *sym)
254 {
255 	free(((void *)sym) - symbol_conf.priv_size);
256 }
257 
258 static size_t symbol__fprintf(struct symbol *sym, FILE *fp)
259 {
260 	return fprintf(fp, " %" PRIx64 "-%" PRIx64 " %c %s\n",
261 		       sym->start, sym->end,
262 		       sym->binding == STB_GLOBAL ? 'g' :
263 		       sym->binding == STB_LOCAL  ? 'l' : 'w',
264 		       sym->name);
265 }
266 
267 size_t symbol__fprintf_symname_offs(const struct symbol *sym,
268 				    const struct addr_location *al, FILE *fp)
269 {
270 	unsigned long offset;
271 	size_t length;
272 
273 	if (sym && sym->name) {
274 		length = fprintf(fp, "%s", sym->name);
275 		if (al) {
276 			offset = al->addr - sym->start;
277 			length += fprintf(fp, "+0x%lx", offset);
278 		}
279 		return length;
280 	} else
281 		return fprintf(fp, "[unknown]");
282 }
283 
284 size_t symbol__fprintf_symname(const struct symbol *sym, FILE *fp)
285 {
286 	return symbol__fprintf_symname_offs(sym, NULL, fp);
287 }
288 
289 void dso__set_long_name(struct dso *dso, char *name)
290 {
291 	if (name == NULL)
292 		return;
293 	dso->long_name = name;
294 	dso->long_name_len = strlen(name);
295 }
296 
297 static void dso__set_short_name(struct dso *dso, const char *name)
298 {
299 	if (name == NULL)
300 		return;
301 	dso->short_name = name;
302 	dso->short_name_len = strlen(name);
303 }
304 
305 static void dso__set_basename(struct dso *dso)
306 {
307 	dso__set_short_name(dso, basename(dso->long_name));
308 }
309 
310 struct dso *dso__new(const char *name)
311 {
312 	struct dso *dso = calloc(1, sizeof(*dso) + strlen(name) + 1);
313 
314 	if (dso != NULL) {
315 		int i;
316 		strcpy(dso->name, name);
317 		dso__set_long_name(dso, dso->name);
318 		dso__set_short_name(dso, dso->name);
319 		for (i = 0; i < MAP__NR_TYPES; ++i)
320 			dso->symbols[i] = dso->symbol_names[i] = RB_ROOT;
321 		dso->symtab_type = SYMTAB__NOT_FOUND;
322 		dso->loaded = 0;
323 		dso->sorted_by_name = 0;
324 		dso->has_build_id = 0;
325 		dso->kernel = DSO_TYPE_USER;
326 		INIT_LIST_HEAD(&dso->node);
327 	}
328 
329 	return dso;
330 }
331 
332 static void symbols__delete(struct rb_root *symbols)
333 {
334 	struct symbol *pos;
335 	struct rb_node *next = rb_first(symbols);
336 
337 	while (next) {
338 		pos = rb_entry(next, struct symbol, rb_node);
339 		next = rb_next(&pos->rb_node);
340 		rb_erase(&pos->rb_node, symbols);
341 		symbol__delete(pos);
342 	}
343 }
344 
345 void dso__delete(struct dso *dso)
346 {
347 	int i;
348 	for (i = 0; i < MAP__NR_TYPES; ++i)
349 		symbols__delete(&dso->symbols[i]);
350 	if (dso->sname_alloc)
351 		free((char *)dso->short_name);
352 	if (dso->lname_alloc)
353 		free(dso->long_name);
354 	free(dso);
355 }
356 
357 void dso__set_build_id(struct dso *dso, void *build_id)
358 {
359 	memcpy(dso->build_id, build_id, sizeof(dso->build_id));
360 	dso->has_build_id = 1;
361 }
362 
363 static void symbols__insert(struct rb_root *symbols, struct symbol *sym)
364 {
365 	struct rb_node **p = &symbols->rb_node;
366 	struct rb_node *parent = NULL;
367 	const u64 ip = sym->start;
368 	struct symbol *s;
369 
370 	while (*p != NULL) {
371 		parent = *p;
372 		s = rb_entry(parent, struct symbol, rb_node);
373 		if (ip < s->start)
374 			p = &(*p)->rb_left;
375 		else
376 			p = &(*p)->rb_right;
377 	}
378 	rb_link_node(&sym->rb_node, parent, p);
379 	rb_insert_color(&sym->rb_node, symbols);
380 }
381 
382 static struct symbol *symbols__find(struct rb_root *symbols, u64 ip)
383 {
384 	struct rb_node *n;
385 
386 	if (symbols == NULL)
387 		return NULL;
388 
389 	n = symbols->rb_node;
390 
391 	while (n) {
392 		struct symbol *s = rb_entry(n, struct symbol, rb_node);
393 
394 		if (ip < s->start)
395 			n = n->rb_left;
396 		else if (ip > s->end)
397 			n = n->rb_right;
398 		else
399 			return s;
400 	}
401 
402 	return NULL;
403 }
404 
405 struct symbol_name_rb_node {
406 	struct rb_node	rb_node;
407 	struct symbol	sym;
408 };
409 
410 static void symbols__insert_by_name(struct rb_root *symbols, struct symbol *sym)
411 {
412 	struct rb_node **p = &symbols->rb_node;
413 	struct rb_node *parent = NULL;
414 	struct symbol_name_rb_node *symn, *s;
415 
416 	symn = container_of(sym, struct symbol_name_rb_node, sym);
417 
418 	while (*p != NULL) {
419 		parent = *p;
420 		s = rb_entry(parent, struct symbol_name_rb_node, rb_node);
421 		if (strcmp(sym->name, s->sym.name) < 0)
422 			p = &(*p)->rb_left;
423 		else
424 			p = &(*p)->rb_right;
425 	}
426 	rb_link_node(&symn->rb_node, parent, p);
427 	rb_insert_color(&symn->rb_node, symbols);
428 }
429 
430 static void symbols__sort_by_name(struct rb_root *symbols,
431 				  struct rb_root *source)
432 {
433 	struct rb_node *nd;
434 
435 	for (nd = rb_first(source); nd; nd = rb_next(nd)) {
436 		struct symbol *pos = rb_entry(nd, struct symbol, rb_node);
437 		symbols__insert_by_name(symbols, pos);
438 	}
439 }
440 
441 static struct symbol *symbols__find_by_name(struct rb_root *symbols,
442 					    const char *name)
443 {
444 	struct rb_node *n;
445 
446 	if (symbols == NULL)
447 		return NULL;
448 
449 	n = symbols->rb_node;
450 
451 	while (n) {
452 		struct symbol_name_rb_node *s;
453 		int cmp;
454 
455 		s = rb_entry(n, struct symbol_name_rb_node, rb_node);
456 		cmp = strcmp(name, s->sym.name);
457 
458 		if (cmp < 0)
459 			n = n->rb_left;
460 		else if (cmp > 0)
461 			n = n->rb_right;
462 		else
463 			return &s->sym;
464 	}
465 
466 	return NULL;
467 }
468 
469 struct symbol *dso__find_symbol(struct dso *dso,
470 				enum map_type type, u64 addr)
471 {
472 	return symbols__find(&dso->symbols[type], addr);
473 }
474 
475 struct symbol *dso__find_symbol_by_name(struct dso *dso, enum map_type type,
476 					const char *name)
477 {
478 	return symbols__find_by_name(&dso->symbol_names[type], name);
479 }
480 
481 void dso__sort_by_name(struct dso *dso, enum map_type type)
482 {
483 	dso__set_sorted_by_name(dso, type);
484 	return symbols__sort_by_name(&dso->symbol_names[type],
485 				     &dso->symbols[type]);
486 }
487 
488 int build_id__sprintf(const u8 *build_id, int len, char *bf)
489 {
490 	char *bid = bf;
491 	const u8 *raw = build_id;
492 	int i;
493 
494 	for (i = 0; i < len; ++i) {
495 		sprintf(bid, "%02x", *raw);
496 		++raw;
497 		bid += 2;
498 	}
499 
500 	return raw - build_id;
501 }
502 
503 size_t dso__fprintf_buildid(struct dso *dso, FILE *fp)
504 {
505 	char sbuild_id[BUILD_ID_SIZE * 2 + 1];
506 
507 	build_id__sprintf(dso->build_id, sizeof(dso->build_id), sbuild_id);
508 	return fprintf(fp, "%s", sbuild_id);
509 }
510 
511 size_t dso__fprintf_symbols_by_name(struct dso *dso,
512 				    enum map_type type, FILE *fp)
513 {
514 	size_t ret = 0;
515 	struct rb_node *nd;
516 	struct symbol_name_rb_node *pos;
517 
518 	for (nd = rb_first(&dso->symbol_names[type]); nd; nd = rb_next(nd)) {
519 		pos = rb_entry(nd, struct symbol_name_rb_node, rb_node);
520 		fprintf(fp, "%s\n", pos->sym.name);
521 	}
522 
523 	return ret;
524 }
525 
526 size_t dso__fprintf(struct dso *dso, enum map_type type, FILE *fp)
527 {
528 	struct rb_node *nd;
529 	size_t ret = fprintf(fp, "dso: %s (", dso->short_name);
530 
531 	if (dso->short_name != dso->long_name)
532 		ret += fprintf(fp, "%s, ", dso->long_name);
533 	ret += fprintf(fp, "%s, %sloaded, ", map_type__name[type],
534 		       dso->loaded ? "" : "NOT ");
535 	ret += dso__fprintf_buildid(dso, fp);
536 	ret += fprintf(fp, ")\n");
537 	for (nd = rb_first(&dso->symbols[type]); nd; nd = rb_next(nd)) {
538 		struct symbol *pos = rb_entry(nd, struct symbol, rb_node);
539 		ret += symbol__fprintf(pos, fp);
540 	}
541 
542 	return ret;
543 }
544 
545 int kallsyms__parse(const char *filename, void *arg,
546 		    int (*process_symbol)(void *arg, const char *name,
547 					  char type, u64 start, u64 end))
548 {
549 	char *line = NULL;
550 	size_t n;
551 	int err = -1;
552 	FILE *file = fopen(filename, "r");
553 
554 	if (file == NULL)
555 		goto out_failure;
556 
557 	err = 0;
558 
559 	while (!feof(file)) {
560 		u64 start;
561 		int line_len, len;
562 		char symbol_type;
563 		char *symbol_name;
564 
565 		line_len = getline(&line, &n, file);
566 		if (line_len < 0 || !line)
567 			break;
568 
569 		line[--line_len] = '\0'; /* \n */
570 
571 		len = hex2u64(line, &start);
572 
573 		len++;
574 		if (len + 2 >= line_len)
575 			continue;
576 
577 		symbol_type = line[len];
578 		len += 2;
579 		symbol_name = line + len;
580 		len = line_len - len;
581 
582 		if (len >= KSYM_NAME_LEN) {
583 			err = -1;
584 			break;
585 		}
586 
587 		/*
588 		 * module symbols are not sorted so we add all
589 		 * symbols with zero length and rely on
590 		 * symbols__fixup_end() to fix it up.
