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 const char * __weak arch__normalize_symbol_name(const char *name)
 105 {
 106         return name;
 107 }
 108
 109 int __weak arch__compare_symbol_names(const char *namea, const char *nameb)
 110 {
 111         return strcmp(namea, nameb);
 112 }
 113
 114 int __weak arch__compare_symbol_names_n(const char *namea, const char *nameb,
 115                                         unsigned int n)
 116 {
 117         return strncmp(namea, nameb, n);
 118 }
 119
 120 int __weak arch__choose_best_symbol(struct symbol *syma,
 121                                     struct symbol *symb __maybe_unused)
 122 {
 123         /* Avoid "SyS" kernel syscall aliases */
 124         if (strlen(syma->name) >= 3 && !strncmp(syma->name, "SyS", 3))
 125                 return SYMBOL_B;
 126         if (strlen(syma->name) >= 10 && !strncmp(syma->name, "compat_SyS", 10))
 127                 return SYMBOL_B;
 128
 129         return SYMBOL_A;
 130 }
 131
 132 static int choose_best_symbol(struct symbol *syma, struct symbol *symb)
 133 {
 134         s64 a;
 135         s64 b;
 136         size_t na, nb;
 137
 138         /* Prefer a symbol with non zero length */
 139         a = syma->end - syma->start;
 140         b = symb->end - symb->start;
 141         if ((b == 0) && (a > 0))
 142                 return SYMBOL_A;
 143         else if ((a == 0) && (b > 0))
 144                 return SYMBOL_B;
 145
 146         /* Prefer a non weak symbol over a weak one */
 147         a = syma->binding == STB_WEAK;
 148         b = symb->binding == STB_WEAK;
 149         if (b && !a)
 150                 return SYMBOL_A;
 151         if (a && !b)
 152                 return SYMBOL_B;
 153
 154         /* Prefer a global symbol over a non global one */
 155         a = syma->binding == STB_GLOBAL;
 156         b = symb->binding == STB_GLOBAL;
 157         if (a && !b)
 158                 return SYMBOL_A;
 159         if (b && !a)
 160                 return SYMBOL_B;
 161
 162         /* Prefer a symbol with less underscores */
 163         a = prefix_underscores_count(syma->name);
 164         b = prefix_underscores_count(symb->name);
 165         if (b > a)
 166                 return SYMBOL_A;
 167         else if (a > b)
 168                 return SYMBOL_B;
 169
 170         /* Choose the symbol with the longest name */
 171         na = strlen(syma->name);
 172         nb = strlen(symb->name);
 173         if (na > nb)
 174                 return SYMBOL_A;
 175         else if (na < nb)
 176                 return SYMBOL_B;
 177
 178         return arch__choose_best_symbol(syma, symb);
 179 }
 180
 181 void symbols__fixup_duplicate(struct rb_root_cached *symbols)
 182 {
 183         struct rb_node *nd;
 184         struct symbol *curr, *next;
 185
 186         if (symbol_conf.allow_aliases)
 187                 return;
 188
 189         nd = rb_first_cached(symbols);
 190
 191         while (nd) {
 192                 curr = rb_entry(nd, struct symbol, rb_node);
 193 again:
 194                 nd = rb_next(&curr->rb_node);
 195                 next = rb_entry(nd, struct symbol, rb_node);
 196
 197                 if (!nd)
 198                         break;
 199
 200                 if (curr->start != next->start)
 201                         continue;
 202
 203                 if (choose_best_symbol(curr, next) == SYMBOL_A) {
 204                         rb_erase_cached(&next->rb_node, symbols);
 205                         symbol__delete(next);
 206                         goto again;
 207                 } else {
 208                         nd = rb_next(&curr->rb_node);
 209                         rb_erase_cached(&curr->rb_node, symbols);
 210                         symbol__delete(curr);
 211                 }
 212         }
 213 }
 214
 215 /* Update zero-sized symbols using the address of the next symbol */
 216 void symbols__fixup_end(struct rb_root_cached *symbols, bool is_kallsyms)
 217 {
 218         struct rb_node *nd, *prevnd = rb_first_cached(symbols);
 219         struct symbol *curr, *prev;
 220
 221         if (prevnd == NULL)
 222                 return;
 223
 224         curr = rb_entry(prevnd, struct symbol, rb_node);
 225
 226         for (nd = rb_next(prevnd); nd; nd = rb_next(nd)) {
 227                 prev = curr;
 228                 curr = rb_entry(nd, struct symbol, rb_node);
 229
 230                 /*
 231                  * On some architecture kernel text segment start is located at
 232                  * some low memory address, while modules are located at high
 233                  * memory addresses (or vice versa).  The gap between end of
 234                  * kernel text segment and beginning of first module's text
 235                  * segment is very big.  Therefore do not fill this gap and do
 236                  * not assign it to the kernel dso map (kallsyms).
 237                  *
 238                  * In kallsyms, it determines module symbols using '[' character
 239                  * like in:
 240                  *   ffffffffc1937000 T hdmi_driver_init  [snd_hda_codec_hdmi]
 241                  */
 242                 if (prev->end == prev->start) {
 243                         /* Last kernel/module symbol mapped to end of page */
 244                         if (is_kallsyms && (!strchr(prev->name, '[') !=
 245                                             !strchr(curr->name, '[')))
 246                                 prev->end = roundup(prev->end + 4096, 4096);
 247                         else
 248                                 prev->end = curr->start;
 249
 250                         pr_debug4("%s sym:%s end:%#" PRIx64 "\n",
 251                                   __func__, prev->name, prev->end);
 252                 }
 253         }
 254
 255         /* Last entry */
 256         if (curr->end == curr->start)
 257                 curr->end = roundup(curr->start, 4096) + 4096;
 258 }
 259
 260 void maps__fixup_end(struct maps *maps)
 261 {
 262         struct map *prev = NULL, *curr;
 263
 264         down_write(&maps->lock);
 265
 266         maps__for_each_entry(maps, curr) {
 267                 if (prev != NULL && !prev->end)
 268                         prev->end = curr->start;
 269
 270                 prev = curr;
 271         }
 272
 273         /*
 274          * We still haven't the actual symbols, so guess the
 275          * last map final address.
 276          */
 277         if (curr && !curr->end)
 278                 curr->end = ~0ULL;
 279
 280         up_write(&maps->lock);
 281 }
 282
 283 struct symbol *symbol__new(u64 start, u64 len, u8 binding, u8 type, const char *name)
 284 {
 285         size_t namelen = strlen(name) + 1;
 286         struct symbol *sym = calloc(1, (symbol_conf.priv_size +
 287                                         sizeof(*sym) + namelen));
 288         if (sym == NULL)
 289                 return NULL;
 290
 291         if (symbol_conf.priv_size) {
 292                 if (symbol_conf.init_annotation) {
 293                         struct annotation *notes = (void *)sym;
 294                         annotation__init(notes);
 295                 }
 296                 sym = ((void *)sym) + symbol_conf.priv_size;
 297         }
 298
 299         sym->start   = start;
 300         sym->end     = len ? start + len : start;
 301         sym->type    = type;
 302         sym->binding = binding;
 303         sym->namelen = namelen - 1;
 304
 305         pr_debug4("%s: %s %#" PRIx64 "-%#" PRIx64 "\n",
 306                   __func__, name, start, sym->end);
 307         memcpy(sym->name, name, namelen);
 308
 309         return sym;
 310 }
 311
 312 void symbol__delete(struct symbol *sym)
 313 {
 314         if (symbol_conf.priv_size) {
 315                 if (symbol_conf.init_annotation) {
 316                         struct annotation *notes = symbol__annotation(sym);
 317
 318                         annotation__exit(notes);
 319                 }
 320         }
 321         free(((void *)sym) - symbol_conf.priv_size);
 322 }
 323
 324 void symbols__delete(struct rb_root_cached *symbols)
 325 {
 326         struct symbol *pos;
 327         struct rb_node *next = rb_first_cached(symbols);
 328
 329         while (next) {
 330                 pos = rb_entry(next, struct symbol, rb_node);
 331                 next = rb_next(&pos->rb_node);
 332                 rb_erase_cached(&pos->rb_node, symbols);
 333                 symbol__delete(pos);
 334         }
 335 }
 336
 337 void __symbols__insert(struct rb_root_cached *symbols,
 338                        struct symbol *sym, bool kernel)
 339 {
 340         struct rb_node **p = &symbols->rb_root.rb_node;
 341         struct rb_node *parent = NULL;
 342         const u64 ip = sym->start;
 343         struct symbol *s;
 344         bool leftmost = true;
 345
 346         if (kernel) {
 347                 const char *name = sym->name;
 348                 /*
 349                  * ppc64 uses function descriptors and appends a '.' to the
 350                  * start of every instruction address. Remove it.
 351                  */
 352                 if (name[0] == '.')
