Merge tag 'linux-kselftest-next-6.6-rc2' of git://git.kernel.org/pub/scm/linux/kernel...
[platform/kernel/linux-starfive.git] / tools / perf / util / symbol.c
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 map_list_node {
52         struct list_head node;
53         struct map *map;
54 };
55
56 struct symbol_conf symbol_conf = {
57         .nanosecs               = false,
58         .use_modules            = true,
59         .try_vmlinux_path       = true,
60         .demangle               = true,
61         .demangle_kernel        = false,
62         .cumulate_callchain     = true,
63         .time_quantum           = 100 * NSEC_PER_MSEC, /* 100ms */
64         .show_hist_headers      = true,
65         .symfs                  = "",
66         .event_group            = true,
67         .inline_name            = true,
68         .res_sample             = 0,
69 };
70
71 static enum dso_binary_type binary_type_symtab[] = {
72         DSO_BINARY_TYPE__KALLSYMS,
73         DSO_BINARY_TYPE__GUEST_KALLSYMS,
74         DSO_BINARY_TYPE__JAVA_JIT,
75         DSO_BINARY_TYPE__DEBUGLINK,
76         DSO_BINARY_TYPE__BUILD_ID_CACHE,
77         DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO,
78         DSO_BINARY_TYPE__FEDORA_DEBUGINFO,
79         DSO_BINARY_TYPE__UBUNTU_DEBUGINFO,
80         DSO_BINARY_TYPE__BUILDID_DEBUGINFO,
81         DSO_BINARY_TYPE__SYSTEM_PATH_DSO,
82         DSO_BINARY_TYPE__GUEST_KMODULE,
83         DSO_BINARY_TYPE__GUEST_KMODULE_COMP,
84         DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE,
85         DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP,
86         DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO,
87         DSO_BINARY_TYPE__MIXEDUP_UBUNTU_DEBUGINFO,
88         DSO_BINARY_TYPE__NOT_FOUND,
89 };
90
91 #define DSO_BINARY_TYPE__SYMTAB_CNT ARRAY_SIZE(binary_type_symtab)
92
93 static struct map_list_node *map_list_node__new(void)
94 {
95         return malloc(sizeof(struct map_list_node));
96 }
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                 next = rb_entry(nd, struct symbol, rb_node);
206
207                 if (!nd)
208                         break;
209
210                 if (curr->start != next->start)
211                         continue;
212
213                 if (choose_best_symbol(curr, next) == SYMBOL_A) {
214                         if (next->type == STT_GNU_IFUNC)
215                                 curr->ifunc_alias = true;
216                         rb_erase_cached(&next->rb_node, symbols);
217                         symbol__delete(next);
218                         goto again;
219                 } else {
220                         if (curr->type == STT_GNU_IFUNC)
221                                 next->ifunc_alias = true;
222                         nd = rb_next(&curr->rb_node);
223                         rb_erase_cached(&curr->rb_node, symbols);
224                         symbol__delete(curr);
225                 }
226         }
227 }
228
229 /* Update zero-sized symbols using the address of the next symbol */
230 void symbols__fixup_end(struct rb_root_cached *symbols, bool is_kallsyms)
231 {
232         struct rb_node *nd, *prevnd = rb_first_cached(symbols);
233         struct symbol *curr, *prev;
234
235         if (prevnd == NULL)
236                 return;
237
238         curr = rb_entry(prevnd, struct symbol, rb_node);
239
240         for (nd = rb_next(prevnd); nd; nd = rb_next(nd)) {
241                 prev = curr;
242                 curr = rb_entry(nd, struct symbol, rb_node);
243
244                 /*
245                  * On some architecture kernel text segment start is located at
246                  * some low memory address, while modules are located at high
247                  * memory addresses (or vice versa).  The gap between end of
248                  * kernel text segment and beginning of first module's text
249                  * segment is very big.  Therefore do not fill this gap and do
250                  * not assign it to the kernel dso map (kallsyms).
251                  *
252                  * In kallsyms, it determines module symbols using '[' character
253                  * like in:
254                  *   ffffffffc1937000 T hdmi_driver_init  [snd_hda_codec_hdmi]
255                  */
256                 if (prev->end == prev->start) {
257                         /* Last kernel/module symbol mapped to end of page */
258                         if (is_kallsyms && (!strchr(prev->name, '[') !=
259                                             !strchr(curr->name, '[')))
260                                 prev->end = roundup(prev->end + 4096, 4096);
261                         else
262                                 prev->end = curr->start;
263
264                         pr_debug4("%s sym:%s end:%#" PRIx64 "\n",
265                                   __func__, prev->name, prev->end);
266                 }
267         }
268
269         /* Last entry */
270         if (curr->end == curr->start)
271                 curr->end = roundup(curr->start, 4096) + 4096;
272 }
273
274 void maps__fixup_end(struct maps *maps)
275 {
276         struct map_rb_node *prev = NULL, *curr;
277
278         down_write(maps__lock(maps));
279
280         maps__for_each_entry(maps, curr) {
281                 if (prev != NULL && !map__end(prev->map))
282                         map__set_end(prev->map, map__start(curr->map));
283
284                 prev = curr;
285         }
286
287         /*
288          * We still haven't the actual symbols, so guess the
289          * last map final address.
290          */
291         if (curr && !map__end(curr->map))
292                 map__set_end(curr->map, ~0ULL);
293
294         up_write(maps__lock(maps));
295 }
296
297 struct symbol *symbol__new(u64 start, u64 len, u8 binding, u8 type, const char *name)
298 {
299         size_t namelen = strlen(name) + 1;
300         struct symbol *sym = calloc(1, (symbol_conf.priv_size +
301                                         sizeof(*sym) + namelen));
302         if (sym == NULL)
303                 return NULL;
304
305         if (symbol_conf.priv_size) {
306                 if (symbol_conf.init_annotation) {
307                         struct annotation *notes = (void *)sym;
308                         annotation__init(notes);
309                 }
310                 sym = ((void *)sym) + symbol_conf.priv_size;
311         }
312
313         sym->start   = start;
314         sym->end     = len ? start + len : start;
315         sym->type    = type;
316         sym->binding = binding;
317         sym->namelen = namelen - 1;
318
319         pr_debug4("%s: %s %#" PRIx64 "-%#" PRIx64 "\n",
320                   __func__, name, start, sym->end);
321         memcpy(sym->name, name, namelen);
322
323         return sym;
324 }
325
326 void symbol__delete(struct symbol *sym)
327 {
328         if (symbol_conf.priv_size) {
329                 if (symbol_conf.init_annotation) {
330                         struct annotation *notes = symbol__annotation(sym);
331
332                         annotation__exit(notes);
333                 }
334         }
335         free(((void *)sym) - symbol_conf.priv_size);
336 }
337
338 void symbols__delete(struct rb_root_cached *symbols)
339 {
340         struct symbol *pos;
341         struct rb_node *next = rb_first_cached(symbols);
342
343         while (next) {
344                 pos = rb_entry(next, struct symbol, rb_node);
345                 next = rb_next(&pos->rb_node);
346                 rb_erase_cached(&pos->rb_node, symbols);
347                 symbol__delete(pos);
348         }
349 }
350
351 void __symbols__insert(struct rb_root_cached *symbols,
352                        struct symbol *sym, bool kernel)
353 {
354         struct rb_node **p = &symbols->rb_root.rb_node;
355         struct rb_node *parent = NULL;
356         const u64 ip = sym->start;
357         struct symbol *s;
358         bool leftmost = true;
359
360         if (kernel) {
361                 const char *name = sym->name;
362                 /*
363                  * ppc64 uses function descriptors and appends a '.' to the
364                  * start of every instruction address. Remove it.
365                  */
366                 if (name[0] == '.')
367                         name++;
368                 sym->idle = symbol__is_idle(name);
369         }
370
371         while (*p != NULL) {
372                 parent = *p;
373                 s = rb_entry(parent, struct symbol, rb_node);
374                 if (ip < s->start)
375                         p = &(*p)->rb_left;
376                 else {
377                         p = &(*p)->rb_right;
378                         leftmost = false;
379                 }
380         }
381         rb_link_node(&sym->rb_node, parent, p);
382         rb_insert_color_cached(&sym->rb_node, symbols, leftmost);
383 }
384
385 void symbols__insert(struct rb_root_cached *symbols, struct symbol *sym)
386 {
387         __symbols__insert(symbols, sym, false);
388 }
389
390 static struct symbol *symbols__find(struct rb_root_cached *symbols, u64 ip)
391 {
392         struct rb_node *n;
393
394         if (symbols == NULL)
395                 return NULL;
396
397         n = symbols->rb_root.rb_node;
398
399         while (n) {
400                 struct symbol *s = rb_entry(n, struct symbol, rb_node);
401
402                 if (ip < s->start)
403                         n = n->rb_left;
404                 else if (ip > s->end || (ip == s->end && ip != s->start))
405                         n = n->rb_right;
406                 else
407                         return s;
408         }
409
410         return NULL;
411 }
412
413 static struct symbol *symbols__first(struct rb_root_cached *symbols)
414 {
415         struct rb_node *n = rb_first_cached(symbols);
416
417         if (n)
418                 return rb_entry(n, struct symbol, rb_node);
419
420         return NULL;
421 }
422
423 static struct symbol *symbols__last(struct rb_root_cached *symbols)
424 {
425         struct rb_node *n = rb_last(&symbols->rb_root);
426
427         if (n)
428                 return rb_entry(n, struct symbol, rb_node);
429
430         return NULL;
431 }
432
433 static struct symbol *symbols__next(struct symbol *sym)
434 {
435         struct rb_node *n = rb_next(&sym->rb_node);
436
437         if (n)
438                 return rb_entry(n, struct symbol, rb_node);
439
440         return NULL;
441 }
442
443 static int symbols__sort_name_cmp(const void *vlhs, const void *vrhs)
444 {
445         const struct symbol *lhs = *((const struct symbol **)vlhs);
446         const struct symbol *rhs = *((const struct symbol **)vrhs);
447
448         return strcmp(lhs->name, rhs->name);
449 }
450
451 static struct symbol **symbols__sort_by_name(struct rb_root_cached *source, size_t *len)
452 {
453         struct rb_node *nd;
454         struct symbol **result;
455         size_t i = 0, size = 0;
456
457         for (nd = rb_first_cached(source); nd; nd = rb_next(nd))
458                 size++;
459
460         result = malloc(sizeof(*result) * size);
461         if (!result)
462                 return NULL;
463
464         for (nd = rb_first_cached(source); nd; nd = rb_next(nd)) {
465                 struct symbol *pos = rb_entry(nd, struct symbol, rb_node);
466
467                 result[i++] = pos;
468         }
469         qsort(result, size, sizeof(*result), symbols__sort_name_cmp);
470         *len = size;
471         return result;
472 }
473
474 int symbol__match_symbol_name(const char *name, const char *str,
475                               enum symbol_tag_include includes)
476 {
477         const char *versioning;
478
479         if (includes == SYMBOL_TAG_INCLUDE__DEFAULT_ONLY &&
480             (versioning = strstr(name, "@@"))) {
481                 int len = strlen(str);
482
483                 if (len < versioning - name)
484                         len = versioning - name;
485
486                 return arch__compare_symbol_names_n(name, str, len);
