xref: /linux/tools/perf/util/symbol.c (revision 1553a1c48281243359a9529a10ddb551f3b967ab)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <dirent.h>
3 #include <errno.h>
4 #include <stdlib.h>
5 #include <stdio.h>
6 #include <string.h>
7 #include <linux/capability.h>
8 #include <linux/kernel.h>
9 #include <linux/mman.h>
10 #include <linux/string.h>
11 #include <linux/time64.h>
12 #include <sys/types.h>
13 #include <sys/stat.h>
14 #include <sys/param.h>
15 #include <fcntl.h>
16 #include <unistd.h>
17 #include <inttypes.h>
18 #include "annotate.h"
19 #include "build-id.h"
20 #include "cap.h"
21 #include "dso.h"
22 #include "util.h" // lsdir()
23 #include "debug.h"
24 #include "event.h"
25 #include "machine.h"
26 #include "map.h"
27 #include "symbol.h"
28 #include "map_symbol.h"
29 #include "mem-events.h"
30 #include "symsrc.h"
31 #include "strlist.h"
32 #include "intlist.h"
33 #include "namespaces.h"
34 #include "header.h"
35 #include "path.h"
36 #include <linux/ctype.h>
37 #include <linux/zalloc.h>
38 
39 #include <elf.h>
40 #include <limits.h>
41 #include <symbol/kallsyms.h>
42 #include <sys/utsname.h>
43 
44 static int dso__load_kernel_sym(struct dso *dso, struct map *map);
45 static int dso__load_guest_kernel_sym(struct dso *dso, struct map *map);
46 static bool symbol__is_idle(const char *name);
47 
48 int vmlinux_path__nr_entries;
49 char **vmlinux_path;
50 
51 struct symbol_conf symbol_conf = {
52 	.nanosecs		= false,
53 	.use_modules		= true,
54 	.try_vmlinux_path	= true,
55 	.demangle		= true,
56 	.demangle_kernel	= false,
57 	.cumulate_callchain	= true,
58 	.time_quantum		= 100 * NSEC_PER_MSEC, /* 100ms */
59 	.show_hist_headers	= true,
60 	.symfs			= "",
61 	.event_group		= true,
62 	.inline_name		= true,
63 	.res_sample		= 0,
64 };
65 
66 struct map_list_node {
67 	struct list_head node;
68 	struct map *map;
69 };
70 
71 static struct map_list_node *map_list_node__new(void)
72 {
73 	return malloc(sizeof(struct map_list_node));
74 }
75 
76 static enum dso_binary_type binary_type_symtab[] = {
77 	DSO_BINARY_TYPE__KALLSYMS,
78 	DSO_BINARY_TYPE__GUEST_KALLSYMS,
79 	DSO_BINARY_TYPE__JAVA_JIT,
80 	DSO_BINARY_TYPE__DEBUGLINK,
81 	DSO_BINARY_TYPE__BUILD_ID_CACHE,
82 	DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO,
83 	DSO_BINARY_TYPE__FEDORA_DEBUGINFO,
84 	DSO_BINARY_TYPE__UBUNTU_DEBUGINFO,
85 	DSO_BINARY_TYPE__BUILDID_DEBUGINFO,
86 	DSO_BINARY_TYPE__SYSTEM_PATH_DSO,
87 	DSO_BINARY_TYPE__GUEST_KMODULE,
88 	DSO_BINARY_TYPE__GUEST_KMODULE_COMP,
89 	DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE,
90 	DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP,
91 	DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO,
92 	DSO_BINARY_TYPE__MIXEDUP_UBUNTU_DEBUGINFO,
93 	DSO_BINARY_TYPE__NOT_FOUND,
94 };
95 
96 #define DSO_BINARY_TYPE__SYMTAB_CNT ARRAY_SIZE(binary_type_symtab)
97 
98 static bool symbol_type__filter(char symbol_type)
99 {
100 	symbol_type = toupper(symbol_type);
101 	return symbol_type == 'T' || symbol_type == 'W' || symbol_type == 'D' || symbol_type == 'B';
102 }
103 
104 static int prefix_underscores_count(const char *str)
105 {
106 	const char *tail = str;
107 
108 	while (*tail == '_')
109 		tail++;
110 
111 	return tail - str;
112 }
113 
114 const char * __weak arch__normalize_symbol_name(const char *name)
115 {
116 	return name;
117 }
118 
119 int __weak arch__compare_symbol_names(const char *namea, const char *nameb)
120 {
121 	return strcmp(namea, nameb);
122 }
123 
124 int __weak arch__compare_symbol_names_n(const char *namea, const char *nameb,
125 					unsigned int n)
126 {
127 	return strncmp(namea, nameb, n);
128 }
129 
130 int __weak arch__choose_best_symbol(struct symbol *syma,
131 				    struct symbol *symb __maybe_unused)
132 {
133 	/* Avoid "SyS" kernel syscall aliases */
134 	if (strlen(syma->name) >= 3 && !strncmp(syma->name, "SyS", 3))
135 		return SYMBOL_B;
136 	if (strlen(syma->name) >= 10 && !strncmp(syma->name, "compat_SyS", 10))
137 		return SYMBOL_B;
138 
139 	return SYMBOL_A;
140 }
141 
142 static int choose_best_symbol(struct symbol *syma, struct symbol *symb)
143 {
144 	s64 a;
145 	s64 b;
146 	size_t na, nb;
147 
148 	/* Prefer a symbol with non zero length */
149 	a = syma->end - syma->start;
150 	b = symb->end - symb->start;
151 	if ((b == 0) && (a > 0))
152 		return SYMBOL_A;
153 	else if ((a == 0) && (b > 0))
154 		return SYMBOL_B;
155 
156 	/* Prefer a non weak symbol over a weak one */
157 	a = syma->binding == STB_WEAK;
158 	b = symb->binding == STB_WEAK;
159 	if (b && !a)
160 		return SYMBOL_A;
161 	if (a && !b)
162 		return SYMBOL_B;
163 
164 	/* Prefer a global symbol over a non global one */
165 	a = syma->binding == STB_GLOBAL;
166 	b = symb->binding == STB_GLOBAL;
167 	if (a && !b)
168 		return SYMBOL_A;
169 	if (b && !a)
170 		return SYMBOL_B;
171 
172 	/* Prefer a symbol with less underscores */
173 	a = prefix_underscores_count(syma->name);
174 	b = prefix_underscores_count(symb->name);
175 	if (b > a)
176 		return SYMBOL_A;
177 	else if (a > b)
178 		return SYMBOL_B;
179 
180 	/* Choose the symbol with the longest name */
181 	na = strlen(syma->name);
182 	nb = strlen(symb->name);
183 	if (na > nb)
184 		return SYMBOL_A;
185 	else if (na < nb)
186 		return SYMBOL_B;
187 
188 	return arch__choose_best_symbol(syma, symb);
189 }
190 
191 void symbols__fixup_duplicate(struct rb_root_cached *symbols)
192 {
193 	struct rb_node *nd;
194 	struct symbol *curr, *next;
195 
196 	if (symbol_conf.allow_aliases)
197 		return;
198 
199 	nd = rb_first_cached(symbols);
200 
201 	while (nd) {
202 		curr = rb_entry(nd, struct symbol, rb_node);
203 again:
204 		nd = rb_next(&curr->rb_node);
205 		if (!nd)
206 			break;
207 
208 		next = rb_entry(nd, struct symbol, rb_node);
209 		if (curr->start != next->start)
210 			continue;
211 
212 		if (choose_best_symbol(curr, next) == SYMBOL_A) {
213 			if (next->type == STT_GNU_IFUNC)
214 				curr->ifunc_alias = true;
215 			rb_erase_cached(&next->rb_node, symbols);
216 			symbol__delete(next);
217 			goto again;
218 		} else {
219 			if (curr->type == STT_GNU_IFUNC)
220 				next->ifunc_alias = true;
221 			nd = rb_next(&curr->rb_node);
222 			rb_erase_cached(&curr->rb_node, symbols);
223 			symbol__delete(curr);
224 		}
225 	}
226 }
227 
228 /* Update zero-sized symbols using the address of the next symbol */
229 void symbols__fixup_end(struct rb_root_cached *symbols, bool is_kallsyms)
230 {
231 	struct rb_node *nd, *prevnd = rb_first_cached(symbols);
232 	struct symbol *curr, *prev;
233 
234 	if (prevnd == NULL)
235 		return;
236 
237 	curr = rb_entry(prevnd, struct symbol, rb_node);
238 
239 	for (nd = rb_next(prevnd); nd; nd = rb_next(nd)) {
240 		prev = curr;
241 		curr = rb_entry(nd, struct symbol, rb_node);
242 
243 		/*
244 		 * On some architecture kernel text segment start is located at
245 		 * some low memory address, while modules are located at high
246 		 * memory addresses (or vice versa).  The gap between end of
247 		 * kernel text segment and beginning of first module's text
248 		 * segment is very big.  Therefore do not fill this gap and do
249 		 * not assign it to the kernel dso map (kallsyms).
250 		 *
251 		 * Also BPF code can be allocated separately from text segments
252 		 * and modules.  So the last entry in a module should not fill
253 		 * the gap too.
254 		 *
255 		 * In kallsyms, it determines module symbols using '[' character
256 		 * like in:
257 		 *   ffffffffc1937000 T hdmi_driver_init  [snd_hda_codec_hdmi]
258 		 */
259 		if (prev->end == prev->start) {
260 			const char *prev_mod;
261 			const char *curr_mod;
262 
263 			if (!is_kallsyms) {
264 				prev->end = curr->start;
265 				continue;
266 			}
267 
268 			prev_mod = strchr(prev->name, '[');
269 			curr_mod = strchr(curr->name, '[');
270 
271 			/* Last kernel/module symbol mapped to end of page */
272 			if (!prev_mod != !curr_mod)
273 				prev->end = roundup(prev->end + 4096, 4096);
274 			/* Last symbol in the previous module */
275 			else if (prev_mod && strcmp(prev_mod, curr_mod))
276 				prev->end = roundup(prev->end + 4096, 4096);
277 			else
278 				prev->end = curr->start;
279 
280 			pr_debug4("%s sym:%s end:%#" PRIx64 "\n",
281 				  __func__, prev->name, prev->end);
282 		}
283 	}
284 
285 	/* Last entry */
286 	if (curr->end == curr->start)
287 		curr->end = roundup(curr->start, 4096) + 4096;
288 }
289 
290 struct symbol *symbol__new(u64 start, u64 len, u8 binding, u8 type, const char *name)
291 {
292 	size_t namelen = strlen(name) + 1;
293 	struct symbol *sym = calloc(1, (symbol_conf.priv_size +
294 					sizeof(*sym) + namelen));
295 	if (sym == NULL)
296 		return NULL;
297 
298 	if (symbol_conf.priv_size) {
299 		if (symbol_conf.init_annotation) {
300 			struct annotation *notes = (void *)sym;
301 			annotation__init(notes);
302 		}
303 		sym = ((void *)sym) + symbol_conf.priv_size;
304 	}
305 
306 	sym->start   = start;
307 	sym->end     = len ? start + len : start;
308 	sym->type    = type;
309 	sym->binding = binding;
310 	sym->namelen = namelen - 1;
311 
312 	pr_debug4("%s: %s %#" PRIx64 "-%#" PRIx64 "\n",
313 		  __func__, name, start, sym->end);
314 	memcpy(sym->name, name, namelen);
315 
316 	return sym;
317 }
318 
319 void symbol__delete(struct symbol *sym)
320 {
321 	if (symbol_conf.priv_size) {
322 		if (symbol_conf.init_annotation) {
323 			struct annotation *notes = symbol__annotation(sym);
324 
325 			annotation__exit(notes);
326 		}
327 	}
328 	free(((void *)sym) - symbol_conf.priv_size);
329 }
330 
331 void symbols__delete(struct rb_root_cached *symbols)
332 {
333 	struct symbol *pos;
334 	struct rb_node *next = rb_first_cached(symbols);
335 
336 	while (next) {
337 		pos = rb_entry(next, struct symbol, rb_node);
338 		next = rb_next(&pos->rb_node);
339 		rb_erase_cached(&pos->rb_node, symbols);
340 		symbol__delete(pos);
341 	}
342 }
343 
344 void __symbols__insert(struct rb_root_cached *symbols,
345 		       struct symbol *sym, bool kernel)
346 {
347 	struct rb_node **p = &symbols->rb_root.rb_node;
348 	struct rb_node *parent = NULL;
349 	const u64 ip = sym->start;
350 	struct symbol *s;
351 	bool leftmost = true;
352 
353 	if (kernel) {
354 		const char *name = sym->name;
355 		/*
356 		 * ppc64 uses function descriptors and appends a '.' to the
357 		 * start of every instruction address. Remove it.