 353                         name++;
 354                 sym->idle = symbol__is_idle(name);
 355         }
 356
 357         while (*p != NULL) {
 358                 parent = *p;
 359                 s = rb_entry(parent, struct symbol, rb_node);
 360                 if (ip < s->start)
 361                         p = &(*p)->rb_left;
 362                 else {
 363                         p = &(*p)->rb_right;
 364                         leftmost = false;
 365                 }
 366         }
 367         rb_link_node(&sym->rb_node, parent, p);
 368         rb_insert_color_cached(&sym->rb_node, symbols, leftmost);
 369 }
 370
 371 void symbols__insert(struct rb_root_cached *symbols, struct symbol *sym)
 372 {
 373         __symbols__insert(symbols, sym, false);
 374 }
 375
 376 static struct symbol *symbols__find(struct rb_root_cached *symbols, u64 ip)
 377 {
 378         struct rb_node *n;
 379
 380         if (symbols == NULL)
 381                 return NULL;
 382
 383         n = symbols->rb_root.rb_node;
 384
 385         while (n) {
 386                 struct symbol *s = rb_entry(n, struct symbol, rb_node);
 387
 388                 if (ip < s->start)
 389                         n = n->rb_left;
 390                 else if (ip > s->end || (ip == s->end && ip != s->start))
 391                         n = n->rb_right;
 392                 else
 393                         return s;
 394         }
 395
 396         return NULL;
 397 }
 398
 399 static struct symbol *symbols__first(struct rb_root_cached *symbols)
 400 {
 401         struct rb_node *n = rb_first_cached(symbols);
 402
 403         if (n)
 404                 return rb_entry(n, struct symbol, rb_node);
 405
 406         return NULL;
 407 }
 408
 409 static struct symbol *symbols__last(struct rb_root_cached *symbols)
 410 {
 411         struct rb_node *n = rb_last(&symbols->rb_root);
 412
 413         if (n)
 414                 return rb_entry(n, struct symbol, rb_node);
 415
 416         return NULL;
 417 }
 418
 419 static struct symbol *symbols__next(struct symbol *sym)
 420 {
 421         struct rb_node *n = rb_next(&sym->rb_node);
 422
 423         if (n)
 424                 return rb_entry(n, struct symbol, rb_node);
 425
 426         return NULL;
 427 }
 428
 429 static void symbols__insert_by_name(struct rb_root_cached *symbols, struct symbol *sym)
 430 {
 431         struct rb_node **p = &symbols->rb_root.rb_node;
 432         struct rb_node *parent = NULL;
 433         struct symbol_name_rb_node *symn, *s;
 434         bool leftmost = true;
 435
 436         symn = container_of(sym, struct symbol_name_rb_node, sym);
 437
 438         while (*p != NULL) {
 439                 parent = *p;
 440                 s = rb_entry(parent, struct symbol_name_rb_node, rb_node);
 441                 if (strcmp(sym->name, s->sym.name) < 0)
 442                         p = &(*p)->rb_left;
 443                 else {
 444                         p = &(*p)->rb_right;
 445                         leftmost = false;
 446                 }
 447         }
 448         rb_link_node(&symn->rb_node, parent, p);
 449         rb_insert_color_cached(&symn->rb_node, symbols, leftmost);
 450 }
 451
 452 static void symbols__sort_by_name(struct rb_root_cached *symbols,
 453                                   struct rb_root_cached *source)
 454 {
 455         struct rb_node *nd;
 456
 457         for (nd = rb_first_cached(source); nd; nd = rb_next(nd)) {
 458                 struct symbol *pos = rb_entry(nd, struct symbol, rb_node);
 459                 symbols__insert_by_name(symbols, pos);
 460         }
 461 }
 462
 463 int symbol__match_symbol_name(const char *name, const char *str,
 464                               enum symbol_tag_include includes)
 465 {
 466         const char *versioning;
 467
 468         if (includes == SYMBOL_TAG_INCLUDE__DEFAULT_ONLY &&
 469             (versioning = strstr(name, "@@"))) {
 470                 int len = strlen(str);
 471
 472                 if (len < versioning - name)
 473                         len = versioning - name;
 474
 475                 return arch__compare_symbol_names_n(name, str, len);
 476         } else
 477                 return arch__compare_symbol_names(name, str);
 478 }
 479
 480 static struct symbol *symbols__find_by_name(struct rb_root_cached *symbols,
 481                                             const char *name,
 482                                             enum symbol_tag_include includes)
 483 {
 484         struct rb_node *n;
 485         struct symbol_name_rb_node *s = NULL;
 486
 487         if (symbols == NULL)
 488                 return NULL;
 489
 490         n = symbols->rb_root.rb_node;
 491
 492         while (n) {
 493                 int cmp;
 494
 495                 s = rb_entry(n, struct symbol_name_rb_node, rb_node);
 496                 cmp = symbol__match_symbol_name(s->sym.name, name, includes);
 497
 498                 if (cmp > 0)
 499                         n = n->rb_left;
 500                 else if (cmp < 0)
 501                         n = n->rb_right;
 502                 else
 503                         break;
 504         }
 505
 506         if (n == NULL)
 507                 return NULL;
 508
 509         if (includes != SYMBOL_TAG_INCLUDE__DEFAULT_ONLY)
 510                 /* return first symbol that has same name (if any) */
 511                 for (n = rb_prev(n); n; n = rb_prev(n)) {
 512                         struct symbol_name_rb_node *tmp;
 513
 514                         tmp = rb_entry(n, struct symbol_name_rb_node, rb_node);
 515                         if (arch__compare_symbol_names(tmp->sym.name, s->sym.name))
 516                                 break;
 517
 518                         s = tmp;
 519                 }
 520
 521         return &s->sym;
 522 }
 523
 524 void dso__reset_find_symbol_cache(struct dso *dso)
 525 {
 526         dso->last_find_result.addr   = 0;
 527         dso->last_find_result.symbol = NULL;
 528 }
 529
 530 void dso__insert_symbol(struct dso *dso, struct symbol *sym)
 531 {
 532         __symbols__insert(&dso->symbols, sym, dso->kernel);
 533
 534         /* update the symbol cache if necessary */
 535         if (dso->last_find_result.addr >= sym->start &&
 536             (dso->last_find_result.addr < sym->end ||
 537             sym->start == sym->end)) {
 538                 dso->last_find_result.symbol = sym;
 539         }
 540 }
 541
 542 void dso__delete_symbol(struct dso *dso, struct symbol *sym)
 543 {
 544         rb_erase_cached(&sym->rb_node, &dso->symbols);
 545         symbol__delete(sym);
 546         dso__reset_find_symbol_cache(dso);
 547 }
 548
 549 struct symbol *dso__find_symbol(struct dso *dso, u64 addr)
 550 {
 551         if (dso->last_find_result.addr != addr || dso->last_find_result.symbol == NULL) {
 552                 dso->last_find_result.addr   = addr;
 553                 dso->last_find_result.symbol = symbols__find(&dso->symbols, addr);
 554         }
 555
 556         return dso->last_find_result.symbol;
 557 }
 558
 559 struct symbol *dso__first_symbol(struct dso *dso)
 560 {
 561         return symbols__first(&dso->symbols);
 562 }
 563
 564 struct symbol *dso__last_symbol(struct dso *dso)
 565 {
 566         return symbols__last(&dso->symbols);
 567 }
 568
 569 struct symbol *dso__next_symbol(struct symbol *sym)
 570 {
 571         return symbols__next(sym);
 572 }
 573
 574 struct symbol *symbol__next_by_name(struct symbol *sym)
 575 {
 576         struct symbol_name_rb_node *s = container_of(sym, struct symbol_name_rb_node, sym);
 577         struct rb_node *n = rb_next(&s->rb_node);
 578
 579         return n ? &rb_entry(n, struct symbol_name_rb_node, rb_node)->sym : NULL;
 580 }
 581
 582  /*
 583   * Returns first symbol that matched with @name.
 584   */
 585 struct symbol *dso__find_symbol_by_name(struct dso *dso, const char *name)
 586 {
 587         struct symbol *s = symbols__find_by_name(&dso->symbol_names, name,
 588                                                  SYMBOL_TAG_INCLUDE__NONE);
 589         if (!s)
 590                 s = symbols__find_by_name(&dso->symbol_names, name,
 591                                           SYMBOL_TAG_INCLUDE__DEFAULT_ONLY);
 592         return s;
 593 }
 594
 595 void dso__sort_by_name(struct dso *dso)
 596 {
 597         dso__set_sorted_by_name(dso);
 598         return symbols__sort_by_name(&dso->symbol_names, &dso->symbols);
 599 }
 600
 601 /*
 602  * While we find nice hex chars, build a long_val.
 603  * Return number of chars processed.
 604  */
 605 static int hex2u64(const char *ptr, u64 *long_val)
 606 {
 607         char *p;
 608
 609         *long_val = strtoull(ptr, &p, 16);
 610
 611         return p - ptr;
 612 }
 613
 614
 615 int modules__parse(const char *filename, void *arg,
 616                    int (*process_module)(void *arg, const char *name,
 617                                          u64 start, u64 size))
 618 {
 619         char *line = NULL;
 620         size_t n;
 621         FILE *file;
 622         int err = 0;
 623
 624         file = fopen(filename, "r");
 625         if (file == NULL)
 626                 return -1;
 627
 628         while (1) {
 629                 char name[PATH_MAX];
 630                 u64 start, size;
 631                 char *sep, *endptr;
 632                 ssize_t line_len;
 633
 634                 line_len = getline(&line, &n, file);
 635                 if (line_len < 0) {
 636                         if (feof(file))
 637                                 break;
 638                         err = -1;
 639                         goto out;
 640                 }
 641
 642                 if (!line) {
 643                         err = -1;
 644                         goto out;
 645                 }
 646
 647                 line[--line_len] = '\0'; /* \n */
 648
 649                 sep = strrchr(line, 'x');
 650                 if (sep == NULL)
 651                         continue;
 652
 653                 hex2u64(sep + 1, &start);
 654
 655                 sep = strchr(line, ' ');
 656                 if (sep == NULL)
 657                         continue;
 658
 659                 *sep = '\0';
 660
 661                 scnprintf(name, sizeof(name), "[%s]", line);
 662
 663                 size = strtoul(sep + 1, &endptr, 0);
 664                 if (*endptr != ' ' && *endptr != '\t')
 665                         continue;
 666
 667                 err = process_module(arg, name, start, size);
 668                 if (err)
 669                         break;
 670         }
 671 out:
 672         free(line);
 673         fclose(file);
 674         return err;
 675 }
 676
 677 /*
 678  * These are symbols in the kernel image, so make sure that
 679  * sym is from a kernel DSO.
 680  */
 681 static bool symbol__is_idle(const char *name)
 682 {
 683         const char * const idle_symbols[] = {
 684                 "acpi_idle_do_entry",
 685                 "acpi_processor_ffh_cstate_enter",
 686                 "arch_cpu_idle",
 687                 "cpu_idle",
 688                 "cpu_startup_entry",
 689                 "idle_cpu",
 690                 "intel_idle",
 691                 "default_idle",
 692                 "native_safe_halt",
 693                 "enter_idle",
 694                 "exit_idle",
 695                 "mwait_idle",
 696                 "mwait_idle_with_hints",
 697                 "mwait_idle_with_hints.constprop.0",
 698                 "poll_idle",
 699                 "ppc64_runlatch_off",
 700                 "pseries_dedicated_idle_sleep",
 701                 "psw_idle",
 702                 "psw_idle_exit",
 703                 NULL
 704         };
 705         int i;
 706         static struct strlist *idle_symbols_list;
 707
 708         if (idle_symbols_list)
 709                 return strlist__has_entry(idle_symbols_list, name);
 710
 711         idle_symbols_list = strlist__new(NULL, NULL);
 712
 713         for (i = 0; idle_symbols[i]; i++)
 714                 strlist__add(idle_symbols_list, idle_symbols[i]);
 715
 716         return strlist__has_entry(idle_symbols_list, name);
 717 }
 718
 719 static int map__process_kallsym_symbol(void *arg, const char *name,
 720                                        char type, u64 start)
 721 {
 722         struct symbol *sym;
 723         struct dso *dso = arg;
 724         struct rb_root_cached *root = &dso->symbols;
 725
 726         if (!symbol_type__filter(type))
 727                 return 0;
 728
 729         /* Ignore local symbols for ARM modules */
 730         if (name[0] == '$')
 731                 return 0;
 732
 733         /*
 734          * module symbols are not sorted so we add all
 735          * symbols, setting length to 0, and rely on
 736          * symbols__fixup_end() to fix it up.
 737          */
 738         sym = symbol__new(start, 0, kallsyms2elf_binding(type), kallsyms2elf_type(type), name);
 739         if (sym == NULL)
 740                 return -ENOMEM;
 741         /*
 742          * We will pass the symbols to the filter later, in
 743          * map__split_kallsyms, when we have split the maps per module
 744          */
 745         __symbols__insert(root, sym, !strchr(name, '['));
 746
 747         return 0;
 748 }
 749
 750 /*
 751  * Loads the function entries in /proc/kallsyms into kernel_map->dso,
 752  * so that we can in the next step set the symbol ->end address and then
 753  * call kernel_maps__split_kallsyms.