487         } else
488                 return arch__compare_symbol_names(name, str);
489 }
490
491 static struct symbol *symbols__find_by_name(struct symbol *symbols[],
492                                             size_t symbols_len,
493                                             const char *name,
494                                             enum symbol_tag_include includes,
495                                             size_t *found_idx)
496 {
497         size_t i, lower = 0, upper = symbols_len;
498         struct symbol *s = NULL;
499
500         if (found_idx)
501                 *found_idx = SIZE_MAX;
502
503         if (!symbols_len)
504                 return NULL;
505
506         while (lower < upper) {
507                 int cmp;
508
509                 i = (lower + upper) / 2;
510                 cmp = symbol__match_symbol_name(symbols[i]->name, name, includes);
511
512                 if (cmp > 0)
513                         upper = i;
514                 else if (cmp < 0)
515                         lower = i + 1;
516                 else {
517                         if (found_idx)
518                                 *found_idx = i;
519                         s = symbols[i];
520                         break;
521                 }
522         }
523         if (s && includes != SYMBOL_TAG_INCLUDE__DEFAULT_ONLY) {
524                 /* return first symbol that has same name (if any) */
525                 for (; i > 0; i--) {
526                         struct symbol *tmp = symbols[i - 1];
527
528                         if (!arch__compare_symbol_names(tmp->name, s->name)) {
529                                 if (found_idx)
530                                         *found_idx = i - 1;
531                                 s = tmp;
532                         } else
533                                 break;
534                 }
535         }
536         assert(!found_idx || !s || s == symbols[*found_idx]);
537         return s;
538 }
539
540 void dso__reset_find_symbol_cache(struct dso *dso)
541 {
542         dso->last_find_result.addr   = 0;
543         dso->last_find_result.symbol = NULL;
544 }
545
546 void dso__insert_symbol(struct dso *dso, struct symbol *sym)
547 {
548         __symbols__insert(&dso->symbols, sym, dso->kernel);
549
550         /* update the symbol cache if necessary */
551         if (dso->last_find_result.addr >= sym->start &&
552             (dso->last_find_result.addr < sym->end ||
553             sym->start == sym->end)) {
554                 dso->last_find_result.symbol = sym;
555         }
556 }
557
558 void dso__delete_symbol(struct dso *dso, struct symbol *sym)
559 {
560         rb_erase_cached(&sym->rb_node, &dso->symbols);
561         symbol__delete(sym);
562         dso__reset_find_symbol_cache(dso);
563 }
564
565 struct symbol *dso__find_symbol(struct dso *dso, u64 addr)
566 {
567         if (dso->last_find_result.addr != addr || dso->last_find_result.symbol == NULL) {
568                 dso->last_find_result.addr   = addr;
569                 dso->last_find_result.symbol = symbols__find(&dso->symbols, addr);
570         }
571
572         return dso->last_find_result.symbol;
573 }
574
575 struct symbol *dso__find_symbol_nocache(struct dso *dso, u64 addr)
576 {
577         return symbols__find(&dso->symbols, addr);
578 }
579
580 struct symbol *dso__first_symbol(struct dso *dso)
581 {
582         return symbols__first(&dso->symbols);
583 }
584
585 struct symbol *dso__last_symbol(struct dso *dso)
586 {
587         return symbols__last(&dso->symbols);
588 }
589
590 struct symbol *dso__next_symbol(struct symbol *sym)
591 {
592         return symbols__next(sym);
593 }
594
595 struct symbol *dso__next_symbol_by_name(struct dso *dso, size_t *idx)
596 {
597         if (*idx + 1 >= dso->symbol_names_len)
598                 return NULL;
599
600         ++*idx;
601         return dso->symbol_names[*idx];
602 }
603
604  /*
605   * Returns first symbol that matched with @name.
606   */
607 struct symbol *dso__find_symbol_by_name(struct dso *dso, const char *name, size_t *idx)
608 {
609         struct symbol *s = symbols__find_by_name(dso->symbol_names, dso->symbol_names_len,
610                                                 name, SYMBOL_TAG_INCLUDE__NONE, idx);
611         if (!s)
612                 s = symbols__find_by_name(dso->symbol_names, dso->symbol_names_len,
613                                         name, SYMBOL_TAG_INCLUDE__DEFAULT_ONLY, idx);
614         return s;
615 }
616
617 void dso__sort_by_name(struct dso *dso)
618 {
619         mutex_lock(&dso->lock);
620         if (!dso__sorted_by_name(dso)) {
621                 size_t len;
622
623                 dso->symbol_names = symbols__sort_by_name(&dso->symbols, &len);
624                 if (dso->symbol_names) {
625                         dso->symbol_names_len = len;
626                         dso__set_sorted_by_name(dso);
627                 }
628         }
629         mutex_unlock(&dso->lock);
630 }
631
632 /*
633  * While we find nice hex chars, build a long_val.
634  * Return number of chars processed.
635  */
636 static int hex2u64(const char *ptr, u64 *long_val)
637 {
638         char *p;
639
640         *long_val = strtoull(ptr, &p, 16);
641
642         return p - ptr;
643 }
644
645
646 int modules__parse(const char *filename, void *arg,
647                    int (*process_module)(void *arg, const char *name,
648                                          u64 start, u64 size))
649 {
650         char *line = NULL;
651         size_t n;
652         FILE *file;
653         int err = 0;
654
655         file = fopen(filename, "r");
656         if (file == NULL)
657                 return -1;
658
659         while (1) {
660                 char name[PATH_MAX];
661                 u64 start, size;
662                 char *sep, *endptr;
663                 ssize_t line_len;
664
665                 line_len = getline(&line, &n, file);
666                 if (line_len < 0) {
667                         if (feof(file))
668                                 break;
669                         err = -1;
670                         goto out;
671                 }
672
673                 if (!line) {
674                         err = -1;
675                         goto out;
676                 }
677
678                 line[--line_len] = '\0'; /* \n */
679
680                 sep = strrchr(line, 'x');
681                 if (sep == NULL)
682                         continue;
683
684                 hex2u64(sep + 1, &start);
685
686                 sep = strchr(line, ' ');
687                 if (sep == NULL)
688                         continue;
689
690                 *sep = '\0';
691
692                 scnprintf(name, sizeof(name), "[%s]", line);
693
694                 size = strtoul(sep + 1, &endptr, 0);
695                 if (*endptr != ' ' && *endptr != '\t')
696                         continue;
697
698                 err = process_module(arg, name, start, size);
699                 if (err)
700                         break;
701         }
702 out:
703         free(line);
704         fclose(file);
705         return err;
706 }
707
708 /*
709  * These are symbols in the kernel image, so make sure that
710  * sym is from a kernel DSO.
711  */
712 static bool symbol__is_idle(const char *name)
713 {
714         const char * const idle_symbols[] = {
715                 "acpi_idle_do_entry",
716                 "acpi_processor_ffh_cstate_enter",
717                 "arch_cpu_idle",
718                 "cpu_idle",
719                 "cpu_startup_entry",
720                 "idle_cpu",
721                 "intel_idle",
722                 "default_idle",
723                 "native_safe_halt",
724                 "enter_idle",
725                 "exit_idle",
726                 "mwait_idle",
727                 "mwait_idle_with_hints",
728                 "mwait_idle_with_hints.constprop.0",
729                 "poll_idle",
730                 "ppc64_runlatch_off",
731                 "pseries_dedicated_idle_sleep",
732                 "psw_idle",
733                 "psw_idle_exit",
734                 NULL
735         };
736         int i;
737         static struct strlist *idle_symbols_list;
738
739         if (idle_symbols_list)
740                 return strlist__has_entry(idle_symbols_list, name);
741
742         idle_symbols_list = strlist__new(NULL, NULL);
743
744         for (i = 0; idle_symbols[i]; i++)
745                 strlist__add(idle_symbols_list, idle_symbols[i]);
746
747         return strlist__has_entry(idle_symbols_list, name);
748 }
749
750 static int map__process_kallsym_symbol(void *arg, const char *name,
751                                        char type, u64 start)
752 {
753         struct symbol *sym;
754         struct dso *dso = arg;
755         struct rb_root_cached *root = &dso->symbols;
756
757         if (!symbol_type__filter(type))
758                 return 0;
759
760         /* Ignore local symbols for ARM modules */
761         if (name[0] == '$')
762                 return 0;
763
764         /*
765          * module symbols are not sorted so we add all
766          * symbols, setting length to 0, and rely on
767          * symbols__fixup_end() to fix it up.
768          */
769         sym = symbol__new(start, 0, kallsyms2elf_binding(type), kallsyms2elf_type(type), name);
770         if (sym == NULL)
771                 return -ENOMEM;
772         /*
773          * We will pass the symbols to the filter later, in
774          * map__split_kallsyms, when we have split the maps per module
775          */
776         __symbols__insert(root, sym, !strchr(name, '['));
777
778         return 0;
779 }
780
781 /*
782  * Loads the function entries in /proc/kallsyms into kernel_map->dso,
783  * so that we can in the next step set the symbol ->end address and then
784  * call kernel_maps__split_kallsyms.
785  */
786 static int dso__load_all_kallsyms(struct dso *dso, const char *filename)
787 {
788         return kallsyms__parse(filename, dso, map__process_kallsym_symbol);
789 }
790
791 static int maps__split_kallsyms_for_kcore(struct maps *kmaps, struct dso *dso)
792 {
793         struct map *curr_map;
794         struct symbol *pos;
795         int count = 0;
796         struct rb_root_cached old_root = dso->symbols;
797         struct rb_root_cached *root = &dso->symbols;
798         struct rb_node *next = rb_first_cached(root);
799
800         if (!kmaps)
801                 return -1;
802
803         *root = RB_ROOT_CACHED;
804
805         while (next) {
806                 struct dso *curr_map_dso;
807                 char *module;
808
809                 pos = rb_entry(next, struct symbol, rb_node);
810                 next = rb_next(&pos->rb_node);
811
812                 rb_erase_cached(&pos->rb_node, &old_root);
813                 RB_CLEAR_NODE(&pos->rb_node);
814                 module = strchr(pos->name, '\t');
815                 if (module)
816                         *module = '\0';
817
818                 curr_map = maps__find(kmaps, pos->start);
819
820                 if (!curr_map) {
821                         symbol__delete(pos);
822                         continue;
823                 }
824                 curr_map_dso = map__dso(curr_map);
825                 pos->start -= map__start(curr_map) - map__pgoff(curr_map);
826                 if (pos->end > map__end(curr_map))
827                         pos->end = map__end(curr_map);
828                 if (pos->end)
829                         pos->end -= map__start(curr_map) - map__pgoff(curr_map);
830                 symbols__insert(&curr_map_dso->symbols, pos);
831                 ++count;
832         }
833
834         /* Symbols have been adjusted */
835         dso->adjust_symbols = 1;
836
837         return count;
838 }
839
840 /*
841  * Split the symbols into maps, making sure there are no overlaps, i.e. the
842  * kernel range is broken in several maps, named [kernel].N, as we don't have
843  * the original ELF section names vmlinux have.