591 		 */
592 		err = process_symbol(arg, symbol_name,
593 				     symbol_type, start, start);
594 		if (err)
595 			break;
596 	}
597 
598 	free(line);
599 	fclose(file);
600 	return err;
601 
602 out_failure:
603 	return -1;
604 }
605 
606 struct process_kallsyms_args {
607 	struct map *map;
608 	struct dso *dso;
609 };
610 
611 static u8 kallsyms2elf_type(char type)
612 {
613 	if (type == 'W')
614 		return STB_WEAK;
615 
616 	return isupper(type) ? STB_GLOBAL : STB_LOCAL;
617 }
618 
619 static int map__process_kallsym_symbol(void *arg, const char *name,
620 				       char type, u64 start, u64 end)
621 {
622 	struct symbol *sym;
623 	struct process_kallsyms_args *a = arg;
624 	struct rb_root *root = &a->dso->symbols[a->map->type];
625 
626 	if (!symbol_type__is_a(type, a->map->type))
627 		return 0;
628 
629 	sym = symbol__new(start, end - start + 1,
630 			  kallsyms2elf_type(type), name);
631 	if (sym == NULL)
632 		return -ENOMEM;
633 	/*
634 	 * We will pass the symbols to the filter later, in
635 	 * map__split_kallsyms, when we have split the maps per module
636 	 */
637 	symbols__insert(root, sym);
638 
639 	return 0;
640 }
641 
642 /*
643  * Loads the function entries in /proc/kallsyms into kernel_map->dso,
644  * so that we can in the next step set the symbol ->end address and then
645  * call kernel_maps__split_kallsyms.
646  */
647 static int dso__load_all_kallsyms(struct dso *dso, const char *filename,
648 				  struct map *map)
649 {
650 	struct process_kallsyms_args args = { .map = map, .dso = dso, };
651 	return kallsyms__parse(filename, &args, map__process_kallsym_symbol);
652 }
653 
654 /*
655  * Split the symbols into maps, making sure there are no overlaps, i.e. the
656  * kernel range is broken in several maps, named [kernel].N, as we don't have
657  * the original ELF section names vmlinux have.
658  */
659 static int dso__split_kallsyms(struct dso *dso, struct map *map,
660 			       symbol_filter_t filter)
661 {
662 	struct map_groups *kmaps = map__kmap(map)->kmaps;
663 	struct machine *machine = kmaps->machine;
664 	struct map *curr_map = map;
665 	struct symbol *pos;
666 	int count = 0, moved = 0;
667 	struct rb_root *root = &dso->symbols[map->type];
668 	struct rb_node *next = rb_first(root);
669 	int kernel_range = 0;
670 
671 	while (next) {
672 		char *module;
673 
674 		pos = rb_entry(next, struct symbol, rb_node);
675 		next = rb_next(&pos->rb_node);
676 
677 		module = strchr(pos->name, '\t');
678 		if (module) {
679 			if (!symbol_conf.use_modules)
680 				goto discard_symbol;
681 
682 			*module++ = '\0';
683 
684 			if (strcmp(curr_map->dso->short_name, module)) {
685 				if (curr_map != map &&
686 				    dso->kernel == DSO_TYPE_GUEST_KERNEL &&
687 				    machine__is_default_guest(machine)) {
688 					/*
689 					 * We assume all symbols of a module are
690 					 * continuous in * kallsyms, so curr_map
691 					 * points to a module and all its
692 					 * symbols are in its kmap. Mark it as
693 					 * loaded.
694 					 */
695 					dso__set_loaded(curr_map->dso,
696 							curr_map->type);
697 				}
698 
699 				curr_map = map_groups__find_by_name(kmaps,
700 							map->type, module);
701 				if (curr_map == NULL) {
702 					pr_debug("%s/proc/{kallsyms,modules} "
703 					         "inconsistency while looking "
704 						 "for \"%s\" module!\n",
705 						 machine->root_dir, module);
706 					curr_map = map;
707 					goto discard_symbol;
708 				}
709 
710 				if (curr_map->dso->loaded &&
711 				    !machine__is_default_guest(machine))
712 					goto discard_symbol;
713 			}
714 			/*
715 			 * So that we look just like we get from .ko files,
716 			 * i.e. not prelinked, relative to map->start.
717 			 */
718 			pos->start = curr_map->map_ip(curr_map, pos->start);
719 			pos->end   = curr_map->map_ip(curr_map, pos->end);
720 		} else if (curr_map != map) {
721 			char dso_name[PATH_MAX];
722 			struct dso *ndso;
723 
724 			if (count == 0) {
725 				curr_map = map;
726 				goto filter_symbol;
727 			}
728 
729 			if (dso->kernel == DSO_TYPE_GUEST_KERNEL)
730 				snprintf(dso_name, sizeof(dso_name),
731 					"[guest.kernel].%d",
732 					kernel_range++);
733 			else
734 				snprintf(dso_name, sizeof(dso_name),
735 					"[kernel].%d",
736 					kernel_range++);
737 
738 			ndso = dso__new(dso_name);
739 			if (ndso == NULL)
740 				return -1;
741 
742 			ndso->kernel = dso->kernel;
743 
744 			curr_map = map__new2(pos->start, ndso, map->type);
745 			if (curr_map == NULL) {
746 				dso__delete(ndso);
747 				return -1;
748 			}
749 
750 			curr_map->map_ip = curr_map->unmap_ip = identity__map_ip;
751 			map_groups__insert(kmaps, curr_map);
752 			++kernel_range;
753 		}
754 filter_symbol:
755 		if (filter && filter(curr_map, pos)) {
756 discard_symbol:		rb_erase(&pos->rb_node, root);
757 			symbol__delete(pos);
758 		} else {
759 			if (curr_map != map) {
760 				rb_erase(&pos->rb_node, root);
761 				symbols__insert(&curr_map->dso->symbols[curr_map->type], pos);
762 				++moved;
763 			} else
764 				++count;
765 		}
766 	}
767 
768 	if (curr_map != map &&
769 	    dso->kernel == DSO_TYPE_GUEST_KERNEL &&
770 	    machine__is_default_guest(kmaps->machine)) {
771 		dso__set_loaded(curr_map->dso, curr_map->type);
772 	}
773 
774 	return count + moved;
775 }
776 
777 static bool symbol__restricted_filename(const char *filename,
778 					const char *restricted_filename)
779 {
780 	bool restricted = false;
781 
782 	if (symbol_conf.kptr_restrict) {
783 		char *r = realpath(filename, NULL);
784 
785 		if (r != NULL) {
786 			restricted = strcmp(r, restricted_filename) == 0;
787 			free(r);
788 			return restricted;
789 		}
790 	}
791 
792 	return restricted;
793 }
794 
795 int dso__load_kallsyms(struct dso *dso, const char *filename,
796 		       struct map *map, symbol_filter_t filter)
797 {
798 	if (symbol__restricted_filename(filename, "/proc/kallsyms"))
799 		return -1;
800 
801 	if (dso__load_all_kallsyms(dso, filename, map) < 0)
802 		return -1;
803 
804 	symbols__fixup_duplicate(&dso->symbols[map->type]);
805 	symbols__fixup_end(&dso->symbols[map->type]);
806 
807 	if (dso->kernel == DSO_TYPE_GUEST_KERNEL)
808 		dso->symtab_type = SYMTAB__GUEST_KALLSYMS;
809 	else
810 		dso->symtab_type = SYMTAB__KALLSYMS;
811 
812 	return dso__split_kallsyms(dso, map, filter);
813 }
814 
815 static int dso__load_perf_map(struct dso *dso, struct map *map,
816 			      symbol_filter_t filter)
817 {
818 	char *line = NULL;
819 	size_t n;
820 	FILE *file;
821 	int nr_syms = 0;
822 
823 	file = fopen(dso->long_name, "r");
824 	if (file == NULL)
825 		goto out_failure;
826 
827 	while (!feof(file)) {
828 		u64 start, size;
829 		struct symbol *sym;
830 		int line_len, len;
831 
832 		line_len = getline(&line, &n, file);
833 		if (line_len < 0)
834 			break;
835 
836 		if (!line)
837 			goto out_failure;
838 
839 		line[--line_len] = '\0'; /* \n */
840 
841 		len = hex2u64(line, &start);
842 
843 		len++;
844 		if (len + 2 >= line_len)
845 			continue;
846 
847 		len += hex2u64(line + len, &size);
848 
849 		len++;
850 		if (len + 2 >= line_len)
851 			continue;
852 
853 		sym = symbol__new(start, size, STB_GLOBAL, line + len);
854 
855 		if (sym == NULL)
856 			goto out_delete_line;
857 
858 		if (filter && filter(map, sym))
859 			symbol__delete(sym);
860 		else {
861 			symbols__insert(&dso->symbols[map->type], sym);
862 			nr_syms++;
863 		}
864 	}
865 
866 	free(line);
867 	fclose(file);
868 
869 	return nr_syms;
870 
871 out_delete_line:
872 	free(line);
873 out_failure:
874 	return -1;
875 }
876 
877 /**
878  * elf_symtab__for_each_symbol - iterate thru all the symbols
879  *
880  * @syms: struct elf_symtab instance to iterate
881  * @idx: uint32_t idx
882  * @sym: GElf_Sym iterator
883  */
884 #define elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) \
885 	for (idx = 0, gelf_getsym(syms, idx, &sym);\
886 	     idx < nr_syms; \
887 	     idx++, gelf_getsym(syms, idx, &sym))
888 
889 static inline uint8_t elf_sym__type(const GElf_Sym *sym)
890 {
891 	return GELF_ST_TYPE(sym->st_info);
892 }
893 
894 static inline int elf_sym__is_function(const GElf_Sym *sym)
895 {
896 	return elf_sym__type(sym) == STT_FUNC &&
897 	       sym->st_name != 0 &&
898 	       sym->st_shndx != SHN_UNDEF;
899 }
900 
901 static inline bool elf_sym__is_object(const GElf_Sym *sym)
902 {
903 	return elf_sym__type(sym) == STT_OBJECT &&
904 		sym->st_name != 0 &&
905 		sym->st_shndx != SHN_UNDEF;
906 }
907 
908 static inline int elf_sym__is_label(const GElf_Sym *sym)
909 {
910 	return elf_sym__type(sym) == STT_NOTYPE &&
911 		sym->st_name != 0 &&
912 		sym->st_shndx != SHN_UNDEF &&
913 		sym->st_shndx != SHN_ABS;
914 }
915 
916 static inline const char *elf_sec__name(const GElf_Shdr *shdr,
917 					const Elf_Data *secstrs)
918 {
919 	return secstrs->d_buf + shdr->sh_name;
920 }
921 
922 static inline int elf_sec__is_text(const GElf_Shdr *shdr,
923 					const Elf_Data *secstrs)
924 {
925 	return strstr(elf_sec__name(shdr, secstrs), "text") != NULL;
926 }
927 
928 static inline bool elf_sec__is_data(const GElf_Shdr *shdr,
929 				    const Elf_Data *secstrs)
930 {
931 	return strstr(elf_sec__name(shdr, secstrs), "data") != NULL;
932 }
933 
934 static inline const char *elf_sym__name(const GElf_Sym *sym,
935 					const Elf_Data *symstrs)
936 {
937 	return symstrs->d_buf + sym->st_name;
938 }
939 
940 static Elf_Scn *elf_section_by_name(Elf *elf, GElf_Ehdr *ep,
941 				    GElf_Shdr *shp, const char *name,
942 				    size_t *idx)
943 {
944 	Elf_Scn *sec = NULL;
945 	size_t cnt = 1;
946 
947 	while ((sec = elf_nextscn(elf, sec)) != NULL) {
948 		char *str;
949 
950 		gelf_getshdr(sec, shp);
951 		str = elf_strptr(elf, ep->e_shstrndx, shp->sh_name);
952 		if (!strcmp(name, str)) {
953 			if (idx)
954 				*idx = cnt;
955 			break;
956 		}
957 		++cnt;
958 	}
959 
960 	return sec;
961 }
962 
963 #define elf_section__for_each_rel(reldata, pos, pos_mem, idx, nr_entries) \
964 	for (idx = 0, pos = gelf_getrel(reldata, 0, &pos_mem); \
965 	     idx < nr_entries; \
966 	     ++idx, pos = gelf_getrel(reldata, idx, &pos_mem))
967 
968 #define elf_section__for_each_rela(reldata, pos, pos_mem, idx, nr_entries) \
969 	for (idx = 0, pos = gelf_getrela(reldata, 0, &pos_mem); \
970 	     idx < nr_entries; \
971 	     ++idx, pos = gelf_getrela(reldata, idx, &pos_mem))
972 
973 /*
974  * We need to check if we have a .dynsym, so that we can handle the
975  * .plt, synthesizing its symbols, that aren't on the symtabs (be it
976  * .dynsym or .symtab).