 754  */
 755 static int dso__load_all_kallsyms(struct dso *dso, const char *filename)
 756 {
 757         return kallsyms__parse(filename, dso, map__process_kallsym_symbol);
 758 }
 759
 760 static int maps__split_kallsyms_for_kcore(struct maps *kmaps, struct dso *dso)
 761 {
 762         struct map *curr_map;
 763         struct symbol *pos;
 764         int count = 0;
 765         struct rb_root_cached old_root = dso->symbols;
 766         struct rb_root_cached *root = &dso->symbols;
 767         struct rb_node *next = rb_first_cached(root);
 768
 769         if (!kmaps)
 770                 return -1;
 771
 772         *root = RB_ROOT_CACHED;
 773
 774         while (next) {
 775                 char *module;
 776
 777                 pos = rb_entry(next, struct symbol, rb_node);
 778                 next = rb_next(&pos->rb_node);
 779
 780                 rb_erase_cached(&pos->rb_node, &old_root);
 781                 RB_CLEAR_NODE(&pos->rb_node);
 782                 module = strchr(pos->name, '\t');
 783                 if (module)
 784                         *module = '\0';
 785
 786                 curr_map = maps__find(kmaps, pos->start);
 787
 788                 if (!curr_map) {
 789                         symbol__delete(pos);
 790                         continue;
 791                 }
 792
 793                 pos->start -= curr_map->start - curr_map->pgoff;
 794                 if (pos->end > curr_map->end)
 795                         pos->end = curr_map->end;
 796                 if (pos->end)
 797                         pos->end -= curr_map->start - curr_map->pgoff;
 798                 symbols__insert(&curr_map->dso->symbols, pos);
 799                 ++count;
 800         }
 801
 802         /* Symbols have been adjusted */
 803         dso->adjust_symbols = 1;
 804
 805         return count;
 806 }
 807
 808 /*
 809  * Split the symbols into maps, making sure there are no overlaps, i.e. the
 810  * kernel range is broken in several maps, named [kernel].N, as we don't have
 811  * the original ELF section names vmlinux have.
 812  */
 813 static int maps__split_kallsyms(struct maps *kmaps, struct dso *dso, u64 delta,
 814                                 struct map *initial_map)
 815 {
 816         struct machine *machine;
 817         struct map *curr_map = initial_map;
 818         struct symbol *pos;
 819         int count = 0, moved = 0;
 820         struct rb_root_cached *root = &dso->symbols;
 821         struct rb_node *next = rb_first_cached(root);
 822         int kernel_range = 0;
 823         bool x86_64;
 824
 825         if (!kmaps)
 826                 return -1;
 827
 828         machine = kmaps->machine;
 829
 830         x86_64 = machine__is(machine, "x86_64");
 831
 832         while (next) {
 833                 char *module;
 834
 835                 pos = rb_entry(next, struct symbol, rb_node);
 836                 next = rb_next(&pos->rb_node);
 837
 838                 module = strchr(pos->name, '\t');
 839                 if (module) {
 840                         if (!symbol_conf.use_modules)
 841                                 goto discard_symbol;
 842
 843                         *module++ = '\0';
 844
 845                         if (strcmp(curr_map->dso->short_name, module)) {
 846                                 if (curr_map != initial_map &&
 847                                     dso->kernel == DSO_SPACE__KERNEL_GUEST &&
 848                                     machine__is_default_guest(machine)) {
 849                                         /*
 850                                          * We assume all symbols of a module are
 851                                          * continuous in * kallsyms, so curr_map
 852                                          * points to a module and all its
 853                                          * symbols are in its kmap. Mark it as
 854                                          * loaded.
 855                                          */
 856                                         dso__set_loaded(curr_map->dso);
 857                                 }
 858
 859                                 curr_map = maps__find_by_name(kmaps, module);
 860                                 if (curr_map == NULL) {
 861                                         pr_debug("%s/proc/{kallsyms,modules} "
 862                                                  "inconsistency while looking "
 863                                                  "for \"%s\" module!\n",
 864                                                  machine->root_dir, module);
 865                                         curr_map = initial_map;
 866                                         goto discard_symbol;
 867                                 }
 868
 869                                 if (curr_map->dso->loaded &&
 870                                     !machine__is_default_guest(machine))
 871                                         goto discard_symbol;
 872                         }
 873                         /*
 874                          * So that we look just like we get from .ko files,
 875                          * i.e. not prelinked, relative to initial_map->start.
 876                          */
 877                         pos->start = curr_map->map_ip(curr_map, pos->start);
 878                         pos->end   = curr_map->map_ip(curr_map, pos->end);
 879                 } else if (x86_64 && is_entry_trampoline(pos->name)) {
 880                         /*
 881                          * These symbols are not needed anymore since the
 882                          * trampoline maps refer to the text section and it's
 883                          * symbols instead. Avoid having to deal with
 884                          * relocations, and the assumption that the first symbol
 885                          * is the start of kernel text, by simply removing the
 886                          * symbols at this point.
 887                          */
 888                         goto discard_symbol;
 889                 } else if (curr_map != initial_map) {
 890                         char dso_name[PATH_MAX];
 891                         struct dso *ndso;
 892
 893                         if (delta) {
 894                                 /* Kernel was relocated at boot time */
 895                                 pos->start -= delta;
 896                                 pos->end -= delta;
 897                         }
 898
 899                         if (count == 0) {
 900                                 curr_map = initial_map;
 901                                 goto add_symbol;
 902                         }
 903
 904                         if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
 905                                 snprintf(dso_name, sizeof(dso_name),
 906                                         "[guest.kernel].%d",
 907                                         kernel_range++);
 908                         else
 909                                 snprintf(dso_name, sizeof(dso_name),
 910                                         "[kernel].%d",
 911                                         kernel_range++);
 912
 913                         ndso = dso__new(dso_name);
 914                         if (ndso == NULL)
 915                                 return -1;
 916
 917                         ndso->kernel = dso->kernel;
 918
 919                         curr_map = map__new2(pos->start, ndso);
 920                         if (curr_map == NULL) {
 921                                 dso__put(ndso);
 922                                 return -1;
 923                         }
 924
 925                         curr_map->map_ip = curr_map->unmap_ip = identity__map_ip;
 926                         maps__insert(kmaps, curr_map);
 927                         ++kernel_range;
 928                 } else if (delta) {
 929                         /* Kernel was relocated at boot time */
 930                         pos->start -= delta;
 931                         pos->end -= delta;
 932                 }
 933 add_symbol:
 934                 if (curr_map != initial_map) {
 935                         rb_erase_cached(&pos->rb_node, root);
 936                         symbols__insert(&curr_map->dso->symbols, pos);
 937                         ++moved;
 938                 } else
 939                         ++count;
 940
 941                 continue;
 942 discard_symbol:
 943                 rb_erase_cached(&pos->rb_node, root);
 944                 symbol__delete(pos);
 945         }
 946
 947         if (curr_map != initial_map &&
 948             dso->kernel == DSO_SPACE__KERNEL_GUEST &&
 949             machine__is_default_guest(kmaps->machine)) {
 950                 dso__set_loaded(curr_map->dso);
 951         }
 952
 953         return count + moved;
 954 }
 955
 956 bool symbol__restricted_filename(const char *filename,
 957                                  const char *restricted_filename)
 958 {
 959         bool restricted = false;
 960
 961         if (symbol_conf.kptr_restrict) {
 962                 char *r = realpath(filename, NULL);
 963
 964                 if (r != NULL) {
 965                         restricted = strcmp(r, restricted_filename) == 0;
 966                         free(r);
 967                         return restricted;
 968                 }
 969         }
 970
 971         return restricted;
 972 }
 973
 974 struct module_info {
 975         struct rb_node rb_node;
 976         char *name;
 977         u64 start;
 978 };
 979
 980 static void add_module(struct module_info *mi, struct rb_root *modules)
 981 {
 982         struct rb_node **p = &modules->rb_node;
 983         struct rb_node *parent = NULL;
 984         struct module_info *m;
 985
 986         while (*p != NULL) {
 987                 parent = *p;
 988                 m = rb_entry(parent, struct module_info, rb_node);
 989                 if (strcmp(mi->name, m->name) < 0)
 990                         p = &(*p)->rb_left;
 991                 else
 992                         p = &(*p)->rb_right;
 993         }
 994         rb_link_node(&mi->rb_node, parent, p);
 995         rb_insert_color(&mi->rb_node, modules);
 996 }
 997
 998 static void delete_modules(struct rb_root *modules)
 999 {
1000         struct module_info *mi;
1001         struct rb_node *next = rb_first(modules);
1002
1003         while (next) {
1004                 mi = rb_entry(next, struct module_info, rb_node);
1005                 next = rb_next(&mi->rb_node);
1006                 rb_erase(&mi->rb_node, modules);
1007                 zfree(&mi->name);
1008                 free(mi);
1009         }
1010 }
1011
1012 static struct module_info *find_module(const char *name,
1013                                        struct rb_root *modules)
1014 {
1015         struct rb_node *n = modules->rb_node;
1016
1017         while (n) {
1018                 struct module_info *m;
1019                 int cmp;
1020
1021                 m = rb_entry(n, struct module_info, rb_node);
1022                 cmp = strcmp(name, m->name);
1023                 if (cmp < 0)
1024                         n = n->rb_left;
1025                 else if (cmp > 0)
1026                         n = n->rb_right;
1027                 else
1028                         return m;
1029         }
1030
1031         return NULL;
1032 }
1033
1034 static int __read_proc_modules(void *arg, const char *name, u64 start,
1035                                u64 size __maybe_unused)
1036 {
1037         struct rb_root *modules = arg;
1038         struct module_info *mi;
1039
1040         mi = zalloc(sizeof(struct module_info));
1041         if (!mi)
1042                 return -ENOMEM;
1043
1044         mi->name = strdup(name);
1045         mi->start = start;
1046
1047         if (!mi->name) {
1048                 free(mi);
1049                 return -ENOMEM;
1050         }
1051
1052         add_module(mi, modules);
1053
1054         return 0;
1055 }
1056
1057 static int read_proc_modules(const char *filename, struct rb_root *modules)
1058 {
1059         if (symbol__restricted_filename(filename, "/proc/modules"))
1060                 return -1;
1061
1062         if (modules__parse(filename, modules, __read_proc_modules)) {
1063                 delete_modules(modules);
1064                 return -1;
1065         }
1066
1067         return 0;
1068 }
1069
1070 int compare_proc_modules(const char *from, const char *to)
1071 {
1072         struct rb_root from_modules = RB_ROOT;
1073         struct rb_root to_modules = RB_ROOT;
1074         struct rb_node *from_node, *to_node;
1075         struct module_info *from_m, *to_m;
1076         int ret = -1;
1077
1078         if (read_proc_modules(from, &from_modules))
1079                 return -1;
1080
1081         if (read_proc_modules(to, &to_modules))
1082                 goto out_delete_from;
1083
1084         from_node = rb_first(&from_modules);
1085         to_node = rb_first(&to_modules);
1086         while (from_node) {
1087                 if (!to_node)
1088                         break;
1089
1090                 from_m = rb_entry(from_node, struct module_info, rb_node);
1091                 to_m = rb_entry(to_node, struct module_info, rb_node);
1092
1093                 if (from_m->start != to_m->start ||
1094                     strcmp(from_m->name, to_m->name))
1095                         break;
1096
1097                 from_node = rb_next(from_node);
1098                 to_node = rb_next(to_node);
1099         }
1100
1101         if (!from_node && !to_node)
1102                 ret = 0;
1103
1104         delete_modules(&to_modules);
1105 out_delete_from:
1106         delete_modules(&from_modules);
1107
1108         return ret;
1109 }
1110
1111 static int do_validate_kcore_modules(const char *filename, struct maps *kmaps)
1112 {
1113         struct rb_root modules = RB_ROOT;
1114         struct map *old_map;
1115         int err;
1116
1117         err = read_proc_modules(filename, &modules);
1118         if (err)
1119                 return err;
1120
1121         maps__for_each_entry(kmaps, old_map) {
1122                 struct module_info *mi;
1123
1124                 if (!__map__is_kmodule(old_map)) {
1125                         continue;
1126                 }
1127
1128                 /* Module must be in memory at the same address */
1129                 mi = find_module(old_map->dso->short_name, &modules);
1130                 if (!mi || mi->start != old_map->start) {
1131                         err = -EINVAL;
1132                         goto out;
1133                 }
1134         }
1135 out:
1136         delete_modules(&modules);
1137         return err;
1138 }
1139
1140 /*
1141  * If kallsyms is referenced by name then we look for filename in the same
1142  * directory.