844  */
845 static int maps__split_kallsyms(struct maps *kmaps, struct dso *dso, u64 delta,
846                                 struct map *initial_map)
847 {
848         struct machine *machine;
849         struct map *curr_map = initial_map;
850         struct symbol *pos;
851         int count = 0, moved = 0;
852         struct rb_root_cached *root = &dso->symbols;
853         struct rb_node *next = rb_first_cached(root);
854         int kernel_range = 0;
855         bool x86_64;
856
857         if (!kmaps)
858                 return -1;
859
860         machine = maps__machine(kmaps);
861
862         x86_64 = machine__is(machine, "x86_64");
863
864         while (next) {
865                 char *module;
866
867                 pos = rb_entry(next, struct symbol, rb_node);
868                 next = rb_next(&pos->rb_node);
869
870                 module = strchr(pos->name, '\t');
871                 if (module) {
872                         struct dso *curr_map_dso;
873
874                         if (!symbol_conf.use_modules)
875                                 goto discard_symbol;
876
877                         *module++ = '\0';
878                         curr_map_dso = map__dso(curr_map);
879                         if (strcmp(curr_map_dso->short_name, module)) {
880                                 if (RC_CHK_ACCESS(curr_map) != RC_CHK_ACCESS(initial_map) &&
881                                     dso->kernel == DSO_SPACE__KERNEL_GUEST &&
882                                     machine__is_default_guest(machine)) {
883                                         /*
884                                          * We assume all symbols of a module are
885                                          * continuous in * kallsyms, so curr_map
886                                          * points to a module and all its
887                                          * symbols are in its kmap. Mark it as
888                                          * loaded.
889                                          */
890                                         dso__set_loaded(curr_map_dso);
891                                 }
892
893                                 curr_map = maps__find_by_name(kmaps, module);
894                                 if (curr_map == NULL) {
895                                         pr_debug("%s/proc/{kallsyms,modules} "
896                                                  "inconsistency while looking "
897                                                  "for \"%s\" module!\n",
898                                                  machine->root_dir, module);
899                                         curr_map = initial_map;
900                                         goto discard_symbol;
901                                 }
902                                 curr_map_dso = map__dso(curr_map);
903                                 if (curr_map_dso->loaded &&
904                                     !machine__is_default_guest(machine))
905                                         goto discard_symbol;
906                         }
907                         /*
908                          * So that we look just like we get from .ko files,
909                          * i.e. not prelinked, relative to initial_map->start.
910                          */
911                         pos->start = map__map_ip(curr_map, pos->start);
912                         pos->end   = map__map_ip(curr_map, pos->end);
913                 } else if (x86_64 && is_entry_trampoline(pos->name)) {
914                         /*
915                          * These symbols are not needed anymore since the
916                          * trampoline maps refer to the text section and it's
917                          * symbols instead. Avoid having to deal with
918                          * relocations, and the assumption that the first symbol
919                          * is the start of kernel text, by simply removing the
920                          * symbols at this point.
921                          */
922                         goto discard_symbol;
923                 } else if (curr_map != initial_map) {
924                         char dso_name[PATH_MAX];
925                         struct dso *ndso;
926
927                         if (delta) {
928                                 /* Kernel was relocated at boot time */
929                                 pos->start -= delta;
930                                 pos->end -= delta;
931                         }
932
933                         if (count == 0) {
934                                 curr_map = initial_map;
935                                 goto add_symbol;
936                         }
937
938                         if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
939                                 snprintf(dso_name, sizeof(dso_name),
940                                         "[guest.kernel].%d",
941                                         kernel_range++);
942                         else
943                                 snprintf(dso_name, sizeof(dso_name),
944                                         "[kernel].%d",
945                                         kernel_range++);
946
947                         ndso = dso__new(dso_name);
948                         if (ndso == NULL)
949                                 return -1;
950
951                         ndso->kernel = dso->kernel;
952
953                         curr_map = map__new2(pos->start, ndso);
954                         if (curr_map == NULL) {
955                                 dso__put(ndso);
956                                 return -1;
957                         }
958
959                         map__set_map_ip(curr_map, identity__map_ip);
960                         map__set_unmap_ip(curr_map, identity__map_ip);
961                         if (maps__insert(kmaps, curr_map)) {
962                                 dso__put(ndso);
963                                 return -1;
964                         }
965                         ++kernel_range;
966                 } else if (delta) {
967                         /* Kernel was relocated at boot time */
968                         pos->start -= delta;
969                         pos->end -= delta;
970                 }
971 add_symbol:
972                 if (curr_map != initial_map) {
973                         struct dso *curr_map_dso = map__dso(curr_map);
974
975                         rb_erase_cached(&pos->rb_node, root);
976                         symbols__insert(&curr_map_dso->symbols, pos);
977                         ++moved;
978                 } else
979                         ++count;
980
981                 continue;
982 discard_symbol:
983                 rb_erase_cached(&pos->rb_node, root);
984                 symbol__delete(pos);
985         }
986
987         if (curr_map != initial_map &&
988             dso->kernel == DSO_SPACE__KERNEL_GUEST &&
989             machine__is_default_guest(maps__machine(kmaps))) {
990                 dso__set_loaded(map__dso(curr_map));
991         }
992
993         return count + moved;
994 }
995
996 bool symbol__restricted_filename(const char *filename,
997                                  const char *restricted_filename)
998 {
999         bool restricted = false;
1000
1001         if (symbol_conf.kptr_restrict) {
1002                 char *r = realpath(filename, NULL);
1003
1004                 if (r != NULL) {
1005                         restricted = strcmp(r, restricted_filename) == 0;
1006                         free(r);
1007                         return restricted;
1008                 }
1009         }
1010
1011         return restricted;
1012 }
1013
1014 struct module_info {
1015         struct rb_node rb_node;
1016         char *name;
1017         u64 start;
1018 };
1019
1020 static void add_module(struct module_info *mi, struct rb_root *modules)
1021 {
1022         struct rb_node **p = &modules->rb_node;
1023         struct rb_node *parent = NULL;
1024         struct module_info *m;
1025
1026         while (*p != NULL) {
1027                 parent = *p;
1028                 m = rb_entry(parent, struct module_info, rb_node);
1029                 if (strcmp(mi->name, m->name) < 0)
1030                         p = &(*p)->rb_left;
1031                 else
1032                         p = &(*p)->rb_right;
1033         }
1034         rb_link_node(&mi->rb_node, parent, p);
1035         rb_insert_color(&mi->rb_node, modules);
1036 }
1037
1038 static void delete_modules(struct rb_root *modules)
1039 {
1040         struct module_info *mi;
1041         struct rb_node *next = rb_first(modules);
1042
1043         while (next) {
1044                 mi = rb_entry(next, struct module_info, rb_node);
1045                 next = rb_next(&mi->rb_node);
1046                 rb_erase(&mi->rb_node, modules);
1047                 zfree(&mi->name);
1048                 free(mi);
1049         }
1050 }
1051
1052 static struct module_info *find_module(const char *name,
1053                                        struct rb_root *modules)
1054 {
1055         struct rb_node *n = modules->rb_node;
1056
1057         while (n) {
1058                 struct module_info *m;
1059                 int cmp;
1060
1061                 m = rb_entry(n, struct module_info, rb_node);
1062                 cmp = strcmp(name, m->name);
1063                 if (cmp < 0)
1064                         n = n->rb_left;
1065                 else if (cmp > 0)
1066                         n = n->rb_right;
1067                 else
1068                         return m;
1069         }
1070
1071         return NULL;
1072 }
1073
1074 static int __read_proc_modules(void *arg, const char *name, u64 start,
1075                                u64 size __maybe_unused)
1076 {
1077         struct rb_root *modules = arg;
1078         struct module_info *mi;
1079
1080         mi = zalloc(sizeof(struct module_info));
1081         if (!mi)
1082                 return -ENOMEM;
1083
1084         mi->name = strdup(name);
1085         mi->start = start;
1086
1087         if (!mi->name) {
1088                 free(mi);
1089                 return -ENOMEM;
1090         }
1091
1092         add_module(mi, modules);
1093
1094         return 0;
1095 }
1096
1097 static int read_proc_modules(const char *filename, struct rb_root *modules)
1098 {
1099         if (symbol__restricted_filename(filename, "/proc/modules"))
1100                 return -1;
1101
1102         if (modules__parse(filename, modules, __read_proc_modules)) {
1103                 delete_modules(modules);
1104                 return -1;
1105         }
1106
1107         return 0;
1108 }
1109
1110 int compare_proc_modules(const char *from, const char *to)
1111 {
1112         struct rb_root from_modules = RB_ROOT;
1113         struct rb_root to_modules = RB_ROOT;
1114         struct rb_node *from_node, *to_node;
1115         struct module_info *from_m, *to_m;
1116         int ret = -1;
1117
1118         if (read_proc_modules(from, &from_modules))
1119                 return -1;
1120
1121         if (read_proc_modules(to, &to_modules))
1122                 goto out_delete_from;
1123
1124         from_node = rb_first(&from_modules);
1125         to_node = rb_first(&to_modules);
1126         while (from_node) {
1127                 if (!to_node)
1128                         break;
1129
1130                 from_m = rb_entry(from_node, struct module_info, rb_node);
1131                 to_m = rb_entry(to_node, struct module_info, rb_node);
1132
1133                 if (from_m->start != to_m->start ||
1134                     strcmp(from_m->name, to_m->name))
1135                         break;
1136
1137                 from_node = rb_next(from_node);
1138                 to_node = rb_next(to_node);
1139         }
1140
1141         if (!from_node && !to_node)
1142                 ret = 0;
1143
1144         delete_modules(&to_modules);
1145 out_delete_from:
1146         delete_modules(&from_modules);
1147
1148         return ret;
1149 }
1150
1151 static int do_validate_kcore_modules(const char *filename, struct maps *kmaps)
1152 {
1153         struct rb_root modules = RB_ROOT;
1154         struct map_rb_node *old_node;
1155         int err;
1156
1157         err = read_proc_modules(filename, &modules);
1158         if (err)
1159                 return err;
1160
1161         maps__for_each_entry(kmaps, old_node) {
1162                 struct map *old_map = old_node->map;
1163                 struct module_info *mi;
1164                 struct dso *dso;
1165
1166                 if (!__map__is_kmodule(old_map)) {
1167                         continue;
1168                 }
1169                 dso = map__dso(old_map);
1170                 /* Module must be in memory at the same address */
1171                 mi = find_module(dso->short_name, &modules);
1172                 if (!mi || mi->start != map__start(old_map)) {
1173                         err = -EINVAL;
1174                         goto out;
1175                 }
1176         }
1177 out:
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 /*
1275  * Merges map into maps by splitting the new map within the existing map
1276  * regions.
1277  */
1278 int maps__merge_in(struct maps *kmaps, struct map *new_map)
1279 {
1280         struct map_rb_node *rb_node;
1281         LIST_HEAD(merged);
1282         int err = 0;
1283
1284         maps__for_each_entry(kmaps, rb_node) {
1285                 struct map *old_map = rb_node->map;
1286
1287                 /* no overload with this one */
1288                 if (map__end(new_map) < map__start(old_map) ||
1289                     map__start(new_map) >= map__end(old_map))
1290                         continue;
1291
1292                 if (map__start(new_map) < map__start(old_map)) {
1293                         /*
1294                          * |new......
