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 symbol_conf symbol_conf = {
  52         .nanosecs               = false,
  53         .use_modules            = true,
  54         .try_vmlinux_path       = true,
  55         .demangle               = true,
  56         .demangle_kernel        = false,
  57         .cumulate_callchain     = true,
  58         .time_quantum           = 100 * NSEC_PER_MSEC, /* 100ms */
  59         .show_hist_headers      = true,
  60         .symfs                  = "",
  61         .event_group            = true,
  62         .inline_name            = true,
  63         .res_sample             = 0,
  64 };
  65
  66 static enum dso_binary_type binary_type_symtab[] = {
  67         DSO_BINARY_TYPE__KALLSYMS,
  68         DSO_BINARY_TYPE__GUEST_KALLSYMS,
  69         DSO_BINARY_TYPE__JAVA_JIT,
  70         DSO_BINARY_TYPE__DEBUGLINK,
  71         DSO_BINARY_TYPE__BUILD_ID_CACHE,
  72         DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO,
  73         DSO_BINARY_TYPE__FEDORA_DEBUGINFO,
  74         DSO_BINARY_TYPE__UBUNTU_DEBUGINFO,
  75         DSO_BINARY_TYPE__BUILDID_DEBUGINFO,
  76         DSO_BINARY_TYPE__SYSTEM_PATH_DSO,
  77         DSO_BINARY_TYPE__GUEST_KMODULE,
  78         DSO_BINARY_TYPE__GUEST_KMODULE_COMP,
  79         DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE,
  80         DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP,
  81         DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO,
  82         DSO_BINARY_TYPE__MIXEDUP_UBUNTU_DEBUGINFO,
  83         DSO_BINARY_TYPE__NOT_FOUND,
  84 };
  85
  86 #define DSO_BINARY_TYPE__SYMTAB_CNT ARRAY_SIZE(binary_type_symtab)
  87
  88 static bool symbol_type__filter(char symbol_type)
  89 {
  90         symbol_type = toupper(symbol_type);
  91         return symbol_type == 'T' || symbol_type == 'W' || symbol_type == 'D' || symbol_type == 'B';
  92 }
  93
  94 static int prefix_underscores_count(const char *str)
  95 {
  96         const char *tail = str;
  97
  98         while (*tail == '_')
  99                 tail++;
 100
 101         return tail - str;
 102 }
 103
 104 void __weak arch__symbols__fixup_end(struct symbol *p, struct symbol *c)
 105 {
 106         p->end = c->start;
 107 }
 108
 109 const char * __weak arch__normalize_symbol_name(const char *name)
 110 {
 111         return name;
 112 }
 113
 114 int __weak arch__compare_symbol_names(const char *namea, const char *nameb)
 115 {
 116         return strcmp(namea, nameb);
 117 }
 118
 119 int __weak arch__compare_symbol_names_n(const char *namea, const char *nameb,
 120                                         unsigned int n)
 121 {
 122         return strncmp(namea, nameb, n);
 123 }
 124
 125 int __weak arch__choose_best_symbol(struct symbol *syma,
 126                                     struct symbol *symb __maybe_unused)
 127 {
 128         /* Avoid "SyS" kernel syscall aliases */
 129         if (strlen(syma->name) >= 3 && !strncmp(syma->name, "SyS", 3))
 130                 return SYMBOL_B;
 131         if (strlen(syma->name) >= 10 && !strncmp(syma->name, "compat_SyS", 10))
 132                 return SYMBOL_B;
 133
 134         return SYMBOL_A;
 135 }
 136
 137 static int choose_best_symbol(struct symbol *syma, struct symbol *symb)
 138 {
 139         s64 a;
 140         s64 b;
 141         size_t na, nb;
 142
 143         /* Prefer a symbol with non zero length */
 144         a = syma->end - syma->start;
 145         b = symb->end - symb->start;
 146         if ((b == 0) && (a > 0))
 147                 return SYMBOL_A;
 148         else if ((a == 0) && (b > 0))
 149                 return SYMBOL_B;
 150
 151         /* Prefer a non weak symbol over a weak one */
 152         a = syma->binding == STB_WEAK;
 153         b = symb->binding == STB_WEAK;
 154         if (b && !a)
 155                 return SYMBOL_A;
 156         if (a && !b)
 157                 return SYMBOL_B;
 158
 159         /* Prefer a global symbol over a non global one */
 160         a = syma->binding == STB_GLOBAL;
 161         b = symb->binding == STB_GLOBAL;
 162         if (a && !b)
 163                 return SYMBOL_A;
 164         if (b && !a)
 165                 return SYMBOL_B;
 166
 167         /* Prefer a symbol with less underscores */
 168         a = prefix_underscores_count(syma->name);
 169         b = prefix_underscores_count(symb->name);
 170         if (b > a)
 171                 return SYMBOL_A;
 172         else if (a > b)
 173                 return SYMBOL_B;
 174
 175         /* Choose the symbol with the longest name */
 176         na = strlen(syma->name);
 177         nb = strlen(symb->name);
 178         if (na > nb)
 179                 return SYMBOL_A;
 180         else if (na < nb)
 181                 return SYMBOL_B;
 182
 183         return arch__choose_best_symbol(syma, symb);
 184 }
 185
 186 void symbols__fixup_duplicate(struct rb_root_cached *symbols)
 187 {
 188         struct rb_node *nd;
 189         struct symbol *curr, *next;
 190
 191         if (symbol_conf.allow_aliases)
 192                 return;
 193
 194         nd = rb_first_cached(symbols);
 195
 196         while (nd) {
 197                 curr = rb_entry(nd, struct symbol, rb_node);
 198 again:
 199                 nd = rb_next(&curr->rb_node);
 200                 next = rb_entry(nd, struct symbol, rb_node);
 201
 202                 if (!nd)
 203                         break;
 204
 205                 if (curr->start != next->start)
 206                         continue;
 207
 208                 if (choose_best_symbol(curr, next) == SYMBOL_A) {
 209                         rb_erase_cached(&next->rb_node, symbols);
 210                         symbol__delete(next);
 211                         goto again;
 212                 } else {
 213                         nd = rb_next(&curr->rb_node);
 214                         rb_erase_cached(&curr->rb_node, symbols);
 215                         symbol__delete(curr);
 216                 }
 217         }
 218 }
 219
 220 void symbols__fixup_end(struct rb_root_cached *symbols)
 221 {
 222         struct rb_node *nd, *prevnd = rb_first_cached(symbols);
 223         struct symbol *curr, *prev;
 224
 225         if (prevnd == NULL)
 226                 return;
 227
 228         curr = rb_entry(prevnd, struct symbol, rb_node);
 229
 230         for (nd = rb_next(prevnd); nd; nd = rb_next(nd)) {
 231                 prev = curr;
 232                 curr = rb_entry(nd, struct symbol, rb_node);
 233
 234                 if (prev->end == prev->start || prev->end != curr->start)
 235                         arch__symbols__fixup_end(prev, curr);
 236         }
 237
 238         /* Last entry */
 239         if (curr->end == curr->start)
 240                 curr->end = roundup(curr->start, 4096) + 4096;
 241 }
 242
 243 void maps__fixup_end(struct maps *maps)
 244 {
 245         struct map *prev = NULL, *curr;
 246
 247         down_write(&maps->lock);
 248
 249         maps__for_each_entry(maps, curr) {
 250                 if (prev != NULL && !prev->end)
 251                         prev->end = curr->start;
 252
 253                 prev = curr;
 254         }
 255
 256         /*
 257          * We still haven't the actual symbols, so guess the
 258          * last map final address.
 259          */
 260         if (curr && !curr->end)
 261                 curr->end = ~0ULL;
 262
 263         up_write(&maps->lock);
 264 }
 265
 266 struct symbol *symbol__new(u64 start, u64 len, u8 binding, u8 type, const char *name)
 267 {
 268         size_t namelen = strlen(name) + 1;
 269         struct symbol *sym = calloc(1, (symbol_conf.priv_size +
 270                                         sizeof(*sym) + namelen));
 271         if (sym == NULL)
 272                 return NULL;
 273
 274         if (symbol_conf.priv_size) {
 275                 if (symbol_conf.init_annotation) {
 276                         struct annotation *notes = (void *)sym;
 277                         annotation__init(notes);
 278                 }
 279                 sym = ((void *)sym) + symbol_conf.priv_size;
 280         }
 281
 282         sym->start   = start;
 283         sym->end     = len ? start + len : start;
 284         sym->type    = type;
 285         sym->binding = binding;
 286         sym->namelen = namelen - 1;
 287
 288         pr_debug4("%s: %s %#" PRIx64 "-%#" PRIx64 "\n",
 289                   __func__, name, start, sym->end);
 290         memcpy(sym->name, name, namelen);
 291
 292         return sym;
 293 }
 294
 295 void symbol__delete(struct symbol *sym)
 296 {
 297         if (symbol_conf.priv_size) {
 298                 if (symbol_conf.init_annotation) {
 299                         struct annotation *notes = symbol__annotation(sym);
 300
 301                         annotation__exit(notes);
 302                 }
 303         }
 304         free(((void *)sym) - symbol_conf.priv_size);
 305 }
 306
 307 void symbols__delete(struct rb_root_cached *symbols)
 308 {
 309         struct symbol *pos;
 310         struct rb_node *next = rb_first_cached(symbols);
 311
 312         while (next) {
 313                 pos = rb_entry(next, struct symbol, rb_node);
 314                 next = rb_next(&pos->rb_node);
 315                 rb_erase_cached(&pos->rb_node, symbols);
 316                 symbol__delete(pos);
 317         }
 318 }
 319
 320 void __symbols__insert(struct rb_root_cached *symbols,
 321                        struct symbol *sym, bool kernel)
 322 {
 323         struct rb_node **p = &symbols->rb_root.rb_node;
 324         struct rb_node *parent = NULL;
 325         const u64 ip = sym->start;
 326         struct symbol *s;
 327         bool leftmost = true;
 328
 329         if (kernel) {
 330                 const char *name = sym->name;
 331                 /*
 332                  * ppc64 uses function descriptors and appends a '.' to the
 333                  * start of every instruction address. Remove it.
 334                  */
 335                 if (name[0] == '.')