358 		 */
359 		if (name[0] == '.')
360 			name++;
361 		sym->idle = symbol__is_idle(name);
362 	}
363 
364 	while (*p != NULL) {
365 		parent = *p;
366 		s = rb_entry(parent, struct symbol, rb_node);
367 		if (ip < s->start)
368 			p = &(*p)->rb_left;
369 		else {
370 			p = &(*p)->rb_right;
371 			leftmost = false;
372 		}
373 	}
374 	rb_link_node(&sym->rb_node, parent, p);
375 	rb_insert_color_cached(&sym->rb_node, symbols, leftmost);
376 }
377 
378 void symbols__insert(struct rb_root_cached *symbols, struct symbol *sym)
379 {
380 	__symbols__insert(symbols, sym, false);
381 }
382 
383 static struct symbol *symbols__find(struct rb_root_cached *symbols, u64 ip)
384 {
385 	struct rb_node *n;
386 
387 	if (symbols == NULL)
388 		return NULL;
389 
390 	n = symbols->rb_root.rb_node;
391 
392 	while (n) {
393 		struct symbol *s = rb_entry(n, struct symbol, rb_node);
394 
395 		if (ip < s->start)
396 			n = n->rb_left;
397 		else if (ip > s->end || (ip == s->end && ip != s->start))
398 			n = n->rb_right;
399 		else
400 			return s;
401 	}
402 
403 	return NULL;
404 }
405 
406 static struct symbol *symbols__first(struct rb_root_cached *symbols)
407 {
408 	struct rb_node *n = rb_first_cached(symbols);
409 
410 	if (n)
411 		return rb_entry(n, struct symbol, rb_node);
412 
413 	return NULL;
414 }
415 
416 static struct symbol *symbols__last(struct rb_root_cached *symbols)
417 {
418 	struct rb_node *n = rb_last(&symbols->rb_root);
419 
420 	if (n)
421 		return rb_entry(n, struct symbol, rb_node);
422 
423 	return NULL;
424 }
425 
426 static struct symbol *symbols__next(struct symbol *sym)
427 {
428 	struct rb_node *n = rb_next(&sym->rb_node);
429 
430 	if (n)
431 		return rb_entry(n, struct symbol, rb_node);
432 
433 	return NULL;
434 }
435 
436 static int symbols__sort_name_cmp(const void *vlhs, const void *vrhs)
437 {
438 	const struct symbol *lhs = *((const struct symbol **)vlhs);
439 	const struct symbol *rhs = *((const struct symbol **)vrhs);
440 
441 	return strcmp(lhs->name, rhs->name);
442 }
443 
444 static struct symbol **symbols__sort_by_name(struct rb_root_cached *source, size_t *len)
445 {
446 	struct rb_node *nd;
447 	struct symbol **result;
448 	size_t i = 0, size = 0;
449 
450 	for (nd = rb_first_cached(source); nd; nd = rb_next(nd))
451 		size++;
452 
453 	result = malloc(sizeof(*result) * size);
454 	if (!result)
455 		return NULL;
456 
457 	for (nd = rb_first_cached(source); nd; nd = rb_next(nd)) {
458 		struct symbol *pos = rb_entry(nd, struct symbol, rb_node);
459 
460 		result[i++] = pos;
461 	}
462 	qsort(result, size, sizeof(*result), symbols__sort_name_cmp);
463 	*len = size;
464 	return result;
465 }
466 
467 int symbol__match_symbol_name(const char *name, const char *str,
468 			      enum symbol_tag_include includes)
469 {
470 	const char *versioning;
471 
472 	if (includes == SYMBOL_TAG_INCLUDE__DEFAULT_ONLY &&
473 	    (versioning = strstr(name, "@@"))) {
474 		int len = strlen(str);
475 
476 		if (len < versioning - name)
477 			len = versioning - name;
478 
479 		return arch__compare_symbol_names_n(name, str, len);
480 	} else
481 		return arch__compare_symbol_names(name, str);
482 }
483 
484 static struct symbol *symbols__find_by_name(struct symbol *symbols[],
485 					    size_t symbols_len,
486 					    const char *name,
487 					    enum symbol_tag_include includes,
488 					    size_t *found_idx)
489 {
490 	size_t i, lower = 0, upper = symbols_len;
491 	struct symbol *s = NULL;
492 
493 	if (found_idx)
494 		*found_idx = SIZE_MAX;
495 
496 	if (!symbols_len)
497 		return NULL;
498 
499 	while (lower < upper) {
500 		int cmp;
501 
502 		i = (lower + upper) / 2;
503 		cmp = symbol__match_symbol_name(symbols[i]->name, name, includes);
504 
505 		if (cmp > 0)
506 			upper = i;
507 		else if (cmp < 0)
508 			lower = i + 1;
509 		else {
510 			if (found_idx)
511 				*found_idx = i;
512 			s = symbols[i];
513 			break;
514 		}
515 	}
516 	if (s && includes != SYMBOL_TAG_INCLUDE__DEFAULT_ONLY) {
517 		/* return first symbol that has same name (if any) */
518 		for (; i > 0; i--) {
519 			struct symbol *tmp = symbols[i - 1];
520 
521 			if (!arch__compare_symbol_names(tmp->name, s->name)) {
522 				if (found_idx)
523 					*found_idx = i - 1;
524 				s = tmp;
525 			} else
526 				break;
527 		}
528 	}
529 	assert(!found_idx || !s || s == symbols[*found_idx]);
530 	return s;
531 }
532 
533 void dso__reset_find_symbol_cache(struct dso *dso)
534 {
535 	dso->last_find_result.addr   = 0;
536 	dso->last_find_result.symbol = NULL;
537 }
538 
539 void dso__insert_symbol(struct dso *dso, struct symbol *sym)
540 {
541 	__symbols__insert(&dso->symbols, sym, dso->kernel);
542 
543 	/* update the symbol cache if necessary */
544 	if (dso->last_find_result.addr >= sym->start &&
545 	    (dso->last_find_result.addr < sym->end ||
546 	    sym->start == sym->end)) {
547 		dso->last_find_result.symbol = sym;
548 	}
549 }
550 
551 void dso__delete_symbol(struct dso *dso, struct symbol *sym)
552 {
553 	rb_erase_cached(&sym->rb_node, &dso->symbols);
554 	symbol__delete(sym);
555 	dso__reset_find_symbol_cache(dso);
556 }
557 
558 struct symbol *dso__find_symbol(struct dso *dso, u64 addr)
559 {
560 	if (dso->last_find_result.addr != addr || dso->last_find_result.symbol == NULL) {
561 		dso->last_find_result.addr   = addr;
562 		dso->last_find_result.symbol = symbols__find(&dso->symbols, addr);
563 	}
564 
565 	return dso->last_find_result.symbol;
566 }
567 
568 struct symbol *dso__find_symbol_nocache(struct dso *dso, u64 addr)
569 {
570 	return symbols__find(&dso->symbols, addr);
571 }
572 
573 struct symbol *dso__first_symbol(struct dso *dso)
574 {
575 	return symbols__first(&dso->symbols);
576 }
577 
578 struct symbol *dso__last_symbol(struct dso *dso)
579 {
580 	return symbols__last(&dso->symbols);
581 }
582 
583 struct symbol *dso__next_symbol(struct symbol *sym)
584 {
585 	return symbols__next(sym);
586 }
587 
588 struct symbol *dso__next_symbol_by_name(struct dso *dso, size_t *idx)
589 {
590 	if (*idx + 1 >= dso->symbol_names_len)
591 		return NULL;
592 
593 	++*idx;
594 	return dso->symbol_names[*idx];
595 }
596 
597  /*
598   * Returns first symbol that matched with @name.
599   */
600 struct symbol *dso__find_symbol_by_name(struct dso *dso, const char *name, size_t *idx)
601 {
602 	struct symbol *s = symbols__find_by_name(dso->symbol_names, dso->symbol_names_len,
603 						name, SYMBOL_TAG_INCLUDE__NONE, idx);
604 	if (!s)
605 		s = symbols__find_by_name(dso->symbol_names, dso->symbol_names_len,
606 					name, SYMBOL_TAG_INCLUDE__DEFAULT_ONLY, idx);
607 	return s;
608 }
609 
610 void dso__sort_by_name(struct dso *dso)
611 {
612 	mutex_lock(&dso->lock);
613 	if (!dso__sorted_by_name(dso)) {
614 		size_t len;
615 
616 		dso->symbol_names = symbols__sort_by_name(&dso->symbols, &len);
617 		if (dso->symbol_names) {
618 			dso->symbol_names_len = len;
619 			dso__set_sorted_by_name(dso);
620 		}
621 	}
622 	mutex_unlock(&dso->lock);
623 }
624 
625 /*
626  * While we find nice hex chars, build a long_val.
627  * Return number of chars processed.
628  */
629 static int hex2u64(const char *ptr, u64 *long_val)
630 {
631 	char *p;
632 
633 	*long_val = strtoull(ptr, &p, 16);
634 
635 	return p - ptr;
636 }
637 
638 
639 int modules__parse(const char *filename, void *arg,
640 		   int (*process_module)(void *arg, const char *name,
641 					 u64 start, u64 size))
642 {
643 	char *line = NULL;
644 	size_t n;
645 	FILE *file;
646 	int err = 0;
647 
648 	file = fopen(filename, "r");
649 	if (file == NULL)
650 		return -1;
651 
652 	while (1) {
653 		char name[PATH_MAX];
654 		u64 start, size;
655 		char *sep, *endptr;
656 		ssize_t line_len;
657 
658 		line_len = getline(&line, &n, file);
659 		if (line_len < 0) {
660 			if (feof(file))
661 				break;
662 			err = -1;
663 			goto out;
664 		}
665 
666 		if (!line) {
667 			err = -1;
668 			goto out;
669 		}
670 
671 		line[--line_len] = '\0'; /* \n */
672 
673 		sep = strrchr(line, 'x');
674 		if (sep == NULL)
675 			continue;
676 
677 		hex2u64(sep + 1, &start);
678 
679 		sep = strchr(line, ' ');
680 		if (sep == NULL)
681 			continue;
682 
683 		*sep = '\0';
684 
685 		scnprintf(name, sizeof(name), "[%s]", line);
686 
687 		size = strtoul(sep + 1, &endptr, 0);
688 		if (*endptr != ' ' && *endptr != '\t')
689 			continue;
690 
691 		err = process_module(arg, name, start, size);
692 		if (err)
693 			break;
694 	}
695 out:
696 	free(line);
697 	fclose(file);
698 	return err;
699 }
700 
701 /*
702  * These are symbols in the kernel image, so make sure that
703  * sym is from a kernel DSO.