1295                          *       |old....
1296                          */
1297                         if (map__end(new_map) < map__end(old_map)) {
1298                                 /*
1299                                  * |new......|     -> |new..|
1300                                  *       |old....| ->       |old....|
1301                                  */
1302                                 map__set_end(new_map, map__start(old_map));
1303                         } else {
1304                                 /*
1305                                  * |new.............| -> |new..|       |new..|
1306                                  *       |old....|    ->       |old....|
1307                                  */
1308                                 struct map_list_node *m = map_list_node__new();
1309
1310                                 if (!m) {
1311                                         err = -ENOMEM;
1312                                         goto out;
1313                                 }
1314
1315                                 m->map = map__clone(new_map);
1316                                 if (!m->map) {
1317                                         free(m);
1318                                         err = -ENOMEM;
1319                                         goto out;
1320                                 }
1321
1322                                 map__set_end(m->map, map__start(old_map));
1323                                 list_add_tail(&m->node, &merged);
1324                                 map__add_pgoff(new_map, map__end(old_map) - map__start(new_map));
1325                                 map__set_start(new_map, map__end(old_map));
1326                         }
1327                 } else {
1328                         /*
1329                          *      |new......
1330                          * |old....
1331                          */
1332                         if (map__end(new_map) < map__end(old_map)) {
1333                                 /*
1334                                  *      |new..|   -> x
1335                                  * |old.........| -> |old.........|
1336                                  */
1337                                 map__put(new_map);
1338                                 new_map = NULL;
1339                                 break;
1340                         } else {
1341                                 /*
1342                                  *      |new......| ->         |new...|
1343                                  * |old....|        -> |old....|
1344                                  */
1345                                 map__add_pgoff(new_map, map__end(old_map) - map__start(new_map));
1346                                 map__set_start(new_map, map__end(old_map));
1347                         }
1348                 }
1349         }
1350
1351 out:
1352         while (!list_empty(&merged)) {
1353                 struct map_list_node *old_node;
1354
1355                 old_node = list_entry(merged.next, struct map_list_node, node);
1356                 list_del_init(&old_node->node);
1357                 if (!err)
1358                         err = maps__insert(kmaps, old_node->map);
1359                 map__put(old_node->map);
1360                 free(old_node);
1361         }
1362
1363         if (new_map) {
1364                 if (!err)
1365                         err = maps__insert(kmaps, new_map);
1366                 map__put(new_map);
1367         }
1368         return err;
1369 }
1370
1371 static int dso__load_kcore(struct dso *dso, struct map *map,
1372                            const char *kallsyms_filename)
1373 {
1374         struct maps *kmaps = map__kmaps(map);
1375         struct kcore_mapfn_data md;
1376         struct map *replacement_map = NULL;
1377         struct map_rb_node *old_node, *next;
1378         struct machine *machine;
1379         bool is_64_bit;
1380         int err, fd;
1381         char kcore_filename[PATH_MAX];
1382         u64 stext;
1383
1384         if (!kmaps)
1385                 return -EINVAL;
1386
1387         machine = maps__machine(kmaps);
1388
1389         /* This function requires that the map is the kernel map */
1390         if (!__map__is_kernel(map))
1391                 return -EINVAL;
1392
1393         if (!filename_from_kallsyms_filename(kcore_filename, "kcore",
1394                                              kallsyms_filename))
1395                 return -EINVAL;
1396
1397         /* Modules and kernel must be present at their original addresses */
1398         if (validate_kcore_addresses(kallsyms_filename, map))
1399                 return -EINVAL;
1400
1401         md.dso = dso;
1402         INIT_LIST_HEAD(&md.maps);
1403
1404         fd = open(kcore_filename, O_RDONLY);
1405         if (fd < 0) {
1406                 pr_debug("Failed to open %s. Note /proc/kcore requires CAP_SYS_RAWIO capability to access.\n",
1407                          kcore_filename);
1408                 return -EINVAL;
1409         }
1410
1411         /* Read new maps into temporary lists */
1412         err = file__read_maps(fd, map__prot(map) & PROT_EXEC, kcore_mapfn, &md,
1413                               &is_64_bit);
1414         if (err)
1415                 goto out_err;
1416         dso->is_64_bit = is_64_bit;
1417
1418         if (list_empty(&md.maps)) {
1419                 err = -EINVAL;
1420                 goto out_err;
1421         }
1422
1423         /* Remove old maps */
1424         maps__for_each_entry_safe(kmaps, old_node, next) {
1425                 struct map *old_map = old_node->map;
1426
1427                 /*
1428                  * We need to preserve eBPF maps even if they are
1429                  * covered by kcore, because we need to access
1430                  * eBPF dso for source data.
1431                  */
1432                 if (old_map != map && !__map__is_bpf_prog(old_map))
1433                         maps__remove(kmaps, old_map);
1434         }
1435         machine->trampolines_mapped = false;
1436
1437         /* Find the kernel map using the '_stext' symbol */
1438         if (!kallsyms__get_function_start(kallsyms_filename, "_stext", &stext)) {
1439                 u64 replacement_size = 0;
1440                 struct map_list_node *new_node;
1441
1442                 list_for_each_entry(new_node, &md.maps, node) {
1443                         struct map *new_map = new_node->map;
1444                         u64 new_size = map__size(new_map);
1445
1446                         if (!(stext >= map__start(new_map) && stext < map__end(new_map)))
1447                                 continue;
1448
1449                         /*
1450                          * On some architectures, ARM64 for example, the kernel
1451                          * text can get allocated inside of the vmalloc segment.
1452                          * Select the smallest matching segment, in case stext
1453                          * falls within more than one in the list.
1454                          */
1455                         if (!replacement_map || new_size < replacement_size) {
1456                                 replacement_map = new_map;
1457                                 replacement_size = new_size;
1458                         }
1459                 }
1460         }
1461
1462         if (!replacement_map)
1463                 replacement_map = list_entry(md.maps.next, struct map_list_node, node)->map;
1464
1465         /* Add new maps */
1466         while (!list_empty(&md.maps)) {
1467                 struct map_list_node *new_node = list_entry(md.maps.next, struct map_list_node, node);
1468                 struct map *new_map = new_node->map;
1469
1470                 list_del_init(&new_node->node);
1471
1472                 if (RC_CHK_ACCESS(new_map) == RC_CHK_ACCESS(replacement_map)) {
1473                         struct map *map_ref;
1474
1475                         map__set_start(map, map__start(new_map));
1476                         map__set_end(map, map__end(new_map));
1477                         map__set_pgoff(map, map__pgoff(new_map));
1478                         map__set_map_ip(map, map__map_ip_ptr(new_map));
1479                         map__set_unmap_ip(map, map__unmap_ip_ptr(new_map));
1480                         /* Ensure maps are correctly ordered */
1481                         map_ref = map__get(map);
1482                         maps__remove(kmaps, map_ref);
1483                         err = maps__insert(kmaps, map_ref);
1484                         map__put(map_ref);
1485                         map__put(new_map);
1486                         if (err)
1487                                 goto out_err;
1488                 } else {
1489                         /*
1490                          * Merge kcore map into existing maps,
1491                          * and ensure that current maps (eBPF)
1492                          * stay intact.
1493                          */
1494                         if (maps__merge_in(kmaps, new_map)) {
1495                                 err = -EINVAL;
1496                                 goto out_err;
1497                         }
1498                 }
1499                 free(new_node);
1500         }
1501
1502         if (machine__is(machine, "x86_64")) {
1503                 u64 addr;
1504
1505                 /*
1506                  * If one of the corresponding symbols is there, assume the
1507                  * entry trampoline maps are too.
1508                  */
1509                 if (!kallsyms__get_function_start(kallsyms_filename,
1510                                                   ENTRY_TRAMPOLINE_NAME,
1511                                                   &addr))
1512                         machine->trampolines_mapped = true;
1513         }
1514
1515         /*
1516          * Set the data type and long name so that kcore can be read via
1517          * dso__data_read_addr().
1518          */
1519         if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
1520                 dso->binary_type = DSO_BINARY_TYPE__GUEST_KCORE;
1521         else
1522                 dso->binary_type = DSO_BINARY_TYPE__KCORE;
1523         dso__set_long_name(dso, strdup(kcore_filename), true);
1524
1525         close(fd);
1526
1527         if (map__prot(map) & PROT_EXEC)
1528                 pr_debug("Using %s for kernel object code\n", kcore_filename);
1529         else
1530                 pr_debug("Using %s for kernel data\n", kcore_filename);
1531
1532         return 0;
1533
1534 out_err:
1535         while (!list_empty(&md.maps)) {
1536                 struct map_list_node *list_node;
1537
1538                 list_node = list_entry(md.maps.next, struct map_list_node, node);
1539                 list_del_init(&list_node->node);
1540                 map__zput(list_node->map);
1541                 free(list_node);
1542         }
1543         close(fd);
1544         return err;
1545 }
1546
1547 /*
1548  * If the kernel is relocated at boot time, kallsyms won't match.  Compute the
1549  * delta based on the relocation reference symbol.