 336                         name++;
 337                 sym->idle = symbol__is_idle(name);
 338         }
 339
 340         while (*p != NULL) {
 341                 parent = *p;
 342                 s = rb_entry(parent, struct symbol, rb_node);
 343                 if (ip < s->start)
 344                         p = &(*p)->rb_left;
 345                 else {
 346                         p = &(*p)->rb_right;
 347                         leftmost = false;
 348                 }
 349         }
 350         rb_link_node(&sym->rb_node, parent, p);
 351         rb_insert_color_cached(&sym->rb_node, symbols, leftmost);
 352 }
 353
 354 void symbols__insert(struct rb_root_cached *symbols, struct symbol *sym)
 355 {
 356         __symbols__insert(symbols, sym, false);
 357 }
 358
 359 static struct symbol *symbols__find(struct rb_root_cached *symbols, u64 ip)
 360 {
 361         struct rb_node *n;
 362
 363         if (symbols == NULL)
 364                 return NULL;
 365
 366         n = symbols->rb_root.rb_node;
 367
 368         while (n) {
 369                 struct symbol *s = rb_entry(n, struct symbol, rb_node);
 370
 371                 if (ip < s->start)
 372                         n = n->rb_left;
 373                 else if (ip > s->end || (ip == s->end && ip != s->start))
 374                         n = n->rb_right;
 375                 else
 376                         return s;
 377         }
 378
 379         return NULL;
 380 }
 381
 382 static struct symbol *symbols__first(struct rb_root_cached *symbols)
 383 {
 384         struct rb_node *n = rb_first_cached(symbols);
 385
 386         if (n)
 387                 return rb_entry(n, struct symbol, rb_node);
 388
 389         return NULL;
 390 }
 391
 392 static struct symbol *symbols__last(struct rb_root_cached *symbols)
 393 {
 394         struct rb_node *n = rb_last(&symbols->rb_root);
 395
 396         if (n)
 397                 return rb_entry(n, struct symbol, rb_node);
 398
 399         return NULL;
 400 }
 401
 402 static struct symbol *symbols__next(struct symbol *sym)
 403 {
 404         struct rb_node *n = rb_next(&sym->rb_node);
 405
 406         if (n)
 407                 return rb_entry(n, struct symbol, rb_node);
 408
 409         return NULL;
 410 }
 411
 412 static void symbols__insert_by_name(struct rb_root_cached *symbols, struct symbol *sym)
 413 {
 414         struct rb_node **p = &symbols->rb_root.rb_node;
 415         struct rb_node *parent = NULL;
 416         struct symbol_name_rb_node *symn, *s;
 417         bool leftmost = true;
 418
 419         symn = container_of(sym, struct symbol_name_rb_node, sym);
 420
 421         while (*p != NULL) {
 422                 parent = *p;
 423                 s = rb_entry(parent, struct symbol_name_rb_node, rb_node);
 424                 if (strcmp(sym->name, s->sym.name) < 0)
 425                         p = &(*p)->rb_left;
 426                 else {
 427                         p = &(*p)->rb_right;
 428                         leftmost = false;
 429                 }
 430         }
 431         rb_link_node(&symn->rb_node, parent, p);
 432         rb_insert_color_cached(&symn->rb_node, symbols, leftmost);
 433 }
 434
 435 static void symbols__sort_by_name(struct rb_root_cached *symbols,
 436                                   struct rb_root_cached *source)
 437 {
 438         struct rb_node *nd;
 439
 440         for (nd = rb_first_cached(source); nd; nd = rb_next(nd)) {
 441                 struct symbol *pos = rb_entry(nd, struct symbol, rb_node);
 442                 symbols__insert_by_name(symbols, pos);
 443         }
 444 }
 445
 446 int symbol__match_symbol_name(const char *name, const char *str,
 447                               enum symbol_tag_include includes)
 448 {
 449         const char *versioning;
 450
 451         if (includes == SYMBOL_TAG_INCLUDE__DEFAULT_ONLY &&
 452             (versioning = strstr(name, "@@"))) {
 453                 int len = strlen(str);
 454
 455                 if (len < versioning - name)
 456                         len = versioning - name;
 457
 458                 return arch__compare_symbol_names_n(name, str, len);
 459         } else
 460                 return arch__compare_symbol_names(name, str);
 461 }
 462
 463 static struct symbol *symbols__find_by_name(struct rb_root_cached *symbols,
 464                                             const char *name,
 465                                             enum symbol_tag_include includes)
 466 {
 467         struct rb_node *n;
 468         struct symbol_name_rb_node *s = NULL;
 469
 470         if (symbols == NULL)
 471                 return NULL;
 472
 473         n = symbols->rb_root.rb_node;
 474
 475         while (n) {
 476                 int cmp;
 477
 478                 s = rb_entry(n, struct symbol_name_rb_node, rb_node);
 479                 cmp = symbol__match_symbol_name(s->sym.name, name, includes);
 480
 481                 if (cmp > 0)
 482                         n = n->rb_left;
 483                 else if (cmp < 0)
 484                         n = n->rb_right;
 485                 else
 486                         break;
 487         }
 488
 489         if (n == NULL)
 490                 return NULL;
 491
 492         if (includes != SYMBOL_TAG_INCLUDE__DEFAULT_ONLY)
 493                 /* return first symbol that has same name (if any) */
 494                 for (n = rb_prev(n); n; n = rb_prev(n)) {
 495                         struct symbol_name_rb_node *tmp;
 496
 497                         tmp = rb_entry(n, struct symbol_name_rb_node, rb_node);
 498                         if (arch__compare_symbol_names(tmp->sym.name, s->sym.name))
 499                                 break;
 500
 501                         s = tmp;
 502                 }
 503
 504         return &s->sym;
 505 }
 506
 507 void dso__reset_find_symbol_cache(struct dso *dso)
 508 {
 509         dso->last_find_result.addr   = 0;
 510         dso->last_find_result.symbol = NULL;
 511 }
 512
 513 void dso__insert_symbol(struct dso *dso, struct symbol *sym)
 514 {
 515         __symbols__insert(&dso->symbols, sym, dso->kernel);
 516
 517         /* update the symbol cache if necessary */
 518         if (dso->last_find_result.addr >= sym->start &&
 519             (dso->last_find_result.addr < sym->end ||
 520             sym->start == sym->end)) {
 521                 dso->last_find_result.symbol = sym;
 522         }
 523 }
 524
 525 void dso__delete_symbol(struct dso *dso, struct symbol *sym)
 526 {
 527         rb_erase_cached(&sym->rb_node, &dso->symbols);
 528         symbol__delete(sym);
 529         dso__reset_find_symbol_cache(dso);
 530 }
 531
 532 struct symbol *dso__find_symbol(struct dso *dso, u64 addr)
 533 {
 534         if (dso->last_find_result.addr != addr || dso->last_find_result.symbol == NULL) {
 535                 dso->last_find_result.addr   = addr;
 536                 dso->last_find_result.symbol = symbols__find(&dso->symbols, addr);
 537         }
 538
 539         return dso->last_find_result.symbol;
 540 }
 541
 542 struct symbol *dso__first_symbol(struct dso *dso)
 543 {
 544         return symbols__first(&dso->symbols);
 545 }
 546
 547 struct symbol *dso__last_symbol(struct dso *dso)
 548 {
 549         return symbols__last(&dso->symbols);
 550 }
 551
 552 struct symbol *dso__next_symbol(struct symbol *sym)
 553 {
 554         return symbols__next(sym);
 555 }
 556
 557 struct symbol *symbol__next_by_name(struct symbol *sym)
 558 {
 559         struct symbol_name_rb_node *s = container_of(sym, struct symbol_name_rb_node, sym);
 560         struct rb_node *n = rb_next(&s->rb_node);
 561
 562         return n ? &rb_entry(n, struct symbol_name_rb_node, rb_node)->sym : NULL;
 563 }
 564
 565  /*
 566   * Returns first symbol that matched with @name.
 567   */
 568 struct symbol *dso__find_symbol_by_name(struct dso *dso, const char *name)
 569 {
 570         struct symbol *s = symbols__find_by_name(&dso->symbol_names, name,
 571                                                  SYMBOL_TAG_INCLUDE__NONE);
 572         if (!s)
 573                 s = symbols__find_by_name(&dso->symbol_names, name,
 574                                           SYMBOL_TAG_INCLUDE__DEFAULT_ONLY);
 575         return s;
 576 }
 577
 578 void dso__sort_by_name(struct dso *dso)
 579 {
 580         dso__set_sorted_by_name(dso);
 581         return symbols__sort_by_name(&dso->symbol_names, &dso->symbols);
 582 }
 583
 584 /*
 585  * While we find nice hex chars, build a long_val.
 586  * Return number of chars processed.
 587  */
 588 static int hex2u64(const char *ptr, u64 *long_val)
 589 {
 590         char *p;
 591
 592         *long_val = strtoull(ptr, &p, 16);
 593
 594         return p - ptr;
 595 }
 596
 597
 598 int modules__parse(const char *filename, void *arg,
 599                    int (*process_module)(void *arg, const char *name,
 600                                          u64 start, u64 size))
 601 {
 602         char *line = NULL;
 603         size_t n;
 604         FILE *file;
 605         int err = 0;
 606
 607         file = fopen(filename, "r");
 608         if (file == NULL)
 609                 return -1;
 610
 611         while (1) {
 612                 char name[PATH_MAX];
 613                 u64 start, size;
 614                 char *sep, *endptr;
 615                 ssize_t line_len;
 616
 617                 line_len = getline(&line, &n, file);
 618                 if (line_len < 0) {
 619                         if (feof(file))
 620                                 break;
 621                         err = -1;
 622                         goto out;
 623                 }
 624
 625                 if (!line) {
 626                         err = -1;
 627                         goto out;
 628                 }
 629
 630                 line[--line_len] = '\0'; /* \n */
 631
 632                 sep = strrchr(line, 'x');
 633                 if (sep == NULL)
 634                         continue;
 635
 636                 hex2u64(sep + 1, &start);
 637
 638                 sep = strchr(line, ' ');
 639                 if (sep == NULL)
 640                         continue;
 641
 642                 *sep = '\0';
 643
 644                 scnprintf(name, sizeof(name), "[%s]", line);
 645
 646                 size = strtoul(sep + 1, &endptr, 0);
 647                 if (*endptr != ' ' && *endptr != '\t')
 648                         continue;
 649
 650                 err = process_module(arg, name, start, size);
 651                 if (err)
 652                         break;
 653         }
 654 out:
 655         free(line);
 656         fclose(file);
 657         return err;
 658 }
 659
 660 /*
 661  * These are symbols in the kernel image, so make sure that
 662  * sym is from a kernel DSO.
 663  */
 664 static bool symbol__is_idle(const char *name)
 665 {
 666         const char * const idle_symbols[] = {
 667                 "acpi_idle_do_entry",
 668                 "acpi_processor_ffh_cstate_enter",
 669                 "arch_cpu_idle",
 670                 "cpu_idle",
 671                 "cpu_startup_entry",
 672                 "idle_cpu",
 673                 "intel_idle",
 674                 "default_idle",
 675                 "native_safe_halt",
 676                 "enter_idle",
 677                 "exit_idle",
 678                 "mwait_idle",
 679                 "mwait_idle_with_hints",
 680                 "mwait_idle_with_hints.constprop.0",
 681                 "poll_idle",
 682                 "ppc64_runlatch_off",
 683                 "pseries_dedicated_idle_sleep",
 684                 "psw_idle",
 685                 "psw_idle_exit",
 686                 NULL
 687         };
 688         int i;
 689         static struct strlist *idle_symbols_list;
 690
 691         if (idle_symbols_list)
 692                 return strlist__has_entry(idle_symbols_list, name);
 693
 694         idle_symbols_list = strlist__new(NULL, NULL);
 695
 696         for (i = 0; idle_symbols[i]; i++)
 697                 strlist__add(idle_symbols_list, idle_symbols[i]);
 698
 699         return strlist__has_entry(idle_symbols_list, name);
 700 }
 701
 702 static int map__process_kallsym_symbol(void *arg, const char *name,
 703                                        char type, u64 start)
 704 {
 705         struct symbol *sym;
 706         struct dso *dso = arg;
 707         struct rb_root_cached *root = &dso->symbols;
 708
 709         if (!symbol_type__filter(type))
 710                 return 0;
 711
 712         /* Ignore local symbols for ARM modules */
 713         if (name[0] == '$')
 714                 return 0;
 715
 716         /*
 717          * module symbols are not sorted so we add all
 718          * symbols, setting length to 0, and rely on
 719          * symbols__fixup_end() to fix it up.