704  */
705 static bool symbol__is_idle(const char *name)
706 {
707 	const char * const idle_symbols[] = {
708 		"acpi_idle_do_entry",
709 		"acpi_processor_ffh_cstate_enter",
710 		"arch_cpu_idle",
711 		"cpu_idle",
712 		"cpu_startup_entry",
713 		"idle_cpu",
714 		"intel_idle",
715 		"intel_idle_ibrs",
716 		"default_idle",
717 		"native_safe_halt",
718 		"enter_idle",
719 		"exit_idle",
720 		"mwait_idle",
721 		"mwait_idle_with_hints",
722 		"mwait_idle_with_hints.constprop.0",
723 		"poll_idle",
724 		"ppc64_runlatch_off",
725 		"pseries_dedicated_idle_sleep",
726 		"psw_idle",
727 		"psw_idle_exit",
728 		NULL
729 	};
730 	int i;
731 	static struct strlist *idle_symbols_list;
732 
733 	if (idle_symbols_list)
734 		return strlist__has_entry(idle_symbols_list, name);
735 
736 	idle_symbols_list = strlist__new(NULL, NULL);
737 
738 	for (i = 0; idle_symbols[i]; i++)
739 		strlist__add(idle_symbols_list, idle_symbols[i]);
740 
741 	return strlist__has_entry(idle_symbols_list, name);
742 }
743 
744 static int map__process_kallsym_symbol(void *arg, const char *name,
745 				       char type, u64 start)
746 {
747 	struct symbol *sym;
748 	struct dso *dso = arg;
749 	struct rb_root_cached *root = &dso->symbols;
750 
751 	if (!symbol_type__filter(type))
752 		return 0;
753 
754 	/* Ignore local symbols for ARM modules */
755 	if (name[0] == '$')
756 		return 0;
757 
758 	/*
759 	 * module symbols are not sorted so we add all
760 	 * symbols, setting length to 0, and rely on
761 	 * symbols__fixup_end() to fix it up.
762 	 */
763 	sym = symbol__new(start, 0, kallsyms2elf_binding(type), kallsyms2elf_type(type), name);
764 	if (sym == NULL)
765 		return -ENOMEM;
766 	/*
767 	 * We will pass the symbols to the filter later, in
768 	 * map__split_kallsyms, when we have split the maps per module
769 	 */
770 	__symbols__insert(root, sym, !strchr(name, '['));
771 
772 	return 0;
773 }
774 
775 /*
776  * Loads the function entries in /proc/kallsyms into kernel_map->dso,
777  * so that we can in the next step set the symbol ->end address and then
778  * call kernel_maps__split_kallsyms.
779  */
780 static int dso__load_all_kallsyms(struct dso *dso, const char *filename)
781 {
782 	return kallsyms__parse(filename, dso, map__process_kallsym_symbol);
783 }
784 
785 static int maps__split_kallsyms_for_kcore(struct maps *kmaps, struct dso *dso)
786 {
787 	struct symbol *pos;
788 	int count = 0;
789 	struct rb_root_cached old_root = dso->symbols;
790 	struct rb_root_cached *root = &dso->symbols;
791 	struct rb_node *next = rb_first_cached(root);
792 
793 	if (!kmaps)
794 		return -1;
795 
796 	*root = RB_ROOT_CACHED;
797 
798 	while (next) {
799 		struct map *curr_map;
800 		struct dso *curr_map_dso;
801 		char *module;
802 
803 		pos = rb_entry(next, struct symbol, rb_node);
804 		next = rb_next(&pos->rb_node);
805 
806 		rb_erase_cached(&pos->rb_node, &old_root);
807 		RB_CLEAR_NODE(&pos->rb_node);
808 		module = strchr(pos->name, '\t');
809 		if (module)
810 			*module = '\0';
811 
812 		curr_map = maps__find(kmaps, pos->start);
813 
814 		if (!curr_map) {
815 			symbol__delete(pos);
816 			continue;
817 		}
818 		curr_map_dso = map__dso(curr_map);
819 		pos->start -= map__start(curr_map) - map__pgoff(curr_map);
820 		if (pos->end > map__end(curr_map))
821 			pos->end = map__end(curr_map);
822 		if (pos->end)
823 			pos->end -= map__start(curr_map) - map__pgoff(curr_map);
824 		symbols__insert(&curr_map_dso->symbols, pos);
825 		++count;
826 		map__put(curr_map);
827 	}
828 
829 	/* Symbols have been adjusted */
830 	dso->adjust_symbols = 1;
831 
832 	return count;
833 }
834 
835 /*
836  * Split the symbols into maps, making sure there are no overlaps, i.e. the
837  * kernel range is broken in several maps, named [kernel].N, as we don't have
838  * the original ELF section names vmlinux have.
839  */
840 static int maps__split_kallsyms(struct maps *kmaps, struct dso *dso, u64 delta,
841 				struct map *initial_map)
842 {
843 	struct machine *machine;
844 	struct map *curr_map = map__get(initial_map);
845 	struct symbol *pos;
846 	int count = 0, moved = 0;
847 	struct rb_root_cached *root = &dso->symbols;
848 	struct rb_node *next = rb_first_cached(root);
849 	int kernel_range = 0;
850 	bool x86_64;
851 
852 	if (!kmaps)
853 		return -1;
854 
855 	machine = maps__machine(kmaps);
856 
857 	x86_64 = machine__is(machine, "x86_64");
858 
859 	while (next) {
860 		char *module;
861 
862 		pos = rb_entry(next, struct symbol, rb_node);
863 		next = rb_next(&pos->rb_node);
864 
865 		module = strchr(pos->name, '\t');
866 		if (module) {
867 			struct dso *curr_map_dso;
868 
869 			if (!symbol_conf.use_modules)
870 				goto discard_symbol;
871 
872 			*module++ = '\0';
873 			curr_map_dso = map__dso(curr_map);
874 			if (strcmp(curr_map_dso->short_name, module)) {
875 				if (!RC_CHK_EQUAL(curr_map, initial_map) &&
876 				    dso->kernel == DSO_SPACE__KERNEL_GUEST &&
877 				    machine__is_default_guest(machine)) {
878 					/*
879 					 * We assume all symbols of a module are
880 					 * continuous in * kallsyms, so curr_map
881 					 * points to a module and all its
882 					 * symbols are in its kmap. Mark it as
883 					 * loaded.
884 					 */
885 					dso__set_loaded(curr_map_dso);
886 				}
887 
888 				map__zput(curr_map);
889 				curr_map = maps__find_by_name(kmaps, module);
890 				if (curr_map == NULL) {
891 					pr_debug("%s/proc/{kallsyms,modules} "
892 					         "inconsistency while looking "
893 						 "for \"%s\" module!\n",
894 						 machine->root_dir, module);
895 					curr_map = map__get(initial_map);
896 					goto discard_symbol;
897 				}
898 				curr_map_dso = map__dso(curr_map);
899 				if (curr_map_dso->loaded &&
900 				    !machine__is_default_guest(machine))
901 					goto discard_symbol;
902 			}
903 			/*
904 			 * So that we look just like we get from .ko files,
905 			 * i.e. not prelinked, relative to initial_map->start.
906 			 */
907 			pos->start = map__map_ip(curr_map, pos->start);
908 			pos->end   = map__map_ip(curr_map, pos->end);
909 		} else if (x86_64 && is_entry_trampoline(pos->name)) {
910 			/*
911 			 * These symbols are not needed anymore since the
912 			 * trampoline maps refer to the text section and it's
913 			 * symbols instead. Avoid having to deal with
914 			 * relocations, and the assumption that the first symbol
915 			 * is the start of kernel text, by simply removing the
916 			 * symbols at this point.
917 			 */
918 			goto discard_symbol;
919 		} else if (!RC_CHK_EQUAL(curr_map, initial_map)) {
920 			char dso_name[PATH_MAX];
921 			struct dso *ndso;
922 
923 			if (delta) {
924 				/* Kernel was relocated at boot time */
925 				pos->start -= delta;
926 				pos->end -= delta;
927 			}
928 
929 			if (count == 0) {
930 				map__zput(curr_map);
931 				curr_map = map__get(initial_map);
932 				goto add_symbol;
933 			}
934 
935 			if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
936 				snprintf(dso_name, sizeof(dso_name),
937 					"[guest.kernel].%d",
938 					kernel_range++);
939 			else
940 				snprintf(dso_name, sizeof(dso_name),
941 					"[kernel].%d",
942 					kernel_range++);
943 
944 			ndso = dso__new(dso_name);
945 			map__zput(curr_map);
946 			if (ndso == NULL)
947 				return -1;
948 
949 			ndso->kernel = dso->kernel;
950 
951 			curr_map = map__new2(pos->start, ndso);
952 			if (curr_map == NULL) {
953 				dso__put(ndso);
954 				return -1;
955 			}
956 
957 			map__set_mapping_type(curr_map, MAPPING_TYPE__IDENTITY);
958 			if (maps__insert(kmaps, curr_map)) {
959 				map__zput(curr_map);
960 				dso__put(ndso);
961 				return -1;
962 			}
963 			++kernel_range;
964 		} else if (delta) {
965 			/* Kernel was relocated at boot time */
966 			pos->start -= delta;
967 			pos->end -= delta;
968 		}
969 add_symbol:
970 		if (!RC_CHK_EQUAL(curr_map, initial_map)) {
971 			struct dso *curr_map_dso = map__dso(curr_map);
972 
973 			rb_erase_cached(&pos->rb_node, root);
974 			symbols__insert(&curr_map_dso->symbols, pos);
975 			++moved;
976 		} else
977 			++count;
978 
979 		continue;
980 discard_symbol:
981 		rb_erase_cached(&pos->rb_node, root);
982 		symbol__delete(pos);
983 	}
984 
985 	if (!RC_CHK_EQUAL(curr_map, initial_map) &&
986 	    dso->kernel == DSO_SPACE__KERNEL_GUEST &&
987 	    machine__is_default_guest(maps__machine(kmaps))) {
988 		dso__set_loaded(map__dso(curr_map));
989 	}
990 	map__put(curr_map);
991 	return count + moved;
992 }
993 
994 bool symbol__restricted_filename(const char *filename,
995 				 const char *restricted_filename)
996 {
997 	bool restricted = false;
998 
999 	if (symbol_conf.kptr_restrict) {
1000 		char *r = realpath(filename, NULL);
1001 
1002 		if (r != NULL) {
1003 			restricted = strcmp(r, restricted_filename) == 0;
1004 			free(r);
1005 			return restricted;
1006 		}
1007 	}
1008 
1009 	return restricted;
1010 }
1011 
1012 struct module_info {
1013 	struct rb_node rb_node;
1014 	char *name;
1015 	u64 start;
1016 };
1017 
1018 static void add_module(struct module_info *mi, struct rb_root *modules)
1019 {
1020 	struct rb_node **p = &modules->rb_node;
1021 	struct rb_node *parent = NULL;
1022 	struct module_info *m;
1023 
1024 	while (*p != NULL) {
1025 		parent = *p;
1026 		m = rb_entry(parent, struct module_info, rb_node);
1027 		if (strcmp(mi->name, m->name) < 0)
1028 			p = &(*p)->rb_left;
1029 		else
1030 			p = &(*p)->rb_right;
1031 	}
1032 	rb_link_node(&mi->rb_node, parent, p);
1033 	rb_insert_color(&mi->rb_node, modules);
1034 }
1035 
1036 static void delete_modules(struct rb_root *modules)
1037 {
1038 	struct module_info *mi;
1039 	struct rb_node *next = rb_first(modules);
1040 
1041 	while (next) {
1042 		mi = rb_entry(next, struct module_info, rb_node);
