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 "mem-info.h"
  31 #include "symsrc.h"
  32 #include "strlist.h"
  33 #include "intlist.h"
  34 #include "namespaces.h"
  35 #include "header.h"
  36 #include "path.h"
  37 #include <linux/ctype.h>
  38 #include <linux/zalloc.h>
  39
  40 #include <elf.h>
  41 #include <limits.h>
  42 #include <symbol/kallsyms.h>
  43 #include <sys/utsname.h>
  44
  45 static int dso__load_kernel_sym(struct dso *dso, struct map *map);
  46 static int dso__load_guest_kernel_sym(struct dso *dso, struct map *map);
  47 static bool symbol__is_idle(const char *name);
  48
  49 int vmlinux_path__nr_entries;
  50 char **vmlinux_path;
  51
  52 struct symbol_conf symbol_conf = {
  53         .nanosecs               = false,
  54         .use_modules            = true,
  55         .try_vmlinux_path       = true,
  56         .demangle               = true,
  57         .demangle_kernel        = false,
  58         .cumulate_callchain     = true,
  59         .time_quantum           = 100 * NSEC_PER_MSEC, /* 100ms */
  60         .show_hist_headers      = true,
  61         .symfs                  = "",
  62         .event_group            = true,
  63         .inline_name            = true,
  64         .res_sample             = 0,
  65 };
  66
  67 struct map_list_node {
  68         struct list_head node;
  69         struct map *map;
  70 };
  71
  72 static struct map_list_node *map_list_node__new(void)
  73 {
  74         return malloc(sizeof(struct map_list_node));
  75 }
  76
  77 static enum dso_binary_type binary_type_symtab[] = {
  78         DSO_BINARY_TYPE__KALLSYMS,
  79         DSO_BINARY_TYPE__GUEST_KALLSYMS,
  80         DSO_BINARY_TYPE__JAVA_JIT,
  81         DSO_BINARY_TYPE__DEBUGLINK,
  82         DSO_BINARY_TYPE__BUILD_ID_CACHE,
  83         DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO,
  84         DSO_BINARY_TYPE__FEDORA_DEBUGINFO,
  85         DSO_BINARY_TYPE__UBUNTU_DEBUGINFO,
  86         DSO_BINARY_TYPE__BUILDID_DEBUGINFO,
  87         DSO_BINARY_TYPE__SYSTEM_PATH_DSO,
  88         DSO_BINARY_TYPE__GUEST_KMODULE,
  89         DSO_BINARY_TYPE__GUEST_KMODULE_COMP,
  90         DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE,
  91         DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP,
  92         DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO,
  93         DSO_BINARY_TYPE__MIXEDUP_UBUNTU_DEBUGINFO,
  94         DSO_BINARY_TYPE__NOT_FOUND,
  95 };
  96
  97 #define DSO_BINARY_TYPE__SYMTAB_CNT ARRAY_SIZE(binary_type_symtab)
  98
  99 static bool symbol_type__filter(char symbol_type)
 100 {
 101         symbol_type = toupper(symbol_type);
 102         return symbol_type == 'T' || symbol_type == 'W' || symbol_type == 'D' || symbol_type == 'B';
 103 }
 104
 105 static int prefix_underscores_count(const char *str)
 106 {
 107         const char *tail = str;
 108
 109         while (*tail == '_')
 110                 tail++;
 111
 112         return tail - str;
 113 }
 114
 115 const char * __weak arch__normalize_symbol_name(const char *name)
 116 {
 117         return name;
 118 }
 119
 120 int __weak arch__compare_symbol_names(const char *namea, const char *nameb)
 121 {
 122         return strcmp(namea, nameb);
 123 }
 124
 125 int __weak arch__compare_symbol_names_n(const char *namea, const char *nameb,
 126                                         unsigned int n)
 127 {
 128         return strncmp(namea, nameb, n);
 129 }
 130
 131 int __weak arch__choose_best_symbol(struct symbol *syma,
 132                                     struct symbol *symb __maybe_unused)
 133 {
 134         /* Avoid "SyS" kernel syscall aliases */
 135         if (strlen(syma->name) >= 3 && !strncmp(syma->name, "SyS", 3))
 136                 return SYMBOL_B;
 137         if (strlen(syma->name) >= 10 && !strncmp(syma->name, "compat_SyS", 10))
 138                 return SYMBOL_B;
 139
 140         return SYMBOL_A;
 141 }
 142
 143 static int choose_best_symbol(struct symbol *syma, struct symbol *symb)
 144 {
 145         s64 a;
 146         s64 b;
 147         size_t na, nb;
 148
 149         /* Prefer a symbol with non zero length */
 150         a = syma->end - syma->start;
 151         b = symb->end - symb->start;
 152         if ((b == 0) && (a > 0))
 153                 return SYMBOL_A;
 154         else if ((a == 0) && (b > 0))
 155                 return SYMBOL_B;
 156
 157         if (syma->type != symb->type) {
 158                 if (syma->type == STT_NOTYPE)
 159                         return SYMBOL_B;
 160                 if (symb->type == STT_NOTYPE)
 161                         return SYMBOL_A;
 162         }
 163
 164         /* Prefer a non weak symbol over a weak one */
 165         a = syma->binding == STB_WEAK;
 166         b = symb->binding == STB_WEAK;
 167         if (b && !a)
 168                 return SYMBOL_A;
 169         if (a && !b)
 170                 return SYMBOL_B;
 171
 172         /* Prefer a global symbol over a non global one */
 173         a = syma->binding == STB_GLOBAL;
 174         b = symb->binding == STB_GLOBAL;
 175         if (a && !b)
 176                 return SYMBOL_A;
 177         if (b && !a)
 178                 return SYMBOL_B;
 179
 180         /* Prefer a symbol with less underscores */
 181         a = prefix_underscores_count(syma->name);
 182         b = prefix_underscores_count(symb->name);
 183         if (b > a)
 184                 return SYMBOL_A;
 185         else if (a > b)
 186                 return SYMBOL_B;
 187
 188         /* Choose the symbol with the longest name */
 189         na = strlen(syma->name);
 190         nb = strlen(symb->name);
 191         if (na > nb)
 192                 return SYMBOL_A;
 193         else if (na < nb)
 194                 return SYMBOL_B;
 195
 196         return arch__choose_best_symbol(syma, symb);
 197 }
 198
 199 void symbols__fixup_duplicate(struct rb_root_cached *symbols)
 200 {
 201         struct rb_node *nd;
 202         struct symbol *curr, *next;
 203
 204         if (symbol_conf.allow_aliases)
 205                 return;
 206
 207         nd = rb_first_cached(symbols);
 208
 209         while (nd) {
 210                 curr = rb_entry(nd, struct symbol, rb_node);
 211 again:
 212                 nd = rb_next(&curr->rb_node);
 213                 if (!nd)
 214                         break;
 215
 216                 next = rb_entry(nd, struct symbol, rb_node);
 217                 if (curr->start != next->start)
 218                         continue;
 219
 220                 if (choose_best_symbol(curr, next) == SYMBOL_A) {
 221                         if (next->type == STT_GNU_IFUNC)
 222                                 curr->ifunc_alias = true;
 223                         rb_erase_cached(&next->rb_node, symbols);
 224                         symbol__delete(next);
 225                         goto again;
 226                 } else {
 227                         if (curr->type == STT_GNU_IFUNC)
 228                                 next->ifunc_alias = true;
 229                         nd = rb_next(&curr->rb_node);
 230                         rb_erase_cached(&curr->rb_node, symbols);
 231                         symbol__delete(curr);
 232                 }
 233         }
 234 }
 235
 236 /* Update zero-sized symbols using the address of the next symbol */
 237 void symbols__fixup_end(struct rb_root_cached *symbols, bool is_kallsyms)
 238 {
 239         struct rb_node *nd, *prevnd = rb_first_cached(symbols);
 240         struct symbol *curr, *prev;
 241
 242         if (prevnd == NULL)
 243                 return;
 244
 245         curr = rb_entry(prevnd, struct symbol, rb_node);
 246
 247         for (nd = rb_next(prevnd); nd; nd = rb_next(nd)) {
 248                 prev = curr;
 249                 curr = rb_entry(nd, struct symbol, rb_node);
 250
 251                 /*
 252                  * On some architecture kernel text segment start is located at
 253                  * some low memory address, while modules are located at high
 254                  * memory addresses (or vice versa).  The gap between end of
 255                  * kernel text segment and beginning of first module's text
 256                  * segment is very big.  Therefore do not fill this gap and do
 257                  * not assign it to the kernel dso map (kallsyms).
 258                  *
 259                  * Also BPF code can be allocated separately from text segments
 260                  * and modules.  So the last entry in a module should not fill
 261                  * the gap too.