1550  */
1551 static int kallsyms__delta(struct kmap *kmap, const char *filename, u64 *delta)
1552 {
1553         u64 addr;
1554
1555         if (!kmap->ref_reloc_sym || !kmap->ref_reloc_sym->name)
1556                 return 0;
1557
1558         if (kallsyms__get_function_start(filename, kmap->ref_reloc_sym->name, &addr))
1559                 return -1;
1560
1561         *delta = addr - kmap->ref_reloc_sym->addr;
1562         return 0;
1563 }
1564
1565 int __dso__load_kallsyms(struct dso *dso, const char *filename,
1566                          struct map *map, bool no_kcore)
1567 {
1568         struct kmap *kmap = map__kmap(map);
1569         u64 delta = 0;
1570
1571         if (symbol__restricted_filename(filename, "/proc/kallsyms"))
1572                 return -1;
1573
1574         if (!kmap || !kmap->kmaps)
1575                 return -1;
1576
1577         if (dso__load_all_kallsyms(dso, filename) < 0)
1578                 return -1;
1579
1580         if (kallsyms__delta(kmap, filename, &delta))
1581                 return -1;
1582
1583         symbols__fixup_end(&dso->symbols, true);
1584         symbols__fixup_duplicate(&dso->symbols);
1585
1586         if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
1587                 dso->symtab_type = DSO_BINARY_TYPE__GUEST_KALLSYMS;
1588         else
1589                 dso->symtab_type = DSO_BINARY_TYPE__KALLSYMS;
1590
1591         if (!no_kcore && !dso__load_kcore(dso, map, filename))
1592                 return maps__split_kallsyms_for_kcore(kmap->kmaps, dso);
1593         else
1594                 return maps__split_kallsyms(kmap->kmaps, dso, delta, map);
1595 }
1596
1597 int dso__load_kallsyms(struct dso *dso, const char *filename,
1598                        struct map *map)
1599 {
1600         return __dso__load_kallsyms(dso, filename, map, false);
1601 }
1602
1603 static int dso__load_perf_map(const char *map_path, struct dso *dso)
1604 {
1605         char *line = NULL;
1606         size_t n;
1607         FILE *file;
1608         int nr_syms = 0;
1609
1610         file = fopen(map_path, "r");
1611         if (file == NULL)
1612                 goto out_failure;
1613
1614         while (!feof(file)) {
1615                 u64 start, size;
1616                 struct symbol *sym;
1617                 int line_len, len;
1618
1619                 line_len = getline(&line, &n, file);
1620                 if (line_len < 0)
1621                         break;
1622
1623                 if (!line)
1624                         goto out_failure;
1625
1626                 line[--line_len] = '\0'; /* \n */
1627
1628                 len = hex2u64(line, &start);
1629
1630                 len++;
1631                 if (len + 2 >= line_len)
1632                         continue;
1633
1634                 len += hex2u64(line + len, &size);
1635
1636                 len++;
1637                 if (len + 2 >= line_len)
1638                         continue;
1639
1640                 sym = symbol__new(start, size, STB_GLOBAL, STT_FUNC, line + len);
1641
1642                 if (sym == NULL)
1643                         goto out_delete_line;
1644
1645                 symbols__insert(&dso->symbols, sym);
1646                 nr_syms++;
1647         }
1648
1649         free(line);
1650         fclose(file);
1651
1652         return nr_syms;
1653
1654 out_delete_line:
1655         free(line);
1656 out_failure:
1657         return -1;
1658 }
1659
1660 #ifdef HAVE_LIBBFD_SUPPORT
1661 #define PACKAGE 'perf'
1662 #include <bfd.h>
1663
1664 static int bfd_symbols__cmpvalue(const void *a, const void *b)
1665 {
1666         const asymbol *as = *(const asymbol **)a, *bs = *(const asymbol **)b;
1667
1668         if (bfd_asymbol_value(as) != bfd_asymbol_value(bs))
1669                 return bfd_asymbol_value(as) - bfd_asymbol_value(bs);
1670
1671         return bfd_asymbol_name(as)[0] - bfd_asymbol_name(bs)[0];
1672 }
1673
1674 static int bfd2elf_binding(asymbol *symbol)
1675 {
1676         if (symbol->flags & BSF_WEAK)
1677                 return STB_WEAK;
1678         if (symbol->flags & BSF_GLOBAL)
1679                 return STB_GLOBAL;
1680         if (symbol->flags & BSF_LOCAL)
1681                 return STB_LOCAL;
1682         return -1;
1683 }
1684
1685 int dso__load_bfd_symbols(struct dso *dso, const char *debugfile)
1686 {
1687         int err = -1;
1688         long symbols_size, symbols_count, i;
1689         asection *section;
1690         asymbol **symbols, *sym;
1691         struct symbol *symbol;
1692         bfd *abfd;
1693         u64 start, len;
1694
1695         abfd = bfd_openr(debugfile, NULL);
1696         if (!abfd)
1697                 return -1;
1698
1699         if (!bfd_check_format(abfd, bfd_object)) {
1700                 pr_debug2("%s: cannot read %s bfd file.\n", __func__,
1701                           dso->long_name);
1702                 goto out_close;
1703         }
1704
1705         if (bfd_get_flavour(abfd) == bfd_target_elf_flavour)
1706                 goto out_close;
1707
1708         symbols_size = bfd_get_symtab_upper_bound(abfd);
1709         if (symbols_size == 0) {
1710                 bfd_close(abfd);
1711                 return 0;
1712         }
1713
1714         if (symbols_size < 0)
1715                 goto out_close;
1716
1717         symbols = malloc(symbols_size);
1718         if (!symbols)
1719                 goto out_close;
1720
1721         symbols_count = bfd_canonicalize_symtab(abfd, symbols);
1722         if (symbols_count < 0)
1723                 goto out_free;
1724
1725         section = bfd_get_section_by_name(abfd, ".text");
1726         if (section) {
1727                 for (i = 0; i < symbols_count; ++i) {
1728                         if (!strcmp(bfd_asymbol_name(symbols[i]), "__ImageBase") ||
1729                             !strcmp(bfd_asymbol_name(symbols[i]), "__image_base__"))
1730                                 break;
1731                 }
1732                 if (i < symbols_count) {
1733                         /* PE symbols can only have 4 bytes, so use .text high bits */
1734                         dso->text_offset = section->vma - (u32)section->vma;
1735                         dso->text_offset += (u32)bfd_asymbol_value(symbols[i]);
1736                 } else {
1737                         dso->text_offset = section->vma - section->filepos;
1738                 }
1739         }
1740
1741         qsort(symbols, symbols_count, sizeof(asymbol *), bfd_symbols__cmpvalue);
1742
1743 #ifdef bfd_get_section
1744 #define bfd_asymbol_section bfd_get_section
1745 #endif
1746         for (i = 0; i < symbols_count; ++i) {
1747                 sym = symbols[i];
1748                 section = bfd_asymbol_section(sym);
1749                 if (bfd2elf_binding(sym) < 0)
1750                         continue;
1751
1752                 while (i + 1 < symbols_count &&
1753                        bfd_asymbol_section(symbols[i + 1]) == section &&
1754                        bfd2elf_binding(symbols[i + 1]) < 0)
1755                         i++;
1756
1757                 if (i + 1 < symbols_count &&
1758                     bfd_asymbol_section(symbols[i + 1]) == section)
1759                         len = symbols[i + 1]->value - sym->value;
1760                 else
1761                         len = section->size - sym->value;
1762
1763                 start = bfd_asymbol_value(sym) - dso->text_offset;
1764                 symbol = symbol__new(start, len, bfd2elf_binding(sym), STT_FUNC,
1765                                      bfd_asymbol_name(sym));
1766                 if (!symbol)
1767                         goto out_free;
1768
1769                 symbols__insert(&dso->symbols, symbol);
1770         }
1771 #ifdef bfd_get_section
1772 #undef bfd_asymbol_section
1773 #endif
1774
1775         symbols__fixup_end(&dso->symbols, false);
1776         symbols__fixup_duplicate(&dso->symbols);
1777         dso->adjust_symbols = 1;
1778
1779         err = 0;
1780 out_free:
1781         free(symbols);
1782 out_close:
1783         bfd_close(abfd);
1784         return err;
1785 }
1786 #endif
1787
1788 static bool dso__is_compatible_symtab_type(struct dso *dso, bool kmod,
1789                                            enum dso_binary_type type)
1790 {
1791         switch (type) {
1792         case DSO_BINARY_TYPE__JAVA_JIT:
1793         case DSO_BINARY_TYPE__DEBUGLINK:
1794         case DSO_BINARY_TYPE__SYSTEM_PATH_DSO:
1795         case DSO_BINARY_TYPE__FEDORA_DEBUGINFO:
1796         case DSO_BINARY_TYPE__UBUNTU_DEBUGINFO:
1797         case DSO_BINARY_TYPE__MIXEDUP_UBUNTU_DEBUGINFO:
1798         case DSO_BINARY_TYPE__BUILDID_DEBUGINFO:
1799         case DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO:
1800                 return !kmod && dso->kernel == DSO_SPACE__USER;
1801
1802         case DSO_BINARY_TYPE__KALLSYMS:
1803         case DSO_BINARY_TYPE__VMLINUX:
1804         case DSO_BINARY_TYPE__KCORE:
1805                 return dso->kernel == DSO_SPACE__KERNEL;
1806
1807         case DSO_BINARY_TYPE__GUEST_KALLSYMS:
1808         case DSO_BINARY_TYPE__GUEST_VMLINUX:
1809         case DSO_BINARY_TYPE__GUEST_KCORE:
1810                 return dso->kernel == DSO_SPACE__KERNEL_GUEST;
1811
1812         case DSO_BINARY_TYPE__GUEST_KMODULE:
1813         case DSO_BINARY_TYPE__GUEST_KMODULE_COMP:
1814         case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE:
1815         case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP:
1816                 /*
1817                  * kernel modules know their symtab type - it's set when
1818                  * creating a module dso in machine__addnew_module_map().
1819                  */
1820                 return kmod && dso->symtab_type == type;
1821
1822         case DSO_BINARY_TYPE__BUILD_ID_CACHE:
1823         case DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO:
1824                 return true;
1825
1826         case DSO_BINARY_TYPE__BPF_PROG_INFO:
1827         case DSO_BINARY_TYPE__BPF_IMAGE:
1828         case DSO_BINARY_TYPE__OOL:
1829         case DSO_BINARY_TYPE__NOT_FOUND:
1830         default:
1831                 return false;
1832         }
1833 }
1834
1835 /* Checks for the existence of the perf-<pid>.map file in two different
1836  * locations.  First, if the process is a separate mount namespace, check in
1837  * that namespace using the pid of the innermost pid namespace.  If's not in a
1838  * namespace, or the file can't be found there, try in the mount namespace of
1839  * the tracing process using our view of its pid.
1840  */
1841 static int dso__find_perf_map(char *filebuf, size_t bufsz,
1842                               struct nsinfo **nsip)
1843 {
1844         struct nscookie nsc;
1845         struct nsinfo *nsi;
1846         struct nsinfo *nnsi;
1847         int rc = -1;
1848
1849         nsi = *nsip;
1850
1851         if (nsinfo__need_setns(nsi)) {
1852                 snprintf(filebuf, bufsz, "/tmp/perf-%d.map", nsinfo__nstgid(nsi));
1853                 nsinfo__mountns_enter(nsi, &nsc);
1854                 rc = access(filebuf, R_OK);
1855                 nsinfo__mountns_exit(&nsc);
1856                 if (rc == 0)
1857                         return rc;
1858         }
1859
1860         nnsi = nsinfo__copy(nsi);
1861         if (nnsi) {
1862                 nsinfo__put(nsi);
1863
1864                 nsinfo__clear_need_setns(nnsi);
1865                 snprintf(filebuf, bufsz, "/tmp/perf-%d.map", nsinfo__tgid(nnsi));
1866                 *nsip = nnsi;
1867                 rc = 0;
1868         }
1869
1870         return rc;
1871 }
1872
1873 int dso__load(struct dso *dso, struct map *map)
1874 {
1875         char *name;
1876         int ret = -1;
1877         u_int i;
1878         struct machine *machine = NULL;
1879         char *root_dir = (char *) "";
1880         int ss_pos = 0;
1881         struct symsrc ss_[2];
1882         struct symsrc *syms_ss = NULL, *runtime_ss = NULL;
1883         bool kmod;
1884         bool perfmap;
1885         struct build_id bid;
1886         struct nscookie nsc;
1887         char newmapname[PATH_MAX];
1888         const char *map_path = dso->long_name;
1889
1890         mutex_lock(&dso->lock);
1891         perfmap = strncmp(dso->name, "/tmp/perf-", 10) == 0;
1892         if (perfmap) {
1893                 if (dso->nsinfo && (dso__find_perf_map(newmapname,
1894                     sizeof(newmapname), &dso->nsinfo) == 0)) {
1895                         map_path = newmapname;
1896                 }
1897         }
1898
1899         nsinfo__mountns_enter(dso->nsinfo, &nsc);
1900
1901         /* check again under the dso->lock */
1902         if (dso__loaded(dso)) {
1903                 ret = 1;
1904                 goto out;
1905         }
1906
1907         kmod = dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1908                 dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP ||
1909                 dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE ||
1910                 dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE_COMP;
1911
1912         if (dso->kernel && !kmod) {
1913                 if (dso->kernel == DSO_SPACE__KERNEL)
1914                         ret = dso__load_kernel_sym(dso, map);
1915                 else if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
1916                         ret = dso__load_guest_kernel_sym(dso, map);
1917
1918                 machine = maps__machine(map__kmaps(map));
1919                 if (machine__is(machine, "x86_64"))
1920                         machine__map_x86_64_entry_trampolines(machine, dso);
1921                 goto out;
1922         }
1923
1924         dso->adjust_symbols = 0;
1925
1926         if (perfmap) {
1927                 ret = dso__load_perf_map(map_path, dso);
1928                 dso->symtab_type = ret > 0 ? DSO_BINARY_TYPE__JAVA_JIT :
1929                                              DSO_BINARY_TYPE__NOT_FOUND;
1930                 goto out;
1931         }
1932
1933         if (machine)
1934                 root_dir = machine->root_dir;
1935
1936         name = malloc(PATH_MAX);
1937         if (!name)
1938                 goto out;
1939
1940         /*
1941          * Read the build id if possible. This is required for
1942          * DSO_BINARY_TYPE__BUILDID_DEBUGINFO to work
1943          */
1944         if (!dso->has_build_id &&
1945             is_regular_file(dso->long_name)) {
1946             __symbol__join_symfs(name, PATH_MAX, dso->long_name);
1947                 if (filename__read_build_id(name, &bid) > 0)
1948                         dso__set_build_id(dso, &bid);
1949         }
1950
1951         /*
1952          * Iterate over candidate debug images.