1043 		next = rb_next(&mi->rb_node);
1044 		rb_erase(&mi->rb_node, modules);
1045 		zfree(&mi->name);
1046 		free(mi);
1047 	}
1048 }
1049 
1050 static struct module_info *find_module(const char *name,
1051 				       struct rb_root *modules)
1052 {
1053 	struct rb_node *n = modules->rb_node;
1054 
1055 	while (n) {
1056 		struct module_info *m;
1057 		int cmp;
1058 
1059 		m = rb_entry(n, struct module_info, rb_node);
1060 		cmp = strcmp(name, m->name);
1061 		if (cmp < 0)
1062 			n = n->rb_left;
1063 		else if (cmp > 0)
1064 			n = n->rb_right;
1065 		else
1066 			return m;
1067 	}
1068 
1069 	return NULL;
1070 }
1071 
1072 static int __read_proc_modules(void *arg, const char *name, u64 start,
1073 			       u64 size __maybe_unused)
1074 {
1075 	struct rb_root *modules = arg;
1076 	struct module_info *mi;
1077 
1078 	mi = zalloc(sizeof(struct module_info));
1079 	if (!mi)
1080 		return -ENOMEM;
1081 
1082 	mi->name = strdup(name);
1083 	mi->start = start;
1084 
1085 	if (!mi->name) {
1086 		free(mi);
1087 		return -ENOMEM;
1088 	}
1089 
1090 	add_module(mi, modules);
1091 
1092 	return 0;
1093 }
1094 
1095 static int read_proc_modules(const char *filename, struct rb_root *modules)
1096 {
1097 	if (symbol__restricted_filename(filename, "/proc/modules"))
1098 		return -1;
1099 
1100 	if (modules__parse(filename, modules, __read_proc_modules)) {
1101 		delete_modules(modules);
1102 		return -1;
1103 	}
1104 
1105 	return 0;
1106 }
1107 
1108 int compare_proc_modules(const char *from, const char *to)
1109 {
1110 	struct rb_root from_modules = RB_ROOT;
1111 	struct rb_root to_modules = RB_ROOT;
1112 	struct rb_node *from_node, *to_node;
1113 	struct module_info *from_m, *to_m;
1114 	int ret = -1;
1115 
1116 	if (read_proc_modules(from, &from_modules))
1117 		return -1;
1118 
1119 	if (read_proc_modules(to, &to_modules))
1120 		goto out_delete_from;
1121 
1122 	from_node = rb_first(&from_modules);
1123 	to_node = rb_first(&to_modules);
1124 	while (from_node) {
1125 		if (!to_node)
1126 			break;
1127 
1128 		from_m = rb_entry(from_node, struct module_info, rb_node);
1129 		to_m = rb_entry(to_node, struct module_info, rb_node);
1130 
1131 		if (from_m->start != to_m->start ||
1132 		    strcmp(from_m->name, to_m->name))
1133 			break;
1134 
1135 		from_node = rb_next(from_node);
1136 		to_node = rb_next(to_node);
1137 	}
1138 
1139 	if (!from_node && !to_node)
1140 		ret = 0;
1141 
1142 	delete_modules(&to_modules);
1143 out_delete_from:
1144 	delete_modules(&from_modules);
1145 
1146 	return ret;
1147 }
1148 
1149 static int do_validate_kcore_modules_cb(struct map *old_map, void *data)
1150 {
1151 	struct rb_root *modules = data;
1152 	struct module_info *mi;
1153 	struct dso *dso;
1154 
1155 	if (!__map__is_kmodule(old_map))
1156 		return 0;
1157 
1158 	dso = map__dso(old_map);
1159 	/* Module must be in memory at the same address */
1160 	mi = find_module(dso->short_name, modules);
1161 	if (!mi || mi->start != map__start(old_map))
1162 		return -EINVAL;
1163 
1164 	return 0;
1165 }
1166 
1167 static int do_validate_kcore_modules(const char *filename, struct maps *kmaps)
1168 {
1169 	struct rb_root modules = RB_ROOT;
1170 	int err;
1171 
1172 	err = read_proc_modules(filename, &modules);
1173 	if (err)
1174 		return err;
1175 
1176 	err = maps__for_each_map(kmaps, do_validate_kcore_modules_cb, &modules);
1177 
1178 	delete_modules(&modules);
1179 	return err;
1180 }
1181 
1182 /*
1183  * If kallsyms is referenced by name then we look for filename in the same
1184  * directory.
1185  */
1186 static bool filename_from_kallsyms_filename(char *filename,
1187 					    const char *base_name,
1188 					    const char *kallsyms_filename)
1189 {
1190 	char *name;
1191 
1192 	strcpy(filename, kallsyms_filename);
1193 	name = strrchr(filename, '/');
1194 	if (!name)
1195 		return false;
1196 
1197 	name += 1;
1198 
1199 	if (!strcmp(name, "kallsyms")) {
1200 		strcpy(name, base_name);
1201 		return true;
1202 	}
1203 
1204 	return false;
1205 }
1206 
1207 static int validate_kcore_modules(const char *kallsyms_filename,
1208 				  struct map *map)
1209 {
1210 	struct maps *kmaps = map__kmaps(map);
1211 	char modules_filename[PATH_MAX];
1212 
1213 	if (!kmaps)
1214 		return -EINVAL;
1215 
1216 	if (!filename_from_kallsyms_filename(modules_filename, "modules",
1217 					     kallsyms_filename))
1218 		return -EINVAL;
1219 
1220 	if (do_validate_kcore_modules(modules_filename, kmaps))
1221 		return -EINVAL;
1222 
1223 	return 0;
1224 }
1225 
1226 static int validate_kcore_addresses(const char *kallsyms_filename,
1227 				    struct map *map)
1228 {
1229 	struct kmap *kmap = map__kmap(map);
1230 
1231 	if (!kmap)
1232 		return -EINVAL;
1233 
1234 	if (kmap->ref_reloc_sym && kmap->ref_reloc_sym->name) {
1235 		u64 start;
1236 
1237 		if (kallsyms__get_function_start(kallsyms_filename,
1238 						 kmap->ref_reloc_sym->name, &start))
1239 			return -ENOENT;
1240 		if (start != kmap->ref_reloc_sym->addr)
1241 			return -EINVAL;
1242 	}
1243 
1244 	return validate_kcore_modules(kallsyms_filename, map);
1245 }
1246 
1247 struct kcore_mapfn_data {
1248 	struct dso *dso;
1249 	struct list_head maps;
1250 };
1251 
1252 static int kcore_mapfn(u64 start, u64 len, u64 pgoff, void *data)
1253 {
1254 	struct kcore_mapfn_data *md = data;
1255 	struct map_list_node *list_node = map_list_node__new();
1256 
1257 	if (!list_node)
1258 		return -ENOMEM;
1259 
1260 	list_node->map = map__new2(start, md->dso);
1261 	if (!list_node->map) {
1262 		free(list_node);
1263 		return -ENOMEM;
1264 	}
1265 
1266 	map__set_end(list_node->map, map__start(list_node->map) + len);
1267 	map__set_pgoff(list_node->map, pgoff);
1268 
1269 	list_add(&list_node->node, &md->maps);
1270 
1271 	return 0;
1272 }
1273 
1274 static bool remove_old_maps(struct map *map, void *data)
1275 {
1276 	const struct map *map_to_save = data;
1277 
1278 	/*
1279 	 * We need to preserve eBPF maps even if they are covered by kcore,
1280 	 * because we need to access eBPF dso for source data.
1281 	 */
1282 	return !RC_CHK_EQUAL(map, map_to_save) && !__map__is_bpf_prog(map);
1283 }
1284 
1285 static int dso__load_kcore(struct dso *dso, struct map *map,
1286 			   const char *kallsyms_filename)
1287 {
1288 	struct maps *kmaps = map__kmaps(map);
1289 	struct kcore_mapfn_data md;
1290 	struct map *replacement_map = NULL;
1291 	struct machine *machine;
1292 	bool is_64_bit;
1293 	int err, fd;
1294 	char kcore_filename[PATH_MAX];
1295 	u64 stext;
1296 
1297 	if (!kmaps)
1298 		return -EINVAL;
1299 
1300 	machine = maps__machine(kmaps);
1301 
1302 	/* This function requires that the map is the kernel map */
1303 	if (!__map__is_kernel(map))
1304 		return -EINVAL;
1305 
1306 	if (!filename_from_kallsyms_filename(kcore_filename, "kcore",
1307 					     kallsyms_filename))
1308 		return -EINVAL;
1309 
1310 	/* Modules and kernel must be present at their original addresses */
1311 	if (validate_kcore_addresses(kallsyms_filename, map))
1312 		return -EINVAL;
1313 
1314 	md.dso = dso;
1315 	INIT_LIST_HEAD(&md.maps);
1316 
1317 	fd = open(kcore_filename, O_RDONLY);
1318 	if (fd < 0) {
1319 		pr_debug("Failed to open %s. Note /proc/kcore requires CAP_SYS_RAWIO capability to access.\n",
1320 			 kcore_filename);
1321 		return -EINVAL;
1322 	}
1323 
1324 	/* Read new maps into temporary lists */
1325 	err = file__read_maps(fd, map__prot(map) & PROT_EXEC, kcore_mapfn, &md,
1326 			      &is_64_bit);
1327 	if (err)
1328 		goto out_err;
1329 	dso->is_64_bit = is_64_bit;
1330 
1331 	if (list_empty(&md.maps)) {
1332 		err = -EINVAL;
1333 		goto out_err;
1334 	}
1335 
1336 	/* Remove old maps */
1337 	maps__remove_maps(kmaps, remove_old_maps, map);
1338 	machine->trampolines_mapped = false;
1339 
1340 	/* Find the kernel map using the '_stext' symbol */
1341 	if (!kallsyms__get_function_start(kallsyms_filename, "_stext", &stext)) {
1342 		u64 replacement_size = 0;
1343 		struct map_list_node *new_node;
1344 
1345 		list_for_each_entry(new_node, &md.maps, node) {
1346 			struct map *new_map = new_node->map;
1347 			u64 new_size = map__size(new_map);
1348 
1349 			if (!(stext >= map__start(new_map) && stext < map__end(new_map)))
1350 				continue;
1351 
1352 			/*
1353 			 * On some architectures, ARM64 for example, the kernel
1354 			 * text can get allocated inside of the vmalloc segment.
1355 			 * Select the smallest matching segment, in case stext
1356 			 * falls within more than one in the list.
1357 			 */
1358 			if (!replacement_map || new_size < replacement_size) {
1359 				replacement_map = new_map;
1360 				replacement_size = new_size;
1361 			}
1362 		}
1363 	}
1364 
1365 	if (!replacement_map)
1366 		replacement_map = list_entry(md.maps.next, struct map_list_node, node)->map;
1367 
1368 	/* Add new maps */
1369 	while (!list_empty(&md.maps)) {
1370 		struct map_list_node *new_node = list_entry(md.maps.next, struct map_list_node, node);
1371 		struct map *new_map = new_node->map;
1372 
1373 		list_del_init(&new_node->node);
1374 
1375 		if (RC_CHK_EQUAL(new_map, replacement_map)) {
1376 			struct map *map_ref;
1377 
1378 			map__set_start(map, map__start(new_map));
1379 			map__set_end(map, map__end(new_map));
1380 			map__set_pgoff(map, map__pgoff(new_map));
1381 			map__set_mapping_type(map, map__mapping_type(new_map));
1382 			/* Ensure maps are correctly ordered */
1383 			map_ref = map__get(map);
1384 			maps__remove(kmaps, map_ref);
1385 			err = maps__insert(kmaps, map_ref);
1386 			map__put(map_ref);
1387 			map__put(new_map);
1388 			if (err)
1389 				goto out_err;
1390 		} else {
1391 			/*
1392 			 * Merge kcore map into existing maps,
1393 			 * and ensure that current maps (eBPF)
1394 			 * stay intact.