 262                  *
 263                  * In kallsyms, it determines module symbols using '[' character
 264                  * like in:
 265                  *   ffffffffc1937000 T hdmi_driver_init  [snd_hda_codec_hdmi]
 266                  */
 267                 if (prev->end == prev->start) {
 268                         const char *prev_mod;
 269                         const char *curr_mod;
 270
 271                         if (!is_kallsyms) {
 272                                 prev->end = curr->start;
 273                                 continue;
 274                         }
 275
 276                         prev_mod = strchr(prev->name, '[');
 277                         curr_mod = strchr(curr->name, '[');
 278
 279                         /* Last kernel/module symbol mapped to end of page */
 280                         if (!prev_mod != !curr_mod)
 281                                 prev->end = roundup(prev->end + 4096, 4096);
 282                         /* Last symbol in the previous module */
 283                         else if (prev_mod && strcmp(prev_mod, curr_mod))
 284                                 prev->end = roundup(prev->end + 4096, 4096);
 285                         else
 286                                 prev->end = curr->start;
 287
 288                         pr_debug4("%s sym:%s end:%#" PRIx64 "\n",
 289                                   __func__, prev->name, prev->end);
 290                 }
 291         }
 292
 293         /* Last entry */
 294         if (curr->end == curr->start)
 295                 curr->end = roundup(curr->start, 4096) + 4096;
 296 }
 297
 298 struct symbol *symbol__new(u64 start, u64 len, u8 binding, u8 type, const char *name)
 299 {
 300         size_t namelen = strlen(name) + 1;
 301         struct symbol *sym = calloc(1, (symbol_conf.priv_size +
 302                                         sizeof(*sym) + namelen));
 303         if (sym == NULL)
 304                 return NULL;
 305
 306         if (symbol_conf.priv_size) {
 307                 if (symbol_conf.init_annotation) {
 308                         struct annotation *notes = (void *)sym;
 309                         annotation__init(notes);
 310                 }
 311                 sym = ((void *)sym) + symbol_conf.priv_size;
 312         }
 313
 314         sym->start   = start;
 315         sym->end     = len ? start + len : start;
 316         sym->type    = type;
 317         sym->binding = binding;
 318         sym->namelen = namelen - 1;
 319
 320         pr_debug4("%s: %s %#" PRIx64 "-%#" PRIx64 "\n",
 321                   __func__, name, start, sym->end);
 322         memcpy(sym->name, name, namelen);
 323
 324         return sym;
 325 }
 326
 327 void symbol__delete(struct symbol *sym)
 328 {
 329         if (symbol_conf.priv_size) {
 330                 if (symbol_conf.init_annotation) {
 331                         struct annotation *notes = symbol__annotation(sym);
 332
 333                         annotation__exit(notes);
 334                 }
 335         }
 336         free(((void *)sym) - symbol_conf.priv_size);
 337 }
 338
 339 void symbols__delete(struct rb_root_cached *symbols)
 340 {
 341         struct symbol *pos;
 342         struct rb_node *next = rb_first_cached(symbols);
 343
 344         while (next) {
 345                 pos = rb_entry(next, struct symbol, rb_node);
 346                 next = rb_next(&pos->rb_node);
 347                 rb_erase_cached(&pos->rb_node, symbols);
 348                 symbol__delete(pos);
 349         }
 350 }
 351
 352 void __symbols__insert(struct rb_root_cached *symbols,
 353                        struct symbol *sym, bool kernel)
 354 {
 355         struct rb_node **p = &symbols->rb_root.rb_node;
 356         struct rb_node *parent = NULL;
 357         const u64 ip = sym->start;
 358         struct symbol *s;
 359         bool leftmost = true;
 360
 361         if (kernel) {
 362                 const char *name = sym->name;
 363                 /*
 364                  * ppc64 uses function descriptors and appends a '.' to the
 365                  * start of every instruction address. Remove it.
 366                  */
 367                 if (name[0] == '.')
 368                         name++;
 369                 sym->idle = symbol__is_idle(name);
 370         }
 371
 372         while (*p != NULL) {
 373                 parent = *p;
 374                 s = rb_entry(parent, struct symbol, rb_node);
 375                 if (ip < s->start)
 376                         p = &(*p)->rb_left;
 377                 else {
 378                         p = &(*p)->rb_right;
 379                         leftmost = false;
 380                 }
 381         }
 382         rb_link_node(&sym->rb_node, parent, p);
 383         rb_insert_color_cached(&sym->rb_node, symbols, leftmost);
 384 }
 385
 386 void symbols__insert(struct rb_root_cached *symbols, struct symbol *sym)
 387 {
 388         __symbols__insert(symbols, sym, false);
 389 }
 390
 391 static struct symbol *symbols__find(struct rb_root_cached *symbols, u64 ip)
 392 {
 393         struct rb_node *n;
 394
 395         if (symbols == NULL)
 396                 return NULL;
 397
 398         n = symbols->rb_root.rb_node;
 399
 400         while (n) {
 401                 struct symbol *s = rb_entry(n, struct symbol, rb_node);
 402
 403                 if (ip < s->start)
 404                         n = n->rb_left;
 405                 else if (ip > s->end || (ip == s->end && ip != s->start))
 406                         n = n->rb_right;
 407                 else
 408                         return s;
 409         }
 410
 411         return NULL;
 412 }
 413
 414 static struct symbol *symbols__first(struct rb_root_cached *symbols)
 415 {
 416         struct rb_node *n = rb_first_cached(symbols);
 417
 418         if (n)
 419                 return rb_entry(n, struct symbol, rb_node);
 420
 421         return NULL;
 422 }
 423
 424 static struct symbol *symbols__last(struct rb_root_cached *symbols)
 425 {
 426         struct rb_node *n = rb_last(&symbols->rb_root);
 427
 428         if (n)
 429                 return rb_entry(n, struct symbol, rb_node);
 430
 431         return NULL;
 432 }
 433
 434 static struct symbol *symbols__next(struct symbol *sym)
 435 {
 436         struct rb_node *n = rb_next(&sym->rb_node);
 437
 438         if (n)
 439                 return rb_entry(n, struct symbol, rb_node);
 440
 441         return NULL;
 442 }
 443
 444 static int symbols__sort_name_cmp(const void *vlhs, const void *vrhs)
 445 {
 446         const struct symbol *lhs = *((const struct symbol **)vlhs);
 447         const struct symbol *rhs = *((const struct symbol **)vrhs);
 448
 449         return strcmp(lhs->name, rhs->name);
 450 }
 451
 452 static struct symbol **symbols__sort_by_name(struct rb_root_cached *source, size_t *len)
 453 {
 454         struct rb_node *nd;
 455         struct symbol **result;
 456         size_t i = 0, size = 0;
 457
 458         for (nd = rb_first_cached(source); nd; nd = rb_next(nd))
 459                 size++;
 460
 461         result = malloc(sizeof(*result) * size);
 462         if (!result)
 463                 return NULL;
 464
 465         for (nd = rb_first_cached(source); nd; nd = rb_next(nd)) {
 466                 struct symbol *pos = rb_entry(nd, struct symbol, rb_node);
 467
 468                 result[i++] = pos;
 469         }
 470         qsort(result, size, sizeof(*result), symbols__sort_name_cmp);
 471         *len = size;
 472         return result;
 473 }
 474
 475 int symbol__match_symbol_name(const char *name, const char *str,
 476                               enum symbol_tag_include includes)
 477 {
 478         const char *versioning;
 479
 480         if (includes == SYMBOL_TAG_INCLUDE__DEFAULT_ONLY &&
 481             (versioning = strstr(name, "@@"))) {
 482                 int len = strlen(str);
 483
 484                 if (len < versioning - name)
 485                         len = versioning - name;
 486
 487                 return arch__compare_symbol_names_n(name, str, len);
 488         } else
 489                 return arch__compare_symbol_names(name, str);
 490 }
 491
 492 static struct symbol *symbols__find_by_name(struct symbol *symbols[],
 493                                             size_t symbols_len,
 494                                             const char *name,
 495                                             enum symbol_tag_include includes,
 496                                             size_t *found_idx)
 497 {
 498         size_t i, lower = 0, upper = symbols_len;
 499         struct symbol *s = NULL;
 500
 501         if (found_idx)
 502                 *found_idx = SIZE_MAX;
 503
 504         if (!symbols_len)
 505                 return NULL;
 506
 507         while (lower < upper) {
 508                 int cmp;
 509
 510                 i = (lower + upper) / 2;
 511                 cmp = symbol__match_symbol_name(symbols[i]->name, name, includes);
 512
 513                 if (cmp > 0)
 514                         upper = i;
 515                 else if (cmp < 0)
 516                         lower = i + 1;
 517                 else {
 518                         if (found_idx)
 519                                 *found_idx = i;
 520                         s = symbols[i];
 521                         break;
 522                 }
 523         }
 524         if (s && includes != SYMBOL_TAG_INCLUDE__DEFAULT_ONLY) {
 525                 /* return first symbol that has same name (if any) */
 526                 for (; i > 0; i--) {
 527                         struct symbol *tmp = symbols[i - 1];
 528
 529                         if (!arch__compare_symbol_names(tmp->name, s->name)) {
 530                                 if (found_idx)
 531                                         *found_idx = i - 1;
 532                                 s = tmp;
 533                         } else
 534                                 break;
 535                 }
 536         }
 537         assert(!found_idx || !s || s == symbols[*found_idx]);
 538         return s;
 539 }
 540
 541 void dso__reset_find_symbol_cache(struct dso *dso)
 542 {
 543         dso__set_last_find_result_addr(dso, 0);
 544         dso__set_last_find_result_symbol(dso, NULL);
 545 }
 546
 547 void dso__insert_symbol(struct dso *dso, struct symbol *sym)
 548 {
 549         __symbols__insert(dso__symbols(dso), sym, dso__kernel(dso));
 550
 551         /* update the symbol cache if necessary */
 552         if (dso__last_find_result_addr(dso) >= sym->start &&
 553             (dso__last_find_result_addr(dso) < sym->end ||
 554             sym->start == sym->end)) {
 555                 dso__set_last_find_result_symbol(dso, sym);
 556         }
 557 }
 558
 559 void dso__delete_symbol(struct dso *dso, struct symbol *sym)
 560 {
 561         rb_erase_cached(&sym->rb_node, dso__symbols(dso));
 562         symbol__delete(sym);
 563         dso__reset_find_symbol_cache(dso);
 564 }
 565
 566 struct symbol *dso__find_symbol(struct dso *dso, u64 addr)
 567 {
 568         if (dso__last_find_result_addr(dso) != addr || dso__last_find_result_symbol(dso) == NULL) {
 569                 dso__set_last_find_result_addr(dso, addr);
 570                 dso__set_last_find_result_symbol(dso, symbols__find(dso__symbols(dso), addr));
 571         }
 572
 573         return dso__last_find_result_symbol(dso);
 574 }
 575
 576 struct symbol *dso__find_symbol_nocache(struct dso *dso, u64 addr)
 577 {
 578         return symbols__find(dso__symbols(dso), addr);
 579 }
 580
 581 struct symbol *dso__first_symbol(struct dso *dso)
 582 {
 583         return symbols__first(dso__symbols(dso));
 584 }
 585
 586 struct symbol *dso__last_symbol(struct dso *dso)
 587 {
 588         return symbols__last(dso__symbols(dso));
 589 }
 590
 591 struct symbol *dso__next_symbol(struct symbol *sym)
 592 {
 593         return symbols__next(sym);
 594 }
 595
 596 struct symbol *dso__next_symbol_by_name(struct dso *dso, size_t *idx)
 597 {
 598         if (*idx + 1 >= dso__symbol_names_len(dso))
 599                 return NULL;
 600
 601         ++*idx;
 602         return dso__symbol_names(dso)[*idx];
 603 }
 604
 605  /*
 606   * Returns first symbol that matched with @name.