1143  */
1144 static bool filename_from_kallsyms_filename(char *filename,
1145                                             const char *base_name,
1146                                             const char *kallsyms_filename)
1147 {
1148         char *name;
1149
1150         strcpy(filename, kallsyms_filename);
1151         name = strrchr(filename, '/');
1152         if (!name)
1153                 return false;
1154
1155         name += 1;
1156
1157         if (!strcmp(name, "kallsyms")) {
1158                 strcpy(name, base_name);
1159                 return true;
1160         }
1161
1162         return false;
1163 }
1164
1165 static int validate_kcore_modules(const char *kallsyms_filename,
1166                                   struct map *map)
1167 {
1168         struct maps *kmaps = map__kmaps(map);
1169         char modules_filename[PATH_MAX];
1170
1171         if (!kmaps)
1172                 return -EINVAL;
1173
1174         if (!filename_from_kallsyms_filename(modules_filename, "modules",
1175                                              kallsyms_filename))
1176                 return -EINVAL;
1177
1178         if (do_validate_kcore_modules(modules_filename, kmaps))
1179                 return -EINVAL;
1180
1181         return 0;
1182 }
1183
1184 static int validate_kcore_addresses(const char *kallsyms_filename,
1185                                     struct map *map)
1186 {
1187         struct kmap *kmap = map__kmap(map);
1188
1189         if (!kmap)
1190                 return -EINVAL;
1191
1192         if (kmap->ref_reloc_sym && kmap->ref_reloc_sym->name) {
1193                 u64 start;
1194
1195                 if (kallsyms__get_function_start(kallsyms_filename,
1196                                                  kmap->ref_reloc_sym->name, &start))
1197                         return -ENOENT;
1198                 if (start != kmap->ref_reloc_sym->addr)
1199                         return -EINVAL;
1200         }
1201
1202         return validate_kcore_modules(kallsyms_filename, map);
1203 }
1204
1205 struct kcore_mapfn_data {
1206         struct dso *dso;
1207         struct list_head maps;
1208 };
1209
1210 static int kcore_mapfn(u64 start, u64 len, u64 pgoff, void *data)
1211 {
1212         struct kcore_mapfn_data *md = data;
1213         struct map *map;
1214
1215         map = map__new2(start, md->dso);
1216         if (map == NULL)
1217                 return -ENOMEM;
1218
1219         map->end = map->start + len;
1220         map->pgoff = pgoff;
1221
1222         list_add(&map->node, &md->maps);
1223
1224         return 0;
1225 }
1226
1227 /*
1228  * Merges map into maps by splitting the new map within the existing map
1229  * regions.
1230  */
1231 int maps__merge_in(struct maps *kmaps, struct map *new_map)
1232 {
1233         struct map *old_map;
1234         LIST_HEAD(merged);
1235
1236         maps__for_each_entry(kmaps, old_map) {
1237                 /* no overload with this one */
1238                 if (new_map->end < old_map->start ||
1239                     new_map->start >= old_map->end)
1240                         continue;
1241
1242                 if (new_map->start < old_map->start) {
1243                         /*
1244                          * |new......
1245                          *       |old....
1246                          */
1247                         if (new_map->end < old_map->end) {
1248                                 /*
1249                                  * |new......|     -> |new..|
1250                                  *       |old....| ->       |old....|
1251                                  */
1252                                 new_map->end = old_map->start;
1253                         } else {
1254                                 /*
1255                                  * |new.............| -> |new..|       |new..|
1256                                  *       |old....|    ->       |old....|
1257                                  */
1258                                 struct map *m = map__clone(new_map);
1259
1260                                 if (!m)
1261                                         return -ENOMEM;
1262
1263                                 m->end = old_map->start;
1264                                 list_add_tail(&m->node, &merged);
1265                                 new_map->pgoff += old_map->end - new_map->start;
1266                                 new_map->start = old_map->end;
1267                         }
1268                 } else {
1269                         /*
1270                          *      |new......
1271                          * |old....
1272                          */
1273                         if (new_map->end < old_map->end) {
1274                                 /*
1275                                  *      |new..|   -> x
1276                                  * |old.........| -> |old.........|
1277                                  */
1278                                 map__put(new_map);
1279                                 new_map = NULL;
1280                                 break;
1281                         } else {
1282                                 /*
1283                                  *      |new......| ->         |new...|
1284                                  * |old....|        -> |old....|
1285                                  */
1286                                 new_map->pgoff += old_map->end - new_map->start;
1287                                 new_map->start = old_map->end;
1288                         }
1289                 }
1290         }
1291
1292         while (!list_empty(&merged)) {
1293                 old_map = list_entry(merged.next, struct map, node);
1294                 list_del_init(&old_map->node);
1295                 maps__insert(kmaps, old_map);
1296                 map__put(old_map);
1297         }
1298
1299         if (new_map) {
1300                 maps__insert(kmaps, new_map);
1301                 map__put(new_map);
1302         }
1303         return 0;
1304 }
1305
1306 static int dso__load_kcore(struct dso *dso, struct map *map,
1307                            const char *kallsyms_filename)
1308 {
1309         struct maps *kmaps = map__kmaps(map);
1310         struct kcore_mapfn_data md;
1311         struct map *old_map, *new_map, *replacement_map = NULL, *next;
1312         struct machine *machine;
1313         bool is_64_bit;
1314         int err, fd;
1315         char kcore_filename[PATH_MAX];
1316         u64 stext;
1317
1318         if (!kmaps)
1319                 return -EINVAL;
1320
1321         machine = kmaps->machine;
1322
1323         /* This function requires that the map is the kernel map */
1324         if (!__map__is_kernel(map))
1325                 return -EINVAL;
1326
1327         if (!filename_from_kallsyms_filename(kcore_filename, "kcore",
1328                                              kallsyms_filename))
1329                 return -EINVAL;
1330
1331         /* Modules and kernel must be present at their original addresses */
1332         if (validate_kcore_addresses(kallsyms_filename, map))
1333                 return -EINVAL;
1334
1335         md.dso = dso;
1336         INIT_LIST_HEAD(&md.maps);
1337
1338         fd = open(kcore_filename, O_RDONLY);
1339         if (fd < 0) {
1340                 pr_debug("Failed to open %s. Note /proc/kcore requires CAP_SYS_RAWIO capability to access.\n",
1341                          kcore_filename);
1342                 return -EINVAL;
1343         }
1344
1345         /* Read new maps into temporary lists */
1346         err = file__read_maps(fd, map->prot & PROT_EXEC, kcore_mapfn, &md,
1347                               &is_64_bit);
1348         if (err)
1349                 goto out_err;
1350         dso->is_64_bit = is_64_bit;
1351
1352         if (list_empty(&md.maps)) {
1353                 err = -EINVAL;
1354                 goto out_err;
1355         }
1356
1357         /* Remove old maps */
1358         maps__for_each_entry_safe(kmaps, old_map, next) {
1359                 /*
1360                  * We need to preserve eBPF maps even if they are
1361                  * covered by kcore, because we need to access
1362                  * eBPF dso for source data.
1363                  */
1364                 if (old_map != map && !__map__is_bpf_prog(old_map))
1365                         maps__remove(kmaps, old_map);
1366         }
1367         machine->trampolines_mapped = false;
1368
1369         /* Find the kernel map using the '_stext' symbol */
1370         if (!kallsyms__get_function_start(kallsyms_filename, "_stext", &stext)) {
1371                 list_for_each_entry(new_map, &md.maps, node) {
1372                         if (stext >= new_map->start && stext < new_map->end) {
1373                                 replacement_map = new_map;
1374                                 break;
1375                         }
1376                 }
1377         }
1378
1379         if (!replacement_map)
1380                 replacement_map = list_entry(md.maps.next, struct map, node);
1381
1382         /* Add new maps */
1383         while (!list_empty(&md.maps)) {
1384                 new_map = list_entry(md.maps.next, struct map, node);
1385                 list_del_init(&new_map->node);
1386                 if (new_map == replacement_map) {
1387                         map->start      = new_map->start;
1388                         map->end        = new_map->end;
1389                         map->pgoff      = new_map->pgoff;
1390                         map->map_ip     = new_map->map_ip;
1391                         map->unmap_ip   = new_map->unmap_ip;
1392                         /* Ensure maps are correctly ordered */
1393                         map__get(map);
1394                         maps__remove(kmaps, map);
1395                         maps__insert(kmaps, map);
1396                         map__put(map);
1397                         map__put(new_map);
1398                 } else {
1399                         /*
1400                          * Merge kcore map into existing maps,
1401                          * and ensure that current maps (eBPF)
1402                          * stay intact.
1403                          */
1404                         if (maps__merge_in(kmaps, new_map))
1405                                 goto out_err;
1406                 }
1407         }
1408
1409         if (machine__is(machine, "x86_64")) {
1410                 u64 addr;
1411
1412                 /*
1413                  * If one of the corresponding symbols is there, assume the
1414                  * entry trampoline maps are too.
1415                  */
1416                 if (!kallsyms__get_function_start(kallsyms_filename,
1417                                                   ENTRY_TRAMPOLINE_NAME,
1418                                                   &addr))
1419                         machine->trampolines_mapped = true;
1420         }
1421
1422         /*
1423          * Set the data type and long name so that kcore can be read via
1424          * dso__data_read_addr().
1425          */
1426         if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
1427                 dso->binary_type = DSO_BINARY_TYPE__GUEST_KCORE;
1428         else
1429                 dso->binary_type = DSO_BINARY_TYPE__KCORE;
1430         dso__set_long_name(dso, strdup(kcore_filename), true);
1431
1432         close(fd);
1433
1434         if (map->prot & PROT_EXEC)
1435                 pr_debug("Using %s for kernel object code\n", kcore_filename);
1436         else
1437                 pr_debug("Using %s for kernel data\n", kcore_filename);
1438
1439         return 0;
1440
1441 out_err:
1442         while (!list_empty(&md.maps)) {
1443                 map = list_entry(md.maps.next, struct map, node);
1444                 list_del_init(&map->node);
1445                 map__put(map);
1446         }
1447         close(fd);
1448         return -EINVAL;
1449 }
1450
1451 /*
1452  * If the kernel is relocated at boot time, kallsyms won't match.  Compute the
1453  * delta based on the relocation reference symbol.