1953          * Keep track of "interesting" ones (those which have a symtab, dynsym,
1954          * and/or opd section) for processing.
1955          */
1956         for (i = 0; i < DSO_BINARY_TYPE__SYMTAB_CNT; i++) {
1957                 struct symsrc *ss = &ss_[ss_pos];
1958                 bool next_slot = false;
1959                 bool is_reg;
1960                 bool nsexit;
1961                 int bfdrc = -1;
1962                 int sirc = -1;
1963
1964                 enum dso_binary_type symtab_type = binary_type_symtab[i];
1965
1966                 nsexit = (symtab_type == DSO_BINARY_TYPE__BUILD_ID_CACHE ||
1967                     symtab_type == DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO);
1968
1969                 if (!dso__is_compatible_symtab_type(dso, kmod, symtab_type))
1970                         continue;
1971
1972                 if (dso__read_binary_type_filename(dso, symtab_type,
1973                                                    root_dir, name, PATH_MAX))
1974                         continue;
1975
1976                 if (nsexit)
1977                         nsinfo__mountns_exit(&nsc);
1978
1979                 is_reg = is_regular_file(name);
1980                 if (!is_reg && errno == ENOENT && dso->nsinfo) {
1981                         char *new_name = dso__filename_with_chroot(dso, name);
1982                         if (new_name) {
1983                                 is_reg = is_regular_file(new_name);
1984                                 strlcpy(name, new_name, PATH_MAX);
1985                                 free(new_name);
1986                         }
1987                 }
1988
1989 #ifdef HAVE_LIBBFD_SUPPORT
1990                 if (is_reg)
1991                         bfdrc = dso__load_bfd_symbols(dso, name);
1992 #endif
1993                 if (is_reg && bfdrc < 0)
1994                         sirc = symsrc__init(ss, dso, name, symtab_type);
1995
1996                 if (nsexit)
1997                         nsinfo__mountns_enter(dso->nsinfo, &nsc);
1998
1999                 if (bfdrc == 0) {
2000                         ret = 0;
2001                         break;
2002                 }
2003
2004                 if (!is_reg || sirc < 0)
2005                         continue;
2006
2007                 if (!syms_ss && symsrc__has_symtab(ss)) {
2008                         syms_ss = ss;
2009                         next_slot = true;
2010                         if (!dso->symsrc_filename)
2011                                 dso->symsrc_filename = strdup(name);
2012                 }
2013
2014                 if (!runtime_ss && symsrc__possibly_runtime(ss)) {
2015                         runtime_ss = ss;
2016                         next_slot = true;
2017                 }
2018
2019                 if (next_slot) {
2020                         ss_pos++;
2021
2022                         if (syms_ss && runtime_ss)
2023                                 break;
2024                 } else {
2025                         symsrc__destroy(ss);
2026                 }
2027
2028         }
2029
2030         if (!runtime_ss && !syms_ss)
2031                 goto out_free;
2032
2033         if (runtime_ss && !syms_ss) {
2034                 syms_ss = runtime_ss;
2035         }
2036
2037         /* We'll have to hope for the best */
2038         if (!runtime_ss && syms_ss)
2039                 runtime_ss = syms_ss;
2040
2041         if (syms_ss)
2042                 ret = dso__load_sym(dso, map, syms_ss, runtime_ss, kmod);
2043         else
2044                 ret = -1;
2045
2046         if (ret > 0) {
2047                 int nr_plt;
2048
2049                 nr_plt = dso__synthesize_plt_symbols(dso, runtime_ss);
2050                 if (nr_plt > 0)
2051                         ret += nr_plt;
2052         }
2053
2054         for (; ss_pos > 0; ss_pos--)
2055                 symsrc__destroy(&ss_[ss_pos - 1]);
2056 out_free:
2057         free(name);
2058         if (ret < 0 && strstr(dso->name, " (deleted)") != NULL)
2059                 ret = 0;
2060 out:
2061         dso__set_loaded(dso);
2062         mutex_unlock(&dso->lock);
2063         nsinfo__mountns_exit(&nsc);
2064
2065         return ret;
2066 }
2067
2068 static int map__strcmp(const void *a, const void *b)
2069 {
2070         const struct map *map_a = *(const struct map **)a;
2071         const struct map *map_b = *(const struct map **)b;
2072         const struct dso *dso_a = map__dso(map_a);
2073         const struct dso *dso_b = map__dso(map_b);
2074         int ret = strcmp(dso_a->short_name, dso_b->short_name);
2075
2076         if (ret == 0 && map_a != map_b) {
2077                 /*
2078                  * Ensure distinct but name equal maps have an order in part to
2079                  * aid reference counting.
2080                  */
2081                 ret = (int)map__start(map_a) - (int)map__start(map_b);
2082                 if (ret == 0)
2083                         ret = (int)((intptr_t)map_a - (intptr_t)map_b);
2084         }
2085
2086         return ret;
2087 }
2088
2089 static int map__strcmp_name(const void *name, const void *b)
2090 {
2091         const struct dso *dso = map__dso(*(const struct map **)b);
2092
2093         return strcmp(name, dso->short_name);
2094 }
2095
2096 void __maps__sort_by_name(struct maps *maps)
2097 {
2098         qsort(maps__maps_by_name(maps), maps__nr_maps(maps), sizeof(struct map *), map__strcmp);
2099 }
2100
2101 static int map__groups__sort_by_name_from_rbtree(struct maps *maps)
2102 {
2103         struct map_rb_node *rb_node;
2104         struct map **maps_by_name = realloc(maps__maps_by_name(maps),
2105                                             maps__nr_maps(maps) * sizeof(struct map *));
2106         int i = 0;
2107
2108         if (maps_by_name == NULL)
2109                 return -1;
2110
2111         up_read(maps__lock(maps));
2112         down_write(maps__lock(maps));
2113
2114         RC_CHK_ACCESS(maps)->maps_by_name = maps_by_name;
2115         RC_CHK_ACCESS(maps)->nr_maps_allocated = maps__nr_maps(maps);
2116
2117         maps__for_each_entry(maps, rb_node)
2118                 maps_by_name[i++] = map__get(rb_node->map);
2119
2120         __maps__sort_by_name(maps);
2121
2122         up_write(maps__lock(maps));
2123         down_read(maps__lock(maps));
2124
2125         return 0;
2126 }
2127
2128 static struct map *__maps__find_by_name(struct maps *maps, const char *name)
2129 {
2130         struct map **mapp;
2131
2132         if (maps__maps_by_name(maps) == NULL &&
2133             map__groups__sort_by_name_from_rbtree(maps))
2134                 return NULL;
2135
2136         mapp = bsearch(name, maps__maps_by_name(maps), maps__nr_maps(maps),
2137                        sizeof(*mapp), map__strcmp_name);
2138         if (mapp)
2139                 return *mapp;
2140         return NULL;
2141 }
2142
2143 struct map *maps__find_by_name(struct maps *maps, const char *name)
2144 {
2145         struct map_rb_node *rb_node;
2146         struct map *map;
2147
2148         down_read(maps__lock(maps));
2149
2150
2151         if (RC_CHK_ACCESS(maps)->last_search_by_name) {
2152                 const struct dso *dso = map__dso(RC_CHK_ACCESS(maps)->last_search_by_name);
2153
2154                 if (strcmp(dso->short_name, name) == 0) {
2155                         map = RC_CHK_ACCESS(maps)->last_search_by_name;
2156                         goto out_unlock;
2157                 }
2158         }
2159         /*
2160          * If we have maps->maps_by_name, then the name isn't in the rbtree,
2161          * as maps->maps_by_name mirrors the rbtree when lookups by name are
2162          * made.
2163          */
2164         map = __maps__find_by_name(maps, name);
2165         if (map || maps__maps_by_name(maps) != NULL)
2166                 goto out_unlock;
2167
2168         /* Fallback to traversing the rbtree... */
2169         maps__for_each_entry(maps, rb_node) {
2170                 struct dso *dso;
2171
2172                 map = rb_node->map;
2173                 dso = map__dso(map);
2174                 if (strcmp(dso->short_name, name) == 0) {
2175                         RC_CHK_ACCESS(maps)->last_search_by_name = map;
2176                         goto out_unlock;
2177                 }
2178         }
2179         map = NULL;
2180
2181 out_unlock:
2182         up_read(maps__lock(maps));
2183         return map;
2184 }
2185
2186 int dso__load_vmlinux(struct dso *dso, struct map *map,
2187                       const char *vmlinux, bool vmlinux_allocated)
2188 {
2189         int err = -1;
2190         struct symsrc ss;
2191         char symfs_vmlinux[PATH_MAX];
2192         enum dso_binary_type symtab_type;
2193
2194         if (vmlinux[0] == '/')
2195                 snprintf(symfs_vmlinux, sizeof(symfs_vmlinux), "%s", vmlinux);
2196         else
2197                 symbol__join_symfs(symfs_vmlinux, vmlinux);
2198
2199         if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
2200                 symtab_type = DSO_BINARY_TYPE__GUEST_VMLINUX;
2201         else
2202                 symtab_type = DSO_BINARY_TYPE__VMLINUX;
2203
2204         if (symsrc__init(&ss, dso, symfs_vmlinux, symtab_type))
2205                 return -1;
2206
2207         /*
2208          * dso__load_sym() may copy 'dso' which will result in the copies having
2209          * an incorrect long name unless we set it here first.