 607   */
 608 struct symbol *dso__find_symbol_by_name(struct dso *dso, const char *name, size_t *idx)
 609 {
 610         struct symbol *s = symbols__find_by_name(dso__symbol_names(dso),
 611                                                  dso__symbol_names_len(dso),
 612                                                  name, SYMBOL_TAG_INCLUDE__NONE, idx);
 613         if (!s) {
 614                 s = symbols__find_by_name(dso__symbol_names(dso), dso__symbol_names_len(dso),
 615                                           name, SYMBOL_TAG_INCLUDE__DEFAULT_ONLY, idx);
 616         }
 617         return s;
 618 }
 619
 620 void dso__sort_by_name(struct dso *dso)
 621 {
 622         mutex_lock(dso__lock(dso));
 623         if (!dso__sorted_by_name(dso)) {
 624                 size_t len;
 625
 626                 dso__set_symbol_names(dso, symbols__sort_by_name(dso__symbols(dso), &len));
 627                 if (dso__symbol_names(dso)) {
 628                         dso__set_symbol_names_len(dso, len);
 629                         dso__set_sorted_by_name(dso);
 630                 }
 631         }
 632         mutex_unlock(dso__lock(dso));
 633 }
 634
 635 /*
 636  * While we find nice hex chars, build a long_val.
 637  * Return number of chars processed.
 638  */
 639 static int hex2u64(const char *ptr, u64 *long_val)
 640 {
 641         char *p;
 642
 643         *long_val = strtoull(ptr, &p, 16);
 644
 645         return p - ptr;
 646 }
 647
 648
 649 int modules__parse(const char *filename, void *arg,
 650                    int (*process_module)(void *arg, const char *name,
 651                                          u64 start, u64 size))
 652 {
 653         char *line = NULL;
 654         size_t n;
 655         FILE *file;
 656         int err = 0;
 657
 658         file = fopen(filename, "r");
 659         if (file == NULL)
 660                 return -1;
 661
 662         while (1) {
 663                 char name[PATH_MAX];
 664                 u64 start, size;
 665                 char *sep, *endptr;
 666                 ssize_t line_len;
 667
 668                 line_len = getline(&line, &n, file);
 669                 if (line_len < 0) {
 670                         if (feof(file))
 671                                 break;
 672                         err = -1;
 673                         goto out;
 674                 }
 675
 676                 if (!line) {
 677                         err = -1;
 678                         goto out;
 679                 }
 680
 681                 line[--line_len] = '\0'; /* \n */
 682
 683                 sep = strrchr(line, 'x');
 684                 if (sep == NULL)
 685                         continue;
 686
 687                 hex2u64(sep + 1, &start);
 688
 689                 sep = strchr(line, ' ');
 690                 if (sep == NULL)
 691                         continue;
 692
 693                 *sep = '\0';
 694
 695                 scnprintf(name, sizeof(name), "[%s]", line);
 696
 697                 size = strtoul(sep + 1, &endptr, 0);
 698                 if (*endptr != ' ' && *endptr != '\t')
 699                         continue;
 700
 701                 err = process_module(arg, name, start, size);
 702                 if (err)
 703                         break;
 704         }
 705 out:
 706         free(line);
 707         fclose(file);
 708         return err;
 709 }
 710
 711 /*
 712  * These are symbols in the kernel image, so make sure that
 713  * sym is from a kernel DSO.
 714  */
 715 static bool symbol__is_idle(const char *name)
 716 {
 717         const char * const idle_symbols[] = {
 718                 "acpi_idle_do_entry",
 719                 "acpi_processor_ffh_cstate_enter",
 720                 "arch_cpu_idle",
 721                 "cpu_idle",
 722                 "cpu_startup_entry",
 723                 "idle_cpu",
 724                 "intel_idle",
 725                 "intel_idle_ibrs",
 726                 "default_idle",
 727                 "native_safe_halt",
 728                 "enter_idle",
 729                 "exit_idle",
 730                 "mwait_idle",
 731                 "mwait_idle_with_hints",
 732                 "mwait_idle_with_hints.constprop.0",
 733                 "poll_idle",
 734                 "ppc64_runlatch_off",
 735                 "pseries_dedicated_idle_sleep",
 736                 "psw_idle",
 737                 "psw_idle_exit",
 738                 NULL
 739         };
 740         int i;
 741         static struct strlist *idle_symbols_list;
 742
 743         if (idle_symbols_list)
 744                 return strlist__has_entry(idle_symbols_list, name);
 745
 746         idle_symbols_list = strlist__new(NULL, NULL);
 747
 748         for (i = 0; idle_symbols[i]; i++)
 749                 strlist__add(idle_symbols_list, idle_symbols[i]);
 750
 751         return strlist__has_entry(idle_symbols_list, name);
 752 }
 753
 754 static int map__process_kallsym_symbol(void *arg, const char *name,
 755                                        char type, u64 start)
 756 {
 757         struct symbol *sym;
 758         struct dso *dso = arg;
 759         struct rb_root_cached *root = dso__symbols(dso);
 760
 761         if (!symbol_type__filter(type))
 762                 return 0;
 763
 764         /* Ignore local symbols for ARM modules */
 765         if (name[0] == '$')
 766                 return 0;
 767
 768         /*
 769          * module symbols are not sorted so we add all
 770          * symbols, setting length to 0, and rely on
 771          * symbols__fixup_end() to fix it up.
 772          */
 773         sym = symbol__new(start, 0, kallsyms2elf_binding(type), kallsyms2elf_type(type), name);
 774         if (sym == NULL)
 775                 return -ENOMEM;
 776         /*
 777          * We will pass the symbols to the filter later, in
 778          * map__split_kallsyms, when we have split the maps per module
 779          */
 780         __symbols__insert(root, sym, !strchr(name, '['));
 781
 782         return 0;
 783 }
 784
 785 /*
 786  * Loads the function entries in /proc/kallsyms into kernel_map->dso,
 787  * so that we can in the next step set the symbol ->end address and then
 788  * call kernel_maps__split_kallsyms.