 720          */
 721         sym = symbol__new(start, 0, kallsyms2elf_binding(type), kallsyms2elf_type(type), name);
 722         if (sym == NULL)
 723                 return -ENOMEM;
 724         /*
 725          * We will pass the symbols to the filter later, in
 726          * map__split_kallsyms, when we have split the maps per module
 727          */
 728         __symbols__insert(root, sym, !strchr(name, '['));
 729
 730         return 0;
 731 }
 732
 733 /*
 734  * Loads the function entries in /proc/kallsyms into kernel_map->dso,
 735  * so that we can in the next step set the symbol ->end address and then
 736  * call kernel_maps__split_kallsyms.
 737  */
 738 static int dso__load_all_kallsyms(struct dso *dso, const char *filename)
 739 {
 740         return kallsyms__parse(filename, dso, map__process_kallsym_symbol);
 741 }
 742
 743 static int maps__split_kallsyms_for_kcore(struct maps *kmaps, struct dso *dso)
 744 {
 745         struct map *curr_map;
 746         struct symbol *pos;
 747         int count = 0;
 748         struct rb_root_cached old_root = dso->symbols;
 749         struct rb_root_cached *root = &dso->symbols;
 750         struct rb_node *next = rb_first_cached(root);
 751
 752         if (!kmaps)
 753                 return -1;
 754
 755         *root = RB_ROOT_CACHED;
 756
 757         while (next) {
 758                 char *module;
 759
 760                 pos = rb_entry(next, struct symbol, rb_node);
 761                 next = rb_next(&pos->rb_node);
 762
 763                 rb_erase_cached(&pos->rb_node, &old_root);
 764                 RB_CLEAR_NODE(&pos->rb_node);
 765                 module = strchr(pos->name, '\t');
 766                 if (module)
 767                         *module = '\0';
 768
 769                 curr_map = maps__find(kmaps, pos->start);
 770
 771                 if (!curr_map) {
 772                         symbol__delete(pos);
 773                         continue;
 774                 }
 775
 776                 pos->start -= curr_map->start - curr_map->pgoff;
 777                 if (pos->end > curr_map->end)
 778                         pos->end = curr_map->end;
 779                 if (pos->end)
 780                         pos->end -= curr_map->start - curr_map->pgoff;
 781                 symbols__insert(&curr_map->dso->symbols, pos);
 782                 ++count;
 783         }
 784
 785         /* Symbols have been adjusted */
 786         dso->adjust_symbols = 1;
 787
 788         return count;
 789 }
 790
 791 /*
 792  * Split the symbols into maps, making sure there are no overlaps, i.e. the
 793  * kernel range is broken in several maps, named [kernel].N, as we don't have
 794  * the original ELF section names vmlinux have.
 795  */
 796 static int maps__split_kallsyms(struct maps *kmaps, struct dso *dso, u64 delta,
 797                                 struct map *initial_map)
 798 {
 799         struct machine *machine;
 800         struct map *curr_map = initial_map;
 801         struct symbol *pos;
 802         int count = 0, moved = 0;
 803         struct rb_root_cached *root = &dso->symbols;
 804         struct rb_node *next = rb_first_cached(root);
 805         int kernel_range = 0;
 806         bool x86_64;
 807
 808         if (!kmaps)
 809                 return -1;
 810
 811         machine = kmaps->machine;
 812
 813         x86_64 = machine__is(machine, "x86_64");
 814
 815         while (next) {
 816                 char *module;
 817
 818                 pos = rb_entry(next, struct symbol, rb_node);
 819                 next = rb_next(&pos->rb_node);
 820
 821                 module = strchr(pos->name, '\t');
 822                 if (module) {
 823                         if (!symbol_conf.use_modules)
 824                                 goto discard_symbol;
 825
 826                         *module++ = '\0';
 827
 828                         if (strcmp(curr_map->dso->short_name, module)) {
 829                                 if (curr_map != initial_map &&
 830                                     dso->kernel == DSO_SPACE__KERNEL_GUEST &&
 831                                     machine__is_default_guest(machine)) {
 832                                         /*
 833                                          * We assume all symbols of a module are
 834                                          * continuous in * kallsyms, so curr_map
 835                                          * points to a module and all its
 836                                          * symbols are in its kmap. Mark it as
 837                                          * loaded.
 838                                          */
 839                                         dso__set_loaded(curr_map->dso);
 840                                 }
 841
 842                                 curr_map = maps__find_by_name(kmaps, module);
 843                                 if (curr_map == NULL) {
 844                                         pr_debug("%s/proc/{kallsyms,modules} "
 845                                                  "inconsistency while looking "
 846                                                  "for \"%s\" module!\n",
 847                                                  machine->root_dir, module);
 848                                         curr_map = initial_map;
 849                                         goto discard_symbol;
 850                                 }
 851
 852                                 if (curr_map->dso->loaded &&
 853                                     !machine__is_default_guest(machine))
 854                                         goto discard_symbol;
 855                         }
 856                         /*
 857                          * So that we look just like we get from .ko files,
 858                          * i.e. not prelinked, relative to initial_map->start.
 859                          */
 860                         pos->start = curr_map->map_ip(curr_map, pos->start);
 861                         pos->end   = curr_map->map_ip(curr_map, pos->end);
 862                 } else if (x86_64 && is_entry_trampoline(pos->name)) {
 863                         /*
 864                          * These symbols are not needed anymore since the
 865                          * trampoline maps refer to the text section and it's
 866                          * symbols instead. Avoid having to deal with
 867                          * relocations, and the assumption that the first symbol
 868                          * is the start of kernel text, by simply removing the
 869                          * symbols at this point.
 870                          */
 871                         goto discard_symbol;
 872                 } else if (curr_map != initial_map) {
 873                         char dso_name[PATH_MAX];
 874                         struct dso *ndso;
 875
 876                         if (delta) {
 877                                 /* Kernel was relocated at boot time */
 878                                 pos->start -= delta;
 879                                 pos->end -= delta;
 880                         }
 881
 882                         if (count == 0) {
 883                                 curr_map = initial_map;
 884                                 goto add_symbol;
 885                         }
 886
 887                         if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
 888                                 snprintf(dso_name, sizeof(dso_name),
 889                                         "[guest.kernel].%d",
 890                                         kernel_range++);
 891                         else
 892                                 snprintf(dso_name, sizeof(dso_name),
 893                                         "[kernel].%d",
 894                                         kernel_range++);
 895
 896                         ndso = dso__new(dso_name);
 897                         if (ndso == NULL)
 898                                 return -1;
 899
 900                         ndso->kernel = dso->kernel;
 901
 902                         curr_map = map__new2(pos->start, ndso);
 903                         if (curr_map == NULL) {
 904                                 dso__put(ndso);
 905                                 return -1;
 906                         }
 907
 908                         curr_map->map_ip = curr_map->unmap_ip = identity__map_ip;
 909                         maps__insert(kmaps, curr_map);
 910                         ++kernel_range;
 911                 } else if (delta) {
 912                         /* Kernel was relocated at boot time */
 913                         pos->start -= delta;
 914                         pos->end -= delta;
 915                 }
 916 add_symbol:
 917                 if (curr_map != initial_map) {
 918                         rb_erase_cached(&pos->rb_node, root);
 919                         symbols__insert(&curr_map->dso->symbols, pos);
 920                         ++moved;
 921                 } else
 922                         ++count;
 923
 924                 continue;
 925 discard_symbol:
 926                 rb_erase_cached(&pos->rb_node, root);
 927                 symbol__delete(pos);
 928         }
 929
 930         if (curr_map != initial_map &&
 931             dso->kernel == DSO_SPACE__KERNEL_GUEST &&
 932             machine__is_default_guest(kmaps->machine)) {
 933                 dso__set_loaded(curr_map->dso);
 934         }
 935
 936         return count + moved;
 937 }
 938
 939 bool symbol__restricted_filename(const char *filename,
 940                                  const char *restricted_filename)
 941 {
 942         bool restricted = false;
 943
 944         if (symbol_conf.kptr_restrict) {
 945                 char *r = realpath(filename, NULL);
 946
 947                 if (r != NULL) {
 948                         restricted = strcmp(r, restricted_filename) == 0;
 949                         free(r);
 950                         return restricted;
 951                 }
 952         }
 953
 954         return restricted;
 955 }
 956
 957 struct module_info {
 958         struct rb_node rb_node;
 959         char *name;
 960         u64 start;
 961 };
 962
 963 static void add_module(struct module_info *mi, struct rb_root *modules)
 964 {
 965         struct rb_node **p = &modules->rb_node;
 966         struct rb_node *parent = NULL;
 967         struct module_info *m;
 968
 969         while (*p != NULL) {
 970                 parent = *p;
 971                 m = rb_entry(parent, struct module_info, rb_node);
 972                 if (strcmp(mi->name, m->name) < 0)
 973                         p = &(*p)->rb_left;
 974                 else
 975                         p = &(*p)->rb_right;
 976         }
 977         rb_link_node(&mi->rb_node, parent, p);
 978         rb_insert_color(&mi->rb_node, modules);
 979 }
 980
 981 static void delete_modules(struct rb_root *modules)
 982 {
 983         struct module_info *mi;
 984         struct rb_node *next = rb_first(modules);
 985
 986         while (next) {
 987                 mi = rb_entry(next, struct module_info, rb_node);
 988                 next = rb_next(&mi->rb_node);
 989                 rb_erase(&mi->rb_node, modules);
 990                 zfree(&mi->name);
 991                 free(mi);
 992         }
 993 }
 994
 995 static struct module_info *find_module(const char *name,
 996                                        struct rb_root *modules)
 997 {
 998         struct rb_node *n = modules->rb_node;
 999
1000         while (n) {
1001                 struct module_info *m;
1002                 int cmp;
1003
1004                 m = rb_entry(n, struct module_info, rb_node);
1005                 cmp = strcmp(name, m->name);
1006                 if (cmp < 0)
1007                         n = n->rb_left;
1008                 else if (cmp > 0)
1009                         n = n->rb_right;
1010                 else
1011                         return m;
1012         }
1013
1014         return NULL;
1015 }
1016
1017 static int __read_proc_modules(void *arg, const char *name, u64 start,
1018                                u64 size __maybe_unused)
1019 {
1020         struct rb_root *modules = arg;
1021         struct module_info *mi;
1022
1023         mi = zalloc(sizeof(struct module_info));
1024         if (!mi)
1025                 return -ENOMEM;
1026
1027         mi->name = strdup(name);
1028         mi->start = start;
1029
1030         if (!mi->name) {
1031                 free(mi);
1032                 return -ENOMEM;
1033         }
1034
1035         add_module(mi, modules);
1036
1037         return 0;
1038 }
1039
1040 static int read_proc_modules(const char *filename, struct rb_root *modules)
1041 {
1042         if (symbol__restricted_filename(filename, "/proc/modules"))
1043                 return -1;
1044
1045         if (modules__parse(filename, modules, __read_proc_modules)) {
1046                 delete_modules(modules);
1047                 return -1;
1048         }
1049
1050         return 0;
1051 }
1052
1053 int compare_proc_modules(const char *from, const char *to)
1054 {
1055         struct rb_root from_modules = RB_ROOT;
1056         struct rb_root to_modules = RB_ROOT;
1057         struct rb_node *from_node, *to_node;
1058         struct module_info *from_m, *to_m;
1059         int ret = -1;
1060
1061         if (read_proc_modules(from, &from_modules))
1062                 return -1;
1063
1064         if (read_proc_modules(to, &to_modules))
1065                 goto out_delete_from;
1066
1067         from_node = rb_first(&from_modules);
1068         to_node = rb_first(&to_modules);
1069         while (from_node) {
1070                 if (!to_node)
1071                         break;
1072
1073                 from_m = rb_entry(from_node, struct module_info, rb_node);
1074                 to_m = rb_entry(to_node, struct module_info, rb_node);
1075
1076                 if (from_m->start != to_m->start ||
1077                     strcmp(from_m->name, to_m->name))
1078                         break;
1079
1080                 from_node = rb_next(from_node);
1081                 to_node = rb_next(to_node);
1082         }
1083
1084         if (!from_node && !to_node)
1085                 ret = 0;
1086
1087         delete_modules(&to_modules);
1088 out_delete_from:
1089         delete_modules(&from_modules);
1090
1091         return ret;
1092 }
1093
1094 static int do_validate_kcore_modules(const char *filename, struct maps *kmaps)
1095 {
1096         struct rb_root modules = RB_ROOT;
1097         struct map *old_map;
1098         int err;
1099
1100         err = read_proc_modules(filename, &modules);
1101         if (err)
1102                 return err;
1103
1104         maps__for_each_entry(kmaps, old_map) {
1105                 struct module_info *mi;
1106
1107                 if (!__map__is_kmodule(old_map)) {
1108                         continue;
1109                 }
1110
1111                 /* Module must be in memory at the same address */
1112                 mi = find_module(old_map->dso->short_name, &modules);
1113                 if (!mi || mi->start != old_map->start) {
1114                         err = -EINVAL;
1115                         goto out;
1116                 }
1117         }
1118 out:
1119         delete_modules(&modules);
1120         return err;
1121 }
1122
1123 /*
1124  * If kallsyms is referenced by name then we look for filename in the same
1125  * directory.