977  * And always look at the original dso, not at debuginfo packages, that
978  * have the PLT data stripped out (shdr_rel_plt.sh_type == SHT_NOBITS).
979  */
980 static int dso__synthesize_plt_symbols(struct  dso *dso, struct map *map,
981 				       symbol_filter_t filter)
982 {
983 	uint32_t nr_rel_entries, idx;
984 	GElf_Sym sym;
985 	u64 plt_offset;
986 	GElf_Shdr shdr_plt;
987 	struct symbol *f;
988 	GElf_Shdr shdr_rel_plt, shdr_dynsym;
989 	Elf_Data *reldata, *syms, *symstrs;
990 	Elf_Scn *scn_plt_rel, *scn_symstrs, *scn_dynsym;
991 	size_t dynsym_idx;
992 	GElf_Ehdr ehdr;
993 	char sympltname[1024];
994 	Elf *elf;
995 	int nr = 0, symidx, fd, err = 0;
996 	char name[PATH_MAX];
997 
998 	snprintf(name, sizeof(name), "%s%s",
999 		 symbol_conf.symfs, dso->long_name);
1000 	fd = open(name, O_RDONLY);
1001 	if (fd < 0)
1002 		goto out;
1003 
1004 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1005 	if (elf == NULL)
1006 		goto out_close;
1007 
1008 	if (gelf_getehdr(elf, &ehdr) == NULL)
1009 		goto out_elf_end;
1010 
1011 	scn_dynsym = elf_section_by_name(elf, &ehdr, &shdr_dynsym,
1012 					 ".dynsym", &dynsym_idx);
1013 	if (scn_dynsym == NULL)
1014 		goto out_elf_end;
1015 
1016 	scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
1017 					  ".rela.plt", NULL);
1018 	if (scn_plt_rel == NULL) {
1019 		scn_plt_rel = elf_section_by_name(elf, &ehdr, &shdr_rel_plt,
1020 						  ".rel.plt", NULL);
1021 		if (scn_plt_rel == NULL)
1022 			goto out_elf_end;
1023 	}
1024 
1025 	err = -1;
1026 
1027 	if (shdr_rel_plt.sh_link != dynsym_idx)
1028 		goto out_elf_end;
1029 
1030 	if (elf_section_by_name(elf, &ehdr, &shdr_plt, ".plt", NULL) == NULL)
1031 		goto out_elf_end;
1032 
1033 	/*
1034 	 * Fetch the relocation section to find the idxes to the GOT
1035 	 * and the symbols in the .dynsym they refer to.
1036 	 */
1037 	reldata = elf_getdata(scn_plt_rel, NULL);
1038 	if (reldata == NULL)
1039 		goto out_elf_end;
1040 
1041 	syms = elf_getdata(scn_dynsym, NULL);
1042 	if (syms == NULL)
1043 		goto out_elf_end;
1044 
1045 	scn_symstrs = elf_getscn(elf, shdr_dynsym.sh_link);
1046 	if (scn_symstrs == NULL)
1047 		goto out_elf_end;
1048 
1049 	symstrs = elf_getdata(scn_symstrs, NULL);
1050 	if (symstrs == NULL)
1051 		goto out_elf_end;
1052 
1053 	nr_rel_entries = shdr_rel_plt.sh_size / shdr_rel_plt.sh_entsize;
1054 	plt_offset = shdr_plt.sh_offset;
1055 
1056 	if (shdr_rel_plt.sh_type == SHT_RELA) {
1057 		GElf_Rela pos_mem, *pos;
1058 
1059 		elf_section__for_each_rela(reldata, pos, pos_mem, idx,
1060 					   nr_rel_entries) {
1061 			symidx = GELF_R_SYM(pos->r_info);
1062 			plt_offset += shdr_plt.sh_entsize;
1063 			gelf_getsym(syms, symidx, &sym);
1064 			snprintf(sympltname, sizeof(sympltname),
1065 				 "%s@plt", elf_sym__name(&sym, symstrs));
1066 
1067 			f = symbol__new(plt_offset, shdr_plt.sh_entsize,
1068 					STB_GLOBAL, sympltname);
1069 			if (!f)
1070 				goto out_elf_end;
1071 
1072 			if (filter && filter(map, f))
1073 				symbol__delete(f);
1074 			else {
1075 				symbols__insert(&dso->symbols[map->type], f);
1076 				++nr;
1077 			}
1078 		}
1079 	} else if (shdr_rel_plt.sh_type == SHT_REL) {
1080 		GElf_Rel pos_mem, *pos;
1081 		elf_section__for_each_rel(reldata, pos, pos_mem, idx,
1082 					  nr_rel_entries) {
1083 			symidx = GELF_R_SYM(pos->r_info);
1084 			plt_offset += shdr_plt.sh_entsize;
1085 			gelf_getsym(syms, symidx, &sym);
1086 			snprintf(sympltname, sizeof(sympltname),
1087 				 "%s@plt", elf_sym__name(&sym, symstrs));
1088 
1089 			f = symbol__new(plt_offset, shdr_plt.sh_entsize,
1090 					STB_GLOBAL, sympltname);
1091 			if (!f)
1092 				goto out_elf_end;
1093 
1094 			if (filter && filter(map, f))
1095 				symbol__delete(f);
1096 			else {
1097 				symbols__insert(&dso->symbols[map->type], f);
1098 				++nr;
1099 			}
1100 		}
1101 	}
1102 
1103 	err = 0;
1104 out_elf_end:
1105 	elf_end(elf);
1106 out_close:
1107 	close(fd);
1108 
1109 	if (err == 0)
1110 		return nr;
1111 out:
1112 	pr_debug("%s: problems reading %s PLT info.\n",
1113 		 __func__, dso->long_name);
1114 	return 0;
1115 }
1116 
1117 static bool elf_sym__is_a(GElf_Sym *sym, enum map_type type)
1118 {
1119 	switch (type) {
1120 	case MAP__FUNCTION:
1121 		return elf_sym__is_function(sym);
1122 	case MAP__VARIABLE:
1123 		return elf_sym__is_object(sym);
1124 	default:
1125 		return false;
1126 	}
1127 }
1128 
1129 static bool elf_sec__is_a(GElf_Shdr *shdr, Elf_Data *secstrs,
1130 			  enum map_type type)
1131 {
1132 	switch (type) {
1133 	case MAP__FUNCTION:
1134 		return elf_sec__is_text(shdr, secstrs);
1135 	case MAP__VARIABLE:
1136 		return elf_sec__is_data(shdr, secstrs);
1137 	default:
1138 		return false;
1139 	}
1140 }
1141 
1142 static size_t elf_addr_to_index(Elf *elf, GElf_Addr addr)
1143 {
1144 	Elf_Scn *sec = NULL;
1145 	GElf_Shdr shdr;
1146 	size_t cnt = 1;
1147 
1148 	while ((sec = elf_nextscn(elf, sec)) != NULL) {
1149 		gelf_getshdr(sec, &shdr);
1150 
1151 		if ((addr >= shdr.sh_addr) &&
1152 		    (addr < (shdr.sh_addr + shdr.sh_size)))
1153 			return cnt;
1154 
1155 		++cnt;
1156 	}
1157 
1158 	return -1;
1159 }
1160 
1161 static int dso__load_sym(struct dso *dso, struct map *map, const char *name,
1162 			 int fd, symbol_filter_t filter, int kmodule,
1163 			 int want_symtab)
1164 {
1165 	struct kmap *kmap = dso->kernel ? map__kmap(map) : NULL;
1166 	struct map *curr_map = map;
1167 	struct dso *curr_dso = dso;
1168 	Elf_Data *symstrs, *secstrs;
1169 	uint32_t nr_syms;
1170 	int err = -1;
1171 	uint32_t idx;
1172 	GElf_Ehdr ehdr;
1173 	GElf_Shdr shdr, opdshdr;
1174 	Elf_Data *syms, *opddata = NULL;
1175 	GElf_Sym sym;
1176 	Elf_Scn *sec, *sec_strndx, *opdsec;
1177 	Elf *elf;
1178 	int nr = 0;
1179 	size_t opdidx = 0;
1180 
1181 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1182 	if (elf == NULL) {
1183 		pr_debug("%s: cannot read %s ELF file.\n", __func__, name);
1184 		goto out_close;
1185 	}
1186 
1187 	if (gelf_getehdr(elf, &ehdr) == NULL) {
1188 		pr_debug("%s: cannot get elf header.\n", __func__);
1189 		goto out_elf_end;
1190 	}
1191 
1192 	/* Always reject images with a mismatched build-id: */
1193 	if (dso->has_build_id) {
1194 		u8 build_id[BUILD_ID_SIZE];
1195 
1196 		if (elf_read_build_id(elf, build_id, BUILD_ID_SIZE) < 0)
1197 			goto out_elf_end;
1198 
1199 		if (!dso__build_id_equal(dso, build_id))
1200 			goto out_elf_end;
1201 	}
1202 
1203 	sec = elf_section_by_name(elf, &ehdr, &shdr, ".symtab", NULL);
1204 	if (sec == NULL) {
1205 		if (want_symtab)
1206 			goto out_elf_end;
1207 
1208 		sec = elf_section_by_name(elf, &ehdr, &shdr, ".dynsym", NULL);
1209 		if (sec == NULL)
1210 			goto out_elf_end;
1211 	}
1212 
1213 	opdsec = elf_section_by_name(elf, &ehdr, &opdshdr, ".opd", &opdidx);
1214 	if (opdshdr.sh_type != SHT_PROGBITS)
1215 		opdsec = NULL;
1216 	if (opdsec)
1217 		opddata = elf_rawdata(opdsec, NULL);
1218 
1219 	syms = elf_getdata(sec, NULL);
1220 	if (syms == NULL)
1221 		goto out_elf_end;
1222 
1223 	sec = elf_getscn(elf, shdr.sh_link);
1224 	if (sec == NULL)
1225 		goto out_elf_end;
1226 
1227 	symstrs = elf_getdata(sec, NULL);
1228 	if (symstrs == NULL)
1229 		goto out_elf_end;
1230 
1231 	sec_strndx = elf_getscn(elf, ehdr.e_shstrndx);
1232 	if (sec_strndx == NULL)
1233 		goto out_elf_end;
1234 
1235 	secstrs = elf_getdata(sec_strndx, NULL);
1236 	if (secstrs == NULL)
1237 		goto out_elf_end;
1238 
1239 	nr_syms = shdr.sh_size / shdr.sh_entsize;
1240 
1241 	memset(&sym, 0, sizeof(sym));
1242 	if (dso->kernel == DSO_TYPE_USER) {
1243 		dso->adjust_symbols = (ehdr.e_type == ET_EXEC ||
1244 				elf_section_by_name(elf, &ehdr, &shdr,
1245 						     ".gnu.prelink_undo",
1246 						     NULL) != NULL);
1247 	} else {
1248 		dso->adjust_symbols = 0;
1249 	}
1250 	elf_symtab__for_each_symbol(syms, nr_syms, idx, sym) {
1251 		struct symbol *f;
1252 		const char *elf_name = elf_sym__name(&sym, symstrs);
1253 		char *demangled = NULL;
1254 		int is_label = elf_sym__is_label(&sym);
1255 		const char *section_name;
1256 
1257 		if (kmap && kmap->ref_reloc_sym && kmap->ref_reloc_sym->name &&
1258 		    strcmp(elf_name, kmap->ref_reloc_sym->name) == 0)
1259 			kmap->ref_reloc_sym->unrelocated_addr = sym.st_value;
1260 
1261 		if (!is_label && !elf_sym__is_a(&sym, map->type))
1262 			continue;
1263 
1264 		/* Reject ARM ELF "mapping symbols": these aren't unique and
1265 		 * don't identify functions, so will confuse the profile
1266 		 * output: */
1267 		if (ehdr.e_machine == EM_ARM) {
1268 			if (!strcmp(elf_name, "$a") ||