 789  */
 790 static int dso__load_all_kallsyms(struct dso *dso, const char *filename)
 791 {
 792         return kallsyms__parse(filename, dso, map__process_kallsym_symbol);
 793 }
 794
 795 static int maps__split_kallsyms_for_kcore(struct maps *kmaps, struct dso *dso)
 796 {
 797         struct symbol *pos;
 798         int count = 0;
 799         struct rb_root_cached *root = dso__symbols(dso);
 800         struct rb_root_cached old_root = *root;
 801         struct rb_node *next = rb_first_cached(root);
 802
 803         if (!kmaps)
 804                 return -1;
 805
 806         *root = RB_ROOT_CACHED;
 807
 808         while (next) {
 809                 struct map *curr_map;
 810                 struct dso *curr_map_dso;
 811                 char *module;
 812
 813                 pos = rb_entry(next, struct symbol, rb_node);
 814                 next = rb_next(&pos->rb_node);
 815
 816                 rb_erase_cached(&pos->rb_node, &old_root);
 817                 RB_CLEAR_NODE(&pos->rb_node);
 818                 module = strchr(pos->name, '\t');
 819                 if (module)
 820                         *module = '\0';
 821
 822                 curr_map = maps__find(kmaps, pos->start);
 823
 824                 if (!curr_map) {
 825                         symbol__delete(pos);
 826                         continue;
 827                 }
 828                 curr_map_dso = map__dso(curr_map);
 829                 pos->start -= map__start(curr_map) - map__pgoff(curr_map);
 830                 if (pos->end > map__end(curr_map))
 831                         pos->end = map__end(curr_map);
 832                 if (pos->end)
 833                         pos->end -= map__start(curr_map) - map__pgoff(curr_map);
 834                 symbols__insert(dso__symbols(curr_map_dso), pos);
 835                 ++count;
 836                 map__put(curr_map);
 837         }
 838
 839         /* Symbols have been adjusted */
 840         dso__set_adjust_symbols(dso, true);
 841
 842         return count;
 843 }
 844
 845 /*
 846  * Split the symbols into maps, making sure there are no overlaps, i.e. the
 847  * kernel range is broken in several maps, named [kernel].N, as we don't have
 848  * the original ELF section names vmlinux have.
 849  */
 850 static int maps__split_kallsyms(struct maps *kmaps, struct dso *dso, u64 delta,
 851                                 struct map *initial_map)
 852 {
 853         struct machine *machine;
 854         struct map *curr_map = map__get(initial_map);
 855         struct symbol *pos;
 856         int count = 0, moved = 0;
 857         struct rb_root_cached *root = dso__symbols(dso);
 858         struct rb_node *next = rb_first_cached(root);
 859         int kernel_range = 0;
 860         bool x86_64;
 861
 862         if (!kmaps)
 863                 return -1;
 864
 865         machine = maps__machine(kmaps);
 866
 867         x86_64 = machine__is(machine, "x86_64");
 868
 869         while (next) {
 870                 char *module;
 871
 872                 pos = rb_entry(next, struct symbol, rb_node);
 873                 next = rb_next(&pos->rb_node);
 874
 875                 module = strchr(pos->name, '\t');
 876                 if (module) {
 877                         struct dso *curr_map_dso;
 878
 879                         if (!symbol_conf.use_modules)
 880                                 goto discard_symbol;
 881
 882                         *module++ = '\0';
 883                         curr_map_dso = map__dso(curr_map);
 884                         if (strcmp(dso__short_name(curr_map_dso), module)) {
 885                                 if (!RC_CHK_EQUAL(curr_map, initial_map) &&
 886                                     dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST &&
 887                                     machine__is_default_guest(machine)) {
 888                                         /*
 889                                          * We assume all symbols of a module are
 890                                          * continuous in * kallsyms, so curr_map
 891                                          * points to a module and all its
 892                                          * symbols are in its kmap. Mark it as
 893                                          * loaded.
 894                                          */
 895                                         dso__set_loaded(curr_map_dso);
 896                                 }
 897
 898                                 map__zput(curr_map);
 899                                 curr_map = maps__find_by_name(kmaps, module);
 900                                 if (curr_map == NULL) {
 901                                         pr_debug("%s/proc/{kallsyms,modules} "
 902                                                  "inconsistency while looking "
 903                                                  "for \"%s\" module!\n",
 904                                                  machine->root_dir, module);
 905                                         curr_map = map__get(initial_map);
 906                                         goto discard_symbol;
 907                                 }
 908                                 curr_map_dso = map__dso(curr_map);
 909                                 if (dso__loaded(curr_map_dso) &&
 910                                     !machine__is_default_guest(machine))
 911                                         goto discard_symbol;
 912                         }
 913                         /*
 914                          * So that we look just like we get from .ko files,
 915                          * i.e. not prelinked, relative to initial_map->start.
 916                          */
 917                         pos->start = map__map_ip(curr_map, pos->start);
 918                         pos->end   = map__map_ip(curr_map, pos->end);
 919                 } else if (x86_64 && is_entry_trampoline(pos->name)) {
 920                         /*
 921                          * These symbols are not needed anymore since the
 922                          * trampoline maps refer to the text section and it's
 923                          * symbols instead. Avoid having to deal with
 924                          * relocations, and the assumption that the first symbol
 925                          * is the start of kernel text, by simply removing the
 926                          * symbols at this point.
 927                          */
 928                         goto discard_symbol;
 929                 } else if (!RC_CHK_EQUAL(curr_map, initial_map)) {
 930                         char dso_name[PATH_MAX];
 931                         struct dso *ndso;
 932
 933                         if (delta) {
 934                                 /* Kernel was relocated at boot time */
 935                                 pos->start -= delta;
 936                                 pos->end -= delta;
 937                         }
 938
 939                         if (count == 0) {
 940                                 map__zput(curr_map);
 941                                 curr_map = map__get(initial_map);
 942                                 goto add_symbol;
 943                         }
 944
 945                         if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST)
 946                                 snprintf(dso_name, sizeof(dso_name),
 947                                         "[guest.kernel].%d",
 948                                         kernel_range++);
 949                         else
 950                                 snprintf(dso_name, sizeof(dso_name),
 951                                         "[kernel].%d",
 952                                         kernel_range++);
 953
 954                         ndso = dso__new(dso_name);
 955                         map__zput(curr_map);
 956                         if (ndso == NULL)
 957                                 return -1;
 958
 959                         dso__set_kernel(ndso, dso__kernel(dso));
 960
 961                         curr_map = map__new2(pos->start, ndso);
 962                         if (curr_map == NULL) {
 963                                 dso__put(ndso);
 964                                 return -1;
 965                         }
 966
 967                         map__set_mapping_type(curr_map, MAPPING_TYPE__IDENTITY);
 968                         if (maps__insert(kmaps, curr_map)) {
 969                                 map__zput(curr_map);
 970                                 dso__put(ndso);
 971                                 return -1;
 972                         }
 973                         ++kernel_range;
 974                 } else if (delta) {
 975                         /* Kernel was relocated at boot time */
 976                         pos->start -= delta;
 977                         pos->end -= delta;
 978                 }
 979 add_symbol:
 980                 if (!RC_CHK_EQUAL(curr_map, initial_map)) {
 981                         struct dso *curr_map_dso = map__dso(curr_map);
 982
 983                         rb_erase_cached(&pos->rb_node, root);
 984                         symbols__insert(dso__symbols(curr_map_dso), pos);
 985                         ++moved;
 986                 } else
 987                         ++count;
 988
 989                 continue;
 990 discard_symbol:
 991                 rb_erase_cached(&pos->rb_node, root);
 992                 symbol__delete(pos);
 993         }
 994
 995         if (!RC_CHK_EQUAL(curr_map, initial_map) &&
 996             dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST &&
 997             machine__is_default_guest(maps__machine(kmaps))) {
 998                 dso__set_loaded(map__dso(curr_map));
 999         }
1000         map__put(curr_map);
1001         return count + moved;
1002 }
1003
1004 bool symbol__restricted_filename(const char *filename,
1005                                  const char *restricted_filename)
1006 {
1007         bool restricted = false;
1008
1009         if (symbol_conf.kptr_restrict) {
1010                 char *r = realpath(filename, NULL);
1011
1012                 if (r != NULL) {
1013                         restricted = strcmp(r, restricted_filename) == 0;
1014                         free(r);
1015                         return restricted;
1016                 }
1017         }
1018
1019         return restricted;
1020 }
1021
1022 struct module_info {
1023         struct rb_node rb_node;
1024         char *name;
1025         u64 start;
1026 };
1027
1028 static void add_module(struct module_info *mi, struct rb_root *modules)
1029 {
1030         struct rb_node **p = &modules->rb_node;
1031         struct rb_node *parent = NULL;
1032         struct module_info *m;
1033
1034         while (*p != NULL) {
1035                 parent = *p;
1036                 m = rb_entry(parent, struct module_info, rb_node);
1037                 if (strcmp(mi->name, m->name) < 0)