1126  */
1127 static bool filename_from_kallsyms_filename(char *filename,
1128                                             const char *base_name,
1129                                             const char *kallsyms_filename)
1130 {
1131         char *name;
1132
1133         strcpy(filename, kallsyms_filename);
1134         name = strrchr(filename, '/');
1135         if (!name)
1136                 return false;
1137
1138         name += 1;
1139
1140         if (!strcmp(name, "kallsyms")) {
1141                 strcpy(name, base_name);
1142                 return true;
1143         }
1144
1145         return false;
1146 }
1147
1148 static int validate_kcore_modules(const char *kallsyms_filename,
1149                                   struct map *map)
1150 {
1151         struct maps *kmaps = map__kmaps(map);
1152         char modules_filename[PATH_MAX];
1153
1154         if (!kmaps)
1155                 return -EINVAL;
1156
1157         if (!filename_from_kallsyms_filename(modules_filename, "modules",
1158                                              kallsyms_filename))
1159                 return -EINVAL;
1160
1161         if (do_validate_kcore_modules(modules_filename, kmaps))
1162                 return -EINVAL;
1163
1164         return 0;
1165 }
1166
1167 static int validate_kcore_addresses(const char *kallsyms_filename,
1168                                     struct map *map)
1169 {
1170         struct kmap *kmap = map__kmap(map);
1171
1172         if (!kmap)
1173                 return -EINVAL;
1174
1175         if (kmap->ref_reloc_sym && kmap->ref_reloc_sym->name) {
1176                 u64 start;
1177
1178                 if (kallsyms__get_function_start(kallsyms_filename,
1179                                                  kmap->ref_reloc_sym->name, &start))
1180                         return -ENOENT;
1181                 if (start != kmap->ref_reloc_sym->addr)
1182                         return -EINVAL;
1183         }
1184
1185         return validate_kcore_modules(kallsyms_filename, map);
1186 }
1187
1188 struct kcore_mapfn_data {
1189         struct dso *dso;
1190         struct list_head maps;
1191 };
1192
1193 static int kcore_mapfn(u64 start, u64 len, u64 pgoff, void *data)
1194 {
1195         struct kcore_mapfn_data *md = data;
1196         struct map *map;
1197
1198         map = map__new2(start, md->dso);
1199         if (map == NULL)
1200                 return -ENOMEM;
1201
1202         map->end = map->start + len;
1203         map->pgoff = pgoff;
1204
1205         list_add(&map->node, &md->maps);
1206
1207         return 0;
1208 }
1209
1210 /*
1211  * Merges map into maps by splitting the new map within the existing map
1212  * regions.
1213  */
1214 int maps__merge_in(struct maps *kmaps, struct map *new_map)
1215 {
1216         struct map *old_map;
1217         LIST_HEAD(merged);
1218
1219         maps__for_each_entry(kmaps, old_map) {
1220                 /* no overload with this one */
1221                 if (new_map->end < old_map->start ||
1222                     new_map->start >= old_map->end)
1223                         continue;
1224
1225                 if (new_map->start < old_map->start) {
1226                         /*
1227                          * |new......
1228                          *       |old....
1229                          */
1230                         if (new_map->end < old_map->end) {
1231                                 /*
1232                                  * |new......|     -> |new..|
1233                                  *       |old....| ->       |old....|
1234                                  */
1235                                 new_map->end = old_map->start;
1236                         } else {
1237                                 /*
1238                                  * |new.............| -> |new..|       |new..|
1239                                  *       |old....|    ->       |old....|
1240                                  */
1241                                 struct map *m = map__clone(new_map);
1242
1243                                 if (!m)
1244                                         return -ENOMEM;
1245
1246                                 m->end = old_map->start;
1247                                 list_add_tail(&m->node, &merged);
1248                                 new_map->pgoff += old_map->end - new_map->start;
1249                                 new_map->start = old_map->end;
1250                         }
1251                 } else {
1252                         /*
1253                          *      |new......
1254                          * |old....
1255                          */
1256                         if (new_map->end < old_map->end) {
1257                                 /*
1258                                  *      |new..|   -> x
1259                                  * |old.........| -> |old.........|
1260                                  */
1261                                 map__put(new_map);
1262                                 new_map = NULL;
1263                                 break;
1264                         } else {
1265                                 /*
1266                                  *      |new......| ->         |new...|
1267                                  * |old....|        -> |old....|
1268                                  */
1269                                 new_map->pgoff += old_map->end - new_map->start;
1270                                 new_map->start = old_map->end;
1271                         }
1272                 }
1273         }
1274
1275         while (!list_empty(&merged)) {
1276                 old_map = list_entry(merged.next, struct map, node);
1277                 list_del_init(&old_map->node);
1278                 maps__insert(kmaps, old_map);
1279                 map__put(old_map);
1280         }
1281
1282         if (new_map) {
1283                 maps__insert(kmaps, new_map);
1284                 map__put(new_map);
1285         }
1286         return 0;
1287 }
1288
1289 static int dso__load_kcore(struct dso *dso, struct map *map,
1290                            const char *kallsyms_filename)
1291 {
1292         struct maps *kmaps = map__kmaps(map);
1293         struct kcore_mapfn_data md;
1294         struct map *old_map, *new_map, *replacement_map = NULL, *next;
1295         struct machine *machine;
1296         bool is_64_bit;
1297         int err, fd;
1298         char kcore_filename[PATH_MAX];
1299         u64 stext;
1300
1301         if (!kmaps)
1302                 return -EINVAL;
1303
1304         machine = kmaps->machine;
1305
1306         /* This function requires that the map is the kernel map */
1307         if (!__map__is_kernel(map))
1308                 return -EINVAL;
1309
1310         if (!filename_from_kallsyms_filename(kcore_filename, "kcore",
1311                                              kallsyms_filename))
1312                 return -EINVAL;
1313
1314         /* Modules and kernel must be present at their original addresses */
1315         if (validate_kcore_addresses(kallsyms_filename, map))
1316                 return -EINVAL;
1317
1318         md.dso = dso;
1319         INIT_LIST_HEAD(&md.maps);
1320
1321         fd = open(kcore_filename, O_RDONLY);
1322         if (fd < 0) {
1323                 pr_debug("Failed to open %s. Note /proc/kcore requires CAP_SYS_RAWIO capability to access.\n",
1324                          kcore_filename);
1325                 return -EINVAL;
1326         }
1327
1328         /* Read new maps into temporary lists */
1329         err = file__read_maps(fd, map->prot & PROT_EXEC, kcore_mapfn, &md,
1330                               &is_64_bit);
1331         if (err)
1332                 goto out_err;
1333         dso->is_64_bit = is_64_bit;
1334
1335         if (list_empty(&md.maps)) {
1336                 err = -EINVAL;
1337                 goto out_err;
1338         }
1339
1340         /* Remove old maps */
1341         maps__for_each_entry_safe(kmaps, old_map, next) {
1342                 /*
1343                  * We need to preserve eBPF maps even if they are
1344                  * covered by kcore, because we need to access
1345                  * eBPF dso for source data.
1346                  */
1347                 if (old_map != map && !__map__is_bpf_prog(old_map))
1348                         maps__remove(kmaps, old_map);
1349         }
1350         machine->trampolines_mapped = false;
1351
1352         /* Find the kernel map using the '_stext' symbol */
1353         if (!kallsyms__get_function_start(kallsyms_filename, "_stext", &stext)) {
1354                 list_for_each_entry(new_map, &md.maps, node) {
1355                         if (stext >= new_map->start && stext < new_map->end) {
1356                                 replacement_map = new_map;
1357                                 break;
1358                         }
1359                 }
1360         }
1361
1362         if (!replacement_map)
1363                 replacement_map = list_entry(md.maps.next, struct map, node);
1364
1365         /* Add new maps */
1366         while (!list_empty(&md.maps)) {
1367                 new_map = list_entry(md.maps.next, struct map, node);
1368                 list_del_init(&new_map->node);
1369                 if (new_map == replacement_map) {
1370                         map->start      = new_map->start;
1371                         map->end        = new_map->end;
1372                         map->pgoff      = new_map->pgoff;
1373                         map->map_ip     = new_map->map_ip;
1374                         map->unmap_ip   = new_map->unmap_ip;
1375                         /* Ensure maps are correctly ordered */
1376                         map__get(map);
1377                         maps__remove(kmaps, map);
1378                         maps__insert(kmaps, map);
1379                         map__put(map);
1380                         map__put(new_map);
1381                 } else {
1382                         /*
1383                          * Merge kcore map into existing maps,
1384                          * and ensure that current maps (eBPF)
1385                          * stay intact.
1386                          */
1387                         if (maps__merge_in(kmaps, new_map))
1388                                 goto out_err;
1389                 }
1390         }
1391
1392         if (machine__is(machine, "x86_64")) {
1393                 u64 addr;
1394
1395                 /*
1396                  * If one of the corresponding symbols is there, assume the
1397                  * entry trampoline maps are too.
1398                  */
1399                 if (!kallsyms__get_function_start(kallsyms_filename,
1400                                                   ENTRY_TRAMPOLINE_NAME,
1401                                                   &addr))
1402                         machine->trampolines_mapped = true;
1403         }
1404
1405         /*
1406          * Set the data type and long name so that kcore can be read via
1407          * dso__data_read_addr().