2210          */
2211         dso__set_long_name(dso, vmlinux, vmlinux_allocated);
2212         if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
2213                 dso->binary_type = DSO_BINARY_TYPE__GUEST_VMLINUX;
2214         else
2215                 dso->binary_type = DSO_BINARY_TYPE__VMLINUX;
2216
2217         err = dso__load_sym(dso, map, &ss, &ss, 0);
2218         symsrc__destroy(&ss);
2219
2220         if (err > 0) {
2221                 dso__set_loaded(dso);
2222                 pr_debug("Using %s for symbols\n", symfs_vmlinux);
2223         }
2224
2225         return err;
2226 }
2227
2228 int dso__load_vmlinux_path(struct dso *dso, struct map *map)
2229 {
2230         int i, err = 0;
2231         char *filename = NULL;
2232
2233         pr_debug("Looking at the vmlinux_path (%d entries long)\n",
2234                  vmlinux_path__nr_entries + 1);
2235
2236         for (i = 0; i < vmlinux_path__nr_entries; ++i) {
2237                 err = dso__load_vmlinux(dso, map, vmlinux_path[i], false);
2238                 if (err > 0)
2239                         goto out;
2240         }
2241
2242         if (!symbol_conf.ignore_vmlinux_buildid)
2243                 filename = dso__build_id_filename(dso, NULL, 0, false);
2244         if (filename != NULL) {
2245                 err = dso__load_vmlinux(dso, map, filename, true);
2246                 if (err > 0)
2247                         goto out;
2248                 free(filename);
2249         }
2250 out:
2251         return err;
2252 }
2253
2254 static bool visible_dir_filter(const char *name, struct dirent *d)
2255 {
2256         if (d->d_type != DT_DIR)
2257                 return false;
2258         return lsdir_no_dot_filter(name, d);
2259 }
2260
2261 static int find_matching_kcore(struct map *map, char *dir, size_t dir_sz)
2262 {
2263         char kallsyms_filename[PATH_MAX];
2264         int ret = -1;
2265         struct strlist *dirs;
2266         struct str_node *nd;
2267
2268         dirs = lsdir(dir, visible_dir_filter);
2269         if (!dirs)
2270                 return -1;
2271
2272         strlist__for_each_entry(nd, dirs) {
2273                 scnprintf(kallsyms_filename, sizeof(kallsyms_filename),
2274                           "%s/%s/kallsyms", dir, nd->s);
2275                 if (!validate_kcore_addresses(kallsyms_filename, map)) {
2276                         strlcpy(dir, kallsyms_filename, dir_sz);
2277                         ret = 0;
2278                         break;
2279                 }
2280         }
2281
2282         strlist__delete(dirs);
2283
2284         return ret;
2285 }
2286
2287 /*
2288  * Use open(O_RDONLY) to check readability directly instead of access(R_OK)
2289  * since access(R_OK) only checks with real UID/GID but open() use effective
2290  * UID/GID and actual capabilities (e.g. /proc/kcore requires CAP_SYS_RAWIO).
2291  */
2292 static bool filename__readable(const char *file)
2293 {
2294         int fd = open(file, O_RDONLY);
2295         if (fd < 0)
2296                 return false;
2297         close(fd);
2298         return true;
2299 }
2300
2301 static char *dso__find_kallsyms(struct dso *dso, struct map *map)
2302 {
2303         struct build_id bid;
2304         char sbuild_id[SBUILD_ID_SIZE];
2305         bool is_host = false;
2306         char path[PATH_MAX];
2307
2308         if (!dso->has_build_id) {
2309                 /*
2310                  * Last resort, if we don't have a build-id and couldn't find
2311                  * any vmlinux file, try the running kernel kallsyms table.
2312                  */
2313                 goto proc_kallsyms;
2314         }
2315
2316         if (sysfs__read_build_id("/sys/kernel/notes", &bid) == 0)
2317                 is_host = dso__build_id_equal(dso, &bid);
2318
2319         /* Try a fast path for /proc/kallsyms if possible */
2320         if (is_host) {
2321                 /*
2322                  * Do not check the build-id cache, unless we know we cannot use
2323                  * /proc/kcore or module maps don't match to /proc/kallsyms.
2324                  * To check readability of /proc/kcore, do not use access(R_OK)
2325                  * since /proc/kcore requires CAP_SYS_RAWIO to read and access
2326                  * can't check it.
2327                  */
2328                 if (filename__readable("/proc/kcore") &&
2329                     !validate_kcore_addresses("/proc/kallsyms", map))
2330                         goto proc_kallsyms;
2331         }
2332
2333         build_id__sprintf(&dso->bid, sbuild_id);
2334
2335         /* Find kallsyms in build-id cache with kcore */
2336         scnprintf(path, sizeof(path), "%s/%s/%s",
2337                   buildid_dir, DSO__NAME_KCORE, sbuild_id);
2338
2339         if (!find_matching_kcore(map, path, sizeof(path)))
2340                 return strdup(path);
2341
2342         /* Use current /proc/kallsyms if possible */
2343         if (is_host) {
2344 proc_kallsyms:
2345                 return strdup("/proc/kallsyms");
2346         }
2347
2348         /* Finally, find a cache of kallsyms */
2349         if (!build_id_cache__kallsyms_path(sbuild_id, path, sizeof(path))) {
2350                 pr_err("No kallsyms or vmlinux with build-id %s was found\n",
2351                        sbuild_id);
2352                 return NULL;
2353         }
2354
2355         return strdup(path);
2356 }
2357
2358 static int dso__load_kernel_sym(struct dso *dso, struct map *map)
2359 {
2360         int err;
2361         const char *kallsyms_filename = NULL;
2362         char *kallsyms_allocated_filename = NULL;
2363         char *filename = NULL;
2364
2365         /*
2366          * Step 1: if the user specified a kallsyms or vmlinux filename, use
2367          * it and only it, reporting errors to the user if it cannot be used.
2368          *
2369          * For instance, try to analyse an ARM perf.data file _without_ a
2370          * build-id, or if the user specifies the wrong path to the right
2371          * vmlinux file, obviously we can't fallback to another vmlinux (a
2372          * x86_86 one, on the machine where analysis is being performed, say),
2373          * or worse, /proc/kallsyms.
2374          *
2375          * If the specified file _has_ a build-id and there is a build-id
2376          * section in the perf.data file, we will still do the expected
2377          * validation in dso__load_vmlinux and will bail out if they don't
2378          * match.
2379          */
2380         if (symbol_conf.kallsyms_name != NULL) {
2381                 kallsyms_filename = symbol_conf.kallsyms_name;
2382                 goto do_kallsyms;
2383         }
2384
2385         if (!symbol_conf.ignore_vmlinux && symbol_conf.vmlinux_name != NULL) {
2386                 return dso__load_vmlinux(dso, map, symbol_conf.vmlinux_name, false);
2387         }
2388
2389         /*
2390          * Before checking on common vmlinux locations, check if it's
2391          * stored as standard build id binary (not kallsyms) under
2392          * .debug cache.
2393          */
2394         if (!symbol_conf.ignore_vmlinux_buildid)
2395                 filename = __dso__build_id_filename(dso, NULL, 0, false, false);
2396         if (filename != NULL) {
2397                 err = dso__load_vmlinux(dso, map, filename, true);
2398                 if (err > 0)
2399                         return err;
2400                 free(filename);
2401         }
2402
2403         if (!symbol_conf.ignore_vmlinux && vmlinux_path != NULL) {
2404                 err = dso__load_vmlinux_path(dso, map);
2405                 if (err > 0)
2406                         return err;
2407         }
2408
2409         /* do not try local files if a symfs was given */
2410         if (symbol_conf.symfs[0] != 0)
2411                 return -1;
2412
2413         kallsyms_allocated_filename = dso__find_kallsyms(dso, map);
2414         if (!kallsyms_allocated_filename)
2415                 return -1;
2416
2417         kallsyms_filename = kallsyms_allocated_filename;
2418
2419 do_kallsyms:
2420         err = dso__load_kallsyms(dso, kallsyms_filename, map);
2421         if (err > 0)
2422                 pr_debug("Using %s for symbols\n", kallsyms_filename);
2423         free(kallsyms_allocated_filename);
2424
2425         if (err > 0 && !dso__is_kcore(dso)) {
2426                 dso->binary_type = DSO_BINARY_TYPE__KALLSYMS;
2427                 dso__set_long_name(dso, DSO__NAME_KALLSYMS, false);
2428                 map__fixup_start(map);
2429                 map__fixup_end(map);
2430         }
2431
2432         return err;
2433 }
2434
2435 static int dso__load_guest_kernel_sym(struct dso *dso, struct map *map)
2436 {
2437         int err;
2438         const char *kallsyms_filename;
2439         struct machine *machine = maps__machine(map__kmaps(map));
2440         char path[PATH_MAX];
2441
2442         if (machine->kallsyms_filename) {
2443                 kallsyms_filename = machine->kallsyms_filename;
2444         } else if (machine__is_default_guest(machine)) {
2445                 /*
2446                  * if the user specified a vmlinux filename, use it and only
2447                  * it, reporting errors to the user if it cannot be used.