1038                         p = &(*p)->rb_left;
1039                 else
1040                         p = &(*p)->rb_right;
1041         }
1042         rb_link_node(&mi->rb_node, parent, p);
1043         rb_insert_color(&mi->rb_node, modules);
1044 }
1045
1046 static void delete_modules(struct rb_root *modules)
1047 {
1048         struct module_info *mi;
1049         struct rb_node *next = rb_first(modules);
1050
1051         while (next) {
1052                 mi = rb_entry(next, struct module_info, rb_node);
1053                 next = rb_next(&mi->rb_node);
1054                 rb_erase(&mi->rb_node, modules);
1055                 zfree(&mi->name);
1056                 free(mi);
1057         }
1058 }
1059
1060 static struct module_info *find_module(const char *name,
1061                                        struct rb_root *modules)
1062 {
1063         struct rb_node *n = modules->rb_node;
1064
1065         while (n) {
1066                 struct module_info *m;
1067                 int cmp;
1068
1069                 m = rb_entry(n, struct module_info, rb_node);
1070                 cmp = strcmp(name, m->name);
1071                 if (cmp < 0)
1072                         n = n->rb_left;
1073                 else if (cmp > 0)
1074                         n = n->rb_right;
1075                 else
1076                         return m;
1077         }
1078
1079         return NULL;
1080 }
1081
1082 static int __read_proc_modules(void *arg, const char *name, u64 start,
1083                                u64 size __maybe_unused)
1084 {
1085         struct rb_root *modules = arg;
1086         struct module_info *mi;
1087
1088         mi = zalloc(sizeof(struct module_info));
1089         if (!mi)
1090                 return -ENOMEM;
1091
1092         mi->name = strdup(name);
1093         mi->start = start;
1094
1095         if (!mi->name) {
1096                 free(mi);
1097                 return -ENOMEM;
1098         }
1099
1100         add_module(mi, modules);
1101
1102         return 0;
1103 }
1104
1105 static int read_proc_modules(const char *filename, struct rb_root *modules)
1106 {
1107         if (symbol__restricted_filename(filename, "/proc/modules"))
1108                 return -1;
1109
1110         if (modules__parse(filename, modules, __read_proc_modules)) {
1111                 delete_modules(modules);
1112                 return -1;
1113         }
1114
1115         return 0;
1116 }
1117
1118 int compare_proc_modules(const char *from, const char *to)
1119 {
1120         struct rb_root from_modules = RB_ROOT;
1121         struct rb_root to_modules = RB_ROOT;
1122         struct rb_node *from_node, *to_node;
1123         struct module_info *from_m, *to_m;
1124         int ret = -1;
1125
1126         if (read_proc_modules(from, &from_modules))
1127                 return -1;
1128
1129         if (read_proc_modules(to, &to_modules))
1130                 goto out_delete_from;
1131
1132         from_node = rb_first(&from_modules);
1133         to_node = rb_first(&to_modules);
1134         while (from_node) {
1135                 if (!to_node)
1136                         break;
1137
1138                 from_m = rb_entry(from_node, struct module_info, rb_node);
1139                 to_m = rb_entry(to_node, struct module_info, rb_node);
1140
1141                 if (from_m->start != to_m->start ||
1142                     strcmp(from_m->name, to_m->name))
1143                         break;
1144
1145                 from_node = rb_next(from_node);
1146                 to_node = rb_next(to_node);
1147         }
1148
1149         if (!from_node && !to_node)
1150                 ret = 0;
1151
1152         delete_modules(&to_modules);
1153 out_delete_from:
1154         delete_modules(&from_modules);
1155
1156         return ret;
1157 }
1158
1159 static int do_validate_kcore_modules_cb(struct map *old_map, void *data)
1160 {
1161         struct rb_root *modules = data;
1162         struct module_info *mi;
1163         struct dso *dso;
1164
1165         if (!__map__is_kmodule(old_map))
1166                 return 0;
1167
1168         dso = map__dso(old_map);
1169         /* Module must be in memory at the same address */
1170         mi = find_module(dso__short_name(dso), modules);
1171         if (!mi || mi->start != map__start(old_map))
1172                 return -EINVAL;
1173
1174         return 0;
1175 }
1176
1177 static int do_validate_kcore_modules(const char *filename, struct maps *kmaps)
1178 {
1179         struct rb_root modules = RB_ROOT;
1180         int err;
1181
1182         err = read_proc_modules(filename, &modules);
1183         if (err)
1184                 return err;
1185
1186         err = maps__for_each_map(kmaps, do_validate_kcore_modules_cb, &modules);
1187
1188         delete_modules(&modules);
1189         return err;
1190 }
1191
1192 /*
1193  * If kallsyms is referenced by name then we look for filename in the same
1194  * directory.
1195  */
1196 static bool filename_from_kallsyms_filename(char *filename,
1197                                             const char *base_name,
1198                                             const char *kallsyms_filename)
1199 {
1200         char *name;
1201
1202         strcpy(filename, kallsyms_filename);
1203         name = strrchr(filename, '/');
1204         if (!name)
1205                 return false;
1206
1207         name += 1;
1208
1209         if (!strcmp(name, "kallsyms")) {
1210                 strcpy(name, base_name);
1211                 return true;
1212         }
1213
1214         return false;
1215 }
1216
1217 static int validate_kcore_modules(const char *kallsyms_filename,
1218                                   struct map *map)
1219 {
1220         struct maps *kmaps = map__kmaps(map);
1221         char modules_filename[PATH_MAX];
1222
1223         if (!kmaps)
1224                 return -EINVAL;
1225
1226         if (!filename_from_kallsyms_filename(modules_filename, "modules",
1227                                              kallsyms_filename))
1228                 return -EINVAL;
1229
1230         if (do_validate_kcore_modules(modules_filename, kmaps))
1231                 return -EINVAL;
1232
1233         return 0;
1234 }
1235
1236 static int validate_kcore_addresses(const char *kallsyms_filename,
1237                                     struct map *map)
1238 {
1239         struct kmap *kmap = map__kmap(map);
1240
1241         if (!kmap)
1242                 return -EINVAL;
1243
1244         if (kmap->ref_reloc_sym && kmap->ref_reloc_sym->name) {
1245                 u64 start;
1246
1247                 if (kallsyms__get_function_start(kallsyms_filename,
1248                                                  kmap->ref_reloc_sym->name, &start))
1249                         return -ENOENT;
1250                 if (start != kmap->ref_reloc_sym->addr)
1251                         return -EINVAL;
1252         }
1253
1254         return validate_kcore_modules(kallsyms_filename, map);
1255 }
1256
1257 struct kcore_mapfn_data {
1258         struct dso *dso;
1259         struct list_head maps;
1260 };
1261
1262 static int kcore_mapfn(u64 start, u64 len, u64 pgoff, void *data)
1263 {
1264         struct kcore_mapfn_data *md = data;
1265         struct map_list_node *list_node = map_list_node__new();
1266
1267         if (!list_node)
1268                 return -ENOMEM;
1269
1270         list_node->map = map__new2(start, md->dso);
1271         if (!list_node->map) {
1272                 free(list_node);
1273                 return -ENOMEM;
1274         }
1275
1276         map__set_end(list_node->map, map__start(list_node->map) + len);
1277         map__set_pgoff(list_node->map, pgoff);
1278
1279         list_add(&list_node->node, &md->maps);
1280
1281         return 0;
1282 }
1283
1284 static bool remove_old_maps(struct map *map, void *data)
1285 {
1286         const struct map *map_to_save = data;
1287
1288         /*
1289          * We need to preserve eBPF maps even if they are covered by kcore,
1290          * because we need to access eBPF dso for source data.
1291          */
1292         return !RC_CHK_EQUAL(map, map_to_save) && !__map__is_bpf_prog(map);
1293 }
1294
1295 static int dso__load_kcore(struct dso *dso, struct map *map,
1296                            const char *kallsyms_filename)
1297 {
1298         struct maps *kmaps = map__kmaps(map);
1299         struct kcore_mapfn_data md;
1300         struct map *map_ref, *replacement_map = NULL;
1301         struct machine *machine;
1302         bool is_64_bit;
1303         int err, fd;
1304         char kcore_filename[PATH_MAX];
1305         u64 stext;
1306
1307         if (!kmaps)
1308                 return -EINVAL;
1309
1310         machine = maps__machine(kmaps);
1311
1312         /* This function requires that the map is the kernel map */
1313         if (!__map__is_kernel(map))
1314                 return -EINVAL;
1315
1316         if (!filename_from_kallsyms_filename(kcore_filename, "kcore",
1317                                              kallsyms_filename))
1318                 return -EINVAL;
1319
1320         /* Modules and kernel must be present at their original addresses */
1321         if (validate_kcore_addresses(kallsyms_filename, map))
1322                 return -EINVAL;
1323
1324         md.dso = dso;
1325         INIT_LIST_HEAD(&md.maps);
1326
1327         fd = open(kcore_filename, O_RDONLY);
1328         if (fd < 0) {
1329                 pr_debug("Failed to open %s. Note /proc/kcore requires CAP_SYS_RAWIO capability to access.\n",
1330                          kcore_filename);
1331                 return -EINVAL;
1332         }
1333
1334         /* Read new maps into temporary lists */
1335         err = file__read_maps(fd, map__prot(map) & PROT_EXEC, kcore_mapfn, &md,
1336                               &is_64_bit);
1337         if (err)
1338                 goto out_err;
1339         dso__set_is_64_bit(dso, is_64_bit);
1340
1341         if (list_empty(&md.maps)) {
1342                 err = -EINVAL;
1343                 goto out_err;
1344         }
1345
1346         /* Remove old maps */
1347         maps__remove_maps(kmaps, remove_old_maps, map);
1348         machine->trampolines_mapped = false;
1349
1350         /* Find the kernel map using the '_stext' symbol */
1351         if (!kallsyms__get_function_start(kallsyms_filename, "_stext", &stext)) {
1352                 u64 replacement_size = 0;
1353                 struct map_list_node *new_node;
1354
1355                 list_for_each_entry(new_node, &md.maps, node) {
1356                         struct map *new_map = new_node->map;
1357                         u64 new_size = map__size(new_map);
1358
1359                         if (!(stext >= map__start(new_map) && stext < map__end(new_map)))
1360                                 continue;
1361
1362                         /*
1363                          * On some architectures, ARM64 for example, the kernel
1364                          * text can get allocated inside of the vmalloc segment.