1454  */
1455 static int kallsyms__delta(struct kmap *kmap, const char *filename, u64 *delta)
1456 {
1457         u64 addr;
1458
1459         if (!kmap->ref_reloc_sym || !kmap->ref_reloc_sym->name)
1460                 return 0;
1461
1462         if (kallsyms__get_function_start(filename, kmap->ref_reloc_sym->name, &addr))
1463                 return -1;
1464
1465         *delta = addr - kmap->ref_reloc_sym->addr;
1466         return 0;
1467 }
1468
1469 int __dso__load_kallsyms(struct dso *dso, const char *filename,
1470                          struct map *map, bool no_kcore)
1471 {
1472         struct kmap *kmap = map__kmap(map);
1473         u64 delta = 0;
1474
1475         if (symbol__restricted_filename(filename, "/proc/kallsyms"))
1476                 return -1;
1477
1478         if (!kmap || !kmap->kmaps)
1479                 return -1;
1480
1481         if (dso__load_all_kallsyms(dso, filename) < 0)
1482                 return -1;
1483
1484         if (kallsyms__delta(kmap, filename, &delta))
1485                 return -1;
1486
1487         symbols__fixup_end(&dso->symbols, true);
1488         symbols__fixup_duplicate(&dso->symbols);
1489
1490         if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
1491                 dso->symtab_type = DSO_BINARY_TYPE__GUEST_KALLSYMS;
1492         else
1493                 dso->symtab_type = DSO_BINARY_TYPE__KALLSYMS;
1494
1495         if (!no_kcore && !dso__load_kcore(dso, map, filename))
1496                 return maps__split_kallsyms_for_kcore(kmap->kmaps, dso);
1497         else
1498                 return maps__split_kallsyms(kmap->kmaps, dso, delta, map);
1499 }
1500
1501 int dso__load_kallsyms(struct dso *dso, const char *filename,
1502                        struct map *map)
1503 {
1504         return __dso__load_kallsyms(dso, filename, map, false);
1505 }
1506
1507 static int dso__load_perf_map(const char *map_path, struct dso *dso)
1508 {
1509         char *line = NULL;
1510         size_t n;
1511         FILE *file;
1512         int nr_syms = 0;
1513
1514         file = fopen(map_path, "r");
1515         if (file == NULL)
1516                 goto out_failure;
1517
1518         while (!feof(file)) {
1519                 u64 start, size;
1520                 struct symbol *sym;
1521                 int line_len, len;
1522
1523                 line_len = getline(&line, &n, file);
1524                 if (line_len < 0)
1525                         break;
1526
1527                 if (!line)
1528                         goto out_failure;
1529
1530                 line[--line_len] = '\0'; /* \n */
1531
1532                 len = hex2u64(line, &start);
1533
1534                 len++;
1535                 if (len + 2 >= line_len)
1536                         continue;
1537
1538                 len += hex2u64(line + len, &size);
1539
1540                 len++;
1541                 if (len + 2 >= line_len)
1542                         continue;
1543
1544                 sym = symbol__new(start, size, STB_GLOBAL, STT_FUNC, line + len);
1545
1546                 if (sym == NULL)
1547                         goto out_delete_line;
1548
1549                 symbols__insert(&dso->symbols, sym);
1550                 nr_syms++;
1551         }
1552
1553         free(line);
1554         fclose(file);
1555
1556         return nr_syms;
1557
1558 out_delete_line:
1559         free(line);
1560 out_failure:
1561         return -1;
1562 }
1563
1564 #ifdef HAVE_LIBBFD_SUPPORT
1565 #define PACKAGE 'perf'
1566 #include <bfd.h>
1567
1568 static int bfd_symbols__cmpvalue(const void *a, const void *b)
1569 {
1570         const asymbol *as = *(const asymbol **)a, *bs = *(const asymbol **)b;
1571
1572         if (bfd_asymbol_value(as) != bfd_asymbol_value(bs))
1573                 return bfd_asymbol_value(as) - bfd_asymbol_value(bs);
1574
1575         return bfd_asymbol_name(as)[0] - bfd_asymbol_name(bs)[0];
1576 }
1577
1578 static int bfd2elf_binding(asymbol *symbol)
1579 {
1580         if (symbol->flags & BSF_WEAK)
1581                 return STB_WEAK;
1582         if (symbol->flags & BSF_GLOBAL)
1583                 return STB_GLOBAL;
1584         if (symbol->flags & BSF_LOCAL)
1585                 return STB_LOCAL;
1586         return -1;
1587 }
1588
1589 int dso__load_bfd_symbols(struct dso *dso, const char *debugfile)
1590 {
1591         int err = -1;
1592         long symbols_size, symbols_count, i;
1593         asection *section;
1594         asymbol **symbols, *sym;
1595         struct symbol *symbol;
1596         bfd *abfd;
1597         u64 start, len;
1598
1599         abfd = bfd_openr(debugfile, NULL);
1600         if (!abfd)
1601                 return -1;
1602
1603         if (!bfd_check_format(abfd, bfd_object)) {
1604                 pr_debug2("%s: cannot read %s bfd file.\n", __func__,
1605                           dso->long_name);
1606                 goto out_close;
1607         }
1608
1609         if (bfd_get_flavour(abfd) == bfd_target_elf_flavour)
1610                 goto out_close;
1611
1612         symbols_size = bfd_get_symtab_upper_bound(abfd);
1613         if (symbols_size == 0) {
1614                 bfd_close(abfd);
1615                 return 0;
1616         }
1617
1618         if (symbols_size < 0)
1619                 goto out_close;
1620
1621         symbols = malloc(symbols_size);
1622         if (!symbols)
1623                 goto out_close;
1624
1625         symbols_count = bfd_canonicalize_symtab(abfd, symbols);
1626         if (symbols_count < 0)
1627                 goto out_free;
1628
1629         section = bfd_get_section_by_name(abfd, ".text");
1630         if (section) {
1631                 for (i = 0; i < symbols_count; ++i) {
1632                         if (!strcmp(bfd_asymbol_name(symbols[i]), "__ImageBase") ||
1633                             !strcmp(bfd_asymbol_name(symbols[i]), "__image_base__"))
1634                                 break;
1635                 }
1636                 if (i < symbols_count) {
1637                         /* PE symbols can only have 4 bytes, so use .text high bits */
1638                         dso->text_offset = section->vma - (u32)section->vma;
1639                         dso->text_offset += (u32)bfd_asymbol_value(symbols[i]);
1640                 } else {
1641                         dso->text_offset = section->vma - section->filepos;
1642                 }
1643         }
1644
1645         qsort(symbols, symbols_count, sizeof(asymbol *), bfd_symbols__cmpvalue);
1646
1647 #ifdef bfd_get_section
1648 #define bfd_asymbol_section bfd_get_section
1649 #endif
1650         for (i = 0; i < symbols_count; ++i) {
1651                 sym = symbols[i];
1652                 section = bfd_asymbol_section(sym);
1653                 if (bfd2elf_binding(sym) < 0)
1654                         continue;
1655
1656                 while (i + 1 < symbols_count &&
1657                        bfd_asymbol_section(symbols[i + 1]) == section &&
1658                        bfd2elf_binding(symbols[i + 1]) < 0)
1659                         i++;
1660
1661                 if (i + 1 < symbols_count &&
1662                     bfd_asymbol_section(symbols[i + 1]) == section)
1663                         len = symbols[i + 1]->value - sym->value;
1664                 else
1665                         len = section->size - sym->value;
1666
1667                 start = bfd_asymbol_value(sym) - dso->text_offset;
1668                 symbol = symbol__new(start, len, bfd2elf_binding(sym), STT_FUNC,
1669                                      bfd_asymbol_name(sym));
1670                 if (!symbol)
1671                         goto out_free;
1672
1673                 symbols__insert(&dso->symbols, symbol);
1674         }
1675 #ifdef bfd_get_section
1676 #undef bfd_asymbol_section
1677 #endif
1678
1679         symbols__fixup_end(&dso->symbols, false);
1680         symbols__fixup_duplicate(&dso->symbols);
1681         dso->adjust_symbols = 1;
1682
1683         err = 0;
1684 out_free:
1685         free(symbols);
1686 out_close:
1687         bfd_close(abfd);
1688         return err;
1689 }
1690 #endif
1691
1692 static bool dso__is_compatible_symtab_type(struct dso *dso, bool kmod,
1693                                            enum dso_binary_type type)
1694 {
1695         switch (type) {
1696         case DSO_BINARY_TYPE__JAVA_JIT:
1697         case DSO_BINARY_TYPE__DEBUGLINK:
1698         case DSO_BINARY_TYPE__SYSTEM_PATH_DSO:
1699         case DSO_BINARY_TYPE__FEDORA_DEBUGINFO:
1700         case DSO_BINARY_TYPE__UBUNTU_DEBUGINFO:
1701         case DSO_BINARY_TYPE__MIXEDUP_UBUNTU_DEBUGINFO:
1702         case DSO_BINARY_TYPE__BUILDID_DEBUGINFO:
1703         case DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO:
1704                 return !kmod && dso->kernel == DSO_SPACE__USER;
1705
1706         case DSO_BINARY_TYPE__KALLSYMS:
1707         case DSO_BINARY_TYPE__VMLINUX:
1708         case DSO_BINARY_TYPE__KCORE:
1709                 return dso->kernel == DSO_SPACE__KERNEL;
1710
1711         case DSO_BINARY_TYPE__GUEST_KALLSYMS:
1712         case DSO_BINARY_TYPE__GUEST_VMLINUX:
1713         case DSO_BINARY_TYPE__GUEST_KCORE:
1714                 return dso->kernel == DSO_SPACE__KERNEL_GUEST;
1715
1716         case DSO_BINARY_TYPE__GUEST_KMODULE:
1717         case DSO_BINARY_TYPE__GUEST_KMODULE_COMP:
1718         case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE:
1719         case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP:
1720                 /*
1721                  * kernel modules know their symtab type - it's set when
1722                  * creating a module dso in machine__addnew_module_map().
1723                  */
1724                 return kmod && dso->symtab_type == type;
1725
1726         case DSO_BINARY_TYPE__BUILD_ID_CACHE:
1727         case DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO:
1728                 return true;
1729
1730         case DSO_BINARY_TYPE__BPF_PROG_INFO:
1731         case DSO_BINARY_TYPE__BPF_IMAGE:
1732         case DSO_BINARY_TYPE__OOL:
1733         case DSO_BINARY_TYPE__NOT_FOUND:
1734         default:
1735                 return false;
1736         }
1737 }
1738
1739 /* Checks for the existence of the perf-<pid>.map file in two different
1740  * locations.  First, if the process is a separate mount namespace, check in
1741  * that namespace using the pid of the innermost pid namespace.  If's not in a
1742  * namespace, or the file can't be found there, try in the mount namespace of
1743  * the tracing process using our view of its pid.