1408          */
1409         if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
1410                 dso->binary_type = DSO_BINARY_TYPE__GUEST_KCORE;
1411         else
1412                 dso->binary_type = DSO_BINARY_TYPE__KCORE;
1413         dso__set_long_name(dso, strdup(kcore_filename), true);
1414
1415         close(fd);
1416
1417         if (map->prot & PROT_EXEC)
1418                 pr_debug("Using %s for kernel object code\n", kcore_filename);
1419         else
1420                 pr_debug("Using %s for kernel data\n", kcore_filename);
1421
1422         return 0;
1423
1424 out_err:
1425         while (!list_empty(&md.maps)) {
1426                 map = list_entry(md.maps.next, struct map, node);
1427                 list_del_init(&map->node);
1428                 map__put(map);
1429         }
1430         close(fd);
1431         return -EINVAL;
1432 }
1433
1434 /*
1435  * If the kernel is relocated at boot time, kallsyms won't match.  Compute the
1436  * delta based on the relocation reference symbol.
1437  */
1438 static int kallsyms__delta(struct kmap *kmap, const char *filename, u64 *delta)
1439 {
1440         u64 addr;
1441
1442         if (!kmap->ref_reloc_sym || !kmap->ref_reloc_sym->name)
1443                 return 0;
1444
1445         if (kallsyms__get_function_start(filename, kmap->ref_reloc_sym->name, &addr))
1446                 return -1;
1447
1448         *delta = addr - kmap->ref_reloc_sym->addr;
1449         return 0;
1450 }
1451
1452 int __dso__load_kallsyms(struct dso *dso, const char *filename,
1453                          struct map *map, bool no_kcore)
1454 {
1455         struct kmap *kmap = map__kmap(map);
1456         u64 delta = 0;
1457
1458         if (symbol__restricted_filename(filename, "/proc/kallsyms"))
1459                 return -1;
1460
1461         if (!kmap || !kmap->kmaps)
1462                 return -1;
1463
1464         if (dso__load_all_kallsyms(dso, filename) < 0)
1465                 return -1;
1466
1467         if (kallsyms__delta(kmap, filename, &delta))
1468                 return -1;
1469
1470         symbols__fixup_end(&dso->symbols);
1471         symbols__fixup_duplicate(&dso->symbols);
1472
1473         if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
1474                 dso->symtab_type = DSO_BINARY_TYPE__GUEST_KALLSYMS;
1475         else
1476                 dso->symtab_type = DSO_BINARY_TYPE__KALLSYMS;
1477
1478         if (!no_kcore && !dso__load_kcore(dso, map, filename))
1479                 return maps__split_kallsyms_for_kcore(kmap->kmaps, dso);
1480         else
1481                 return maps__split_kallsyms(kmap->kmaps, dso, delta, map);
1482 }
1483
1484 int dso__load_kallsyms(struct dso *dso, const char *filename,
1485                        struct map *map)
1486 {
1487         return __dso__load_kallsyms(dso, filename, map, false);
1488 }
1489
1490 static int dso__load_perf_map(const char *map_path, struct dso *dso)
1491 {
1492         char *line = NULL;
1493         size_t n;
1494         FILE *file;
1495         int nr_syms = 0;
1496
1497         file = fopen(map_path, "r");
1498         if (file == NULL)
1499                 goto out_failure;
1500
1501         while (!feof(file)) {
1502                 u64 start, size;
1503                 struct symbol *sym;
1504                 int line_len, len;
1505
1506                 line_len = getline(&line, &n, file);
1507                 if (line_len < 0)
1508                         break;
1509
1510                 if (!line)
1511                         goto out_failure;
1512
1513                 line[--line_len] = '\0'; /* \n */
1514
1515                 len = hex2u64(line, &start);
1516
1517                 len++;
1518                 if (len + 2 >= line_len)
1519                         continue;
1520
1521                 len += hex2u64(line + len, &size);
1522
1523                 len++;
1524                 if (len + 2 >= line_len)
1525                         continue;
1526
1527                 sym = symbol__new(start, size, STB_GLOBAL, STT_FUNC, line + len);
1528
1529                 if (sym == NULL)
1530                         goto out_delete_line;
1531
1532                 symbols__insert(&dso->symbols, sym);
1533                 nr_syms++;
1534         }
1535
1536         free(line);
1537         fclose(file);
1538
1539         return nr_syms;
1540
1541 out_delete_line:
1542         free(line);
1543 out_failure:
1544         return -1;
1545 }
1546
1547 #ifdef HAVE_LIBBFD_SUPPORT
1548 #define PACKAGE 'perf'
1549 #include <bfd.h>
1550
1551 static int bfd_symbols__cmpvalue(const void *a, const void *b)
1552 {
1553         const asymbol *as = *(const asymbol **)a, *bs = *(const asymbol **)b;
1554
1555         if (bfd_asymbol_value(as) != bfd_asymbol_value(bs))
1556                 return bfd_asymbol_value(as) - bfd_asymbol_value(bs);
1557
1558         return bfd_asymbol_name(as)[0] - bfd_asymbol_name(bs)[0];
1559 }
1560
1561 static int bfd2elf_binding(asymbol *symbol)
1562 {
1563         if (symbol->flags & BSF_WEAK)
1564                 return STB_WEAK;
1565         if (symbol->flags & BSF_GLOBAL)
1566                 return STB_GLOBAL;
1567         if (symbol->flags & BSF_LOCAL)
1568                 return STB_LOCAL;
1569         return -1;
1570 }
1571
1572 int dso__load_bfd_symbols(struct dso *dso, const char *debugfile)
1573 {
1574         int err = -1;
1575         long symbols_size, symbols_count, i;
1576         asection *section;
1577         asymbol **symbols, *sym;
1578         struct symbol *symbol;
1579         bfd *abfd;
1580         u64 start, len;
1581
1582         abfd = bfd_openr(debugfile, NULL);
1583         if (!abfd)
1584                 return -1;
1585
1586         if (!bfd_check_format(abfd, bfd_object)) {
1587                 pr_debug2("%s: cannot read %s bfd file.\n", __func__,
1588                           dso->long_name);
1589                 goto out_close;
1590         }
1591
1592         if (bfd_get_flavour(abfd) == bfd_target_elf_flavour)
1593                 goto out_close;
1594
1595         symbols_size = bfd_get_symtab_upper_bound(abfd);
1596         if (symbols_size == 0) {
1597                 bfd_close(abfd);
1598                 return 0;
1599         }
1600
1601         if (symbols_size < 0)
1602                 goto out_close;
1603
1604         symbols = malloc(symbols_size);
1605         if (!symbols)
1606                 goto out_close;
1607
1608         symbols_count = bfd_canonicalize_symtab(abfd, symbols);
1609         if (symbols_count < 0)
1610                 goto out_free;
1611
1612         section = bfd_get_section_by_name(abfd, ".text");
1613         if (section) {
1614                 for (i = 0; i < symbols_count; ++i) {
1615                         if (!strcmp(bfd_asymbol_name(symbols[i]), "__ImageBase") ||
1616                             !strcmp(bfd_asymbol_name(symbols[i]), "__image_base__"))
1617                                 break;
1618                 }
1619                 if (i < symbols_count) {
1620                         /* PE symbols can only have 4 bytes, so use .text high bits */
1621                         dso->text_offset = section->vma - (u32)section->vma;
1622                         dso->text_offset += (u32)bfd_asymbol_value(symbols[i]);
1623                 } else {
1624                         dso->text_offset = section->vma - section->filepos;
1625                 }
1626         }
1627
1628         qsort(symbols, symbols_count, sizeof(asymbol *), bfd_symbols__cmpvalue);
1629
1630 #ifdef bfd_get_section
1631 #define bfd_asymbol_section bfd_get_section
1632 #endif
1633         for (i = 0; i < symbols_count; ++i) {
1634                 sym = symbols[i];
1635                 section = bfd_asymbol_section(sym);
1636                 if (bfd2elf_binding(sym) < 0)
1637                         continue;
1638
1639                 while (i + 1 < symbols_count &&
1640                        bfd_asymbol_section(symbols[i + 1]) == section &&
1641                        bfd2elf_binding(symbols[i + 1]) < 0)
1642                         i++;
1643
1644                 if (i + 1 < symbols_count &&
1645                     bfd_asymbol_section(symbols[i + 1]) == section)
1646                         len = symbols[i + 1]->value - sym->value;
1647                 else
1648                         len = section->size - sym->value;
1649
1650                 start = bfd_asymbol_value(sym) - dso->text_offset;
1651                 symbol = symbol__new(start, len, bfd2elf_binding(sym), STT_FUNC,
1652                                      bfd_asymbol_name(sym));
1653                 if (!symbol)
1654                         goto out_free;
1655
1656                 symbols__insert(&dso->symbols, symbol);
1657         }
1658 #ifdef bfd_get_section
1659 #undef bfd_asymbol_section
1660 #endif
1661
1662         symbols__fixup_end(&dso->symbols);
1663         symbols__fixup_duplicate(&dso->symbols);
1664         dso->adjust_symbols = 1;
1665
1666         err = 0;
1667 out_free:
1668         free(symbols);
1669 out_close:
1670         bfd_close(abfd);
1671         return err;
1672 }
1673 #endif
1674
1675 static bool dso__is_compatible_symtab_type(struct dso *dso, bool kmod,
1676                                            enum dso_binary_type type)
1677 {
1678         switch (type) {
1679         case DSO_BINARY_TYPE__JAVA_JIT:
1680         case DSO_BINARY_TYPE__DEBUGLINK:
1681         case DSO_BINARY_TYPE__SYSTEM_PATH_DSO:
1682         case DSO_BINARY_TYPE__FEDORA_DEBUGINFO:
1683         case DSO_BINARY_TYPE__UBUNTU_DEBUGINFO:
1684         case DSO_BINARY_TYPE__MIXEDUP_UBUNTU_DEBUGINFO:
1685         case DSO_BINARY_TYPE__BUILDID_DEBUGINFO:
1686         case DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO:
1687                 return !kmod && dso->kernel == DSO_SPACE__USER;
1688
1689         case DSO_BINARY_TYPE__KALLSYMS:
1690         case DSO_BINARY_TYPE__VMLINUX:
1691         case DSO_BINARY_TYPE__KCORE:
1692                 return dso->kernel == DSO_SPACE__KERNEL;
1693
1694         case DSO_BINARY_TYPE__GUEST_KALLSYMS:
1695         case DSO_BINARY_TYPE__GUEST_VMLINUX:
1696         case DSO_BINARY_TYPE__GUEST_KCORE:
1697                 return dso->kernel == DSO_SPACE__KERNEL_GUEST;
1698
1699         case DSO_BINARY_TYPE__GUEST_KMODULE:
1700         case DSO_BINARY_TYPE__GUEST_KMODULE_COMP:
1701         case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE:
1702         case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP:
1703                 /*
1704                  * kernel modules know their symtab type - it's set when
1705                  * creating a module dso in machine__addnew_module_map().
1706                  */
1707                 return kmod && dso->symtab_type == type;
1708
1709         case DSO_BINARY_TYPE__BUILD_ID_CACHE:
1710         case DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO:
1711                 return true;
1712
1713         case DSO_BINARY_TYPE__BPF_PROG_INFO:
1714         case DSO_BINARY_TYPE__BPF_IMAGE:
1715         case DSO_BINARY_TYPE__OOL:
1716         case DSO_BINARY_TYPE__NOT_FOUND:
1717         default:
1718                 return false;
1719         }
1720 }
1721
1722 /* Checks for the existence of the perf-<pid>.map file in two different
1723  * locations.  First, if the process is a separate mount namespace, check in
1724  * that namespace using the pid of the innermost pid namespace.  If's not in a
1725  * namespace, or the file can't be found there, try in the mount namespace of
1726  * the tracing process using our view of its pid.