1365                          * Select the smallest matching segment, in case stext
1366                          * falls within more than one in the list.
1367                          */
1368                         if (!replacement_map || new_size < replacement_size) {
1369                                 replacement_map = new_map;
1370                                 replacement_size = new_size;
1371                         }
1372                 }
1373         }
1374
1375         if (!replacement_map)
1376                 replacement_map = list_entry(md.maps.next, struct map_list_node, node)->map;
1377
1378         /*
1379          * Update addresses of vmlinux map. Re-insert it to ensure maps are
1380          * correctly ordered. Do this before using maps__merge_in() for the
1381          * remaining maps so vmlinux gets split if necessary.
1382          */
1383         map_ref = map__get(map);
1384         maps__remove(kmaps, map_ref);
1385
1386         map__set_start(map_ref, map__start(replacement_map));
1387         map__set_end(map_ref, map__end(replacement_map));
1388         map__set_pgoff(map_ref, map__pgoff(replacement_map));
1389         map__set_mapping_type(map_ref, map__mapping_type(replacement_map));
1390
1391         err = maps__insert(kmaps, map_ref);
1392         map__put(map_ref);
1393         if (err)
1394                 goto out_err;
1395
1396         /* Add new maps */
1397         while (!list_empty(&md.maps)) {
1398                 struct map_list_node *new_node = list_entry(md.maps.next, struct map_list_node, node);
1399                 struct map *new_map = new_node->map;
1400
1401                 list_del_init(&new_node->node);
1402
1403                 /* skip if replacement_map, already inserted above */
1404                 if (!RC_CHK_EQUAL(new_map, replacement_map)) {
1405                         /*
1406                          * Merge kcore map into existing maps,
1407                          * and ensure that current maps (eBPF)
1408                          * stay intact.
1409                          */
1410                         if (maps__merge_in(kmaps, new_map)) {
1411                                 err = -EINVAL;
1412                                 goto out_err;
1413                         }
1414                 }
1415                 free(new_node);
1416         }
1417
1418         if (machine__is(machine, "x86_64")) {
1419                 u64 addr;
1420
1421                 /*
1422                  * If one of the corresponding symbols is there, assume the
1423                  * entry trampoline maps are too.
1424                  */
1425                 if (!kallsyms__get_function_start(kallsyms_filename,
1426                                                   ENTRY_TRAMPOLINE_NAME,
1427                                                   &addr))
1428                         machine->trampolines_mapped = true;
1429         }
1430
1431         /*
1432          * Set the data type and long name so that kcore can be read via
1433          * dso__data_read_addr().
1434          */
1435         if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST)
1436                 dso__set_binary_type(dso, DSO_BINARY_TYPE__GUEST_KCORE);
1437         else
1438                 dso__set_binary_type(dso, DSO_BINARY_TYPE__KCORE);
1439         dso__set_long_name(dso, strdup(kcore_filename), true);
1440
1441         close(fd);
1442
1443         if (map__prot(map) & PROT_EXEC)
1444                 pr_debug("Using %s for kernel object code\n", kcore_filename);
1445         else
1446                 pr_debug("Using %s for kernel data\n", kcore_filename);
1447
1448         return 0;
1449
1450 out_err:
1451         while (!list_empty(&md.maps)) {
1452                 struct map_list_node *list_node;
1453
1454                 list_node = list_entry(md.maps.next, struct map_list_node, node);
1455                 list_del_init(&list_node->node);
1456                 map__zput(list_node->map);
1457                 free(list_node);
1458         }
1459         close(fd);
1460         return err;
1461 }
1462
1463 /*
1464  * If the kernel is relocated at boot time, kallsyms won't match.  Compute the
1465  * delta based on the relocation reference symbol.
1466  */
1467 static int kallsyms__delta(struct kmap *kmap, const char *filename, u64 *delta)
1468 {
1469         u64 addr;
1470
1471         if (!kmap->ref_reloc_sym || !kmap->ref_reloc_sym->name)
1472                 return 0;
1473
1474         if (kallsyms__get_function_start(filename, kmap->ref_reloc_sym->name, &addr))
1475                 return -1;
1476
1477         *delta = addr - kmap->ref_reloc_sym->addr;
1478         return 0;
1479 }
1480
1481 int __dso__load_kallsyms(struct dso *dso, const char *filename,
1482                          struct map *map, bool no_kcore)
1483 {
1484         struct kmap *kmap = map__kmap(map);
1485         u64 delta = 0;
1486
1487         if (symbol__restricted_filename(filename, "/proc/kallsyms"))
1488                 return -1;
1489
1490         if (!kmap || !kmap->kmaps)
1491                 return -1;
1492
1493         if (dso__load_all_kallsyms(dso, filename) < 0)
1494                 return -1;
1495
1496         if (kallsyms__delta(kmap, filename, &delta))
1497                 return -1;
1498
1499         symbols__fixup_end(dso__symbols(dso), true);
1500         symbols__fixup_duplicate(dso__symbols(dso));
1501
1502         if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST)
1503                 dso__set_symtab_type(dso, DSO_BINARY_TYPE__GUEST_KALLSYMS);
1504         else
1505                 dso__set_symtab_type(dso, DSO_BINARY_TYPE__KALLSYMS);
1506
1507         if (!no_kcore && !dso__load_kcore(dso, map, filename))
1508                 return maps__split_kallsyms_for_kcore(kmap->kmaps, dso);
1509         else
1510                 return maps__split_kallsyms(kmap->kmaps, dso, delta, map);
1511 }
1512
1513 int dso__load_kallsyms(struct dso *dso, const char *filename,
1514                        struct map *map)
1515 {
1516         return __dso__load_kallsyms(dso, filename, map, false);
1517 }
1518
1519 static int dso__load_perf_map(const char *map_path, struct dso *dso)
1520 {
1521         char *line = NULL;
1522         size_t n;
1523         FILE *file;
1524         int nr_syms = 0;
1525
1526         file = fopen(map_path, "r");
1527         if (file == NULL)
1528                 goto out_failure;
1529
1530         while (!feof(file)) {
1531                 u64 start, size;
1532                 struct symbol *sym;
1533                 int line_len, len;
1534
1535                 line_len = getline(&line, &n, file);
1536                 if (line_len < 0)
1537                         break;
1538
1539                 if (!line)
1540                         goto out_failure;
1541
1542                 line[--line_len] = '\0'; /* \n */
1543
1544                 len = hex2u64(line, &start);
1545
1546                 len++;
1547                 if (len + 2 >= line_len)
1548                         continue;
1549
1550                 len += hex2u64(line + len, &size);
1551
1552                 len++;
1553                 if (len + 2 >= line_len)
1554                         continue;
1555
1556                 sym = symbol__new(start, size, STB_GLOBAL, STT_FUNC, line + len);
1557
1558                 if (sym == NULL)
1559                         goto out_delete_line;
1560
1561                 symbols__insert(dso__symbols(dso), sym);
1562                 nr_syms++;
1563         }
1564
1565         free(line);
1566         fclose(file);
1567
1568         return nr_syms;
1569
1570 out_delete_line:
1571         free(line);
1572 out_failure:
1573         return -1;
1574 }
1575
1576 #ifdef HAVE_LIBBFD_SUPPORT
1577 #define PACKAGE 'perf'
1578 #include <bfd.h>
1579
1580 static int bfd_symbols__cmpvalue(const void *a, const void *b)
1581 {
1582         const asymbol *as = *(const asymbol **)a, *bs = *(const asymbol **)b;
1583
1584         if (bfd_asymbol_value(as) != bfd_asymbol_value(bs))
1585                 return bfd_asymbol_value(as) - bfd_asymbol_value(bs);
1586
1587         return bfd_asymbol_name(as)[0] - bfd_asymbol_name(bs)[0];
1588 }
1589
1590 static int bfd2elf_binding(asymbol *symbol)
1591 {
1592         if (symbol->flags & BSF_WEAK)
1593                 return STB_WEAK;
1594         if (symbol->flags & BSF_GLOBAL)
1595                 return STB_GLOBAL;
1596         if (symbol->flags & BSF_LOCAL)
1597                 return STB_LOCAL;
1598         return -1;
1599 }
1600
1601 int dso__load_bfd_symbols(struct dso *dso, const char *debugfile)
1602 {
1603         int err = -1;
1604         long symbols_size, symbols_count, i;
1605         asection *section;
1606         asymbol **symbols, *sym;
1607         struct symbol *symbol;
1608         bfd *abfd;
1609         u64 start, len;
1610
1611         abfd = bfd_openr(debugfile, NULL);
1612         if (!abfd)
1613                 return -1;
1614
1615         if (!bfd_check_format(abfd, bfd_object)) {
1616                 pr_debug2("%s: cannot read %s bfd file.\n", __func__,
1617                           dso__long_name(dso));
1618                 goto out_close;
1619         }
1620
1621         if (bfd_get_flavour(abfd) == bfd_target_elf_flavour)
1622                 goto out_close;
1623
1624         symbols_size = bfd_get_symtab_upper_bound(abfd);
1625         if (symbols_size == 0) {
1626                 bfd_close(abfd);
1627                 return 0;
1628         }
1629
1630         if (symbols_size < 0)
1631                 goto out_close;
1632