1269 			    !strcmp(elf_name, "$d") ||
1270 			    !strcmp(elf_name, "$t"))
1271 				continue;
1272 		}
1273 
1274 		if (opdsec && sym.st_shndx == opdidx) {
1275 			u32 offset = sym.st_value - opdshdr.sh_addr;
1276 			u64 *opd = opddata->d_buf + offset;
1277 			sym.st_value = *opd;
1278 			sym.st_shndx = elf_addr_to_index(elf, sym.st_value);
1279 		}
1280 
1281 		sec = elf_getscn(elf, sym.st_shndx);
1282 		if (!sec)
1283 			goto out_elf_end;
1284 
1285 		gelf_getshdr(sec, &shdr);
1286 
1287 		if (is_label && !elf_sec__is_a(&shdr, secstrs, map->type))
1288 			continue;
1289 
1290 		section_name = elf_sec__name(&shdr, secstrs);
1291 
1292 		/* On ARM, symbols for thumb functions have 1 added to
1293 		 * the symbol address as a flag - remove it */
1294 		if ((ehdr.e_machine == EM_ARM) &&
1295 		    (map->type == MAP__FUNCTION) &&
1296 		    (sym.st_value & 1))
1297 			--sym.st_value;
1298 
1299 		if (dso->kernel != DSO_TYPE_USER || kmodule) {
1300 			char dso_name[PATH_MAX];
1301 
1302 			if (strcmp(section_name,
1303 				   (curr_dso->short_name +
1304 				    dso->short_name_len)) == 0)
1305 				goto new_symbol;
1306 
1307 			if (strcmp(section_name, ".text") == 0) {
1308 				curr_map = map;
1309 				curr_dso = dso;
1310 				goto new_symbol;
1311 			}
1312 
1313 			snprintf(dso_name, sizeof(dso_name),
1314 				 "%s%s", dso->short_name, section_name);
1315 
1316 			curr_map = map_groups__find_by_name(kmap->kmaps, map->type, dso_name);
1317 			if (curr_map == NULL) {
1318 				u64 start = sym.st_value;
1319 
1320 				if (kmodule)
1321 					start += map->start + shdr.sh_offset;
1322 
1323 				curr_dso = dso__new(dso_name);
1324 				if (curr_dso == NULL)
1325 					goto out_elf_end;
1326 				curr_dso->kernel = dso->kernel;
1327 				curr_dso->long_name = dso->long_name;
1328 				curr_dso->long_name_len = dso->long_name_len;
1329 				curr_map = map__new2(start, curr_dso,
1330 						     map->type);
1331 				if (curr_map == NULL) {
1332 					dso__delete(curr_dso);
1333 					goto out_elf_end;
1334 				}
1335 				curr_map->map_ip = identity__map_ip;
1336 				curr_map->unmap_ip = identity__map_ip;
1337 				curr_dso->symtab_type = dso->symtab_type;
1338 				map_groups__insert(kmap->kmaps, curr_map);
1339 				dsos__add(&dso->node, curr_dso);
1340 				dso__set_loaded(curr_dso, map->type);
1341 			} else
1342 				curr_dso = curr_map->dso;
1343 
1344 			goto new_symbol;
1345 		}
1346 
1347 		if (curr_dso->adjust_symbols) {
1348 			pr_debug4("%s: adjusting symbol: st_value: %#" PRIx64 " "
1349 				  "sh_addr: %#" PRIx64 " sh_offset: %#" PRIx64 "\n", __func__,
1350 				  (u64)sym.st_value, (u64)shdr.sh_addr,
1351 				  (u64)shdr.sh_offset);
1352 			sym.st_value -= shdr.sh_addr - shdr.sh_offset;
1353 		}
1354 		/*
1355 		 * We need to figure out if the object was created from C++ sources
1356 		 * DWARF DW_compile_unit has this, but we don't always have access
1357 		 * to it...
1358 		 */
1359 		demangled = bfd_demangle(NULL, elf_name, DMGL_PARAMS | DMGL_ANSI);
1360 		if (demangled != NULL)
1361 			elf_name = demangled;
1362 new_symbol:
1363 		f = symbol__new(sym.st_value, sym.st_size,
1364 				GELF_ST_BIND(sym.st_info), elf_name);
1365 		free(demangled);
1366 		if (!f)
1367 			goto out_elf_end;
1368 
1369 		if (filter && filter(curr_map, f))
1370 			symbol__delete(f);
1371 		else {
1372 			symbols__insert(&curr_dso->symbols[curr_map->type], f);
1373 			nr++;
1374 		}
1375 	}
1376 
1377 	/*
1378 	 * For misannotated, zeroed, ASM function sizes.
1379 	 */
1380 	if (nr > 0) {
1381 		symbols__fixup_duplicate(&dso->symbols[map->type]);
1382 		symbols__fixup_end(&dso->symbols[map->type]);
1383 		if (kmap) {
1384 			/*
1385 			 * We need to fixup this here too because we create new
1386 			 * maps here, for things like vsyscall sections.
1387 			 */
1388 			__map_groups__fixup_end(kmap->kmaps, map->type);
1389 		}
1390 	}
1391 	err = nr;
1392 out_elf_end:
1393 	elf_end(elf);
1394 out_close:
1395 	return err;
1396 }
1397 
1398 static bool dso__build_id_equal(const struct dso *dso, u8 *build_id)
1399 {
1400 	return memcmp(dso->build_id, build_id, sizeof(dso->build_id)) == 0;
1401 }
1402 
1403 bool __dsos__read_build_ids(struct list_head *head, bool with_hits)
1404 {
1405 	bool have_build_id = false;
1406 	struct dso *pos;
1407 
1408 	list_for_each_entry(pos, head, node) {
1409 		if (with_hits && !pos->hit)
1410 			continue;
1411 		if (pos->has_build_id) {
1412 			have_build_id = true;
1413 			continue;
1414 		}
1415 		if (filename__read_build_id(pos->long_name, pos->build_id,
1416 					    sizeof(pos->build_id)) > 0) {
1417 			have_build_id	  = true;
1418 			pos->has_build_id = true;
1419 		}
1420 	}
1421 
1422 	return have_build_id;
1423 }
1424 
1425 /*
1426  * Align offset to 4 bytes as needed for note name and descriptor data.
1427  */
1428 #define NOTE_ALIGN(n) (((n) + 3) & -4U)
1429 
1430 static int elf_read_build_id(Elf *elf, void *bf, size_t size)
1431 {
1432 	int err = -1;
1433 	GElf_Ehdr ehdr;
1434 	GElf_Shdr shdr;
1435 	Elf_Data *data;
1436 	Elf_Scn *sec;
1437 	Elf_Kind ek;
1438 	void *ptr;
1439 
1440 	if (size < BUILD_ID_SIZE)
1441 		goto out;
1442 
1443 	ek = elf_kind(elf);
1444 	if (ek != ELF_K_ELF)
1445 		goto out;
1446 
1447 	if (gelf_getehdr(elf, &ehdr) == NULL) {
1448 		pr_err("%s: cannot get elf header.\n", __func__);
1449 		goto out;
1450 	}
1451 
1452 	sec = elf_section_by_name(elf, &ehdr, &shdr,
1453 				  ".note.gnu.build-id", NULL);
1454 	if (sec == NULL) {
1455 		sec = elf_section_by_name(elf, &ehdr, &shdr,
1456 					  ".notes", NULL);
1457 		if (sec == NULL)
1458 			goto out;
1459 	}
1460 
1461 	data = elf_getdata(sec, NULL);
1462 	if (data == NULL)
1463 		goto out;
1464 
1465 	ptr = data->d_buf;
1466 	while (ptr < (data->d_buf + data->d_size)) {
1467 		GElf_Nhdr *nhdr = ptr;
1468 		size_t namesz = NOTE_ALIGN(nhdr->n_namesz),
1469 		       descsz = NOTE_ALIGN(nhdr->n_descsz);
1470 		const char *name;
1471 
1472 		ptr += sizeof(*nhdr);
1473 		name = ptr;
1474 		ptr += namesz;
1475 		if (nhdr->n_type == NT_GNU_BUILD_ID &&
1476 		    nhdr->n_namesz == sizeof("GNU")) {
1477 			if (memcmp(name, "GNU", sizeof("GNU")) == 0) {
1478 				size_t sz = min(size, descsz);
1479 				memcpy(bf, ptr, sz);
1480 				memset(bf + sz, 0, size - sz);
1481 				err = descsz;
1482 				break;
1483 			}
1484 		}
1485 		ptr += descsz;
1486 	}
1487 
1488 out:
1489 	return err;
1490 }
1491 
1492 int filename__read_build_id(const char *filename, void *bf, size_t size)
1493 {
1494 	int fd, err = -1;
1495 	Elf *elf;
1496 
1497 	if (size < BUILD_ID_SIZE)
1498 		goto out;
1499 
1500 	fd = open(filename, O_RDONLY);
1501 	if (fd < 0)
1502 		goto out;
1503 
1504 	elf = elf_begin(fd, PERF_ELF_C_READ_MMAP, NULL);
1505 	if (elf == NULL) {
1506 		pr_debug2("%s: cannot read %s ELF file.\n", __func__, filename);
1507 		goto out_close;
1508 	}
1509 
1510 	err = elf_read_build_id(elf, bf, size);
1511 
1512 	elf_end(elf);
1513 out_close:
1514 	close(fd);
1515 out:
1516 	return err;
1517 }
1518 
1519 int sysfs__read_build_id(const char *filename, void *build_id, size_t size)
1520 {
1521 	int fd, err = -1;
1522 
1523 	if (size < BUILD_ID_SIZE)
1524 		goto out;
1525 
1526 	fd = open(filename, O_RDONLY);
1527 	if (fd < 0)
1528 		goto out;
1529 
1530 	while (1) {
1531 		char bf[BUFSIZ];
1532 		GElf_Nhdr nhdr;
1533 		size_t namesz, descsz;
1534 
1535 		if (read(fd, &nhdr, sizeof(nhdr)) != sizeof(nhdr))
1536 			break;
1537 
1538 		namesz = NOTE_ALIGN(nhdr.n_namesz);
1539 		descsz = NOTE_ALIGN(nhdr.n_descsz);
1540 		if (nhdr.n_type == NT_GNU_BUILD_ID &&
1541 		    nhdr.n_namesz == sizeof("GNU")) {
1542 			if (read(fd, bf, namesz) != (ssize_t)namesz)
1543 				break;
1544 			if (memcmp(bf, "GNU", sizeof("GNU")) == 0) {
1545 				size_t sz = min(descsz, size);
1546 				if (read(fd, build_id, sz) == (ssize_t)sz) {
1547 					memset(build_id + sz, 0, size - sz);
1548 					err = 0;
1549 					break;
1550 				}
1551 			} else if (read(fd, bf, descsz) != (ssize_t)descsz)
1552 				break;
1553 		} else {
1554 			int n = namesz + descsz;
1555 			if (read(fd, bf, n) != n)
1556 				break;
1557 		}
1558 	}
1559 	close(fd);
1560 out:
1561 	return err;
1562 }
1563 
1564 char dso__symtab_origin(const struct dso *dso)
1565 {
1566 	static const char origin[] = {
1567 		[SYMTAB__KALLSYMS]	      = 'k',
1568 		[SYMTAB__JAVA_JIT]	      = 'j',
1569 		[SYMTAB__BUILD_ID_CACHE]      = 'B',
1570 		[SYMTAB__FEDORA_DEBUGINFO]    = 'f',
1571 		[SYMTAB__UBUNTU_DEBUGINFO]    = 'u',
1572 		[SYMTAB__BUILDID_DEBUGINFO]   = 'b',