1395 			 */
1396 			if (maps__merge_in(kmaps, new_map)) {
1397 				err = -EINVAL;
1398 				goto out_err;
1399 			}
1400 		}
1401 		free(new_node);
1402 	}
1403 
1404 	if (machine__is(machine, "x86_64")) {
1405 		u64 addr;
1406 
1407 		/*
1408 		 * If one of the corresponding symbols is there, assume the
1409 		 * entry trampoline maps are too.
1410 		 */
1411 		if (!kallsyms__get_function_start(kallsyms_filename,
1412 						  ENTRY_TRAMPOLINE_NAME,
1413 						  &addr))
1414 			machine->trampolines_mapped = true;
1415 	}
1416 
1417 	/*
1418 	 * Set the data type and long name so that kcore can be read via
1419 	 * dso__data_read_addr().
1420 	 */
1421 	if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
1422 		dso->binary_type = DSO_BINARY_TYPE__GUEST_KCORE;
1423 	else
1424 		dso->binary_type = DSO_BINARY_TYPE__KCORE;
1425 	dso__set_long_name(dso, strdup(kcore_filename), true);
1426 
1427 	close(fd);
1428 
1429 	if (map__prot(map) & PROT_EXEC)
1430 		pr_debug("Using %s for kernel object code\n", kcore_filename);
1431 	else
1432 		pr_debug("Using %s for kernel data\n", kcore_filename);
1433 
1434 	return 0;
1435 
1436 out_err:
1437 	while (!list_empty(&md.maps)) {
1438 		struct map_list_node *list_node;
1439 
1440 		list_node = list_entry(md.maps.next, struct map_list_node, node);
1441 		list_del_init(&list_node->node);
1442 		map__zput(list_node->map);
1443 		free(list_node);
1444 	}
1445 	close(fd);
1446 	return err;
1447 }
1448 
1449 /*
1450  * If the kernel is relocated at boot time, kallsyms won't match.  Compute the
1451  * delta based on the relocation reference symbol.
1452  */
1453 static int kallsyms__delta(struct kmap *kmap, const char *filename, u64 *delta)
1454 {
1455 	u64 addr;
1456 
1457 	if (!kmap->ref_reloc_sym || !kmap->ref_reloc_sym->name)
1458 		return 0;
1459 
1460 	if (kallsyms__get_function_start(filename, kmap->ref_reloc_sym->name, &addr))
1461 		return -1;
1462 
1463 	*delta = addr - kmap->ref_reloc_sym->addr;
1464 	return 0;
1465 }
1466 
1467 int __dso__load_kallsyms(struct dso *dso, const char *filename,
1468 			 struct map *map, bool no_kcore)
1469 {
1470 	struct kmap *kmap = map__kmap(map);
1471 	u64 delta = 0;
1472 
1473 	if (symbol__restricted_filename(filename, "/proc/kallsyms"))
1474 		return -1;
1475 
1476 	if (!kmap || !kmap->kmaps)
1477 		return -1;
1478 
1479 	if (dso__load_all_kallsyms(dso, filename) < 0)
1480 		return -1;
1481 
1482 	if (kallsyms__delta(kmap, filename, &delta))
1483 		return -1;
1484 
1485 	symbols__fixup_end(&dso->symbols, true);
1486 	symbols__fixup_duplicate(&dso->symbols);
1487 
1488 	if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
1489 		dso->symtab_type = DSO_BINARY_TYPE__GUEST_KALLSYMS;
1490 	else
1491 		dso->symtab_type = DSO_BINARY_TYPE__KALLSYMS;
1492 
1493 	if (!no_kcore && !dso__load_kcore(dso, map, filename))
1494 		return maps__split_kallsyms_for_kcore(kmap->kmaps, dso);
1495 	else
1496 		return maps__split_kallsyms(kmap->kmaps, dso, delta, map);
1497 }
1498 
1499 int dso__load_kallsyms(struct dso *dso, const char *filename,
1500 		       struct map *map)
1501 {
1502 	return __dso__load_kallsyms(dso, filename, map, false);
1503 }
1504 
1505 static int dso__load_perf_map(const char *map_path, struct dso *dso)
1506 {
1507 	char *line = NULL;
1508 	size_t n;
1509 	FILE *file;
1510 	int nr_syms = 0;
1511 
1512 	file = fopen(map_path, "r");
1513 	if (file == NULL)
1514 		goto out_failure;
1515 
1516 	while (!feof(file)) {
1517 		u64 start, size;
1518 		struct symbol *sym;
1519 		int line_len, len;
1520 
1521 		line_len = getline(&line, &n, file);
1522 		if (line_len < 0)
1523 			break;
1524 
1525 		if (!line)
1526 			goto out_failure;
1527 
1528 		line[--line_len] = '\0'; /* \n */
1529 
1530 		len = hex2u64(line, &start);
1531 
1532 		len++;
1533 		if (len + 2 >= line_len)
1534 			continue;
1535 
1536 		len += hex2u64(line + len, &size);
1537 
1538 		len++;
1539 		if (len + 2 >= line_len)
1540 			continue;
1541 
1542 		sym = symbol__new(start, size, STB_GLOBAL, STT_FUNC, line + len);
1543 
1544 		if (sym == NULL)
1545 			goto out_delete_line;
1546 
1547 		symbols__insert(&dso->symbols, sym);
1548 		nr_syms++;
1549 	}
1550 
1551 	free(line);
1552 	fclose(file);
1553 
1554 	return nr_syms;
1555 
1556 out_delete_line:
1557 	free(line);
1558 out_failure:
1559 	return -1;
1560 }
1561 
1562 #ifdef HAVE_LIBBFD_SUPPORT
1563 #define PACKAGE 'perf'
1564 #include <bfd.h>
1565 
1566 static int bfd_symbols__cmpvalue(const void *a, const void *b)
1567 {
1568 	const asymbol *as = *(const asymbol **)a, *bs = *(const asymbol **)b;
1569 
1570 	if (bfd_asymbol_value(as) != bfd_asymbol_value(bs))
1571 		return bfd_asymbol_value(as) - bfd_asymbol_value(bs);
1572 
1573 	return bfd_asymbol_name(as)[0] - bfd_asymbol_name(bs)[0];
1574 }
1575 
1576 static int bfd2elf_binding(asymbol *symbol)
1577 {
1578 	if (symbol->flags & BSF_WEAK)
1579 		return STB_WEAK;
1580 	if (symbol->flags & BSF_GLOBAL)
1581 		return STB_GLOBAL;
1582 	if (symbol->flags & BSF_LOCAL)
1583 		return STB_LOCAL;
1584 	return -1;
1585 }
1586 
1587 int dso__load_bfd_symbols(struct dso *dso, const char *debugfile)
1588 {
1589 	int err = -1;
1590 	long symbols_size, symbols_count, i;
1591 	asection *section;
1592 	asymbol **symbols, *sym;
1593 	struct symbol *symbol;
1594 	bfd *abfd;
1595 	u64 start, len;
1596 
1597 	abfd = bfd_openr(debugfile, NULL);
1598 	if (!abfd)
1599 		return -1;
1600 
1601 	if (!bfd_check_format(abfd, bfd_object)) {
1602 		pr_debug2("%s: cannot read %s bfd file.\n", __func__,
1603 			  dso->long_name);
1604 		goto out_close;
1605 	}
1606 
1607 	if (bfd_get_flavour(abfd) == bfd_target_elf_flavour)
1608 		goto out_close;
1609 
1610 	symbols_size = bfd_get_symtab_upper_bound(abfd);
1611 	if (symbols_size == 0) {
1612 		bfd_close(abfd);
1613 		return 0;
1614 	}
1615 
1616 	if (symbols_size < 0)
1617 		goto out_close;
1618 
1619 	symbols = malloc(symbols_size);
1620 	if (!symbols)
1621 		goto out_close;
1622 
1623 	symbols_count = bfd_canonicalize_symtab(abfd, symbols);
1624 	if (symbols_count < 0)
1625 		goto out_free;
1626 
1627 	section = bfd_get_section_by_name(abfd, ".text");
1628 	if (section) {
1629 		for (i = 0; i < symbols_count; ++i) {
1630 			if (!strcmp(bfd_asymbol_name(symbols[i]), "__ImageBase") ||
1631 			    !strcmp(bfd_asymbol_name(symbols[i]), "__image_base__"))
1632 				break;
1633 		}
1634 		if (i < symbols_count) {
1635 			/* PE symbols can only have 4 bytes, so use .text high bits */
1636 			dso->text_offset = section->vma - (u32)section->vma;
1637 			dso->text_offset += (u32)bfd_asymbol_value(symbols[i]);
1638 			dso->text_end = (section->vma - dso->text_offset) + section->size;
1639 		} else {
1640 			dso->text_offset = section->vma - section->filepos;
1641 			dso->text_end = section->filepos + section->size;
1642 		}
1643 	}
1644 
1645 	qsort(symbols, symbols_count, sizeof(asymbol *), bfd_symbols__cmpvalue);
1646 
1647 #ifdef bfd_get_section
1648 #define bfd_asymbol_section bfd_get_section
1649 #endif
1650 	for (i = 0; i < symbols_count; ++i) {
1651 		sym = symbols[i];
1652 		section = bfd_asymbol_section(sym);
1653 		if (bfd2elf_binding(sym) < 0)
1654 			continue;
1655 
1656 		while (i + 1 < symbols_count &&
1657 		       bfd_asymbol_section(symbols[i + 1]) == section &&
1658 		       bfd2elf_binding(symbols[i + 1]) < 0)
1659 			i++;
1660 
1661 		if (i + 1 < symbols_count &&
1662 		    bfd_asymbol_section(symbols[i + 1]) == section)
1663 			len = symbols[i + 1]->value - sym->value;
1664 		else
1665 			len = section->size - sym->value;
1666 
1667 		start = bfd_asymbol_value(sym) - dso->text_offset;
1668 		symbol = symbol__new(start, len, bfd2elf_binding(sym), STT_FUNC,
1669 				     bfd_asymbol_name(sym));
1670 		if (!symbol)
1671 			goto out_free;
1672 
1673 		symbols__insert(&dso->symbols, symbol);
1674 	}
1675 #ifdef bfd_get_section
1676 #undef bfd_asymbol_section
1677 #endif
1678 
1679 	symbols__fixup_end(&dso->symbols, false);
1680 	symbols__fixup_duplicate(&dso->symbols);
1681 	dso->adjust_symbols = 1;
1682 
1683 	err = 0;
1684 out_free:
1685 	free(symbols);
1686 out_close:
1687 	bfd_close(abfd);
1688 	return err;
1689 }
1690 #endif
1691 
1692 static bool dso__is_compatible_symtab_type(struct dso *dso, bool kmod,
1693 					   enum dso_binary_type type)
1694 {
1695 	switch (type) {
1696 	case DSO_BINARY_TYPE__JAVA_JIT:
1697 	case DSO_BINARY_TYPE__DEBUGLINK:
1698 	case DSO_BINARY_TYPE__SYSTEM_PATH_DSO:
1699 	case DSO_BINARY_TYPE__FEDORA_DEBUGINFO:
1700 	case DSO_BINARY_TYPE__UBUNTU_DEBUGINFO:
1701 	case DSO_BINARY_TYPE__MIXEDUP_UBUNTU_DEBUGINFO:
1702 	case DSO_BINARY_TYPE__BUILDID_DEBUGINFO:
1703 	case DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO:
1704 		return !kmod && dso->kernel == DSO_SPACE__USER;
1705 
1706 	case DSO_BINARY_TYPE__KALLSYMS:
1707 	case DSO_BINARY_TYPE__VMLINUX:
1708 	case DSO_BINARY_TYPE__KCORE:
1709 		return dso->kernel == DSO_SPACE__KERNEL;
1710 
1711 	case DSO_BINARY_TYPE__GUEST_KALLSYMS:
1712 	case DSO_BINARY_TYPE__GUEST_VMLINUX:
1713 	case DSO_BINARY_TYPE__GUEST_KCORE:
1714 		return dso->kernel == DSO_SPACE__KERNEL_GUEST;
1715 
1716 	case DSO_BINARY_TYPE__GUEST_KMODULE:
1717 	case DSO_BINARY_TYPE__GUEST_KMODULE_COMP:
1718 	case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE:
1719 	case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP:
1720 		/*
1721 		 * kernel modules know their symtab type - it's set when
1722 		 * creating a module dso in machine__addnew_module_map().