2448                  * Or use file guest_kallsyms inputted by user on commandline
2449                  */
2450                 if (symbol_conf.default_guest_vmlinux_name != NULL) {
2451                         err = dso__load_vmlinux(dso, map,
2452                                                 symbol_conf.default_guest_vmlinux_name,
2453                                                 false);
2454                         return err;
2455                 }
2456
2457                 kallsyms_filename = symbol_conf.default_guest_kallsyms;
2458                 if (!kallsyms_filename)
2459                         return -1;
2460         } else {
2461                 sprintf(path, "%s/proc/kallsyms", machine->root_dir);
2462                 kallsyms_filename = path;
2463         }
2464
2465         err = dso__load_kallsyms(dso, kallsyms_filename, map);
2466         if (err > 0)
2467                 pr_debug("Using %s for symbols\n", kallsyms_filename);
2468         if (err > 0 && !dso__is_kcore(dso)) {
2469                 dso->binary_type = DSO_BINARY_TYPE__GUEST_KALLSYMS;
2470                 dso__set_long_name(dso, machine->mmap_name, false);
2471                 map__fixup_start(map);
2472                 map__fixup_end(map);
2473         }
2474
2475         return err;
2476 }
2477
2478 static void vmlinux_path__exit(void)
2479 {
2480         while (--vmlinux_path__nr_entries >= 0)
2481                 zfree(&vmlinux_path[vmlinux_path__nr_entries]);
2482         vmlinux_path__nr_entries = 0;
2483
2484         zfree(&vmlinux_path);
2485 }
2486
2487 static const char * const vmlinux_paths[] = {
2488         "vmlinux",
2489         "/boot/vmlinux"
2490 };
2491
2492 static const char * const vmlinux_paths_upd[] = {
2493         "/boot/vmlinux-%s",
2494         "/usr/lib/debug/boot/vmlinux-%s",
2495         "/lib/modules/%s/build/vmlinux",
2496         "/usr/lib/debug/lib/modules/%s/vmlinux",
2497         "/usr/lib/debug/boot/vmlinux-%s.debug"
2498 };
2499
2500 static int vmlinux_path__add(const char *new_entry)
2501 {
2502         vmlinux_path[vmlinux_path__nr_entries] = strdup(new_entry);
2503         if (vmlinux_path[vmlinux_path__nr_entries] == NULL)
2504                 return -1;
2505         ++vmlinux_path__nr_entries;
2506
2507         return 0;
2508 }
2509
2510 static int vmlinux_path__init(struct perf_env *env)
2511 {
2512         struct utsname uts;
2513         char bf[PATH_MAX];
2514         char *kernel_version;
2515         unsigned int i;
2516
2517         vmlinux_path = malloc(sizeof(char *) * (ARRAY_SIZE(vmlinux_paths) +
2518                               ARRAY_SIZE(vmlinux_paths_upd)));
2519         if (vmlinux_path == NULL)
2520                 return -1;
2521
2522         for (i = 0; i < ARRAY_SIZE(vmlinux_paths); i++)
2523                 if (vmlinux_path__add(vmlinux_paths[i]) < 0)
2524                         goto out_fail;
2525
2526         /* only try kernel version if no symfs was given */
2527         if (symbol_conf.symfs[0] != 0)
2528                 return 0;
2529
2530         if (env) {
2531                 kernel_version = env->os_release;
2532         } else {
2533                 if (uname(&uts) < 0)
2534                         goto out_fail;
2535
2536                 kernel_version = uts.release;
2537         }
2538
2539         for (i = 0; i < ARRAY_SIZE(vmlinux_paths_upd); i++) {
2540                 snprintf(bf, sizeof(bf), vmlinux_paths_upd[i], kernel_version);
2541                 if (vmlinux_path__add(bf) < 0)
2542                         goto out_fail;
2543         }
2544
2545         return 0;
2546
2547 out_fail:
2548         vmlinux_path__exit();
2549         return -1;
2550 }
2551
2552 int setup_list(struct strlist **list, const char *list_str,
2553                       const char *list_name)
2554 {
2555         if (list_str == NULL)
2556                 return 0;
2557
2558         *list = strlist__new(list_str, NULL);
2559         if (!*list) {
2560                 pr_err("problems parsing %s list\n", list_name);
2561                 return -1;
2562         }
2563
2564         symbol_conf.has_filter = true;
2565         return 0;
2566 }
2567
2568 int setup_intlist(struct intlist **list, const char *list_str,
2569                   const char *list_name)
2570 {
2571         if (list_str == NULL)
2572                 return 0;
2573
2574         *list = intlist__new(list_str);
2575         if (!*list) {
2576                 pr_err("problems parsing %s list\n", list_name);
2577                 return -1;
2578         }
2579         return 0;
2580 }
2581
2582 static int setup_addrlist(struct intlist **addr_list, struct strlist *sym_list)
2583 {
2584         struct str_node *pos, *tmp;
2585         unsigned long val;
2586         char *sep;
2587         const char *end;
2588         int i = 0, err;
2589
2590         *addr_list = intlist__new(NULL);
2591         if (!*addr_list)
2592                 return -1;
2593
2594         strlist__for_each_entry_safe(pos, tmp, sym_list) {
2595                 errno = 0;
2596                 val = strtoul(pos->s, &sep, 16);
2597                 if (errno || (sep == pos->s))
2598                         continue;
2599
2600                 if (*sep != '\0') {
2601                         end = pos->s + strlen(pos->s) - 1;
2602                         while (end >= sep && isspace(*end))
2603                                 end--;
2604
2605                         if (end >= sep)
2606                                 continue;
2607                 }
2608
2609                 err = intlist__add(*addr_list, val);
2610                 if (err)
2611                         break;
2612
2613                 strlist__remove(sym_list, pos);
2614                 i++;
2615         }
2616
2617         if (i == 0) {
2618                 intlist__delete(*addr_list);
2619                 *addr_list = NULL;
2620         }
2621
2622         return 0;
2623 }
2624
2625 static bool symbol__read_kptr_restrict(void)
2626 {
2627         bool value = false;
2628         FILE *fp = fopen("/proc/sys/kernel/kptr_restrict", "r");
2629
2630         if (fp != NULL) {
2631                 char line[8];
2632
2633                 if (fgets(line, sizeof(line), fp) != NULL)
2634                         value = perf_cap__capable(CAP_SYSLOG) ?
2635                                         (atoi(line) >= 2) :
2636                                         (atoi(line) != 0);
2637
2638                 fclose(fp);
2639         }
2640
2641         /* Per kernel/kallsyms.c:
2642          * we also restrict when perf_event_paranoid > 1 w/o CAP_SYSLOG
2643          */
2644         if (perf_event_paranoid() > 1 && !perf_cap__capable(CAP_SYSLOG))
2645                 value = true;
2646
2647         return value;
2648 }
2649
2650 int symbol__annotation_init(void)
2651 {
2652         if (symbol_conf.init_annotation)
2653                 return 0;
2654
2655         if (symbol_conf.initialized) {
2656                 pr_err("Annotation needs to be init before symbol__init()\n");
2657                 return -1;
2658         }
2659
2660         symbol_conf.priv_size += sizeof(struct annotation);
2661         symbol_conf.init_annotation = true;
2662         return 0;
2663 }
2664
2665 int symbol__init(struct perf_env *env)
2666 {
2667         const char *symfs;
2668
2669         if (symbol_conf.initialized)
2670                 return 0;
2671
2672         symbol_conf.priv_size = PERF_ALIGN(symbol_conf.priv_size, sizeof(u64));
2673
2674         symbol__elf_init();
2675
2676         if (symbol_conf.try_vmlinux_path && vmlinux_path__init(env) < 0)
2677                 return -1;
2678
2679         if (symbol_conf.field_sep && *symbol_conf.field_sep == '.') {
2680                 pr_err("'.' is the only non valid --field-separator argument\n");
2681                 return -1;
2682         }
2683
2684         if (setup_list(&symbol_conf.dso_list,
2685                        symbol_conf.dso_list_str, "dso") < 0)
2686                 return -1;
2687
2688         if (setup_list(&symbol_conf.comm_list,
2689                        symbol_conf.comm_list_str, "comm") < 0)
2690                 goto out_free_dso_list;
2691
2692         if (setup_intlist(&symbol_conf.pid_list,
2693                        symbol_conf.pid_list_str, "pid") < 0)
2694                 goto out_free_comm_list;
2695
2696         if (setup_intlist(&symbol_conf.tid_list,
2697                        symbol_conf.tid_list_str, "tid") < 0)
2698                 goto out_free_pid_list;
2699
2700         if (setup_list(&symbol_conf.sym_list,
2701                        symbol_conf.sym_list_str, "symbol") < 0)
2702                 goto out_free_tid_list;
2703
2704         if (symbol_conf.sym_list &&
2705             setup_addrlist(&symbol_conf.addr_list, symbol_conf.sym_list) < 0)
2706                 goto out_free_sym_list;
2707
2708         if (setup_list(&symbol_conf.bt_stop_list,
2709                        symbol_conf.bt_stop_list_str, "symbol") < 0)
2710                 goto out_free_sym_list;
2711
2712         /*
2713          * A path to symbols of "/" is identical to ""
2714          * reset here for simplicity.
2715          */
2716         symfs = realpath(symbol_conf.symfs, NULL);
2717         if (symfs == NULL)
2718                 symfs = symbol_conf.symfs;
2719         if (strcmp(symfs, "/") == 0)
2720                 symbol_conf.symfs = "";
2721         if (symfs != symbol_conf.symfs)
2722                 free((void *)symfs);
2723
2724         symbol_conf.kptr_restrict = symbol__read_kptr_restrict();
2725
2726         symbol_conf.initialized = true;
2727         return 0;
2728
2729 out_free_sym_list:
2730         strlist__delete(symbol_conf.sym_list);
2731         intlist__delete(symbol_conf.addr_list);
2732 out_free_tid_list:
2733         intlist__delete(symbol_conf.tid_list);
2734 out_free_pid_list:
2735         intlist__delete(symbol_conf.pid_list);
2736 out_free_comm_list:
2737         strlist__delete(symbol_conf.comm_list);
2738 out_free_dso_list:
2739         strlist__delete(symbol_conf.dso_list);
2740         return -1;
2741 }
2742
2743 void symbol__exit(void)
2744 {
2745         if (!symbol_conf.initialized)
2746                 return;
2747         strlist__delete(symbol_conf.bt_stop_list);
2748         strlist__delete(symbol_conf.sym_list);
2749         strlist__delete(symbol_conf.dso_list);
2750         strlist__delete(symbol_conf.comm_list);
2751         intlist__delete(symbol_conf.tid_list);
2752         intlist__delete(symbol_conf.pid_list);
2753         intlist__delete(symbol_conf.addr_list);
2754         vmlinux_path__exit();
2755         symbol_conf.sym_list = symbol_conf.dso_list = symbol_conf.comm_list = NULL;
2756         symbol_conf.bt_stop_list = NULL;
2757         symbol_conf.initialized = false;
2758 }
2759
2760 int symbol__config_symfs(const struct option *opt __maybe_unused,
2761                          const char *dir, int unset __maybe_unused)
2762 {
2763         char *bf = NULL;
2764         int ret;
2765
2766         symbol_conf.symfs = strdup(dir);
2767         if (symbol_conf.symfs == NULL)
2768                 return -ENOMEM;
2769
2770         /* skip the locally configured cache if a symfs is given, and
2771          * config buildid dir to symfs/.debug
2772          */
2773         ret = asprintf(&bf, "%s/%s", dir, ".debug");
2774         if (ret < 0)
2775                 return -ENOMEM;
2776
2777         set_buildid_dir(bf);
2778
2779         free(bf);
2780         return 0;
2781 }
2782
2783 struct mem_info *mem_info__get(struct mem_info *mi)
2784 {
2785         if (mi)
2786                 refcount_inc(&mi->refcnt);
2787         return mi;
2788 }
2789
2790 void mem_info__put(struct mem_info *mi)
2791 {
2792         if (mi && refcount_dec_and_test(&mi->refcnt))
2793                 free(mi);
2794 }
2795
2796 struct mem_info *mem_info__new(void)
2797 {
2798         struct mem_info *mi = zalloc(sizeof(*mi));
2799
2800         if (mi)
2801                 refcount_set(&mi->refcnt, 1);
2802         return mi;
2803 }
2804
2805 /*
2806  * Checks that user supplied symbol kernel files are accessible because
2807  * the default mechanism for accessing elf files fails silently. i.e. if
2808  * debug syms for a build ID aren't found perf carries on normally. When
2809  * they are user supplied we should assume that the user doesn't want to
2810  * silently fail.
2811  */
2812 int symbol__validate_sym_arguments(void)
2813 {
2814         if (symbol_conf.vmlinux_name &&
2815             access(symbol_conf.vmlinux_name, R_OK)) {
2816                 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
2817                 return -EINVAL;
2818         }
2819         if (symbol_conf.kallsyms_name &&
2820             access(symbol_conf.kallsyms_name, R_OK)) {
2821                 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
2822                 return -EINVAL;
2823         }
2824         return 0;
2825 }