1744  */
1745 static int dso__find_perf_map(char *filebuf, size_t bufsz,
1746                               struct nsinfo **nsip)
1747 {
1748         struct nscookie nsc;
1749         struct nsinfo *nsi;
1750         struct nsinfo *nnsi;
1751         int rc = -1;
1752
1753         nsi = *nsip;
1754
1755         if (nsinfo__need_setns(nsi)) {
1756                 snprintf(filebuf, bufsz, "/tmp/perf-%d.map", nsinfo__nstgid(nsi));
1757                 nsinfo__mountns_enter(nsi, &nsc);
1758                 rc = access(filebuf, R_OK);
1759                 nsinfo__mountns_exit(&nsc);
1760                 if (rc == 0)
1761                         return rc;
1762         }
1763
1764         nnsi = nsinfo__copy(nsi);
1765         if (nnsi) {
1766                 nsinfo__put(nsi);
1767
1768                 nsinfo__clear_need_setns(nnsi);
1769                 snprintf(filebuf, bufsz, "/tmp/perf-%d.map", nsinfo__tgid(nnsi));
1770                 *nsip = nnsi;
1771                 rc = 0;
1772         }
1773
1774         return rc;
1775 }
1776
1777 int dso__load(struct dso *dso, struct map *map)
1778 {
1779         char *name;
1780         int ret = -1;
1781         u_int i;
1782         struct machine *machine = NULL;
1783         char *root_dir = (char *) "";
1784         int ss_pos = 0;
1785         struct symsrc ss_[2];
1786         struct symsrc *syms_ss = NULL, *runtime_ss = NULL;
1787         bool kmod;
1788         bool perfmap;
1789         struct build_id bid;
1790         struct nscookie nsc;
1791         char newmapname[PATH_MAX];
1792         const char *map_path = dso->long_name;
1793
1794         perfmap = strncmp(dso->name, "/tmp/perf-", 10) == 0;
1795         if (perfmap) {
1796                 if (dso->nsinfo && (dso__find_perf_map(newmapname,
1797                     sizeof(newmapname), &dso->nsinfo) == 0)) {
1798                         map_path = newmapname;
1799                 }
1800         }
1801
1802         nsinfo__mountns_enter(dso->nsinfo, &nsc);
1803         pthread_mutex_lock(&dso->lock);
1804
1805         /* check again under the dso->lock */
1806         if (dso__loaded(dso)) {
1807                 ret = 1;
1808                 goto out;
1809         }
1810
1811         kmod = dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1812                 dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP ||
1813                 dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE ||
1814                 dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE_COMP;
1815
1816         if (dso->kernel && !kmod) {
1817                 if (dso->kernel == DSO_SPACE__KERNEL)
1818                         ret = dso__load_kernel_sym(dso, map);
1819                 else if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
1820                         ret = dso__load_guest_kernel_sym(dso, map);
1821
1822                 machine = map__kmaps(map)->machine;
1823                 if (machine__is(machine, "x86_64"))
1824                         machine__map_x86_64_entry_trampolines(machine, dso);
1825                 goto out;
1826         }
1827
1828         dso->adjust_symbols = 0;
1829
1830         if (perfmap) {
1831                 ret = dso__load_perf_map(map_path, dso);
1832                 dso->symtab_type = ret > 0 ? DSO_BINARY_TYPE__JAVA_JIT :
1833                                              DSO_BINARY_TYPE__NOT_FOUND;
1834                 goto out;
1835         }
1836
1837         if (machine)
1838                 root_dir = machine->root_dir;
1839
1840         name = malloc(PATH_MAX);
1841         if (!name)
1842                 goto out;
1843
1844         /*
1845          * Read the build id if possible. This is required for
1846          * DSO_BINARY_TYPE__BUILDID_DEBUGINFO to work
1847          */
1848         if (!dso->has_build_id &&
1849             is_regular_file(dso->long_name)) {
1850             __symbol__join_symfs(name, PATH_MAX, dso->long_name);
1851                 if (filename__read_build_id(name, &bid) > 0)
1852                         dso__set_build_id(dso, &bid);
1853         }
1854
1855         /*
1856          * Iterate over candidate debug images.
1857          * Keep track of "interesting" ones (those which have a symtab, dynsym,
1858          * and/or opd section) for processing.
1859          */
1860         for (i = 0; i < DSO_BINARY_TYPE__SYMTAB_CNT; i++) {
1861                 struct symsrc *ss = &ss_[ss_pos];
1862                 bool next_slot = false;
1863                 bool is_reg;
1864                 bool nsexit;
1865                 int bfdrc = -1;
1866                 int sirc = -1;
1867
1868                 enum dso_binary_type symtab_type = binary_type_symtab[i];
1869
1870                 nsexit = (symtab_type == DSO_BINARY_TYPE__BUILD_ID_CACHE ||
1871                     symtab_type == DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO);
1872
1873                 if (!dso__is_compatible_symtab_type(dso, kmod, symtab_type))
1874                         continue;
1875
1876                 if (dso__read_binary_type_filename(dso, symtab_type,
1877                                                    root_dir, name, PATH_MAX))
1878                         continue;
1879
1880                 if (nsexit)
1881                         nsinfo__mountns_exit(&nsc);
1882
1883                 is_reg = is_regular_file(name);
1884                 if (!is_reg && errno == ENOENT && dso->nsinfo) {
1885                         char *new_name = filename_with_chroot(dso->nsinfo->pid,
1886                                                               name);
1887                         if (new_name) {
1888                                 is_reg = is_regular_file(new_name);
1889                                 strlcpy(name, new_name, PATH_MAX);
1890                                 free(new_name);
1891                         }
1892                 }
1893
1894 #ifdef HAVE_LIBBFD_SUPPORT
1895                 if (is_reg)
1896                         bfdrc = dso__load_bfd_symbols(dso, name);
1897 #endif
1898                 if (is_reg && bfdrc < 0)
1899                         sirc = symsrc__init(ss, dso, name, symtab_type);
1900
1901                 if (nsexit)
1902                         nsinfo__mountns_enter(dso->nsinfo, &nsc);
1903
1904                 if (bfdrc == 0) {
1905                         ret = 0;
1906                         break;
1907                 }
1908
1909                 if (!is_reg || sirc < 0)
1910                         continue;
1911
1912                 if (!syms_ss && symsrc__has_symtab(ss)) {
1913                         syms_ss = ss;
1914                         next_slot = true;
1915                         if (!dso->symsrc_filename)
1916                                 dso->symsrc_filename = strdup(name);
1917                 }
1918
1919                 if (!runtime_ss && symsrc__possibly_runtime(ss)) {
1920                         runtime_ss = ss;
1921                         next_slot = true;
1922                 }
1923
1924                 if (next_slot) {
1925                         ss_pos++;
1926
1927                         if (syms_ss && runtime_ss)
1928                                 break;
1929                 } else {
1930                         symsrc__destroy(ss);
1931                 }
1932
1933         }
1934
1935         if (!runtime_ss && !syms_ss)
1936                 goto out_free;
1937
1938         if (runtime_ss && !syms_ss) {
1939                 syms_ss = runtime_ss;
1940         }
1941
1942         /* We'll have to hope for the best */
1943         if (!runtime_ss && syms_ss)
1944                 runtime_ss = syms_ss;
1945
1946         if (syms_ss)
1947                 ret = dso__load_sym(dso, map, syms_ss, runtime_ss, kmod);
1948         else
1949                 ret = -1;
1950
1951         if (ret > 0) {
1952                 int nr_plt;
1953
1954                 nr_plt = dso__synthesize_plt_symbols(dso, runtime_ss);
1955                 if (nr_plt > 0)
1956                         ret += nr_plt;
1957         }
1958
1959         for (; ss_pos > 0; ss_pos--)
1960                 symsrc__destroy(&ss_[ss_pos - 1]);
1961 out_free:
1962         free(name);
1963         if (ret < 0 && strstr(dso->name, " (deleted)") != NULL)
1964                 ret = 0;
1965 out:
1966         dso__set_loaded(dso);
1967         pthread_mutex_unlock(&dso->lock);
1968         nsinfo__mountns_exit(&nsc);
1969
1970         return ret;
1971 }
1972
1973 static int map__strcmp(const void *a, const void *b)
1974 {
1975         const struct map *ma = *(const struct map **)a, *mb = *(const struct map **)b;
1976         return strcmp(ma->dso->short_name, mb->dso->short_name);
1977 }
1978
1979 static int map__strcmp_name(const void *name, const void *b)
1980 {
1981         const struct map *map = *(const struct map **)b;
1982         return strcmp(name, map->dso->short_name);
1983 }
1984
1985 void __maps__sort_by_name(struct maps *maps)
1986 {
1987         qsort(maps->maps_by_name, maps->nr_maps, sizeof(struct map *), map__strcmp);
1988 }
1989
1990 static int map__groups__sort_by_name_from_rbtree(struct maps *maps)
1991 {
1992         struct map *map;
1993         struct map **maps_by_name = realloc(maps->maps_by_name, maps->nr_maps * sizeof(map));
1994         int i = 0;
1995
1996         if (maps_by_name == NULL)
1997                 return -1;
1998
1999         maps->maps_by_name = maps_by_name;
2000         maps->nr_maps_allocated = maps->nr_maps;
2001
2002         maps__for_each_entry(maps, map)
2003                 maps_by_name[i++] = map;
2004
2005         __maps__sort_by_name(maps);
2006         return 0;
2007 }
2008
2009 static struct map *__maps__find_by_name(struct maps *maps, const char *name)
2010 {
2011         struct map **mapp;
2012
2013         if (maps->maps_by_name == NULL &&
2014             map__groups__sort_by_name_from_rbtree(maps))
2015                 return NULL;
2016
2017         mapp = bsearch(name, maps->maps_by_name, maps->nr_maps, sizeof(*mapp), map__strcmp_name);
2018         if (mapp)
2019                 return *mapp;
2020         return NULL;
2021 }
2022
2023 struct map *maps__find_by_name(struct maps *maps, const char *name)
2024 {
2025         struct map *map;
2026
2027         down_read(&maps->lock);
2028
2029         if (maps->last_search_by_name && strcmp(maps->last_search_by_name->dso->short_name, name) == 0) {
2030                 map = maps->last_search_by_name;
2031                 goto out_unlock;
2032         }
2033         /*
2034          * If we have maps->maps_by_name, then the name isn't in the rbtree,
2035          * as maps->maps_by_name mirrors the rbtree when lookups by name are
2036          * made.