1723 		 */
1724 		return kmod && dso->symtab_type == type;
1725 
1726 	case DSO_BINARY_TYPE__BUILD_ID_CACHE:
1727 	case DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO:
1728 		return true;
1729 
1730 	case DSO_BINARY_TYPE__BPF_PROG_INFO:
1731 	case DSO_BINARY_TYPE__BPF_IMAGE:
1732 	case DSO_BINARY_TYPE__OOL:
1733 	case DSO_BINARY_TYPE__NOT_FOUND:
1734 	default:
1735 		return false;
1736 	}
1737 }
1738 
1739 /* Checks for the existence of the perf-<pid>.map file in two different
1740  * locations.  First, if the process is a separate mount namespace, check in
1741  * that namespace using the pid of the innermost pid namespace.  If's not in a
1742  * namespace, or the file can't be found there, try in the mount namespace of
1743  * the tracing process using our view of its pid.
1744  */
1745 static int dso__find_perf_map(char *filebuf, size_t bufsz,
1746 			      struct nsinfo **nsip)
1747 {
1748 	struct nscookie nsc;
1749 	struct nsinfo *nsi;
1750 	struct nsinfo *nnsi;
1751 	int rc = -1;
1752 
1753 	nsi = *nsip;
1754 
1755 	if (nsinfo__need_setns(nsi)) {
1756 		snprintf(filebuf, bufsz, "/tmp/perf-%d.map", nsinfo__nstgid(nsi));
1757 		nsinfo__mountns_enter(nsi, &nsc);
1758 		rc = access(filebuf, R_OK);
1759 		nsinfo__mountns_exit(&nsc);
1760 		if (rc == 0)
1761 			return rc;
1762 	}
1763 
1764 	nnsi = nsinfo__copy(nsi);
1765 	if (nnsi) {
1766 		nsinfo__put(nsi);
1767 
1768 		nsinfo__clear_need_setns(nnsi);
1769 		snprintf(filebuf, bufsz, "/tmp/perf-%d.map", nsinfo__tgid(nnsi));
1770 		*nsip = nnsi;
1771 		rc = 0;
1772 	}
1773 
1774 	return rc;
1775 }
1776 
1777 int dso__load(struct dso *dso, struct map *map)
1778 {
1779 	char *name;
1780 	int ret = -1;
1781 	u_int i;
1782 	struct machine *machine = NULL;
1783 	char *root_dir = (char *) "";
1784 	int ss_pos = 0;
1785 	struct symsrc ss_[2];
1786 	struct symsrc *syms_ss = NULL, *runtime_ss = NULL;
1787 	bool kmod;
1788 	bool perfmap;
1789 	struct build_id bid;
1790 	struct nscookie nsc;
1791 	char newmapname[PATH_MAX];
1792 	const char *map_path = dso->long_name;
1793 
1794 	mutex_lock(&dso->lock);
1795 	perfmap = strncmp(dso->name, "/tmp/perf-", 10) == 0;
1796 	if (perfmap) {
1797 		if (dso->nsinfo && (dso__find_perf_map(newmapname,
1798 		    sizeof(newmapname), &dso->nsinfo) == 0)) {
1799 			map_path = newmapname;
1800 		}
1801 	}
1802 
1803 	nsinfo__mountns_enter(dso->nsinfo, &nsc);
1804 
1805 	/* check again under the dso->lock */
1806 	if (dso__loaded(dso)) {
1807 		ret = 1;
1808 		goto out;
1809 	}
1810 
1811 	kmod = dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1812 		dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP ||
1813 		dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE ||
1814 		dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE_COMP;
1815 
1816 	if (dso->kernel && !kmod) {
1817 		if (dso->kernel == DSO_SPACE__KERNEL)
1818 			ret = dso__load_kernel_sym(dso, map);
1819 		else if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
1820 			ret = dso__load_guest_kernel_sym(dso, map);
1821 
1822 		machine = maps__machine(map__kmaps(map));
1823 		if (machine__is(machine, "x86_64"))
1824 			machine__map_x86_64_entry_trampolines(machine, dso);
1825 		goto out;
1826 	}
1827 
1828 	dso->adjust_symbols = 0;
1829 
1830 	if (perfmap) {
1831 		ret = dso__load_perf_map(map_path, dso);
1832 		dso->symtab_type = ret > 0 ? DSO_BINARY_TYPE__JAVA_JIT :
1833 					     DSO_BINARY_TYPE__NOT_FOUND;
1834 		goto out;
1835 	}
1836 
1837 	if (machine)
1838 		root_dir = machine->root_dir;
1839 
1840 	name = malloc(PATH_MAX);
1841 	if (!name)
1842 		goto out;
1843 
1844 	/*
1845 	 * Read the build id if possible. This is required for
1846 	 * DSO_BINARY_TYPE__BUILDID_DEBUGINFO to work
1847 	 */
1848 	if (!dso->has_build_id &&
1849 	    is_regular_file(dso->long_name)) {
1850 	    __symbol__join_symfs(name, PATH_MAX, dso->long_name);
1851 		if (filename__read_build_id(name, &bid) > 0)
1852 			dso__set_build_id(dso, &bid);
1853 	}
1854 
1855 	/*
1856 	 * Iterate over candidate debug images.
1857 	 * Keep track of "interesting" ones (those which have a symtab, dynsym,
1858 	 * and/or opd section) for processing.
1859 	 */
1860 	for (i = 0; i < DSO_BINARY_TYPE__SYMTAB_CNT; i++) {
1861 		struct symsrc *ss = &ss_[ss_pos];
1862 		bool next_slot = false;
1863 		bool is_reg;
1864 		bool nsexit;
1865 		int bfdrc = -1;
1866 		int sirc = -1;
1867 
1868 		enum dso_binary_type symtab_type = binary_type_symtab[i];
1869 
1870 		nsexit = (symtab_type == DSO_BINARY_TYPE__BUILD_ID_CACHE ||
1871 		    symtab_type == DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO);
1872 
1873 		if (!dso__is_compatible_symtab_type(dso, kmod, symtab_type))
1874 			continue;
1875 
1876 		if (dso__read_binary_type_filename(dso, symtab_type,
1877 						   root_dir, name, PATH_MAX))
1878 			continue;
1879 
1880 		if (nsexit)
1881 			nsinfo__mountns_exit(&nsc);
1882 
1883 		is_reg = is_regular_file(name);
1884 		if (!is_reg && errno == ENOENT && dso->nsinfo) {
1885 			char *new_name = dso__filename_with_chroot(dso, name);
1886 			if (new_name) {
1887 				is_reg = is_regular_file(new_name);
1888 				strlcpy(name, new_name, PATH_MAX);
1889 				free(new_name);
1890 			}
1891 		}
1892 
1893 #ifdef HAVE_LIBBFD_SUPPORT
1894 		if (is_reg)
1895 			bfdrc = dso__load_bfd_symbols(dso, name);
1896 #endif
1897 		if (is_reg && bfdrc < 0)
1898 			sirc = symsrc__init(ss, dso, name, symtab_type);
1899 
1900 		if (nsexit)
1901 			nsinfo__mountns_enter(dso->nsinfo, &nsc);
1902 
1903 		if (bfdrc == 0) {
1904 			ret = 0;
1905 			break;
1906 		}
1907 
1908 		if (!is_reg || sirc < 0)
1909 			continue;
1910 
1911 		if (!syms_ss && symsrc__has_symtab(ss)) {
1912 			syms_ss = ss;
1913 			next_slot = true;
1914 			if (!dso->symsrc_filename)
1915 				dso->symsrc_filename = strdup(name);
1916 		}
1917 
1918 		if (!runtime_ss && symsrc__possibly_runtime(ss)) {
1919 			runtime_ss = ss;
1920 			next_slot = true;
1921 		}
1922 
1923 		if (next_slot) {
1924 			ss_pos++;
1925 
1926 			if (syms_ss && runtime_ss)
1927 				break;
1928 		} else {
1929 			symsrc__destroy(ss);
1930 		}
1931 
1932 	}
1933 
1934 	if (!runtime_ss && !syms_ss)
1935 		goto out_free;
1936 
1937 	if (runtime_ss && !syms_ss) {
1938 		syms_ss = runtime_ss;
1939 	}
1940 
1941 	/* We'll have to hope for the best */
1942 	if (!runtime_ss && syms_ss)
1943 		runtime_ss = syms_ss;
1944 
1945 	if (syms_ss)
1946 		ret = dso__load_sym(dso, map, syms_ss, runtime_ss, kmod);
1947 	else
1948 		ret = -1;
1949 
1950 	if (ret > 0) {
1951 		int nr_plt;
1952 
1953 		nr_plt = dso__synthesize_plt_symbols(dso, runtime_ss);
1954 		if (nr_plt > 0)
1955 			ret += nr_plt;
1956 	}
1957 
1958 	for (; ss_pos > 0; ss_pos--)
1959 		symsrc__destroy(&ss_[ss_pos - 1]);
1960 out_free:
1961 	free(name);
1962 	if (ret < 0 && strstr(dso->name, " (deleted)") != NULL)
1963 		ret = 0;
1964 out:
1965 	dso__set_loaded(dso);
1966 	mutex_unlock(&dso->lock);
1967 	nsinfo__mountns_exit(&nsc);
1968 
1969 	return ret;
1970 }
1971 
1972 int dso__load_vmlinux(struct dso *dso, struct map *map,
1973 		      const char *vmlinux, bool vmlinux_allocated)
1974 {
1975 	int err = -1;
1976 	struct symsrc ss;
1977 	char symfs_vmlinux[PATH_MAX];
1978 	enum dso_binary_type symtab_type;
1979 
1980 	if (vmlinux[0] == '/')
1981 		snprintf(symfs_vmlinux, sizeof(symfs_vmlinux), "%s", vmlinux);
1982 	else
1983 		symbol__join_symfs(symfs_vmlinux, vmlinux);
1984 
1985 	if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
1986 		symtab_type = DSO_BINARY_TYPE__GUEST_VMLINUX;
1987 	else
1988 		symtab_type = DSO_BINARY_TYPE__VMLINUX;
1989 
1990 	if (symsrc__init(&ss, dso, symfs_vmlinux, symtab_type))
1991 		return -1;
1992 
1993 	/*
1994 	 * dso__load_sym() may copy 'dso' which will result in the copies having
1995 	 * an incorrect long name unless we set it here first.