1727  */
1728 static int dso__find_perf_map(char *filebuf, size_t bufsz,
1729                               struct nsinfo **nsip)
1730 {
1731         struct nscookie nsc;
1732         struct nsinfo *nsi;
1733         struct nsinfo *nnsi;
1734         int rc = -1;
1735
1736         nsi = *nsip;
1737
1738         if (nsinfo__need_setns(nsi)) {
1739                 snprintf(filebuf, bufsz, "/tmp/perf-%d.map", nsinfo__nstgid(nsi));
1740                 nsinfo__mountns_enter(nsi, &nsc);
1741                 rc = access(filebuf, R_OK);
1742                 nsinfo__mountns_exit(&nsc);
1743                 if (rc == 0)
1744                         return rc;
1745         }
1746
1747         nnsi = nsinfo__copy(nsi);
1748         if (nnsi) {
1749                 nsinfo__put(nsi);
1750
1751                 nsinfo__clear_need_setns(nnsi);
1752                 snprintf(filebuf, bufsz, "/tmp/perf-%d.map", nsinfo__tgid(nnsi));
1753                 *nsip = nnsi;
1754                 rc = 0;
1755         }
1756
1757         return rc;
1758 }
1759
1760 int dso__load(struct dso *dso, struct map *map)
1761 {
1762         char *name;
1763         int ret = -1;
1764         u_int i;
1765         struct machine *machine = NULL;
1766         char *root_dir = (char *) "";
1767         int ss_pos = 0;
1768         struct symsrc ss_[2];
1769         struct symsrc *syms_ss = NULL, *runtime_ss = NULL;
1770         bool kmod;
1771         bool perfmap;
1772         struct build_id bid;
1773         struct nscookie nsc;
1774         char newmapname[PATH_MAX];
1775         const char *map_path = dso->long_name;
1776
1777         perfmap = strncmp(dso->name, "/tmp/perf-", 10) == 0;
1778         if (perfmap) {
1779                 if (dso->nsinfo && (dso__find_perf_map(newmapname,
1780                     sizeof(newmapname), &dso->nsinfo) == 0)) {
1781                         map_path = newmapname;
1782                 }
1783         }
1784
1785         nsinfo__mountns_enter(dso->nsinfo, &nsc);
1786         pthread_mutex_lock(&dso->lock);
1787
1788         /* check again under the dso->lock */
1789         if (dso__loaded(dso)) {
1790                 ret = 1;
1791                 goto out;
1792         }
1793
1794         kmod = dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1795                 dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP ||
1796                 dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE ||
1797                 dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE_COMP;
1798
1799         if (dso->kernel && !kmod) {
1800                 if (dso->kernel == DSO_SPACE__KERNEL)
1801                         ret = dso__load_kernel_sym(dso, map);
1802                 else if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
1803                         ret = dso__load_guest_kernel_sym(dso, map);
1804
1805                 machine = map__kmaps(map)->machine;
1806                 if (machine__is(machine, "x86_64"))
1807                         machine__map_x86_64_entry_trampolines(machine, dso);
1808                 goto out;
1809         }
1810
1811         dso->adjust_symbols = 0;
1812
1813         if (perfmap) {
1814                 ret = dso__load_perf_map(map_path, dso);
1815                 dso->symtab_type = ret > 0 ? DSO_BINARY_TYPE__JAVA_JIT :
1816                                              DSO_BINARY_TYPE__NOT_FOUND;
1817                 goto out;
1818         }
1819
1820         if (machine)
1821                 root_dir = machine->root_dir;
1822
1823         name = malloc(PATH_MAX);
1824         if (!name)
1825                 goto out;
1826
1827         /*
1828          * Read the build id if possible. This is required for
1829          * DSO_BINARY_TYPE__BUILDID_DEBUGINFO to work
1830          */
1831         if (!dso->has_build_id &&
1832             is_regular_file(dso->long_name)) {
1833             __symbol__join_symfs(name, PATH_MAX, dso->long_name);
1834                 if (filename__read_build_id(name, &bid) > 0)
1835                         dso__set_build_id(dso, &bid);
1836         }
1837
1838         /*
1839          * Iterate over candidate debug images.
1840          * Keep track of "interesting" ones (those which have a symtab, dynsym,
1841          * and/or opd section) for processing.
1842          */
1843         for (i = 0; i < DSO_BINARY_TYPE__SYMTAB_CNT; i++) {
1844                 struct symsrc *ss = &ss_[ss_pos];
1845                 bool next_slot = false;
1846                 bool is_reg;
1847                 bool nsexit;
1848                 int bfdrc = -1;
1849                 int sirc = -1;
1850
1851                 enum dso_binary_type symtab_type = binary_type_symtab[i];
1852
1853                 nsexit = (symtab_type == DSO_BINARY_TYPE__BUILD_ID_CACHE ||
1854                     symtab_type == DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO);
1855
1856                 if (!dso__is_compatible_symtab_type(dso, kmod, symtab_type))
1857                         continue;
1858
1859                 if (dso__read_binary_type_filename(dso, symtab_type,
1860                                                    root_dir, name, PATH_MAX))
1861                         continue;
1862
1863                 if (nsexit)
1864                         nsinfo__mountns_exit(&nsc);
1865
1866                 is_reg = is_regular_file(name);
1867                 if (!is_reg && errno == ENOENT && dso->nsinfo) {
1868                         char *new_name = filename_with_chroot(dso->nsinfo->pid,
1869                                                               name);
1870                         if (new_name) {
1871                                 is_reg = is_regular_file(new_name);
1872                                 strlcpy(name, new_name, PATH_MAX);
1873                                 free(new_name);
1874                         }
1875                 }
1876
1877 #ifdef HAVE_LIBBFD_SUPPORT
1878                 if (is_reg)
1879                         bfdrc = dso__load_bfd_symbols(dso, name);
1880 #endif
1881                 if (is_reg && bfdrc < 0)
1882                         sirc = symsrc__init(ss, dso, name, symtab_type);
1883
1884                 if (nsexit)
1885                         nsinfo__mountns_enter(dso->nsinfo, &nsc);
1886
1887                 if (bfdrc == 0) {
1888                         ret = 0;
1889                         break;
1890                 }
1891
1892                 if (!is_reg || sirc < 0)
1893                         continue;
1894
1895                 if (!syms_ss && symsrc__has_symtab(ss)) {
1896                         syms_ss = ss;
1897                         next_slot = true;
1898                         if (!dso->symsrc_filename)
1899                                 dso->symsrc_filename = strdup(name);
1900                 }
1901
1902                 if (!runtime_ss && symsrc__possibly_runtime(ss)) {
1903                         runtime_ss = ss;
1904                         next_slot = true;
1905                 }
1906
1907                 if (next_slot) {
1908                         ss_pos++;
1909
1910                         if (syms_ss && runtime_ss)
1911                                 break;
1912                 } else {
1913                         symsrc__destroy(ss);
1914                 }
1915
1916         }
1917
1918         if (!runtime_ss && !syms_ss)
1919                 goto out_free;
1920
1921         if (runtime_ss && !syms_ss) {
1922                 syms_ss = runtime_ss;
1923         }
1924
1925         /* We'll have to hope for the best */
1926         if (!runtime_ss && syms_ss)
1927                 runtime_ss = syms_ss;
1928
1929         if (syms_ss)
1930                 ret = dso__load_sym(dso, map, syms_ss, runtime_ss, kmod);
1931         else
1932                 ret = -1;
1933
1934         if (ret > 0) {
1935                 int nr_plt;
1936
1937                 nr_plt = dso__synthesize_plt_symbols(dso, runtime_ss);
1938                 if (nr_plt > 0)
1939                         ret += nr_plt;
1940         }
1941
1942         for (; ss_pos > 0; ss_pos--)
1943                 symsrc__destroy(&ss_[ss_pos - 1]);
1944 out_free:
1945         free(name);
1946         if (ret < 0 && strstr(dso->name, " (deleted)") != NULL)
1947                 ret = 0;
1948 out:
1949         dso__set_loaded(dso);
1950         pthread_mutex_unlock(&dso->lock);
1951         nsinfo__mountns_exit(&nsc);
1952
1953         return ret;
1954 }
1955
1956 static int map__strcmp(const void *a, const void *b)
1957 {
1958         const struct map *ma = *(const struct map **)a, *mb = *(const struct map **)b;
1959         return strcmp(ma->dso->short_name, mb->dso->short_name);
1960 }
1961
1962 static int map__strcmp_name(const void *name, const void *b)
1963 {
1964         const struct map *map = *(const struct map **)b;
1965         return strcmp(name, map->dso->short_name);
1966 }
1967
1968 void __maps__sort_by_name(struct maps *maps)
1969 {
1970         qsort(maps->maps_by_name, maps->nr_maps, sizeof(struct map *), map__strcmp);
1971 }
1972
1973 static int map__groups__sort_by_name_from_rbtree(struct maps *maps)
1974 {
1975         struct map *map;
1976         struct map **maps_by_name = realloc(maps->maps_by_name, maps->nr_maps * sizeof(map));
1977         int i = 0;
1978
1979         if (maps_by_name == NULL)
1980                 return -1;
1981
1982         maps->maps_by_name = maps_by_name;
1983         maps->nr_maps_allocated = maps->nr_maps;
1984
1985         maps__for_each_entry(maps, map)
1986                 maps_by_name[i++] = map;
1987
1988         __maps__sort_by_name(maps);
1989         return 0;
1990 }
1991
1992 static struct map *__maps__find_by_name(struct maps *maps, const char *name)
1993 {
1994         struct map **mapp;
1995
1996         if (maps->maps_by_name == NULL &&
1997             map__groups__sort_by_name_from_rbtree(maps))
1998                 return NULL;
1999
2000         mapp = bsearch(name, maps->maps_by_name, maps->nr_maps, sizeof(*mapp), map__strcmp_name);
2001         if (mapp)
2002                 return *mapp;
2003         return NULL;
2004 }
2005
2006 struct map *maps__find_by_name(struct maps *maps, const char *name)
2007 {
2008         struct map *map;
2009
2010         down_read(&maps->lock);
2011
2012         if (maps->last_search_by_name && strcmp(maps->last_search_by_name->dso->short_name, name) == 0) {
2013                 map = maps->last_search_by_name;
2014                 goto out_unlock;
2015         }
2016         /*
2017          * If we have maps->maps_by_name, then the name isn't in the rbtree,
2018          * as maps->maps_by_name mirrors the rbtree when lookups by name are
2019          * made.