1633         symbols = malloc(symbols_size);
1634         if (!symbols)
1635                 goto out_close;
1636
1637         symbols_count = bfd_canonicalize_symtab(abfd, symbols);
1638         if (symbols_count < 0)
1639                 goto out_free;
1640
1641         section = bfd_get_section_by_name(abfd, ".text");
1642         if (section) {
1643                 for (i = 0; i < symbols_count; ++i) {
1644                         if (!strcmp(bfd_asymbol_name(symbols[i]), "__ImageBase") ||
1645                             !strcmp(bfd_asymbol_name(symbols[i]), "__image_base__"))
1646                                 break;
1647                 }
1648                 if (i < symbols_count) {
1649                         /* PE symbols can only have 4 bytes, so use .text high bits */
1650                         u64 text_offset = (section->vma - (u32)section->vma)
1651                                 + (u32)bfd_asymbol_value(symbols[i]);
1652                         dso__set_text_offset(dso, text_offset);
1653                         dso__set_text_end(dso, (section->vma - text_offset) + section->size);
1654                 } else {
1655                         dso__set_text_offset(dso, section->vma - section->filepos);
1656                         dso__set_text_end(dso, section->filepos + section->size);
1657                 }
1658         }
1659
1660         qsort(symbols, symbols_count, sizeof(asymbol *), bfd_symbols__cmpvalue);
1661
1662 #ifdef bfd_get_section
1663 #define bfd_asymbol_section bfd_get_section
1664 #endif
1665         for (i = 0; i < symbols_count; ++i) {
1666                 sym = symbols[i];
1667                 section = bfd_asymbol_section(sym);
1668                 if (bfd2elf_binding(sym) < 0)
1669                         continue;
1670
1671                 while (i + 1 < symbols_count &&
1672                        bfd_asymbol_section(symbols[i + 1]) == section &&
1673                        bfd2elf_binding(symbols[i + 1]) < 0)
1674                         i++;
1675
1676                 if (i + 1 < symbols_count &&
1677                     bfd_asymbol_section(symbols[i + 1]) == section)
1678                         len = symbols[i + 1]->value - sym->value;
1679                 else
1680                         len = section->size - sym->value;
1681
1682                 start = bfd_asymbol_value(sym) - dso__text_offset(dso);
1683                 symbol = symbol__new(start, len, bfd2elf_binding(sym), STT_FUNC,
1684                                      bfd_asymbol_name(sym));
1685                 if (!symbol)
1686                         goto out_free;
1687
1688                 symbols__insert(dso__symbols(dso), symbol);
1689         }
1690 #ifdef bfd_get_section
1691 #undef bfd_asymbol_section
1692 #endif
1693
1694         symbols__fixup_end(dso__symbols(dso), false);
1695         symbols__fixup_duplicate(dso__symbols(dso));
1696         dso__set_adjust_symbols(dso, true);
1697
1698         err = 0;
1699 out_free:
1700         free(symbols);
1701 out_close:
1702         bfd_close(abfd);
1703         return err;
1704 }
1705 #endif
1706
1707 static bool dso__is_compatible_symtab_type(struct dso *dso, bool kmod,
1708                                            enum dso_binary_type type)
1709 {
1710         switch (type) {
1711         case DSO_BINARY_TYPE__JAVA_JIT:
1712         case DSO_BINARY_TYPE__DEBUGLINK:
1713         case DSO_BINARY_TYPE__SYSTEM_PATH_DSO:
1714         case DSO_BINARY_TYPE__FEDORA_DEBUGINFO:
1715         case DSO_BINARY_TYPE__UBUNTU_DEBUGINFO:
1716         case DSO_BINARY_TYPE__MIXEDUP_UBUNTU_DEBUGINFO:
1717         case DSO_BINARY_TYPE__BUILDID_DEBUGINFO:
1718         case DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO:
1719                 return !kmod && dso__kernel(dso) == DSO_SPACE__USER;
1720
1721         case DSO_BINARY_TYPE__KALLSYMS:
1722         case DSO_BINARY_TYPE__VMLINUX:
1723         case DSO_BINARY_TYPE__KCORE:
1724                 return dso__kernel(dso) == DSO_SPACE__KERNEL;
1725
1726         case DSO_BINARY_TYPE__GUEST_KALLSYMS:
1727         case DSO_BINARY_TYPE__GUEST_VMLINUX:
1728         case DSO_BINARY_TYPE__GUEST_KCORE:
1729                 return dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST;
1730
1731         case DSO_BINARY_TYPE__GUEST_KMODULE:
1732         case DSO_BINARY_TYPE__GUEST_KMODULE_COMP:
1733         case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE:
1734         case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP:
1735                 /*
1736                  * kernel modules know their symtab type - it's set when
1737                  * creating a module dso in machine__addnew_module_map().
1738                  */
1739                 return kmod && dso__symtab_type(dso) == type;
1740
1741         case DSO_BINARY_TYPE__BUILD_ID_CACHE:
1742         case DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO:
1743                 return true;
1744
1745         case DSO_BINARY_TYPE__BPF_PROG_INFO:
1746         case DSO_BINARY_TYPE__BPF_IMAGE:
1747         case DSO_BINARY_TYPE__OOL:
1748         case DSO_BINARY_TYPE__NOT_FOUND:
1749         default:
1750                 return false;
1751         }
1752 }
1753
1754 /* Checks for the existence of the perf-<pid>.map file in two different
1755  * locations.  First, if the process is a separate mount namespace, check in
1756  * that namespace using the pid of the innermost pid namespace.  If's not in a
1757  * namespace, or the file can't be found there, try in the mount namespace of
1758  * the tracing process using our view of its pid.
1759  */
1760 static int dso__find_perf_map(char *filebuf, size_t bufsz,
1761                               struct nsinfo **nsip)
1762 {
1763         struct nscookie nsc;
1764         struct nsinfo *nsi;
1765         struct nsinfo *nnsi;
1766         int rc = -1;
1767
1768         nsi = *nsip;
1769
1770         if (nsinfo__need_setns(nsi)) {
1771                 snprintf(filebuf, bufsz, "/tmp/perf-%d.map", nsinfo__nstgid(nsi));
1772                 nsinfo__mountns_enter(nsi, &nsc);
1773                 rc = access(filebuf, R_OK);
1774                 nsinfo__mountns_exit(&nsc);
1775                 if (rc == 0)
1776                         return rc;
1777         }
1778
1779         nnsi = nsinfo__copy(nsi);
1780         if (nnsi) {
1781                 nsinfo__put(nsi);
1782
1783                 nsinfo__clear_need_setns(nnsi);
1784                 snprintf(filebuf, bufsz, "/tmp/perf-%d.map", nsinfo__tgid(nnsi));
1785                 *nsip = nnsi;
1786                 rc = 0;
1787         }
1788
1789         return rc;
1790 }
1791
1792 int dso__load(struct dso *dso, struct map *map)
1793 {
1794         char *name;
1795         int ret = -1;
1796         u_int i;
1797         struct machine *machine = NULL;
1798         char *root_dir = (char *) "";
1799         int ss_pos = 0;
1800         struct symsrc ss_[2];
1801         struct symsrc *syms_ss = NULL, *runtime_ss = NULL;
1802         bool kmod;
1803         bool perfmap;
1804         struct build_id bid;
1805         struct nscookie nsc;
1806         char newmapname[PATH_MAX];
1807         const char *map_path = dso__long_name(dso);
1808
1809         mutex_lock(dso__lock(dso));
1810         perfmap = is_perf_pid_map_name(map_path);
1811
1812         if (perfmap) {
1813                 if (dso__nsinfo(dso) &&
1814                     (dso__find_perf_map(newmapname, sizeof(newmapname),
1815                                         dso__nsinfo_ptr(dso)) == 0)) {
1816                         map_path = newmapname;
1817                 }
1818         }
1819
1820         nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
1821
1822         /* check again under the dso->lock */
1823         if (dso__loaded(dso)) {
1824                 ret = 1;
1825                 goto out;
1826         }
1827
1828         kmod = dso__is_kmod(dso);
1829
1830         if (dso__kernel(dso) && !kmod) {
1831                 if (dso__kernel(dso) == DSO_SPACE__KERNEL)
1832                         ret = dso__load_kernel_sym(dso, map);
1833                 else if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST)
1834                         ret = dso__load_guest_kernel_sym(dso, map);
1835
1836                 machine = maps__machine(map__kmaps(map));
1837                 if (machine__is(machine, "x86_64"))
1838                         machine__map_x86_64_entry_trampolines(machine, dso);
1839                 goto out;
1840         }
1841
1842         dso__set_adjust_symbols(dso, false);
1843
1844         if (perfmap) {
1845                 ret = dso__load_perf_map(map_path, dso);
1846                 dso__set_symtab_type(dso, ret > 0
1847                                 ? DSO_BINARY_TYPE__JAVA_JIT
1848                                 : DSO_BINARY_TYPE__NOT_FOUND);
1849                 goto out;
1850         }
1851
1852         if (machine)
1853                 root_dir = machine->root_dir;
1854
1855         name = malloc(PATH_MAX);
1856         if (!name)
1857                 goto out;
1858
1859         /*
1860          * Read the build id if possible. This is required for
1861          * DSO_BINARY_TYPE__BUILDID_DEBUGINFO to work
1862          */
1863         if (!dso__has_build_id(dso) &&
1864             is_regular_file(dso__long_name(dso))) {
1865                 __symbol__join_symfs(name, PATH_MAX, dso__long_name(dso));
1866                 if (filename__read_build_id(name, &bid) > 0)
1867                         dso__set_build_id(dso, &bid);
1868         }
1869
1870         /*
1871          * Iterate over candidate debug images.