2037          */
2038         map = __maps__find_by_name(maps, name);
2039         if (map || maps->maps_by_name != NULL)
2040                 goto out_unlock;
2041
2042         /* Fallback to traversing the rbtree... */
2043         maps__for_each_entry(maps, map)
2044                 if (strcmp(map->dso->short_name, name) == 0) {
2045                         maps->last_search_by_name = map;
2046                         goto out_unlock;
2047                 }
2048
2049         map = NULL;
2050
2051 out_unlock:
2052         up_read(&maps->lock);
2053         return map;
2054 }
2055
2056 int dso__load_vmlinux(struct dso *dso, struct map *map,
2057                       const char *vmlinux, bool vmlinux_allocated)
2058 {
2059         int err = -1;
2060         struct symsrc ss;
2061         char symfs_vmlinux[PATH_MAX];
2062         enum dso_binary_type symtab_type;
2063
2064         if (vmlinux[0] == '/')
2065                 snprintf(symfs_vmlinux, sizeof(symfs_vmlinux), "%s", vmlinux);
2066         else
2067                 symbol__join_symfs(symfs_vmlinux, vmlinux);
2068
2069         if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
2070                 symtab_type = DSO_BINARY_TYPE__GUEST_VMLINUX;
2071         else
2072                 symtab_type = DSO_BINARY_TYPE__VMLINUX;
2073
2074         if (symsrc__init(&ss, dso, symfs_vmlinux, symtab_type))
2075                 return -1;
2076
2077         err = dso__load_sym(dso, map, &ss, &ss, 0);
2078         symsrc__destroy(&ss);
2079
2080         if (err > 0) {
2081                 if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
2082                         dso->binary_type = DSO_BINARY_TYPE__GUEST_VMLINUX;
2083                 else
2084                         dso->binary_type = DSO_BINARY_TYPE__VMLINUX;
2085                 dso__set_long_name(dso, vmlinux, vmlinux_allocated);
2086                 dso__set_loaded(dso);
2087                 pr_debug("Using %s for symbols\n", symfs_vmlinux);
2088         }
2089
2090         return err;
2091 }
2092
2093 int dso__load_vmlinux_path(struct dso *dso, struct map *map)
2094 {
2095         int i, err = 0;
2096         char *filename = NULL;
2097
2098         pr_debug("Looking at the vmlinux_path (%d entries long)\n",
2099                  vmlinux_path__nr_entries + 1);
2100
2101         for (i = 0; i < vmlinux_path__nr_entries; ++i) {
2102                 err = dso__load_vmlinux(dso, map, vmlinux_path[i], false);
2103                 if (err > 0)
2104                         goto out;
2105         }
2106
2107         if (!symbol_conf.ignore_vmlinux_buildid)
2108                 filename = dso__build_id_filename(dso, NULL, 0, false);
2109         if (filename != NULL) {
2110                 err = dso__load_vmlinux(dso, map, filename, true);
2111                 if (err > 0)
2112                         goto out;
2113                 free(filename);
2114         }
2115 out:
2116         return err;
2117 }
2118
2119 static bool visible_dir_filter(const char *name, struct dirent *d)
2120 {
2121         if (d->d_type != DT_DIR)
2122                 return false;
2123         return lsdir_no_dot_filter(name, d);
2124 }
2125
2126 static int find_matching_kcore(struct map *map, char *dir, size_t dir_sz)
2127 {
2128         char kallsyms_filename[PATH_MAX];
2129         int ret = -1;
2130         struct strlist *dirs;
2131         struct str_node *nd;
2132
2133         dirs = lsdir(dir, visible_dir_filter);
2134         if (!dirs)
2135                 return -1;
2136
2137         strlist__for_each_entry(nd, dirs) {
2138                 scnprintf(kallsyms_filename, sizeof(kallsyms_filename),
2139                           "%s/%s/kallsyms", dir, nd->s);
2140                 if (!validate_kcore_addresses(kallsyms_filename, map)) {
2141                         strlcpy(dir, kallsyms_filename, dir_sz);
2142                         ret = 0;
2143                         break;
2144                 }
2145         }
2146
2147         strlist__delete(dirs);
2148
2149         return ret;
2150 }
2151
2152 /*
2153  * Use open(O_RDONLY) to check readability directly instead of access(R_OK)
2154  * since access(R_OK) only checks with real UID/GID but open() use effective
2155  * UID/GID and actual capabilities (e.g. /proc/kcore requires CAP_SYS_RAWIO).
2156  */
2157 static bool filename__readable(const char *file)
2158 {
2159         int fd = open(file, O_RDONLY);
2160         if (fd < 0)
2161                 return false;
2162         close(fd);
2163         return true;
2164 }
2165
2166 static char *dso__find_kallsyms(struct dso *dso, struct map *map)
2167 {
2168         struct build_id bid;
2169         char sbuild_id[SBUILD_ID_SIZE];
2170         bool is_host = false;
2171         char path[PATH_MAX];
2172
2173         if (!dso->has_build_id) {
2174                 /*
2175                  * Last resort, if we don't have a build-id and couldn't find
2176                  * any vmlinux file, try the running kernel kallsyms table.
2177                  */
2178                 goto proc_kallsyms;
2179         }
2180
2181         if (sysfs__read_build_id("/sys/kernel/notes", &bid) == 0)
2182                 is_host = dso__build_id_equal(dso, &bid);
2183
2184         /* Try a fast path for /proc/kallsyms if possible */
2185         if (is_host) {
2186                 /*
2187                  * Do not check the build-id cache, unless we know we cannot use
2188                  * /proc/kcore or module maps don't match to /proc/kallsyms.
2189                  * To check readability of /proc/kcore, do not use access(R_OK)
2190                  * since /proc/kcore requires CAP_SYS_RAWIO to read and access
2191                  * can't check it.
2192                  */
2193                 if (filename__readable("/proc/kcore") &&
2194                     !validate_kcore_addresses("/proc/kallsyms", map))
2195                         goto proc_kallsyms;
2196         }
2197
2198         build_id__sprintf(&dso->bid, sbuild_id);
2199
2200         /* Find kallsyms in build-id cache with kcore */
2201         scnprintf(path, sizeof(path), "%s/%s/%s",
2202                   buildid_dir, DSO__NAME_KCORE, sbuild_id);
2203
2204         if (!find_matching_kcore(map, path, sizeof(path)))
2205                 return strdup(path);
2206
2207         /* Use current /proc/kallsyms if possible */
2208         if (is_host) {
2209 proc_kallsyms:
2210                 return strdup("/proc/kallsyms");
2211         }
2212
2213         /* Finally, find a cache of kallsyms */
2214         if (!build_id_cache__kallsyms_path(sbuild_id, path, sizeof(path))) {
2215                 pr_err("No kallsyms or vmlinux with build-id %s was found\n",
2216                        sbuild_id);
2217                 return NULL;
2218         }
2219
2220         return strdup(path);
2221 }
2222
2223 static int dso__load_kernel_sym(struct dso *dso, struct map *map)
2224 {
2225         int err;
2226         const char *kallsyms_filename = NULL;
2227         char *kallsyms_allocated_filename = NULL;
2228         char *filename = NULL;
2229
2230         /*
2231          * Step 1: if the user specified a kallsyms or vmlinux filename, use
2232          * it and only it, reporting errors to the user if it cannot be used.
2233          *
2234          * For instance, try to analyse an ARM perf.data file _without_ a
2235          * build-id, or if the user specifies the wrong path to the right
2236          * vmlinux file, obviously we can't fallback to another vmlinux (a
2237          * x86_86 one, on the machine where analysis is being performed, say),
2238          * or worse, /proc/kallsyms.
2239          *
2240          * If the specified file _has_ a build-id and there is a build-id
2241          * section in the perf.data file, we will still do the expected
2242          * validation in dso__load_vmlinux and will bail out if they don't
2243          * match.
2244          */
2245         if (symbol_conf.kallsyms_name != NULL) {
2246                 kallsyms_filename = symbol_conf.kallsyms_name;
2247                 goto do_kallsyms;
2248         }
2249
2250         if (!symbol_conf.ignore_vmlinux && symbol_conf.vmlinux_name != NULL) {
2251                 return dso__load_vmlinux(dso, map, symbol_conf.vmlinux_name, false);
2252         }
2253
2254         /*
2255          * Before checking on common vmlinux locations, check if it's
2256          * stored as standard build id binary (not kallsyms) under
2257          * .debug cache.
2258          */
2259         if (!symbol_conf.ignore_vmlinux_buildid)
2260                 filename = __dso__build_id_filename(dso, NULL, 0, false, false);
2261         if (filename != NULL) {
2262                 err = dso__load_vmlinux(dso, map, filename, true);
2263                 if (err > 0)
2264                         return err;
2265                 free(filename);
2266         }
2267
2268         if (!symbol_conf.ignore_vmlinux && vmlinux_path != NULL) {
2269                 err = dso__load_vmlinux_path(dso, map);
2270                 if (err > 0)
2271                         return err;
2272         }
2273
2274         /* do not try local files if a symfs was given */
2275         if (symbol_conf.symfs[0] != 0)
2276                 return -1;
2277
2278         kallsyms_allocated_filename = dso__find_kallsyms(dso, map);
2279         if (!kallsyms_allocated_filename)
2280                 return -1;
2281
2282         kallsyms_filename = kallsyms_allocated_filename;
2283
2284 do_kallsyms:
2285         err = dso__load_kallsyms(dso, kallsyms_filename, map);
2286         if (err > 0)
2287                 pr_debug("Using %s for symbols\n", kallsyms_filename);
2288         free(kallsyms_allocated_filename);
2289
2290         if (err > 0 && !dso__is_kcore(dso)) {
2291                 dso->binary_type = DSO_BINARY_TYPE__KALLSYMS;
2292                 dso__set_long_name(dso, DSO__NAME_KALLSYMS, false);
2293                 map__fixup_start(map);
2294                 map__fixup_end(map);
2295         }
2296
2297         return err;
2298 }
2299
2300 static int dso__load_guest_kernel_sym(struct dso *dso, struct map *map)
2301 {
2302         int err;
2303         const char *kallsyms_filename = NULL;
2304         struct machine *machine = map__kmaps(map)->machine;
2305         char path[PATH_MAX];
2306
2307         if (machine__is_default_guest(machine)) {
2308                 /*
2309                  * if the user specified a vmlinux filename, use it and only
2310                  * it, reporting errors to the user if it cannot be used.