2020          */
2021         map = __maps__find_by_name(maps, name);
2022         if (map || maps->maps_by_name != NULL)
2023                 goto out_unlock;
2024
2025         /* Fallback to traversing the rbtree... */
2026         maps__for_each_entry(maps, map)
2027                 if (strcmp(map->dso->short_name, name) == 0) {
2028                         maps->last_search_by_name = map;
2029                         goto out_unlock;
2030                 }
2031
2032         map = NULL;
2033
2034 out_unlock:
2035         up_read(&maps->lock);
2036         return map;
2037 }
2038
2039 int dso__load_vmlinux(struct dso *dso, struct map *map,
2040                       const char *vmlinux, bool vmlinux_allocated)
2041 {
2042         int err = -1;
2043         struct symsrc ss;
2044         char symfs_vmlinux[PATH_MAX];
2045         enum dso_binary_type symtab_type;
2046
2047         if (vmlinux[0] == '/')
2048                 snprintf(symfs_vmlinux, sizeof(symfs_vmlinux), "%s", vmlinux);
2049         else
2050                 symbol__join_symfs(symfs_vmlinux, vmlinux);
2051
2052         if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
2053                 symtab_type = DSO_BINARY_TYPE__GUEST_VMLINUX;
2054         else
2055                 symtab_type = DSO_BINARY_TYPE__VMLINUX;
2056
2057         if (symsrc__init(&ss, dso, symfs_vmlinux, symtab_type))
2058                 return -1;
2059
2060         err = dso__load_sym(dso, map, &ss, &ss, 0);
2061         symsrc__destroy(&ss);
2062
2063         if (err > 0) {
2064                 if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
2065                         dso->binary_type = DSO_BINARY_TYPE__GUEST_VMLINUX;
2066                 else
2067                         dso->binary_type = DSO_BINARY_TYPE__VMLINUX;
2068                 dso__set_long_name(dso, vmlinux, vmlinux_allocated);
2069                 dso__set_loaded(dso);
2070                 pr_debug("Using %s for symbols\n", symfs_vmlinux);
2071         }
2072
2073         return err;
2074 }
2075
2076 int dso__load_vmlinux_path(struct dso *dso, struct map *map)
2077 {
2078         int i, err = 0;
2079         char *filename = NULL;
2080
2081         pr_debug("Looking at the vmlinux_path (%d entries long)\n",
2082                  vmlinux_path__nr_entries + 1);
2083
2084         for (i = 0; i < vmlinux_path__nr_entries; ++i) {
2085                 err = dso__load_vmlinux(dso, map, vmlinux_path[i], false);
2086                 if (err > 0)
2087                         goto out;
2088         }
2089
2090         if (!symbol_conf.ignore_vmlinux_buildid)
2091                 filename = dso__build_id_filename(dso, NULL, 0, false);
2092         if (filename != NULL) {
2093                 err = dso__load_vmlinux(dso, map, filename, true);
2094                 if (err > 0)
2095                         goto out;
2096                 free(filename);
2097         }
2098 out:
2099         return err;
2100 }
2101
2102 static bool visible_dir_filter(const char *name, struct dirent *d)
2103 {
2104         if (d->d_type != DT_DIR)
2105                 return false;
2106         return lsdir_no_dot_filter(name, d);
2107 }
2108
2109 static int find_matching_kcore(struct map *map, char *dir, size_t dir_sz)
2110 {
2111         char kallsyms_filename[PATH_MAX];
2112         int ret = -1;
2113         struct strlist *dirs;
2114         struct str_node *nd;
2115
2116         dirs = lsdir(dir, visible_dir_filter);
2117         if (!dirs)
2118                 return -1;
2119
2120         strlist__for_each_entry(nd, dirs) {
2121                 scnprintf(kallsyms_filename, sizeof(kallsyms_filename),
2122                           "%s/%s/kallsyms", dir, nd->s);
2123                 if (!validate_kcore_addresses(kallsyms_filename, map)) {
2124                         strlcpy(dir, kallsyms_filename, dir_sz);
2125                         ret = 0;
2126                         break;
2127                 }
2128         }
2129
2130         strlist__delete(dirs);
2131
2132         return ret;
2133 }
2134
2135 /*
2136  * Use open(O_RDONLY) to check readability directly instead of access(R_OK)
2137  * since access(R_OK) only checks with real UID/GID but open() use effective
2138  * UID/GID and actual capabilities (e.g. /proc/kcore requires CAP_SYS_RAWIO).
2139  */
2140 static bool filename__readable(const char *file)
2141 {
2142         int fd = open(file, O_RDONLY);
2143         if (fd < 0)
2144                 return false;
2145         close(fd);
2146         return true;
2147 }
2148
2149 static char *dso__find_kallsyms(struct dso *dso, struct map *map)
2150 {
2151         struct build_id bid;
2152         char sbuild_id[SBUILD_ID_SIZE];
2153         bool is_host = false;
2154         char path[PATH_MAX];
2155
2156         if (!dso->has_build_id) {
2157                 /*
2158                  * Last resort, if we don't have a build-id and couldn't find
2159                  * any vmlinux file, try the running kernel kallsyms table.
2160                  */
2161                 goto proc_kallsyms;
2162         }
2163
2164         if (sysfs__read_build_id("/sys/kernel/notes", &bid) == 0)
2165                 is_host = dso__build_id_equal(dso, &bid);
2166
2167         /* Try a fast path for /proc/kallsyms if possible */
2168         if (is_host) {
2169                 /*
2170                  * Do not check the build-id cache, unless we know we cannot use
2171                  * /proc/kcore or module maps don't match to /proc/kallsyms.
2172                  * To check readability of /proc/kcore, do not use access(R_OK)
2173                  * since /proc/kcore requires CAP_SYS_RAWIO to read and access
2174                  * can't check it.
2175                  */
2176                 if (filename__readable("/proc/kcore") &&
2177                     !validate_kcore_addresses("/proc/kallsyms", map))
2178                         goto proc_kallsyms;
2179         }
2180
2181         build_id__sprintf(&dso->bid, sbuild_id);
2182
2183         /* Find kallsyms in build-id cache with kcore */
2184         scnprintf(path, sizeof(path), "%s/%s/%s",
2185                   buildid_dir, DSO__NAME_KCORE, sbuild_id);
2186
2187         if (!find_matching_kcore(map, path, sizeof(path)))
2188                 return strdup(path);
2189
2190         /* Use current /proc/kallsyms if possible */
2191         if (is_host) {
2192 proc_kallsyms:
2193                 return strdup("/proc/kallsyms");
2194         }
2195
2196         /* Finally, find a cache of kallsyms */
2197         if (!build_id_cache__kallsyms_path(sbuild_id, path, sizeof(path))) {
2198                 pr_err("No kallsyms or vmlinux with build-id %s was found\n",
2199                        sbuild_id);
2200                 return NULL;
2201         }
2202
2203         return strdup(path);
2204 }
2205
2206 static int dso__load_kernel_sym(struct dso *dso, struct map *map)
2207 {
2208         int err;
2209         const char *kallsyms_filename = NULL;
2210         char *kallsyms_allocated_filename = NULL;
2211         char *filename = NULL;
2212
2213         /*
2214          * Step 1: if the user specified a kallsyms or vmlinux filename, use
2215          * it and only it, reporting errors to the user if it cannot be used.
2216          *
2217          * For instance, try to analyse an ARM perf.data file _without_ a
2218          * build-id, or if the user specifies the wrong path to the right
2219          * vmlinux file, obviously we can't fallback to another vmlinux (a
2220          * x86_86 one, on the machine where analysis is being performed, say),
2221          * or worse, /proc/kallsyms.
2222          *
2223          * If the specified file _has_ a build-id and there is a build-id
2224          * section in the perf.data file, we will still do the expected
2225          * validation in dso__load_vmlinux and will bail out if they don't
2226          * match.
2227          */
2228         if (symbol_conf.kallsyms_name != NULL) {
2229                 kallsyms_filename = symbol_conf.kallsyms_name;
2230                 goto do_kallsyms;
2231         }
2232
2233         if (!symbol_conf.ignore_vmlinux && symbol_conf.vmlinux_name != NULL) {
2234                 return dso__load_vmlinux(dso, map, symbol_conf.vmlinux_name, false);
2235         }
2236
2237         /*
2238          * Before checking on common vmlinux locations, check if it's
2239          * stored as standard build id binary (not kallsyms) under
2240          * .debug cache.
2241          */
2242         if (!symbol_conf.ignore_vmlinux_buildid)
2243                 filename = __dso__build_id_filename(dso, NULL, 0, false, false);
2244         if (filename != NULL) {
2245                 err = dso__load_vmlinux(dso, map, filename, true);
2246                 if (err > 0)
2247                         return err;
2248                 free(filename);
2249         }
2250
2251         if (!symbol_conf.ignore_vmlinux && vmlinux_path != NULL) {
2252                 err = dso__load_vmlinux_path(dso, map);
2253                 if (err > 0)
2254                         return err;
2255         }
2256
2257         /* do not try local files if a symfs was given */
2258         if (symbol_conf.symfs[0] != 0)
2259                 return -1;
2260
2261         kallsyms_allocated_filename = dso__find_kallsyms(dso, map);
2262         if (!kallsyms_allocated_filename)
2263                 return -1;
2264
2265         kallsyms_filename = kallsyms_allocated_filename;
2266
2267 do_kallsyms:
2268         err = dso__load_kallsyms(dso, kallsyms_filename, map);
2269         if (err > 0)
2270                 pr_debug("Using %s for symbols\n", kallsyms_filename);
2271         free(kallsyms_allocated_filename);
2272
2273         if (err > 0 && !dso__is_kcore(dso)) {
2274                 dso->binary_type = DSO_BINARY_TYPE__KALLSYMS;
2275                 dso__set_long_name(dso, DSO__NAME_KALLSYMS, false);
2276                 map__fixup_start(map);
2277                 map__fixup_end(map);
2278         }
2279
2280         return err;
2281 }
2282
2283 static int dso__load_guest_kernel_sym(struct dso *dso, struct map *map)
2284 {
2285         int err;
2286         const char *kallsyms_filename = NULL;
2287         struct machine *machine = map__kmaps(map)->machine;
2288         char path[PATH_MAX];
2289
2290         if (machine__is_default_guest(machine)) {
2291                 /*
2292                  * if the user specified a vmlinux filename, use it and only
2293                  * it, reporting errors to the user if it cannot be used.