1573 		[SYMTAB__SYSTEM_PATH_DSO]     = 'd',
1574 		[SYMTAB__SYSTEM_PATH_KMODULE] = 'K',
1575 		[SYMTAB__GUEST_KALLSYMS]      =  'g',
1576 		[SYMTAB__GUEST_KMODULE]	      =  'G',
1577 	};
1578 
1579 	if (dso == NULL || dso->symtab_type == SYMTAB__NOT_FOUND)
1580 		return '!';
1581 	return origin[dso->symtab_type];
1582 }
1583 
1584 int dso__load(struct dso *dso, struct map *map, symbol_filter_t filter)
1585 {
1586 	int size = PATH_MAX;
1587 	char *name;
1588 	int ret = -1;
1589 	int fd;
1590 	struct machine *machine;
1591 	const char *root_dir;
1592 	int want_symtab;
1593 
1594 	dso__set_loaded(dso, map->type);
1595 
1596 	if (dso->kernel == DSO_TYPE_KERNEL)
1597 		return dso__load_kernel_sym(dso, map, filter);
1598 	else if (dso->kernel == DSO_TYPE_GUEST_KERNEL)
1599 		return dso__load_guest_kernel_sym(dso, map, filter);
1600 
1601 	if (map->groups && map->groups->machine)
1602 		machine = map->groups->machine;
1603 	else
1604 		machine = NULL;
1605 
1606 	name = malloc(size);
1607 	if (!name)
1608 		return -1;
1609 
1610 	dso->adjust_symbols = 0;
1611 
1612 	if (strncmp(dso->name, "/tmp/perf-", 10) == 0) {
1613 		struct stat st;
1614 
1615 		if (lstat(dso->name, &st) < 0)
1616 			return -1;
1617 
1618 		if (st.st_uid && (st.st_uid != geteuid())) {
1619 			pr_warning("File %s not owned by current user or root, "
1620 				"ignoring it.\n", dso->name);
1621 			return -1;
1622 		}
1623 
1624 		ret = dso__load_perf_map(dso, map, filter);
1625 		dso->symtab_type = ret > 0 ? SYMTAB__JAVA_JIT :
1626 					      SYMTAB__NOT_FOUND;
1627 		return ret;
1628 	}
1629 
1630 	/* Iterate over candidate debug images.
1631 	 * On the first pass, only load images if they have a full symtab.
1632 	 * Failing that, do a second pass where we accept .dynsym also
1633 	 */
1634 	want_symtab = 1;
1635 restart:
1636 	for (dso->symtab_type = SYMTAB__BUILD_ID_CACHE;
1637 	     dso->symtab_type != SYMTAB__NOT_FOUND;
1638 	     dso->symtab_type++) {
1639 		switch (dso->symtab_type) {
1640 		case SYMTAB__BUILD_ID_CACHE:
1641 			/* skip the locally configured cache if a symfs is given */
1642 			if (symbol_conf.symfs[0] ||
1643 			    (dso__build_id_filename(dso, name, size) == NULL)) {
1644 				continue;
1645 			}
1646 			break;
1647 		case SYMTAB__FEDORA_DEBUGINFO:
1648 			snprintf(name, size, "%s/usr/lib/debug%s.debug",
1649 				 symbol_conf.symfs, dso->long_name);
1650 			break;
1651 		case SYMTAB__UBUNTU_DEBUGINFO:
1652 			snprintf(name, size, "%s/usr/lib/debug%s",
1653 				 symbol_conf.symfs, dso->long_name);
1654 			break;
1655 		case SYMTAB__BUILDID_DEBUGINFO: {
1656 			char build_id_hex[BUILD_ID_SIZE * 2 + 1];
1657 
1658 			if (!dso->has_build_id)
1659 				continue;
1660 
1661 			build_id__sprintf(dso->build_id,
1662 					  sizeof(dso->build_id),
1663 					  build_id_hex);
1664 			snprintf(name, size,
1665 				 "%s/usr/lib/debug/.build-id/%.2s/%s.debug",
1666 				 symbol_conf.symfs, build_id_hex, build_id_hex + 2);
1667 			}
1668 			break;
1669 		case SYMTAB__SYSTEM_PATH_DSO:
1670 			snprintf(name, size, "%s%s",
1671 			     symbol_conf.symfs, dso->long_name);
1672 			break;
1673 		case SYMTAB__GUEST_KMODULE:
1674 			if (map->groups && machine)
1675 				root_dir = machine->root_dir;
1676 			else
1677 				root_dir = "";
1678 			snprintf(name, size, "%s%s%s", symbol_conf.symfs,
1679 				 root_dir, dso->long_name);
1680 			break;
1681 
1682 		case SYMTAB__SYSTEM_PATH_KMODULE:
1683 			snprintf(name, size, "%s%s", symbol_conf.symfs,
1684 				 dso->long_name);
1685 			break;
1686 		default:;
1687 		}
1688 
1689 		/* Name is now the name of the next image to try */
1690 		fd = open(name, O_RDONLY);
1691 		if (fd < 0)
1692 			continue;
1693 
1694 		ret = dso__load_sym(dso, map, name, fd, filter, 0,
1695 				    want_symtab);
1696 		close(fd);
1697 
1698 		/*
1699 		 * Some people seem to have debuginfo files _WITHOUT_ debug
1700 		 * info!?!?
1701 		 */
1702 		if (!ret)
1703 			continue;
1704 
1705 		if (ret > 0) {
1706 			int nr_plt = dso__synthesize_plt_symbols(dso, map,
1707 								 filter);
1708 			if (nr_plt > 0)
1709 				ret += nr_plt;
1710 			break;
1711 		}
1712 	}
1713 
1714 	/*
1715 	 * If we wanted a full symtab but no image had one,
1716 	 * relax our requirements and repeat the search.
1717 	 */
1718 	if (ret <= 0 && want_symtab) {
1719 		want_symtab = 0;
1720 		goto restart;
1721 	}
1722 
1723 	free(name);
1724 	if (ret < 0 && strstr(dso->name, " (deleted)") != NULL)
1725 		return 0;
1726 	return ret;
1727 }
1728 
1729 struct map *map_groups__find_by_name(struct map_groups *mg,
1730 				     enum map_type type, const char *name)
1731 {
1732 	struct rb_node *nd;
1733 
1734 	for (nd = rb_first(&mg->maps[type]); nd; nd = rb_next(nd)) {
1735 		struct map *map = rb_entry(nd, struct map, rb_node);
1736 
1737 		if (map->dso && strcmp(map->dso->short_name, name) == 0)
1738 			return map;
1739 	}
1740 
1741 	return NULL;
1742 }
1743 
1744 static int dso__kernel_module_get_build_id(struct dso *dso,
1745 					   const char *root_dir)
1746 {
1747 	char filename[PATH_MAX];
1748 	/*
1749 	 * kernel module short names are of the form "[module]" and
1750 	 * we need just "module" here.
1751 	 */
1752 	const char *name = dso->short_name + 1;
1753 
1754 	snprintf(filename, sizeof(filename),
1755 		 "%s/sys/module/%.*s/notes/.note.gnu.build-id",
1756 		 root_dir, (int)strlen(name) - 1, name);
1757 
1758 	if (sysfs__read_build_id(filename, dso->build_id,
1759 				 sizeof(dso->build_id)) == 0)
1760 		dso->has_build_id = true;
1761 
1762 	return 0;
1763 }
1764 
1765 static int map_groups__set_modules_path_dir(struct map_groups *mg,
1766 				const char *dir_name)
1767 {
1768 	struct dirent *dent;
1769 	DIR *dir = opendir(dir_name);
1770 	int ret = 0;
1771 
1772 	if (!dir) {
1773 		pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
1774 		return -1;
1775 	}
1776 
1777 	while ((dent = readdir(dir)) != NULL) {
1778 		char path[PATH_MAX];
1779 		struct stat st;
1780 
1781 		/*sshfs might return bad dent->d_type, so we have to stat*/
1782 		snprintf(path, sizeof(path), "%s/%s", dir_name, dent->d_name);
1783 		if (stat(path, &st))
1784 			continue;
1785 
1786 		if (S_ISDIR(st.st_mode)) {
1787 			if (!strcmp(dent->d_name, ".") ||
1788 			    !strcmp(dent->d_name, ".."))
1789 				continue;
1790 
1791 			ret = map_groups__set_modules_path_dir(mg, path);
1792 			if (ret < 0)
1793 				goto out;
1794 		} else {
1795 			char *dot = strrchr(dent->d_name, '.'),
1796 			     dso_name[PATH_MAX];
1797 			struct map *map;
1798 			char *long_name;
1799 
1800 			if (dot == NULL || strcmp(dot, ".ko"))
1801 				continue;
1802 			snprintf(dso_name, sizeof(dso_name), "[%.*s]",
1803 				 (int)(dot - dent->d_name), dent->d_name);
1804 
1805 			strxfrchar(dso_name, '-', '_');
1806 			map = map_groups__find_by_name(mg, MAP__FUNCTION,
1807 						       dso_name);
1808 			if (map == NULL)
1809 				continue;
1810 
1811 			long_name = strdup(path);
1812 			if (long_name == NULL) {
1813 				ret = -1;
1814 				goto out;
1815 			}
1816 			dso__set_long_name(map->dso, long_name);
1817 			map->dso->lname_alloc = 1;
1818 			dso__kernel_module_get_build_id(map->dso, "");
1819 		}
1820 	}
1821 
1822 out:
1823 	closedir(dir);
1824 	return ret;
1825 }
1826 
1827 static char *get_kernel_version(const char *root_dir)
1828 {
1829 	char version[PATH_MAX];
1830 	FILE *file;
1831 	char *name, *tmp;
1832 	const char *prefix = "Linux version ";
1833 
1834 	sprintf(version, "%s/proc/version", root_dir);
1835 	file = fopen(version, "r");
1836 	if (!file)
1837 		return NULL;
1838 
1839 	version[0] = '\0';
1840 	tmp = fgets(version, sizeof(version), file);
1841 	fclose(file);
1842 
1843 	name = strstr(version, prefix);
1844 	if (!name)
1845 		return NULL;
1846 	name += strlen(prefix);
1847 	tmp = strchr(name, ' ');
1848 	if (tmp)
1849 		*tmp = '\0';
1850 
1851 	return strdup(name);
1852 }
1853 
1854 static int machine__set_modules_path(struct machine *machine)
1855 {
1856 	char *version;
1857 	char modules_path[PATH_MAX];
1858 
1859 	version = get_kernel_version(machine->root_dir);
1860 	if (!version)
1861 		return -1;
1862 
1863 	snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s/kernel",
1864 		 machine->root_dir, version);
1865 	free(version);
1866 
1867 	return map_groups__set_modules_path_dir(&machine->kmaps, modules_path);
1868 }
1869 
1870 /*
1871  * Constructor variant for modules (where we know from /proc/modules where
1872  * they are loaded) and for vmlinux, where only after we load all the
1873  * symbols we'll know where it starts and ends.