1996 	 */
1997 	dso__set_long_name(dso, vmlinux, vmlinux_allocated);
1998 	if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
1999 		dso->binary_type = DSO_BINARY_TYPE__GUEST_VMLINUX;
2000 	else
2001 		dso->binary_type = DSO_BINARY_TYPE__VMLINUX;
2002 
2003 	err = dso__load_sym(dso, map, &ss, &ss, 0);
2004 	symsrc__destroy(&ss);
2005 
2006 	if (err > 0) {
2007 		dso__set_loaded(dso);
2008 		pr_debug("Using %s for symbols\n", symfs_vmlinux);
2009 	}
2010 
2011 	return err;
2012 }
2013 
2014 int dso__load_vmlinux_path(struct dso *dso, struct map *map)
2015 {
2016 	int i, err = 0;
2017 	char *filename = NULL;
2018 
2019 	pr_debug("Looking at the vmlinux_path (%d entries long)\n",
2020 		 vmlinux_path__nr_entries + 1);
2021 
2022 	for (i = 0; i < vmlinux_path__nr_entries; ++i) {
2023 		err = dso__load_vmlinux(dso, map, vmlinux_path[i], false);
2024 		if (err > 0)
2025 			goto out;
2026 	}
2027 
2028 	if (!symbol_conf.ignore_vmlinux_buildid)
2029 		filename = dso__build_id_filename(dso, NULL, 0, false);
2030 	if (filename != NULL) {
2031 		err = dso__load_vmlinux(dso, map, filename, true);
2032 		if (err > 0)
2033 			goto out;
2034 		free(filename);
2035 	}
2036 out:
2037 	return err;
2038 }
2039 
2040 static bool visible_dir_filter(const char *name, struct dirent *d)
2041 {
2042 	if (d->d_type != DT_DIR)
2043 		return false;
2044 	return lsdir_no_dot_filter(name, d);
2045 }
2046 
2047 static int find_matching_kcore(struct map *map, char *dir, size_t dir_sz)
2048 {
2049 	char kallsyms_filename[PATH_MAX];
2050 	int ret = -1;
2051 	struct strlist *dirs;
2052 	struct str_node *nd;
2053 
2054 	dirs = lsdir(dir, visible_dir_filter);
2055 	if (!dirs)
2056 		return -1;
2057 
2058 	strlist__for_each_entry(nd, dirs) {
2059 		scnprintf(kallsyms_filename, sizeof(kallsyms_filename),
2060 			  "%s/%s/kallsyms", dir, nd->s);
2061 		if (!validate_kcore_addresses(kallsyms_filename, map)) {
2062 			strlcpy(dir, kallsyms_filename, dir_sz);
2063 			ret = 0;
2064 			break;
2065 		}
2066 	}
2067 
2068 	strlist__delete(dirs);
2069 
2070 	return ret;
2071 }
2072 
2073 /*
2074  * Use open(O_RDONLY) to check readability directly instead of access(R_OK)
2075  * since access(R_OK) only checks with real UID/GID but open() use effective
2076  * UID/GID and actual capabilities (e.g. /proc/kcore requires CAP_SYS_RAWIO).
2077  */
2078 static bool filename__readable(const char *file)
2079 {
2080 	int fd = open(file, O_RDONLY);
2081 	if (fd < 0)
2082 		return false;
2083 	close(fd);
2084 	return true;
2085 }
2086 
2087 static char *dso__find_kallsyms(struct dso *dso, struct map *map)
2088 {
2089 	struct build_id bid;
2090 	char sbuild_id[SBUILD_ID_SIZE];
2091 	bool is_host = false;
2092 	char path[PATH_MAX];
2093 
2094 	if (!dso->has_build_id) {
2095 		/*
2096 		 * Last resort, if we don't have a build-id and couldn't find
2097 		 * any vmlinux file, try the running kernel kallsyms table.
2098 		 */
2099 		goto proc_kallsyms;
2100 	}
2101 
2102 	if (sysfs__read_build_id("/sys/kernel/notes", &bid) == 0)
2103 		is_host = dso__build_id_equal(dso, &bid);
2104 
2105 	/* Try a fast path for /proc/kallsyms if possible */
2106 	if (is_host) {
2107 		/*
2108 		 * Do not check the build-id cache, unless we know we cannot use
2109 		 * /proc/kcore or module maps don't match to /proc/kallsyms.
2110 		 * To check readability of /proc/kcore, do not use access(R_OK)
2111 		 * since /proc/kcore requires CAP_SYS_RAWIO to read and access
2112 		 * can't check it.
2113 		 */
2114 		if (filename__readable("/proc/kcore") &&
2115 		    !validate_kcore_addresses("/proc/kallsyms", map))
2116 			goto proc_kallsyms;
2117 	}
2118 
2119 	build_id__sprintf(&dso->bid, sbuild_id);
2120 
2121 	/* Find kallsyms in build-id cache with kcore */
2122 	scnprintf(path, sizeof(path), "%s/%s/%s",
2123 		  buildid_dir, DSO__NAME_KCORE, sbuild_id);
2124 
2125 	if (!find_matching_kcore(map, path, sizeof(path)))
2126 		return strdup(path);
2127 
2128 	/* Use current /proc/kallsyms if possible */
2129 	if (is_host) {
2130 proc_kallsyms:
2131 		return strdup("/proc/kallsyms");
2132 	}
2133 
2134 	/* Finally, find a cache of kallsyms */
2135 	if (!build_id_cache__kallsyms_path(sbuild_id, path, sizeof(path))) {
2136 		pr_err("No kallsyms or vmlinux with build-id %s was found\n",
2137 		       sbuild_id);
2138 		return NULL;
2139 	}
2140 
2141 	return strdup(path);
2142 }
2143 
2144 static int dso__load_kernel_sym(struct dso *dso, struct map *map)
2145 {
2146 	int err;
2147 	const char *kallsyms_filename = NULL;
2148 	char *kallsyms_allocated_filename = NULL;
2149 	char *filename = NULL;
2150 
2151 	/*
2152 	 * Step 1: if the user specified a kallsyms or vmlinux filename, use
2153 	 * it and only it, reporting errors to the user if it cannot be used.
2154 	 *
2155 	 * For instance, try to analyse an ARM perf.data file _without_ a
2156 	 * build-id, or if the user specifies the wrong path to the right
2157 	 * vmlinux file, obviously we can't fallback to another vmlinux (a
2158 	 * x86_86 one, on the machine where analysis is being performed, say),
2159 	 * or worse, /proc/kallsyms.
2160 	 *
2161 	 * If the specified file _has_ a build-id and there is a build-id
2162 	 * section in the perf.data file, we will still do the expected
2163 	 * validation in dso__load_vmlinux and will bail out if they don't
2164 	 * match.
2165 	 */
2166 	if (symbol_conf.kallsyms_name != NULL) {
2167 		kallsyms_filename = symbol_conf.kallsyms_name;
2168 		goto do_kallsyms;
2169 	}
2170 
2171 	if (!symbol_conf.ignore_vmlinux && symbol_conf.vmlinux_name != NULL) {
2172 		return dso__load_vmlinux(dso, map, symbol_conf.vmlinux_name, false);
2173 	}
2174 
2175 	/*
2176 	 * Before checking on common vmlinux locations, check if it's
2177 	 * stored as standard build id binary (not kallsyms) under
2178 	 * .debug cache.
2179 	 */
2180 	if (!symbol_conf.ignore_vmlinux_buildid)
2181 		filename = __dso__build_id_filename(dso, NULL, 0, false, false);
2182 	if (filename != NULL) {
2183 		err = dso__load_vmlinux(dso, map, filename, true);
2184 		if (err > 0)
2185 			return err;
2186 		free(filename);
2187 	}
2188 
2189 	if (!symbol_conf.ignore_vmlinux && vmlinux_path != NULL) {
2190 		err = dso__load_vmlinux_path(dso, map);
2191 		if (err > 0)
2192 			return err;
2193 	}
2194 
2195 	/* do not try local files if a symfs was given */
2196 	if (symbol_conf.symfs[0] != 0)
2197 		return -1;
2198 
2199 	kallsyms_allocated_filename = dso__find_kallsyms(dso, map);
2200 	if (!kallsyms_allocated_filename)
2201 		return -1;
2202 
2203 	kallsyms_filename = kallsyms_allocated_filename;
2204 
2205 do_kallsyms:
2206 	err = dso__load_kallsyms(dso, kallsyms_filename, map);
2207 	if (err > 0)
2208 		pr_debug("Using %s for symbols\n", kallsyms_filename);
2209 	free(kallsyms_allocated_filename);
2210 
2211 	if (err > 0 && !dso__is_kcore(dso)) {
2212 		dso->binary_type = DSO_BINARY_TYPE__KALLSYMS;
2213 		dso__set_long_name(dso, DSO__NAME_KALLSYMS, false);
2214 		map__fixup_start(map);
2215 		map__fixup_end(map);
2216 	}
2217 
2218 	return err;
2219 }
2220 
2221 static int dso__load_guest_kernel_sym(struct dso *dso, struct map *map)
2222 {
2223 	int err;
2224 	const char *kallsyms_filename;
2225 	struct machine *machine = maps__machine(map__kmaps(map));
2226 	char path[PATH_MAX];
2227 
2228 	if (machine->kallsyms_filename) {
2229 		kallsyms_filename = machine->kallsyms_filename;
2230 	} else if (machine__is_default_guest(machine)) {
2231 		/*
2232 		 * if the user specified a vmlinux filename, use it and only
2233 		 * it, reporting errors to the user if it cannot be used.