2311                  * Or use file guest_kallsyms inputted by user on commandline
2312                  */
2313                 if (symbol_conf.default_guest_vmlinux_name != NULL) {
2314                         err = dso__load_vmlinux(dso, map,
2315                                                 symbol_conf.default_guest_vmlinux_name,
2316                                                 false);
2317                         return err;
2318                 }
2319
2320                 kallsyms_filename = symbol_conf.default_guest_kallsyms;
2321                 if (!kallsyms_filename)
2322                         return -1;
2323         } else {
2324                 sprintf(path, "%s/proc/kallsyms", machine->root_dir);
2325                 kallsyms_filename = path;
2326         }
2327
2328         err = dso__load_kallsyms(dso, kallsyms_filename, map);
2329         if (err > 0)
2330                 pr_debug("Using %s for symbols\n", kallsyms_filename);
2331         if (err > 0 && !dso__is_kcore(dso)) {
2332                 dso->binary_type = DSO_BINARY_TYPE__GUEST_KALLSYMS;
2333                 dso__set_long_name(dso, machine->mmap_name, false);
2334                 map__fixup_start(map);
2335                 map__fixup_end(map);
2336         }
2337
2338         return err;
2339 }
2340
2341 static void vmlinux_path__exit(void)
2342 {
2343         while (--vmlinux_path__nr_entries >= 0)
2344                 zfree(&vmlinux_path[vmlinux_path__nr_entries]);
2345         vmlinux_path__nr_entries = 0;
2346
2347         zfree(&vmlinux_path);
2348 }
2349
2350 static const char * const vmlinux_paths[] = {
2351         "vmlinux",
2352         "/boot/vmlinux"
2353 };
2354
2355 static const char * const vmlinux_paths_upd[] = {
2356         "/boot/vmlinux-%s",
2357         "/usr/lib/debug/boot/vmlinux-%s",
2358         "/lib/modules/%s/build/vmlinux",
2359         "/usr/lib/debug/lib/modules/%s/vmlinux",
2360         "/usr/lib/debug/boot/vmlinux-%s.debug"
2361 };
2362
2363 static int vmlinux_path__add(const char *new_entry)
2364 {
2365         vmlinux_path[vmlinux_path__nr_entries] = strdup(new_entry);
2366         if (vmlinux_path[vmlinux_path__nr_entries] == NULL)
2367                 return -1;
2368         ++vmlinux_path__nr_entries;
2369
2370         return 0;
2371 }
2372
2373 static int vmlinux_path__init(struct perf_env *env)
2374 {
2375         struct utsname uts;
2376         char bf[PATH_MAX];
2377         char *kernel_version;
2378         unsigned int i;
2379
2380         vmlinux_path = malloc(sizeof(char *) * (ARRAY_SIZE(vmlinux_paths) +
2381                               ARRAY_SIZE(vmlinux_paths_upd)));
2382         if (vmlinux_path == NULL)
2383                 return -1;
2384
2385         for (i = 0; i < ARRAY_SIZE(vmlinux_paths); i++)
2386                 if (vmlinux_path__add(vmlinux_paths[i]) < 0)
2387                         goto out_fail;
2388
2389         /* only try kernel version if no symfs was given */
2390         if (symbol_conf.symfs[0] != 0)
2391                 return 0;
2392
2393         if (env) {
2394                 kernel_version = env->os_release;
2395         } else {
2396                 if (uname(&uts) < 0)
2397                         goto out_fail;
2398
2399                 kernel_version = uts.release;
2400         }
2401
2402         for (i = 0; i < ARRAY_SIZE(vmlinux_paths_upd); i++) {
2403                 snprintf(bf, sizeof(bf), vmlinux_paths_upd[i], kernel_version);
2404                 if (vmlinux_path__add(bf) < 0)
2405                         goto out_fail;
2406         }
2407
2408         return 0;
2409
2410 out_fail:
2411         vmlinux_path__exit();
2412         return -1;
2413 }
2414
2415 int setup_list(struct strlist **list, const char *list_str,
2416                       const char *list_name)
2417 {
2418         if (list_str == NULL)
2419                 return 0;
2420
2421         *list = strlist__new(list_str, NULL);
2422         if (!*list) {
2423                 pr_err("problems parsing %s list\n", list_name);
2424                 return -1;
2425         }
2426
2427         symbol_conf.has_filter = true;
2428         return 0;
2429 }
2430
2431 int setup_intlist(struct intlist **list, const char *list_str,
2432                   const char *list_name)
2433 {
2434         if (list_str == NULL)
2435                 return 0;
2436
2437         *list = intlist__new(list_str);
2438         if (!*list) {
2439                 pr_err("problems parsing %s list\n", list_name);
2440                 return -1;
2441         }
2442         return 0;
2443 }
2444
2445 static int setup_addrlist(struct intlist **addr_list, struct strlist *sym_list)
2446 {
2447         struct str_node *pos, *tmp;
2448         unsigned long val;
2449         char *sep;
2450         const char *end;
2451         int i = 0, err;
2452
2453         *addr_list = intlist__new(NULL);
2454         if (!*addr_list)
2455                 return -1;
2456
2457         strlist__for_each_entry_safe(pos, tmp, sym_list) {
2458                 errno = 0;
2459                 val = strtoul(pos->s, &sep, 16);
2460                 if (errno || (sep == pos->s))
2461                         continue;
2462
2463                 if (*sep != '\0') {
2464                         end = pos->s + strlen(pos->s) - 1;
2465                         while (end >= sep && isspace(*end))
2466                                 end--;
2467
2468                         if (end >= sep)
2469                                 continue;
2470                 }
2471
2472                 err = intlist__add(*addr_list, val);
2473                 if (err)
2474                         break;
2475
2476                 strlist__remove(sym_list, pos);
2477                 i++;
2478         }
2479
2480         if (i == 0) {
2481                 intlist__delete(*addr_list);
2482                 *addr_list = NULL;
2483         }
2484
2485         return 0;
2486 }
2487
2488 static bool symbol__read_kptr_restrict(void)
2489 {
2490         bool value = false;
2491         FILE *fp = fopen("/proc/sys/kernel/kptr_restrict", "r");
2492
2493         if (fp != NULL) {
2494                 char line[8];
2495
2496                 if (fgets(line, sizeof(line), fp) != NULL)
2497                         value = perf_cap__capable(CAP_SYSLOG) ?
2498                                         (atoi(line) >= 2) :
2499                                         (atoi(line) != 0);
2500
2501                 fclose(fp);
2502         }
2503
2504         /* Per kernel/kallsyms.c:
2505          * we also restrict when perf_event_paranoid > 1 w/o CAP_SYSLOG
2506          */
2507         if (perf_event_paranoid() > 1 && !perf_cap__capable(CAP_SYSLOG))
2508                 value = true;
2509
2510         return value;
2511 }
2512
2513 int symbol__annotation_init(void)
2514 {
2515         if (symbol_conf.init_annotation)
2516                 return 0;
2517
2518         if (symbol_conf.initialized) {
2519                 pr_err("Annotation needs to be init before symbol__init()\n");
2520                 return -1;
2521         }
2522
2523         symbol_conf.priv_size += sizeof(struct annotation);
2524         symbol_conf.init_annotation = true;
2525         return 0;
2526 }
2527
2528 int symbol__init(struct perf_env *env)
2529 {
2530         const char *symfs;
2531
2532         if (symbol_conf.initialized)
2533                 return 0;
2534
2535         symbol_conf.priv_size = PERF_ALIGN(symbol_conf.priv_size, sizeof(u64));
2536
2537         symbol__elf_init();
2538
2539         if (symbol_conf.sort_by_name)
2540                 symbol_conf.priv_size += (sizeof(struct symbol_name_rb_node) -
2541                                           sizeof(struct symbol));
2542
2543         if (symbol_conf.try_vmlinux_path && vmlinux_path__init(env) < 0)
2544                 return -1;
2545
2546         if (symbol_conf.field_sep && *symbol_conf.field_sep == '.') {
2547                 pr_err("'.' is the only non valid --field-separator argument\n");
2548                 return -1;
2549         }
2550
2551         if (setup_list(&symbol_conf.dso_list,
2552                        symbol_conf.dso_list_str, "dso") < 0)
2553                 return -1;
2554
2555         if (setup_list(&symbol_conf.comm_list,
2556                        symbol_conf.comm_list_str, "comm") < 0)
2557                 goto out_free_dso_list;
2558
2559         if (setup_intlist(&symbol_conf.pid_list,
2560                        symbol_conf.pid_list_str, "pid") < 0)
2561                 goto out_free_comm_list;
2562
2563         if (setup_intlist(&symbol_conf.tid_list,
2564                        symbol_conf.tid_list_str, "tid") < 0)
2565                 goto out_free_pid_list;
2566
2567         if (setup_list(&symbol_conf.sym_list,
2568                        symbol_conf.sym_list_str, "symbol") < 0)
2569                 goto out_free_tid_list;
2570
2571         if (symbol_conf.sym_list &&
2572             setup_addrlist(&symbol_conf.addr_list, symbol_conf.sym_list) < 0)
2573                 goto out_free_sym_list;
2574
2575         if (setup_list(&symbol_conf.bt_stop_list,
2576                        symbol_conf.bt_stop_list_str, "symbol") < 0)
2577                 goto out_free_sym_list;
2578
2579         /*
2580          * A path to symbols of "/" is identical to ""
2581          * reset here for simplicity.
2582          */
2583         symfs = realpath(symbol_conf.symfs, NULL);
2584         if (symfs == NULL)
2585                 symfs = symbol_conf.symfs;
2586         if (strcmp(symfs, "/") == 0)
2587                 symbol_conf.symfs = "";
2588         if (symfs != symbol_conf.symfs)
2589                 free((void *)symfs);
2590
2591         symbol_conf.kptr_restrict = symbol__read_kptr_restrict();
2592
2593         symbol_conf.initialized = true;
2594         return 0;
2595
2596 out_free_sym_list:
2597         strlist__delete(symbol_conf.sym_list);
2598         intlist__delete(symbol_conf.addr_list);
2599 out_free_tid_list:
2600         intlist__delete(symbol_conf.tid_list);
2601 out_free_pid_list:
2602         intlist__delete(symbol_conf.pid_list);
2603 out_free_comm_list:
2604         strlist__delete(symbol_conf.comm_list);
2605 out_free_dso_list:
2606         strlist__delete(symbol_conf.dso_list);
2607         return -1;
2608 }
2609
2610 void symbol__exit(void)
2611 {
2612         if (!symbol_conf.initialized)
2613                 return;
2614         strlist__delete(symbol_conf.bt_stop_list);
2615         strlist__delete(symbol_conf.sym_list);
2616         strlist__delete(symbol_conf.dso_list);
2617         strlist__delete(symbol_conf.comm_list);
2618         intlist__delete(symbol_conf.tid_list);
2619         intlist__delete(symbol_conf.pid_list);
2620         intlist__delete(symbol_conf.addr_list);
2621         vmlinux_path__exit();
2622         symbol_conf.sym_list = symbol_conf.dso_list = symbol_conf.comm_list = NULL;
2623         symbol_conf.bt_stop_list = NULL;
2624         symbol_conf.initialized = false;
2625 }
2626
2627 int symbol__config_symfs(const struct option *opt __maybe_unused,
2628                          const char *dir, int unset __maybe_unused)
2629 {
2630         char *bf = NULL;
2631         int ret;
2632
2633         symbol_conf.symfs = strdup(dir);
2634         if (symbol_conf.symfs == NULL)
2635                 return -ENOMEM;
2636
2637         /* skip the locally configured cache if a symfs is given, and
2638          * config buildid dir to symfs/.debug
2639          */
2640         ret = asprintf(&bf, "%s/%s", dir, ".debug");
2641         if (ret < 0)
2642                 return -ENOMEM;
2643
2644         set_buildid_dir(bf);
2645
2646         free(bf);
2647         return 0;
2648 }
2649
2650 struct mem_info *mem_info__get(struct mem_info *mi)
2651 {
2652         if (mi)
2653                 refcount_inc(&mi->refcnt);
2654         return mi;
2655 }
2656
2657 void mem_info__put(struct mem_info *mi)
2658 {
2659         if (mi && refcount_dec_and_test(&mi->refcnt))
2660                 free(mi);
2661 }
2662
2663 struct mem_info *mem_info__new(void)
2664 {
2665         struct mem_info *mi = zalloc(sizeof(*mi));
2666
2667         if (mi)
2668                 refcount_set(&mi->refcnt, 1);
2669         return mi;
2670 }
2671
2672 /*
2673  * Checks that user supplied symbol kernel files are accessible because
2674  * the default mechanism for accessing elf files fails silently. i.e. if
2675  * debug syms for a build ID aren't found perf carries on normally. When
2676  * they are user supplied we should assume that the user doesn't want to
2677  * silently fail.
2678  */
2679 int symbol__validate_sym_arguments(void)
2680 {
2681         if (symbol_conf.vmlinux_name &&
2682             access(symbol_conf.vmlinux_name, R_OK)) {
2683                 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
2684                 return -EINVAL;
2685         }
2686         if (symbol_conf.kallsyms_name &&
2687             access(symbol_conf.kallsyms_name, R_OK)) {
2688                 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
2689                 return -EINVAL;
2690         }
2691         return 0;
2692 }