1874  */
1875 static struct map *map__new2(u64 start, struct dso *dso, enum map_type type)
1876 {
1877 	struct map *map = calloc(1, (sizeof(*map) +
1878 				     (dso->kernel ? sizeof(struct kmap) : 0)));
1879 	if (map != NULL) {
1880 		/*
1881 		 * ->end will be filled after we load all the symbols
1882 		 */
1883 		map__init(map, type, start, 0, 0, dso);
1884 	}
1885 
1886 	return map;
1887 }
1888 
1889 struct map *machine__new_module(struct machine *machine, u64 start,
1890 				const char *filename)
1891 {
1892 	struct map *map;
1893 	struct dso *dso = __dsos__findnew(&machine->kernel_dsos, filename);
1894 
1895 	if (dso == NULL)
1896 		return NULL;
1897 
1898 	map = map__new2(start, dso, MAP__FUNCTION);
1899 	if (map == NULL)
1900 		return NULL;
1901 
1902 	if (machine__is_host(machine))
1903 		dso->symtab_type = SYMTAB__SYSTEM_PATH_KMODULE;
1904 	else
1905 		dso->symtab_type = SYMTAB__GUEST_KMODULE;
1906 	map_groups__insert(&machine->kmaps, map);
1907 	return map;
1908 }
1909 
1910 static int machine__create_modules(struct machine *machine)
1911 {
1912 	char *line = NULL;
1913 	size_t n;
1914 	FILE *file;
1915 	struct map *map;
1916 	const char *modules;
1917 	char path[PATH_MAX];
1918 
1919 	if (machine__is_default_guest(machine))
1920 		modules = symbol_conf.default_guest_modules;
1921 	else {
1922 		sprintf(path, "%s/proc/modules", machine->root_dir);
1923 		modules = path;
1924 	}
1925 
1926 	if (symbol__restricted_filename(path, "/proc/modules"))
1927 		return -1;
1928 
1929 	file = fopen(modules, "r");
1930 	if (file == NULL)
1931 		return -1;
1932 
1933 	while (!feof(file)) {
1934 		char name[PATH_MAX];
1935 		u64 start;
1936 		char *sep;
1937 		int line_len;
1938 
1939 		line_len = getline(&line, &n, file);
1940 		if (line_len < 0)
1941 			break;
1942 
1943 		if (!line)
1944 			goto out_failure;
1945 
1946 		line[--line_len] = '\0'; /* \n */
1947 
1948 		sep = strrchr(line, 'x');
1949 		if (sep == NULL)
1950 			continue;
1951 
1952 		hex2u64(sep + 1, &start);
1953 
1954 		sep = strchr(line, ' ');
1955 		if (sep == NULL)
1956 			continue;
1957 
1958 		*sep = '\0';
1959 
1960 		snprintf(name, sizeof(name), "[%s]", line);
1961 		map = machine__new_module(machine, start, name);
1962 		if (map == NULL)
1963 			goto out_delete_line;
1964 		dso__kernel_module_get_build_id(map->dso, machine->root_dir);
1965 	}
1966 
1967 	free(line);
1968 	fclose(file);
1969 
1970 	return machine__set_modules_path(machine);
1971 
1972 out_delete_line:
1973 	free(line);
1974 out_failure:
1975 	return -1;
1976 }
1977 
1978 int dso__load_vmlinux(struct dso *dso, struct map *map,
1979 		      const char *vmlinux, symbol_filter_t filter)
1980 {
1981 	int err = -1, fd;
1982 	char symfs_vmlinux[PATH_MAX];
1983 
1984 	snprintf(symfs_vmlinux, sizeof(symfs_vmlinux), "%s%s",
1985 		 symbol_conf.symfs, vmlinux);
1986 	fd = open(symfs_vmlinux, O_RDONLY);
1987 	if (fd < 0)
1988 		return -1;
1989 
1990 	dso__set_long_name(dso, (char *)vmlinux);
1991 	dso__set_loaded(dso, map->type);
1992 	err = dso__load_sym(dso, map, symfs_vmlinux, fd, filter, 0, 0);
1993 	close(fd);
1994 
1995 	if (err > 0)
1996 		pr_debug("Using %s for symbols\n", symfs_vmlinux);
1997 
1998 	return err;
1999 }
2000 
2001 int dso__load_vmlinux_path(struct dso *dso, struct map *map,
2002 			   symbol_filter_t filter)
2003 {
2004 	int i, err = 0;
2005 	char *filename;
2006 
2007 	pr_debug("Looking at the vmlinux_path (%d entries long)\n",
2008 		 vmlinux_path__nr_entries + 1);
2009 
2010 	filename = dso__build_id_filename(dso, NULL, 0);
2011 	if (filename != NULL) {
2012 		err = dso__load_vmlinux(dso, map, filename, filter);
2013 		if (err > 0) {
2014 			dso__set_long_name(dso, filename);
2015 			goto out;
2016 		}
2017 		free(filename);
2018 	}
2019 
2020 	for (i = 0; i < vmlinux_path__nr_entries; ++i) {
2021 		err = dso__load_vmlinux(dso, map, vmlinux_path[i], filter);
2022 		if (err > 0) {
2023 			dso__set_long_name(dso, strdup(vmlinux_path[i]));
2024 			break;
2025 		}
2026 	}
2027 out:
2028 	return err;
2029 }
2030 
2031 static int dso__load_kernel_sym(struct dso *dso, struct map *map,
2032 				symbol_filter_t filter)
2033 {
2034 	int err;
2035 	const char *kallsyms_filename = NULL;
2036 	char *kallsyms_allocated_filename = NULL;
2037 	/*
2038 	 * Step 1: if the user specified a kallsyms or vmlinux filename, use
2039 	 * it and only it, reporting errors to the user if it cannot be used.
2040 	 *
2041 	 * For instance, try to analyse an ARM perf.data file _without_ a
2042 	 * build-id, or if the user specifies the wrong path to the right
2043 	 * vmlinux file, obviously we can't fallback to another vmlinux (a
2044 	 * x86_86 one, on the machine where analysis is being performed, say),
2045 	 * or worse, /proc/kallsyms.
2046 	 *
2047 	 * If the specified file _has_ a build-id and there is a build-id
2048 	 * section in the perf.data file, we will still do the expected
2049 	 * validation in dso__load_vmlinux and will bail out if they don't
2050 	 * match.
2051 	 */
2052 	if (symbol_conf.kallsyms_name != NULL) {
2053 		kallsyms_filename = symbol_conf.kallsyms_name;
2054 		goto do_kallsyms;
2055 	}
2056 
2057 	if (symbol_conf.vmlinux_name != NULL) {
2058 		err = dso__load_vmlinux(dso, map,
2059 					symbol_conf.vmlinux_name, filter);
2060 		if (err > 0) {
2061 			dso__set_long_name(dso,
2062 					   strdup(symbol_conf.vmlinux_name));
2063 			goto out_fixup;
2064 		}
2065 		return err;
2066 	}
2067 
2068 	if (vmlinux_path != NULL) {
2069 		err = dso__load_vmlinux_path(dso, map, filter);
2070 		if (err > 0)
2071 			goto out_fixup;
2072 	}
2073 
2074 	/* do not try local files if a symfs was given */
2075 	if (symbol_conf.symfs[0] != 0)
2076 		return -1;
2077 
2078 	/*
2079 	 * Say the kernel DSO was created when processing the build-id header table,
2080 	 * we have a build-id, so check if it is the same as the running kernel,
2081 	 * using it if it is.
2082 	 */
2083 	if (dso->has_build_id) {
2084 		u8 kallsyms_build_id[BUILD_ID_SIZE];
2085 		char sbuild_id[BUILD_ID_SIZE * 2 + 1];
2086 
2087 		if (sysfs__read_build_id("/sys/kernel/notes", kallsyms_build_id,
2088 					 sizeof(kallsyms_build_id)) == 0) {
2089 			if (dso__build_id_equal(dso, kallsyms_build_id)) {
2090 				kallsyms_filename = "/proc/kallsyms";
2091 				goto do_kallsyms;
2092 			}
2093 		}
2094 		/*
2095 		 * Now look if we have it on the build-id cache in
2096 		 * $HOME/.debug/[kernel.kallsyms].
2097 		 */
2098 		build_id__sprintf(dso->build_id, sizeof(dso->build_id),
2099 				  sbuild_id);
2100 
2101 		if (asprintf(&kallsyms_allocated_filename,
2102 			     "%s/.debug/[kernel.kallsyms]/%s",
2103 			     getenv("HOME"), sbuild_id) == -1) {
2104 			pr_err("Not enough memory for kallsyms file lookup\n");
2105 			return -1;
2106 		}
2107 
2108 		kallsyms_filename = kallsyms_allocated_filename;
2109 
2110 		if (access(kallsyms_filename, F_OK)) {
2111 			pr_err("No kallsyms or vmlinux with build-id %s "
2112 			       "was found\n", sbuild_id);
2113 			free(kallsyms_allocated_filename);
2114 			return -1;
2115 		}
2116 	} else {
2117 		/*
2118 		 * Last resort, if we don't have a build-id and couldn't find
2119 		 * any vmlinux file, try the running kernel kallsyms table.
2120 		 */
2121 		kallsyms_filename = "/proc/kallsyms";
2122 	}
2123 
2124 do_kallsyms:
2125 	err = dso__load_kallsyms(dso, kallsyms_filename, map, filter);
2126 	if (err > 0)
2127 		pr_debug("Using %s for symbols\n", kallsyms_filename);
2128 	free(kallsyms_allocated_filename);
2129 
2130 	if (err > 0) {
2131 out_fixup:
2132 		if (kallsyms_filename != NULL)
2133 			dso__set_long_name(dso, strdup("[kernel.kallsyms]"));
2134 		map__fixup_start(map);
2135 		map__fixup_end(map);
2136 	}
2137 
2138 	return err;
2139 }
2140 
2141 static int dso__load_guest_kernel_sym(struct dso *dso, struct map *map,
2142 				      symbol_filter_t filter)
2143 {
2144 	int err;
2145 	const char *kallsyms_filename = NULL;
2146 	struct machine *machine;
2147 	char path[PATH_MAX];
2148 
2149 	if (!map->groups) {
2150 		pr_debug("Guest kernel map hasn't the point to groups\n");
2151 		return -1;
2152 	}
2153 	machine = map->groups->machine;
2154 
2155 	if (machine__is_default_guest(machine)) {
2156 		/*
2157 		 * if the user specified a vmlinux filename, use it and only
2158 		 * it, reporting errors to the user if it cannot be used.