2234 		 * Or use file guest_kallsyms inputted by user on commandline
2235 		 */
2236 		if (symbol_conf.default_guest_vmlinux_name != NULL) {
2237 			err = dso__load_vmlinux(dso, map,
2238 						symbol_conf.default_guest_vmlinux_name,
2239 						false);
2240 			return err;
2241 		}
2242 
2243 		kallsyms_filename = symbol_conf.default_guest_kallsyms;
2244 		if (!kallsyms_filename)
2245 			return -1;
2246 	} else {
2247 		sprintf(path, "%s/proc/kallsyms", machine->root_dir);
2248 		kallsyms_filename = path;
2249 	}
2250 
2251 	err = dso__load_kallsyms(dso, kallsyms_filename, map);
2252 	if (err > 0)
2253 		pr_debug("Using %s for symbols\n", kallsyms_filename);
2254 	if (err > 0 && !dso__is_kcore(dso)) {
2255 		dso->binary_type = DSO_BINARY_TYPE__GUEST_KALLSYMS;
2256 		dso__set_long_name(dso, machine->mmap_name, false);
2257 		map__fixup_start(map);
2258 		map__fixup_end(map);
2259 	}
2260 
2261 	return err;
2262 }
2263 
2264 static void vmlinux_path__exit(void)
2265 {
2266 	while (--vmlinux_path__nr_entries >= 0)
2267 		zfree(&vmlinux_path[vmlinux_path__nr_entries]);
2268 	vmlinux_path__nr_entries = 0;
2269 
2270 	zfree(&vmlinux_path);
2271 }
2272 
2273 static const char * const vmlinux_paths[] = {
2274 	"vmlinux",
2275 	"/boot/vmlinux"
2276 };
2277 
2278 static const char * const vmlinux_paths_upd[] = {
2279 	"/boot/vmlinux-%s",
2280 	"/usr/lib/debug/boot/vmlinux-%s",
2281 	"/lib/modules/%s/build/vmlinux",
2282 	"/usr/lib/debug/lib/modules/%s/vmlinux",
2283 	"/usr/lib/debug/boot/vmlinux-%s.debug"
2284 };
2285 
2286 static int vmlinux_path__add(const char *new_entry)
2287 {
2288 	vmlinux_path[vmlinux_path__nr_entries] = strdup(new_entry);
2289 	if (vmlinux_path[vmlinux_path__nr_entries] == NULL)
2290 		return -1;
2291 	++vmlinux_path__nr_entries;
2292 
2293 	return 0;
2294 }
2295 
2296 static int vmlinux_path__init(struct perf_env *env)
2297 {
2298 	struct utsname uts;
2299 	char bf[PATH_MAX];
2300 	char *kernel_version;
2301 	unsigned int i;
2302 
2303 	vmlinux_path = malloc(sizeof(char *) * (ARRAY_SIZE(vmlinux_paths) +
2304 			      ARRAY_SIZE(vmlinux_paths_upd)));
2305 	if (vmlinux_path == NULL)
2306 		return -1;
2307 
2308 	for (i = 0; i < ARRAY_SIZE(vmlinux_paths); i++)
2309 		if (vmlinux_path__add(vmlinux_paths[i]) < 0)
2310 			goto out_fail;
2311 
2312 	/* only try kernel version if no symfs was given */
2313 	if (symbol_conf.symfs[0] != 0)
2314 		return 0;
2315 
2316 	if (env) {
2317 		kernel_version = env->os_release;
2318 	} else {
2319 		if (uname(&uts) < 0)
2320 			goto out_fail;
2321 
2322 		kernel_version = uts.release;
2323 	}
2324 
2325 	for (i = 0; i < ARRAY_SIZE(vmlinux_paths_upd); i++) {
2326 		snprintf(bf, sizeof(bf), vmlinux_paths_upd[i], kernel_version);
2327 		if (vmlinux_path__add(bf) < 0)
2328 			goto out_fail;
2329 	}
2330 
2331 	return 0;
2332 
2333 out_fail:
2334 	vmlinux_path__exit();
2335 	return -1;
2336 }
2337 
2338 int setup_list(struct strlist **list, const char *list_str,
2339 		      const char *list_name)
2340 {
2341 	if (list_str == NULL)
2342 		return 0;
2343 
2344 	*list = strlist__new(list_str, NULL);
2345 	if (!*list) {
2346 		pr_err("problems parsing %s list\n", list_name);
2347 		return -1;
2348 	}
2349 
2350 	symbol_conf.has_filter = true;
2351 	return 0;
2352 }
2353 
2354 int setup_intlist(struct intlist **list, const char *list_str,
2355 		  const char *list_name)
2356 {
2357 	if (list_str == NULL)
2358 		return 0;
2359 
2360 	*list = intlist__new(list_str);
2361 	if (!*list) {
2362 		pr_err("problems parsing %s list\n", list_name);
2363 		return -1;
2364 	}
2365 	return 0;
2366 }
2367 
2368 static int setup_addrlist(struct intlist **addr_list, struct strlist *sym_list)
2369 {
2370 	struct str_node *pos, *tmp;
2371 	unsigned long val;
2372 	char *sep;
2373 	const char *end;
2374 	int i = 0, err;
2375 
2376 	*addr_list = intlist__new(NULL);
2377 	if (!*addr_list)
2378 		return -1;
2379 
2380 	strlist__for_each_entry_safe(pos, tmp, sym_list) {
2381 		errno = 0;
2382 		val = strtoul(pos->s, &sep, 16);
2383 		if (errno || (sep == pos->s))
2384 			continue;
2385 
2386 		if (*sep != '\0') {
2387 			end = pos->s + strlen(pos->s) - 1;
2388 			while (end >= sep && isspace(*end))
2389 				end--;
2390 
2391 			if (end >= sep)
2392 				continue;
2393 		}
2394 
2395 		err = intlist__add(*addr_list, val);
2396 		if (err)
2397 			break;
2398 
2399 		strlist__remove(sym_list, pos);
2400 		i++;
2401 	}
2402 
2403 	if (i == 0) {
2404 		intlist__delete(*addr_list);
2405 		*addr_list = NULL;
2406 	}
2407 
2408 	return 0;
2409 }
2410 
2411 static bool symbol__read_kptr_restrict(void)
2412 {
2413 	bool value = false;
2414 	FILE *fp = fopen("/proc/sys/kernel/kptr_restrict", "r");
2415 
2416 	if (fp != NULL) {
2417 		char line[8];
2418 
2419 		if (fgets(line, sizeof(line), fp) != NULL)
2420 			value = perf_cap__capable(CAP_SYSLOG) ?
2421 					(atoi(line) >= 2) :
2422 					(atoi(line) != 0);
2423 
2424 		fclose(fp);
2425 	}
2426 
2427 	/* Per kernel/kallsyms.c:
2428 	 * we also restrict when perf_event_paranoid > 1 w/o CAP_SYSLOG
2429 	 */
2430 	if (perf_event_paranoid() > 1 && !perf_cap__capable(CAP_SYSLOG))
2431 		value = true;
2432 
2433 	return value;
2434 }
2435 
2436 int symbol__annotation_init(void)
2437 {
2438 	if (symbol_conf.init_annotation)
2439 		return 0;
2440 
2441 	if (symbol_conf.initialized) {
2442 		pr_err("Annotation needs to be init before symbol__init()\n");
2443 		return -1;
2444 	}
2445 
2446 	symbol_conf.priv_size += sizeof(struct annotation);
2447 	symbol_conf.init_annotation = true;
2448 	return 0;
2449 }
2450 
2451 int symbol__init(struct perf_env *env)
2452 {
2453 	const char *symfs;
2454 
2455 	if (symbol_conf.initialized)
2456 		return 0;
2457 
2458 	symbol_conf.priv_size = PERF_ALIGN(symbol_conf.priv_size, sizeof(u64));
2459 
2460 	symbol__elf_init();
2461 
2462 	if (symbol_conf.try_vmlinux_path && vmlinux_path__init(env) < 0)
2463 		return -1;
2464 
2465 	if (symbol_conf.field_sep && *symbol_conf.field_sep == '.') {
2466 		pr_err("'.' is the only non valid --field-separator argument\n");
2467 		return -1;
2468 	}
2469 
2470 	if (setup_list(&symbol_conf.dso_list,
2471 		       symbol_conf.dso_list_str, "dso") < 0)
2472 		return -1;
2473 
2474 	if (setup_list(&symbol_conf.comm_list,
2475 		       symbol_conf.comm_list_str, "comm") < 0)
2476 		goto out_free_dso_list;
2477 
2478 	if (setup_intlist(&symbol_conf.pid_list,
2479 		       symbol_conf.pid_list_str, "pid") < 0)
2480 		goto out_free_comm_list;
2481 
2482 	if (setup_intlist(&symbol_conf.tid_list,
2483 		       symbol_conf.tid_list_str, "tid") < 0)
2484 		goto out_free_pid_list;
2485 
2486 	if (setup_list(&symbol_conf.sym_list,
2487 		       symbol_conf.sym_list_str, "symbol") < 0)
2488 		goto out_free_tid_list;
2489 
2490 	if (symbol_conf.sym_list &&
2491 	    setup_addrlist(&symbol_conf.addr_list, symbol_conf.sym_list) < 0)
2492 		goto out_free_sym_list;
2493 
2494 	if (setup_list(&symbol_conf.bt_stop_list,
2495 		       symbol_conf.bt_stop_list_str, "symbol") < 0)
2496 		goto out_free_sym_list;
2497 
2498 	/*
2499 	 * A path to symbols of "/" is identical to ""
2500 	 * reset here for simplicity.
2501 	 */
2502 	symfs = realpath(symbol_conf.symfs, NULL);
2503 	if (symfs == NULL)
2504 		symfs = symbol_conf.symfs;
2505 	if (strcmp(symfs, "/") == 0)
2506 		symbol_conf.symfs = "";
2507 	if (symfs != symbol_conf.symfs)
2508 		free((void *)symfs);
2509 
2510 	symbol_conf.kptr_restrict = symbol__read_kptr_restrict();
2511 
2512 	symbol_conf.initialized = true;
2513 	return 0;
2514 
2515 out_free_sym_list:
2516 	strlist__delete(symbol_conf.sym_list);
2517 	intlist__delete(symbol_conf.addr_list);
2518 out_free_tid_list:
2519 	intlist__delete(symbol_conf.tid_list);
2520 out_free_pid_list:
2521 	intlist__delete(symbol_conf.pid_list);
2522 out_free_comm_list:
2523 	strlist__delete(symbol_conf.comm_list);
2524 out_free_dso_list:
2525 	strlist__delete(symbol_conf.dso_list);
2526 	return -1;
2527 }
2528 
2529 void symbol__exit(void)
2530 {
2531 	if (!symbol_conf.initialized)
2532 		return;
2533 	strlist__delete(symbol_conf.bt_stop_list);
2534 	strlist__delete(symbol_conf.sym_list);
2535 	strlist__delete(symbol_conf.dso_list);
2536 	strlist__delete(symbol_conf.comm_list);
2537 	intlist__delete(symbol_conf.tid_list);
2538 	intlist__delete(symbol_conf.pid_list);
2539 	intlist__delete(symbol_conf.addr_list);
2540 	vmlinux_path__exit();
2541 	symbol_conf.sym_list = symbol_conf.dso_list = symbol_conf.comm_list = NULL;
2542 	symbol_conf.bt_stop_list = NULL;
2543 	symbol_conf.initialized = false;
2544 }
2545 
2546 int symbol__config_symfs(const struct option *opt __maybe_unused,
2547 			 const char *dir, int unset __maybe_unused)
2548 {
2549 	char *bf = NULL;
2550 	int ret;
2551 
2552 	symbol_conf.symfs = strdup(dir);
2553 	if (symbol_conf.symfs == NULL)
2554 		return -ENOMEM;
2555 
2556 	/* skip the locally configured cache if a symfs is given, and
2557 	 * config buildid dir to symfs/.debug
2558 	 */
2559 	ret = asprintf(&bf, "%s/%s", dir, ".debug");
2560 	if (ret < 0)
2561 		return -ENOMEM;
2562 
2563 	set_buildid_dir(bf);
2564 
2565 	free(bf);
2566 	return 0;
2567 }
2568 
2569 struct mem_info *mem_info__get(struct mem_info *mi)
2570 {
2571 	if (mi)
2572 		refcount_inc(&mi->refcnt);
2573 	return mi;
2574 }
2575 
2576 void mem_info__put(struct mem_info *mi)
2577 {
2578 	if (mi && refcount_dec_and_test(&mi->refcnt)) {
2579 		addr_map_symbol__exit(&mi->iaddr);
2580 		addr_map_symbol__exit(&mi->daddr);
2581 		free(mi);
2582 	}
2583 }
2584 
2585 struct mem_info *mem_info__new(void)
2586 {
2587 	struct mem_info *mi = zalloc(sizeof(*mi));
2588 
2589 	if (mi)
2590 		refcount_set(&mi->refcnt, 1);
2591 	return mi;
2592 }
2593 
2594 /*
2595  * Checks that user supplied symbol kernel files are accessible because
2596  * the default mechanism for accessing elf files fails silently. i.e. if
2597  * debug syms for a build ID aren't found perf carries on normally. When
2598  * they are user supplied we should assume that the user doesn't want to
2599  * silently fail.
2600  */
2601 int symbol__validate_sym_arguments(void)
2602 {
2603 	if (symbol_conf.vmlinux_name &&
2604 	    access(symbol_conf.vmlinux_name, R_OK)) {
2605 		pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
2606 		return -EINVAL;
2607 	}
2608 	if (symbol_conf.kallsyms_name &&
2609 	    access(symbol_conf.kallsyms_name, R_OK)) {
2610 		pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
2611 		return -EINVAL;
2612 	}
2613 	return 0;
2614 }
2615