2294                  * Or use file guest_kallsyms inputted by user on commandline
2295                  */
2296                 if (symbol_conf.default_guest_vmlinux_name != NULL) {
2297                         err = dso__load_vmlinux(dso, map,
2298                                                 symbol_conf.default_guest_vmlinux_name,
2299                                                 false);
2300                         return err;
2301                 }
2302
2303                 kallsyms_filename = symbol_conf.default_guest_kallsyms;
2304                 if (!kallsyms_filename)
2305                         return -1;
2306         } else {
2307                 sprintf(path, "%s/proc/kallsyms", machine->root_dir);
2308                 kallsyms_filename = path;
2309         }
2310
2311         err = dso__load_kallsyms(dso, kallsyms_filename, map);
2312         if (err > 0)
2313                 pr_debug("Using %s for symbols\n", kallsyms_filename);
2314         if (err > 0 && !dso__is_kcore(dso)) {
2315                 dso->binary_type = DSO_BINARY_TYPE__GUEST_KALLSYMS;
2316                 dso__set_long_name(dso, machine->mmap_name, false);
2317                 map__fixup_start(map);
2318                 map__fixup_end(map);
2319         }
2320
2321         return err;
2322 }
2323
2324 static void vmlinux_path__exit(void)
2325 {
2326         while (--vmlinux_path__nr_entries >= 0)
2327                 zfree(&vmlinux_path[vmlinux_path__nr_entries]);
2328         vmlinux_path__nr_entries = 0;
2329
2330         zfree(&vmlinux_path);
2331 }
2332
2333 static const char * const vmlinux_paths[] = {
2334         "vmlinux",
2335         "/boot/vmlinux"
2336 };
2337
2338 static const char * const vmlinux_paths_upd[] = {
2339         "/boot/vmlinux-%s",
2340         "/usr/lib/debug/boot/vmlinux-%s",
2341         "/lib/modules/%s/build/vmlinux",
2342         "/usr/lib/debug/lib/modules/%s/vmlinux",
2343         "/usr/lib/debug/boot/vmlinux-%s.debug"
2344 };
2345
2346 static int vmlinux_path__add(const char *new_entry)
2347 {
2348         vmlinux_path[vmlinux_path__nr_entries] = strdup(new_entry);
2349         if (vmlinux_path[vmlinux_path__nr_entries] == NULL)
2350                 return -1;
2351         ++vmlinux_path__nr_entries;
2352
2353         return 0;
2354 }
2355
2356 static int vmlinux_path__init(struct perf_env *env)
2357 {
2358         struct utsname uts;
2359         char bf[PATH_MAX];
2360         char *kernel_version;
2361         unsigned int i;
2362
2363         vmlinux_path = malloc(sizeof(char *) * (ARRAY_SIZE(vmlinux_paths) +
2364                               ARRAY_SIZE(vmlinux_paths_upd)));
2365         if (vmlinux_path == NULL)
2366                 return -1;
2367
2368         for (i = 0; i < ARRAY_SIZE(vmlinux_paths); i++)
2369                 if (vmlinux_path__add(vmlinux_paths[i]) < 0)
2370                         goto out_fail;
2371
2372         /* only try kernel version if no symfs was given */
2373         if (symbol_conf.symfs[0] != 0)
2374                 return 0;
2375
2376         if (env) {
2377                 kernel_version = env->os_release;
2378         } else {
2379                 if (uname(&uts) < 0)
2380                         goto out_fail;
2381
2382                 kernel_version = uts.release;
2383         }
2384
2385         for (i = 0; i < ARRAY_SIZE(vmlinux_paths_upd); i++) {
2386                 snprintf(bf, sizeof(bf), vmlinux_paths_upd[i], kernel_version);
2387                 if (vmlinux_path__add(bf) < 0)
2388                         goto out_fail;
2389         }
2390
2391         return 0;
2392
2393 out_fail:
2394         vmlinux_path__exit();
2395         return -1;
2396 }
2397
2398 int setup_list(struct strlist **list, const char *list_str,
2399                       const char *list_name)
2400 {
2401         if (list_str == NULL)
2402                 return 0;
2403
2404         *list = strlist__new(list_str, NULL);
2405         if (!*list) {
2406                 pr_err("problems parsing %s list\n", list_name);
2407                 return -1;
2408         }
2409
2410         symbol_conf.has_filter = true;
2411         return 0;
2412 }
2413
2414 int setup_intlist(struct intlist **list, const char *list_str,
2415                   const char *list_name)
2416 {
2417         if (list_str == NULL)
2418                 return 0;
2419
2420         *list = intlist__new(list_str);
2421         if (!*list) {
2422                 pr_err("problems parsing %s list\n", list_name);
2423                 return -1;
2424         }
2425         return 0;
2426 }
2427
2428 static int setup_addrlist(struct intlist **addr_list, struct strlist *sym_list)
2429 {
2430         struct str_node *pos, *tmp;
2431         unsigned long val;
2432         char *sep;
2433         const char *end;
2434         int i = 0, err;
2435
2436         *addr_list = intlist__new(NULL);
2437         if (!*addr_list)
2438                 return -1;
2439
2440         strlist__for_each_entry_safe(pos, tmp, sym_list) {
2441                 errno = 0;
2442                 val = strtoul(pos->s, &sep, 16);
2443                 if (errno || (sep == pos->s))
2444                         continue;
2445
2446                 if (*sep != '\0') {
2447                         end = pos->s + strlen(pos->s) - 1;
2448                         while (end >= sep && isspace(*end))
2449                                 end--;
2450
2451                         if (end >= sep)
2452                                 continue;
2453                 }
2454
2455                 err = intlist__add(*addr_list, val);
2456                 if (err)
2457                         break;
2458
2459                 strlist__remove(sym_list, pos);
2460                 i++;
2461         }
2462
2463         if (i == 0) {
2464                 intlist__delete(*addr_list);
2465                 *addr_list = NULL;
2466         }
2467
2468         return 0;
2469 }
2470
2471 static bool symbol__read_kptr_restrict(void)
2472 {
2473         bool value = false;
2474         FILE *fp = fopen("/proc/sys/kernel/kptr_restrict", "r");
2475
2476         if (fp != NULL) {
2477                 char line[8];
2478
2479                 if (fgets(line, sizeof(line), fp) != NULL)
2480                         value = perf_cap__capable(CAP_SYSLOG) ?
2481                                         (atoi(line) >= 2) :
2482                                         (atoi(line) != 0);
2483
2484                 fclose(fp);
2485         }
2486
2487         /* Per kernel/kallsyms.c:
2488          * we also restrict when perf_event_paranoid > 1 w/o CAP_SYSLOG
2489          */
2490         if (perf_event_paranoid() > 1 && !perf_cap__capable(CAP_SYSLOG))
2491                 value = true;
2492
2493         return value;
2494 }
2495
2496 int symbol__annotation_init(void)
2497 {
2498         if (symbol_conf.init_annotation)
2499                 return 0;
2500
2501         if (symbol_conf.initialized) {
2502                 pr_err("Annotation needs to be init before symbol__init()\n");
2503                 return -1;
2504         }
2505
2506         symbol_conf.priv_size += sizeof(struct annotation);
2507         symbol_conf.init_annotation = true;
2508         return 0;
2509 }
2510
2511 int symbol__init(struct perf_env *env)
2512 {
2513         const char *symfs;
2514
2515         if (symbol_conf.initialized)
2516                 return 0;
2517
2518         symbol_conf.priv_size = PERF_ALIGN(symbol_conf.priv_size, sizeof(u64));
2519
2520         symbol__elf_init();
2521
2522         if (symbol_conf.sort_by_name)
2523                 symbol_conf.priv_size += (sizeof(struct symbol_name_rb_node) -
2524                                           sizeof(struct symbol));
2525
2526         if (symbol_conf.try_vmlinux_path && vmlinux_path__init(env) < 0)
2527                 return -1;
2528
2529         if (symbol_conf.field_sep && *symbol_conf.field_sep == '.') {
2530                 pr_err("'.' is the only non valid --field-separator argument\n");
2531                 return -1;
2532         }
2533
2534         if (setup_list(&symbol_conf.dso_list,
2535                        symbol_conf.dso_list_str, "dso") < 0)
2536                 return -1;
2537
2538         if (setup_list(&symbol_conf.comm_list,
2539                        symbol_conf.comm_list_str, "comm") < 0)
2540                 goto out_free_dso_list;
2541
2542         if (setup_intlist(&symbol_conf.pid_list,
2543                        symbol_conf.pid_list_str, "pid") < 0)
2544                 goto out_free_comm_list;
2545
2546         if (setup_intlist(&symbol_conf.tid_list,
2547                        symbol_conf.tid_list_str, "tid") < 0)
2548                 goto out_free_pid_list;
2549
2550         if (setup_list(&symbol_conf.sym_list,
2551                        symbol_conf.sym_list_str, "symbol") < 0)
2552                 goto out_free_tid_list;
2553
2554         if (symbol_conf.sym_list &&
2555             setup_addrlist(&symbol_conf.addr_list, symbol_conf.sym_list) < 0)
2556                 goto out_free_sym_list;
2557
2558         if (setup_list(&symbol_conf.bt_stop_list,
2559                        symbol_conf.bt_stop_list_str, "symbol") < 0)
2560                 goto out_free_sym_list;
2561
2562         /*
2563          * A path to symbols of "/" is identical to ""
2564          * reset here for simplicity.
2565          */
2566         symfs = realpath(symbol_conf.symfs, NULL);
2567         if (symfs == NULL)
2568                 symfs = symbol_conf.symfs;
2569         if (strcmp(symfs, "/") == 0)
2570                 symbol_conf.symfs = "";
2571         if (symfs != symbol_conf.symfs)
2572                 free((void *)symfs);
2573
2574         symbol_conf.kptr_restrict = symbol__read_kptr_restrict();
2575
2576         symbol_conf.initialized = true;
2577         return 0;
2578
2579 out_free_sym_list:
2580         strlist__delete(symbol_conf.sym_list);
2581         intlist__delete(symbol_conf.addr_list);
2582 out_free_tid_list:
2583         intlist__delete(symbol_conf.tid_list);
2584 out_free_pid_list:
2585         intlist__delete(symbol_conf.pid_list);
2586 out_free_comm_list:
2587         strlist__delete(symbol_conf.comm_list);
2588 out_free_dso_list:
2589         strlist__delete(symbol_conf.dso_list);
2590         return -1;
2591 }
2592
2593 void symbol__exit(void)
2594 {
2595         if (!symbol_conf.initialized)
2596                 return;
2597         strlist__delete(symbol_conf.bt_stop_list);
2598         strlist__delete(symbol_conf.sym_list);
2599         strlist__delete(symbol_conf.dso_list);
2600         strlist__delete(symbol_conf.comm_list);
2601         intlist__delete(symbol_conf.tid_list);
2602         intlist__delete(symbol_conf.pid_list);
2603         intlist__delete(symbol_conf.addr_list);
2604         vmlinux_path__exit();
2605         symbol_conf.sym_list = symbol_conf.dso_list = symbol_conf.comm_list = NULL;
2606         symbol_conf.bt_stop_list = NULL;
2607         symbol_conf.initialized = false;
2608 }
2609
2610 int symbol__config_symfs(const struct option *opt __maybe_unused,
2611                          const char *dir, int unset __maybe_unused)
2612 {
2613         char *bf = NULL;
2614         int ret;
2615
2616         symbol_conf.symfs = strdup(dir);
2617         if (symbol_conf.symfs == NULL)
2618                 return -ENOMEM;
2619
2620         /* skip the locally configured cache if a symfs is given, and
2621          * config buildid dir to symfs/.debug
2622          */
2623         ret = asprintf(&bf, "%s/%s", dir, ".debug");
2624         if (ret < 0)
2625                 return -ENOMEM;
2626
2627         set_buildid_dir(bf);
2628
2629         free(bf);
2630         return 0;
2631 }
2632
2633 struct mem_info *mem_info__get(struct mem_info *mi)
2634 {
2635         if (mi)
2636                 refcount_inc(&mi->refcnt);
2637         return mi;
2638 }
2639
2640 void mem_info__put(struct mem_info *mi)
2641 {
2642         if (mi && refcount_dec_and_test(&mi->refcnt))
2643                 free(mi);
2644 }
2645
2646 struct mem_info *mem_info__new(void)
2647 {
2648         struct mem_info *mi = zalloc(sizeof(*mi));
2649
2650         if (mi)
2651                 refcount_set(&mi->refcnt, 1);
2652         return mi;
2653 }
2654
2655 /*
2656  * Checks that user supplied symbol kernel files are accessible because
2657  * the default mechanism for accessing elf files fails silently. i.e. if
2658  * debug syms for a build ID aren't found perf carries on normally. When
2659  * they are user supplied we should assume that the user doesn't want to
2660  * silently fail.
2661  */
2662 int symbol__validate_sym_arguments(void)
2663 {
2664         if (symbol_conf.vmlinux_name &&
2665             access(symbol_conf.vmlinux_name, R_OK)) {
2666                 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
2667                 return -EINVAL;
2668         }
2669         if (symbol_conf.kallsyms_name &&
2670             access(symbol_conf.kallsyms_name, R_OK)) {
2671                 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
2672                 return -EINVAL;
2673         }
2674         return 0;
2675 }