2159 		 * Or use file guest_kallsyms inputted by user on commandline
2160 		 */
2161 		if (symbol_conf.default_guest_vmlinux_name != NULL) {
2162 			err = dso__load_vmlinux(dso, map,
2163 				symbol_conf.default_guest_vmlinux_name, filter);
2164 			goto out_try_fixup;
2165 		}
2166 
2167 		kallsyms_filename = symbol_conf.default_guest_kallsyms;
2168 		if (!kallsyms_filename)
2169 			return -1;
2170 	} else {
2171 		sprintf(path, "%s/proc/kallsyms", machine->root_dir);
2172 		kallsyms_filename = path;
2173 	}
2174 
2175 	err = dso__load_kallsyms(dso, kallsyms_filename, map, filter);
2176 	if (err > 0)
2177 		pr_debug("Using %s for symbols\n", kallsyms_filename);
2178 
2179 out_try_fixup:
2180 	if (err > 0) {
2181 		if (kallsyms_filename != NULL) {
2182 			machine__mmap_name(machine, path, sizeof(path));
2183 			dso__set_long_name(dso, strdup(path));
2184 		}
2185 		map__fixup_start(map);
2186 		map__fixup_end(map);
2187 	}
2188 
2189 	return err;
2190 }
2191 
2192 static void dsos__add(struct list_head *head, struct dso *dso)
2193 {
2194 	list_add_tail(&dso->node, head);
2195 }
2196 
2197 static struct dso *dsos__find(struct list_head *head, const char *name)
2198 {
2199 	struct dso *pos;
2200 
2201 	list_for_each_entry(pos, head, node)
2202 		if (strcmp(pos->long_name, name) == 0)
2203 			return pos;
2204 	return NULL;
2205 }
2206 
2207 struct dso *__dsos__findnew(struct list_head *head, const char *name)
2208 {
2209 	struct dso *dso = dsos__find(head, name);
2210 
2211 	if (!dso) {
2212 		dso = dso__new(name);
2213 		if (dso != NULL) {
2214 			dsos__add(head, dso);
2215 			dso__set_basename(dso);
2216 		}
2217 	}
2218 
2219 	return dso;
2220 }
2221 
2222 size_t __dsos__fprintf(struct list_head *head, FILE *fp)
2223 {
2224 	struct dso *pos;
2225 	size_t ret = 0;
2226 
2227 	list_for_each_entry(pos, head, node) {
2228 		int i;
2229 		for (i = 0; i < MAP__NR_TYPES; ++i)
2230 			ret += dso__fprintf(pos, i, fp);
2231 	}
2232 
2233 	return ret;
2234 }
2235 
2236 size_t machines__fprintf_dsos(struct rb_root *machines, FILE *fp)
2237 {
2238 	struct rb_node *nd;
2239 	size_t ret = 0;
2240 
2241 	for (nd = rb_first(machines); nd; nd = rb_next(nd)) {
2242 		struct machine *pos = rb_entry(nd, struct machine, rb_node);
2243 		ret += __dsos__fprintf(&pos->kernel_dsos, fp);
2244 		ret += __dsos__fprintf(&pos->user_dsos, fp);
2245 	}
2246 
2247 	return ret;
2248 }
2249 
2250 static size_t __dsos__fprintf_buildid(struct list_head *head, FILE *fp,
2251 				      bool with_hits)
2252 {
2253 	struct dso *pos;
2254 	size_t ret = 0;
2255 
2256 	list_for_each_entry(pos, head, node) {
2257 		if (with_hits && !pos->hit)
2258 			continue;
2259 		ret += dso__fprintf_buildid(pos, fp);
2260 		ret += fprintf(fp, " %s\n", pos->long_name);
2261 	}
2262 	return ret;
2263 }
2264 
2265 size_t machine__fprintf_dsos_buildid(struct machine *machine, FILE *fp,
2266 				     bool with_hits)
2267 {
2268 	return __dsos__fprintf_buildid(&machine->kernel_dsos, fp, with_hits) +
2269 	       __dsos__fprintf_buildid(&machine->user_dsos, fp, with_hits);
2270 }
2271 
2272 size_t machines__fprintf_dsos_buildid(struct rb_root *machines,
2273 				      FILE *fp, bool with_hits)
2274 {
2275 	struct rb_node *nd;
2276 	size_t ret = 0;
2277 
2278 	for (nd = rb_first(machines); nd; nd = rb_next(nd)) {
2279 		struct machine *pos = rb_entry(nd, struct machine, rb_node);
2280 		ret += machine__fprintf_dsos_buildid(pos, fp, with_hits);
2281 	}
2282 	return ret;
2283 }
2284 
2285 static struct dso*
2286 dso__kernel_findnew(struct machine *machine, const char *name,
2287 		    const char *short_name, int dso_type)
2288 {
2289 	/*
2290 	 * The kernel dso could be created by build_id processing.
2291 	 */
2292 	struct dso *dso = __dsos__findnew(&machine->kernel_dsos, name);
2293 
2294 	/*
2295 	 * We need to run this in all cases, since during the build_id
2296 	 * processing we had no idea this was the kernel dso.
2297 	 */
2298 	if (dso != NULL) {
2299 		dso__set_short_name(dso, short_name);
2300 		dso->kernel = dso_type;
2301 	}
2302 
2303 	return dso;
2304 }
2305 
2306 void dso__read_running_kernel_build_id(struct dso *dso, struct machine *machine)
2307 {
2308 	char path[PATH_MAX];
2309 
2310 	if (machine__is_default_guest(machine))
2311 		return;
2312 	sprintf(path, "%s/sys/kernel/notes", machine->root_dir);
2313 	if (sysfs__read_build_id(path, dso->build_id,
2314 				 sizeof(dso->build_id)) == 0)
2315 		dso->has_build_id = true;
2316 }
2317 
2318 static struct dso *machine__get_kernel(struct machine *machine)
2319 {
2320 	const char *vmlinux_name = NULL;
2321 	struct dso *kernel;
2322 
2323 	if (machine__is_host(machine)) {
2324 		vmlinux_name = symbol_conf.vmlinux_name;
2325 		if (!vmlinux_name)
2326 			vmlinux_name = "[kernel.kallsyms]";
2327 
2328 		kernel = dso__kernel_findnew(machine, vmlinux_name,
2329 					     "[kernel]",
2330 					     DSO_TYPE_KERNEL);
2331 	} else {
2332 		char bf[PATH_MAX];
2333 
2334 		if (machine__is_default_guest(machine))
2335 			vmlinux_name = symbol_conf.default_guest_vmlinux_name;
2336 		if (!vmlinux_name)
2337 			vmlinux_name = machine__mmap_name(machine, bf,
2338 							  sizeof(bf));
2339 
2340 		kernel = dso__kernel_findnew(machine, vmlinux_name,
2341 					     "[guest.kernel]",
2342 					     DSO_TYPE_GUEST_KERNEL);
2343 	}
2344 
2345 	if (kernel != NULL && (!kernel->has_build_id))
2346 		dso__read_running_kernel_build_id(kernel, machine);
2347 
2348 	return kernel;
2349 }
2350 
2351 struct process_args {
2352 	u64 start;
2353 };
2354 
2355 static int symbol__in_kernel(void *arg, const char *name,
2356 			     char type __used, u64 start, u64 end __used)
2357 {
2358 	struct process_args *args = arg;
2359 
2360 	if (strchr(name, '['))
2361 		return 0;
2362 
2363 	args->start = start;
2364 	return 1;
2365 }
2366 
2367 /* Figure out the start address of kernel map from /proc/kallsyms */
2368 static u64 machine__get_kernel_start_addr(struct machine *machine)
2369 {
2370 	const char *filename;
2371 	char path[PATH_MAX];
2372 	struct process_args args;
2373 
2374 	if (machine__is_host(machine)) {
2375 		filename = "/proc/kallsyms";
2376 	} else {
2377 		if (machine__is_default_guest(machine))
2378 			filename = (char *)symbol_conf.default_guest_kallsyms;
2379 		else {
2380 			sprintf(path, "%s/proc/kallsyms", machine->root_dir);
2381 			filename = path;
2382 		}
2383 	}
2384 
2385 	if (symbol__restricted_filename(filename, "/proc/kallsyms"))
2386 		return 0;
2387 
2388 	if (kallsyms__parse(filename, &args, symbol__in_kernel) <= 0)
2389 		return 0;
2390 
2391 	return args.start;
2392 }
2393 
2394 int __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
2395 {
2396 	enum map_type type;
2397 	u64 start = machine__get_kernel_start_addr(machine);
2398 
2399 	for (type = 0; type < MAP__NR_TYPES; ++type) {
2400 		struct kmap *kmap;
2401 
2402 		machine->vmlinux_maps[type] = map__new2(start, kernel, type);
2403 		if (machine->vmlinux_maps[type] == NULL)
2404 			return -1;
2405 
2406 		machine->vmlinux_maps[type]->map_ip =
2407 			machine->vmlinux_maps[type]->unmap_ip =
2408 				identity__map_ip;
2409 		kmap = map__kmap(machine->vmlinux_maps[type]);
2410 		kmap->kmaps = &machine->kmaps;
2411 		map_groups__insert(&machine->kmaps,
2412 				   machine->vmlinux_maps[type]);
2413 	}
2414 
2415 	return 0;
2416 }
2417 
2418 void machine__destroy_kernel_maps(struct machine *machine)
2419 {
2420 	enum map_type type;
2421 
2422 	for (type = 0; type < MAP__NR_TYPES; ++type) {
2423 		struct kmap *kmap;
2424 
2425 		if (machine->vmlinux_maps[type] == NULL)
2426 			continue;
2427 
2428 		kmap = map__kmap(machine->vmlinux_maps[type]);
2429 		map_groups__remove(&machine->kmaps,
2430 				   machine->vmlinux_maps[type]);
2431 		if (kmap->ref_reloc_sym) {
2432 			/*
2433 			 * ref_reloc_sym is shared among all maps, so free just
2434 			 * on one of them.
2435 			 */
2436 			if (type == MAP__FUNCTION) {
2437 				free((char *)kmap->ref_reloc_sym->name);
2438 				kmap->ref_reloc_sym->name = NULL;
2439 				free(kmap->ref_reloc_sym);
2440 			}
2441 			kmap->ref_reloc_sym = NULL;
2442 		}
2443 
2444 		map__delete(machine->vmlinux_maps[type]);
2445 		machine->vmlinux_maps[type] = NULL;
2446 	}
2447 }
2448 
2449 int machine__create_kernel_maps(struct machine *machine)
2450 {
2451 	struct dso *kernel = machine__get_kernel(machine);
2452 
2453 	if (kernel == NULL ||
2454 	    __machine__create_kernel_maps(machine, kernel) < 0)
2455 		return -1;
2456 
2457 	if (symbol_conf.use_modules && machine__create_modules(machine) < 0)
2458 		pr_debug("Problems creating module maps, continuing anyway...\n");
2459 	/*
2460 	 * Now that we have all the maps created, just set the ->end of them:
2461 	 */
2462 	map_groups__fixup_end(&machine->kmaps);
2463 	return 0;
2464 }
2465 
2466 static void vmlinux_path__exit(void)
2467 {
2468 	while (--vmlinux_path__nr_entries >= 0) {
2469 		free(vmlinux_path[vmlinux_path__nr_entries]);
2470 		vmlinux_path[vmlinux_path__nr_entries] = NULL;
2471 	}
2472 
2473 	free(vmlinux_path);
2474 	vmlinux_path = NULL;
2475 }
2476 
2477 static int vmlinux_path__init(void)
2478 {
2479 	struct utsname uts;
2480 	char bf[PATH_MAX];
2481 
2482 	vmlinux_path = malloc(sizeof(char *) * 5);
2483 	if (vmlinux_path == NULL)
2484 		return -1;
2485 
2486 	vmlinux_path[vmlinux_path__nr_entries] = strdup("vmlinux");
2487 	if (vmlinux_path[vmlinux_path__nr_entries] == NULL)
2488 		goto out_fail;
2489 	++vmlinux_path__nr_entries;
2490 	vmlinux_path[vmlinux_path__nr_entries] = strdup("/boot/vmlinux");
2491 	if (vmlinux_path[vmlinux_path__nr_entries] == NULL)
2492 		goto out_fail;
2493 	++vmlinux_path__nr_entries;
2494 
2495 	/* only try running kernel version if no symfs was given */
2496 	if (symbol_conf.symfs[0] != 0)
2497 		return 0;
2498 
2499 	if (uname(&uts) < 0)
2500 		return -1;
2501 
2502 	snprintf(bf, sizeof(bf), "/boot/vmlinux-%s", uts.release);
2503 	vmlinux_path[vmlinux_path__nr_entries] = strdup(bf);
2504 	if (vmlinux_path[vmlinux_path__nr_entries] == NULL)
2505 		goto out_fail;
2506 	++vmlinux_path__nr_entries;
2507 	snprintf(bf, sizeof(bf), "/lib/modules/%s/build/vmlinux", uts.release);
2508 	vmlinux_path[vmlinux_path__nr_entries] = strdup(bf);
2509 	if (vmlinux_path[vmlinux_path__nr_entries] == NULL)
2510 		goto out_fail;
2511 	++vmlinux_path__nr_entries;
2512 	snprintf(bf, sizeof(bf), "/usr/lib/debug/lib/modules/%s/vmlinux",
2513 		 uts.release);
2514 	vmlinux_path[vmlinux_path__nr_entries] = strdup(bf);
2515 	if (vmlinux_path[vmlinux_path__nr_entries] == NULL)
2516 		goto out_fail;
2517 	++vmlinux_path__nr_entries;
2518 
2519 	return 0;
2520 
2521 out_fail:
2522 	vmlinux_path__exit();
2523 	return -1;
2524 }
2525 
2526 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
2527 {
2528 	int i;
2529 	size_t printed = 0;
2530 	struct dso *kdso = machine->vmlinux_maps[MAP__FUNCTION]->dso;
2531 
2532 	if (kdso->has_build_id) {
2533 		char filename[PATH_MAX];
2534 		if (dso__build_id_filename(kdso, filename, sizeof(filename)))
2535 			printed += fprintf(fp, "[0] %s\n", filename);
2536 	}
2537 
2538 	for (i = 0; i < vmlinux_path__nr_entries; ++i)
2539 		printed += fprintf(fp, "[%d] %s\n",
2540 				   i + kdso->has_build_id, vmlinux_path[i]);
2541 
2542 	return printed;
2543 }
2544 
2545 static int setup_list(struct strlist **list, const char *list_str,
2546 		      const char *list_name)
2547 {
2548 	if (list_str == NULL)
2549 		return 0;
2550 
2551 	*list = strlist__new(true, list_str);
2552 	if (!*list) {
2553 		pr_err("problems parsing %s list\n", list_name);
2554 		return -1;
2555 	}
2556 	return 0;
2557 }
2558 
2559 static bool symbol__read_kptr_restrict(void)
2560 {
2561 	bool value = false;
2562 
2563 	if (geteuid() != 0) {
2564 		FILE *fp = fopen("/proc/sys/kernel/kptr_restrict", "r");
2565 		if (fp != NULL) {
2566 			char line[8];
2567 
2568 			if (fgets(line, sizeof(line), fp) != NULL)
2569 				value = atoi(line) != 0;
2570 
2571 			fclose(fp);
2572 		}
2573 	}
2574 
2575 	return value;
2576 }
2577 
2578 int symbol__init(void)
2579 {
2580 	const char *symfs;
2581 
2582 	if (symbol_conf.initialized)
2583 		return 0;
2584 
2585 	symbol_conf.priv_size = ALIGN(symbol_conf.priv_size, sizeof(u64));
2586 
2587 	elf_version(EV_CURRENT);
2588 	if (symbol_conf.sort_by_name)
2589 		symbol_conf.priv_size += (sizeof(struct symbol_name_rb_node) -
2590 					  sizeof(struct symbol));
2591 
2592 	if (symbol_conf.try_vmlinux_path && vmlinux_path__init() < 0)
2593 		return -1;
2594 
2595 	if (symbol_conf.field_sep && *symbol_conf.field_sep == '.') {
2596 		pr_err("'.' is the only non valid --field-separator argument\n");
2597 		return -1;
2598 	}
2599 
2600 	if (setup_list(&symbol_conf.dso_list,
2601 		       symbol_conf.dso_list_str, "dso") < 0)
2602 		return -1;
2603 
2604 	if (setup_list(&symbol_conf.comm_list,
2605 		       symbol_conf.comm_list_str, "comm") < 0)
2606 		goto out_free_dso_list;
2607 
2608 	if (setup_list(&symbol_conf.sym_list,
2609 		       symbol_conf.sym_list_str, "symbol") < 0)
2610 		goto out_free_comm_list;
2611 
2612 	/*
2613 	 * A path to symbols of "/" is identical to ""
2614 	 * reset here for simplicity.
2615 	 */
2616 	symfs = realpath(symbol_conf.symfs, NULL);
2617 	if (symfs == NULL)
2618 		symfs = symbol_conf.symfs;
2619 	if (strcmp(symfs, "/") == 0)
2620 		symbol_conf.symfs = "";
2621 	if (symfs != symbol_conf.symfs)
2622 		free((void *)symfs);
2623 
2624 	symbol_conf.kptr_restrict = symbol__read_kptr_restrict();
2625 
2626 	symbol_conf.initialized = true;
2627 	return 0;
2628 
2629 out_free_comm_list:
2630 	strlist__delete(symbol_conf.comm_list);
2631 out_free_dso_list:
2632 	strlist__delete(symbol_conf.dso_list);
2633 	return -1;
2634 }
2635 
2636 void symbol__exit(void)
2637 {
2638 	if (!symbol_conf.initialized)
2639 		return;
2640 	strlist__delete(symbol_conf.sym_list);
2641 	strlist__delete(symbol_conf.dso_list);
2642 	strlist__delete(symbol_conf.comm_list);
2643 	vmlinux_path__exit();
2644 	symbol_conf.sym_list = symbol_conf.dso_list = symbol_conf.comm_list = NULL;
2645 	symbol_conf.initialized = false;
2646 }
2647 
2648 int machines__create_kernel_maps(struct rb_root *machines, pid_t pid)
2649 {
2650 	struct machine *machine = machines__findnew(machines, pid);
2651 
2652 	if (machine == NULL)
2653 		return -1;
2654 
2655 	return machine__create_kernel_maps(machine);
2656 }
2657 
2658 static int hex(char ch)
2659 {
2660 	if ((ch >= '0') && (ch <= '9'))
2661 		return ch - '0';
2662 	if ((ch >= 'a') && (ch <= 'f'))
2663 		return ch - 'a' + 10;
2664 	if ((ch >= 'A') && (ch <= 'F'))
2665 		return ch - 'A' + 10;
2666 	return -1;
2667 }
2668 
2669 /*
2670  * While we find nice hex chars, build a long_val.
2671  * Return number of chars processed.
2672  */
2673 int hex2u64(const char *ptr, u64 *long_val)
2674 {
2675 	const char *p = ptr;
2676 	*long_val = 0;
2677 
2678 	while (*p) {
2679 		const int hex_val = hex(*p);
2680 
2681 		if (hex_val < 0)
2682 			break;
2683 
2684 		*long_val = (*long_val << 4) | hex_val;
2685 		p++;
2686 	}
2687 
2688 	return p - ptr;
2689 }
2690 
2691 char *strxfrchar(char *s, char from, char to)
2692 {
2693 	char *p = s;
2694 
2695 	while ((p = strchr(p, from)) != NULL)
2696 		*p++ = to;
2697 
2698 	return s;
2699 }
2700 
2701 int machines__create_guest_kernel_maps(struct rb_root *machines)
2702 {
2703 	int ret = 0;
2704 	struct dirent **namelist = NULL;
2705 	int i, items = 0;
2706 	char path[PATH_MAX];
2707 	pid_t pid;
2708 
2709 	if (symbol_conf.default_guest_vmlinux_name ||
2710 	    symbol_conf.default_guest_modules ||
2711 	    symbol_conf.default_guest_kallsyms) {
2712 		machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
2713 	}
2714 
2715 	if (symbol_conf.guestmount) {
2716 		items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
2717 		if (items <= 0)
2718 			return -ENOENT;
2719 		for (i = 0; i < items; i++) {
2720 			if (!isdigit(namelist[i]->d_name[0])) {
2721 				/* Filter out . and .. */
2722 				continue;
2723 			}
2724 			pid = atoi(namelist[i]->d_name);
2725 			sprintf(path, "%s/%s/proc/kallsyms",
2726 				symbol_conf.guestmount,
2727 				namelist[i]->d_name);
2728 			ret = access(path, R_OK);
2729 			if (ret) {
2730 				pr_debug("Can't access file %s\n", path);
2731 				goto failure;
2732 			}
2733 			machines__create_kernel_maps(machines, pid);
2734 		}
2735 failure:
2736 		free(namelist);
2737 	}
2738 
2739 	return ret;
2740 }
2741 
2742 void machines__destroy_guest_kernel_maps(struct rb_root *machines)
2743 {
2744 	struct rb_node *next = rb_first(machines);
2745 
2746 	while (next) {
2747 		struct machine *pos = rb_entry(next, struct machine, rb_node);
2748 
2749 		next = rb_next(&pos->rb_node);
2750 		rb_erase(&pos->rb_node, machines);
2751 		machine__delete(pos);
2752 	}
2753 }
2754 
2755 int machine__load_kallsyms(struct machine *machine, const char *filename,
2756 			   enum map_type type, symbol_filter_t filter)
2757 {
2758 	struct map *map = machine->vmlinux_maps[type];
2759 	int ret = dso__load_kallsyms(map->dso, filename, map, filter);
2760 
2761 	if (ret > 0) {
2762 		dso__set_loaded(map->dso, type);
2763 		/*
2764 		 * Since /proc/kallsyms will have multiple sessions for the
2765 		 * kernel, with modules between them, fixup the end of all
2766 		 * sections.
2767 		 */
2768 		__map_groups__fixup_end(&machine->kmaps, type);
2769 	}
2770 
2771 	return ret;
2772 }
2773 
2774 int machine__load_vmlinux_path(struct machine *machine, enum map_type type,
2775 			       symbol_filter_t filter)
2776 {
2777 	struct map *map = machine->vmlinux_maps[type];
2778 	int ret = dso__load_vmlinux_path(map->dso, map, filter);
2779 
2780 	if (ret > 0) {
2781 		dso__set_loaded(map->dso, type);
2782 		map__reloc_vmlinux(map);
2783 	}
2784 
2785 	return ret;
2786 }
2787