1872          * Keep track of "interesting" ones (those which have a symtab, dynsym,
1873          * and/or opd section) for processing.
1874          */
1875         for (i = 0; i < DSO_BINARY_TYPE__SYMTAB_CNT; i++) {
1876                 struct symsrc *ss = &ss_[ss_pos];
1877                 bool next_slot = false;
1878                 bool is_reg;
1879                 bool nsexit;
1880                 int bfdrc = -1;
1881                 int sirc = -1;
1882
1883                 enum dso_binary_type symtab_type = binary_type_symtab[i];
1884
1885                 nsexit = (symtab_type == DSO_BINARY_TYPE__BUILD_ID_CACHE ||
1886                     symtab_type == DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO);
1887
1888                 if (!dso__is_compatible_symtab_type(dso, kmod, symtab_type))
1889                         continue;
1890
1891                 if (dso__read_binary_type_filename(dso, symtab_type,
1892                                                    root_dir, name, PATH_MAX))
1893                         continue;
1894
1895                 if (nsexit)
1896                         nsinfo__mountns_exit(&nsc);
1897
1898                 is_reg = is_regular_file(name);
1899                 if (!is_reg && errno == ENOENT && dso__nsinfo(dso)) {
1900                         char *new_name = dso__filename_with_chroot(dso, name);
1901                         if (new_name) {
1902                                 is_reg = is_regular_file(new_name);
1903                                 strlcpy(name, new_name, PATH_MAX);
1904                                 free(new_name);
1905                         }
1906                 }
1907
1908 #ifdef HAVE_LIBBFD_SUPPORT
1909                 if (is_reg)
1910                         bfdrc = dso__load_bfd_symbols(dso, name);
1911 #endif
1912                 if (is_reg && bfdrc < 0)
1913                         sirc = symsrc__init(ss, dso, name, symtab_type);
1914
1915                 if (nsexit)
1916                         nsinfo__mountns_enter(dso__nsinfo(dso), &nsc);
1917
1918                 if (bfdrc == 0) {
1919                         ret = 0;
1920                         break;
1921                 }
1922
1923                 if (!is_reg || sirc < 0)
1924                         continue;
1925
1926                 if (!syms_ss && symsrc__has_symtab(ss)) {
1927                         syms_ss = ss;
1928                         next_slot = true;
1929                         if (!dso__symsrc_filename(dso))
1930                                 dso__set_symsrc_filename(dso, strdup(name));
1931                 }
1932
1933                 if (!runtime_ss && symsrc__possibly_runtime(ss)) {
1934                         runtime_ss = ss;
1935                         next_slot = true;
1936                 }
1937
1938                 if (next_slot) {
1939                         ss_pos++;
1940
1941                         if (dso__binary_type(dso) == DSO_BINARY_TYPE__NOT_FOUND)
1942                                 dso__set_binary_type(dso, symtab_type);
1943
1944                         if (syms_ss && runtime_ss)
1945                                 break;
1946                 } else {
1947                         symsrc__destroy(ss);
1948                 }
1949
1950         }
1951
1952         if (!runtime_ss && !syms_ss)
1953                 goto out_free;
1954
1955         if (runtime_ss && !syms_ss) {
1956                 syms_ss = runtime_ss;
1957         }
1958
1959         /* We'll have to hope for the best */
1960         if (!runtime_ss && syms_ss)
1961                 runtime_ss = syms_ss;
1962
1963         if (syms_ss)
1964                 ret = dso__load_sym(dso, map, syms_ss, runtime_ss, kmod);
1965         else
1966                 ret = -1;
1967
1968         if (ret > 0) {
1969                 int nr_plt;
1970
1971                 nr_plt = dso__synthesize_plt_symbols(dso, runtime_ss);
1972                 if (nr_plt > 0)
1973                         ret += nr_plt;
1974         }
1975
1976         for (; ss_pos > 0; ss_pos--)
1977                 symsrc__destroy(&ss_[ss_pos - 1]);
1978 out_free:
1979         free(name);
1980         if (ret < 0 && strstr(dso__name(dso), " (deleted)") != NULL)
1981                 ret = 0;
1982 out:
1983         dso__set_loaded(dso);
1984         mutex_unlock(dso__lock(dso));
1985         nsinfo__mountns_exit(&nsc);
1986
1987         return ret;
1988 }
1989
1990 /*
1991  * Always takes ownership of vmlinux when vmlinux_allocated == true, even if
1992  * it returns an error.
1993  */
1994 int dso__load_vmlinux(struct dso *dso, struct map *map,
1995                       const char *vmlinux, bool vmlinux_allocated)
1996 {
1997         int err = -1;
1998         struct symsrc ss;
1999         char symfs_vmlinux[PATH_MAX];
2000         enum dso_binary_type symtab_type;
2001
2002         if (vmlinux[0] == '/')
2003                 snprintf(symfs_vmlinux, sizeof(symfs_vmlinux), "%s", vmlinux);
2004         else
2005                 symbol__join_symfs(symfs_vmlinux, vmlinux);
2006
2007         if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST)
2008                 symtab_type = DSO_BINARY_TYPE__GUEST_VMLINUX;
2009         else
2010                 symtab_type = DSO_BINARY_TYPE__VMLINUX;
2011
2012         if (symsrc__init(&ss, dso, symfs_vmlinux, symtab_type)) {
2013                 if (vmlinux_allocated)
2014                         free((char *) vmlinux);
2015                 return -1;
2016         }
2017
2018         /*
2019          * dso__load_sym() may copy 'dso' which will result in the copies having
2020          * an incorrect long name unless we set it here first.
2021          */
2022         dso__set_long_name(dso, vmlinux, vmlinux_allocated);
2023         if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST)
2024                 dso__set_binary_type(dso, DSO_BINARY_TYPE__GUEST_VMLINUX);
2025         else
2026                 dso__set_binary_type(dso, DSO_BINARY_TYPE__VMLINUX);
2027
2028         err = dso__load_sym(dso, map, &ss, &ss, 0);
2029         symsrc__destroy(&ss);
2030
2031         if (err > 0) {
2032                 dso__set_loaded(dso);
2033                 pr_debug("Using %s for symbols\n", symfs_vmlinux);
2034         }
2035
2036         return err;
2037 }
2038
2039 int dso__load_vmlinux_path(struct dso *dso, struct map *map)
2040 {
2041         int i, err = 0;
2042         char *filename = NULL;
2043
2044         pr_debug("Looking at the vmlinux_path (%d entries long)\n",
2045                  vmlinux_path__nr_entries + 1);
2046
2047         for (i = 0; i < vmlinux_path__nr_entries; ++i) {
2048                 err = dso__load_vmlinux(dso, map, vmlinux_path[i], false);
2049                 if (err > 0)
2050                         goto out;
2051         }
2052
2053         if (!symbol_conf.ignore_vmlinux_buildid)
2054                 filename = dso__build_id_filename(dso, NULL, 0, false);
2055         if (filename != NULL) {
2056                 err = dso__load_vmlinux(dso, map, filename, true);
2057                 if (err > 0)
2058                         goto out;
2059         }
2060 out:
2061         return err;
2062 }
2063
2064 static bool visible_dir_filter(const char *name, struct dirent *d)
2065 {
2066         if (d->d_type != DT_DIR)
2067                 return false;
2068         return lsdir_no_dot_filter(name, d);
2069 }
2070
2071 static int find_matching_kcore(struct map *map, char *dir, size_t dir_sz)
2072 {
2073         char kallsyms_filename[PATH_MAX];
2074         int ret = -1;
2075         struct strlist *dirs;
2076         struct str_node *nd;
2077
2078         dirs = lsdir(dir, visible_dir_filter);
2079         if (!dirs)
2080                 return -1;
2081
2082         strlist__for_each_entry(nd, dirs) {
2083                 scnprintf(kallsyms_filename, sizeof(kallsyms_filename),
2084                           "%s/%s/kallsyms", dir, nd->s);
2085                 if (!validate_kcore_addresses(kallsyms_filename, map)) {
2086                         strlcpy(dir, kallsyms_filename, dir_sz);
2087                         ret = 0;
2088                         break;
2089                 }
2090         }
2091
2092         strlist__delete(dirs);
2093
2094         return ret;
2095 }
2096
2097 /*
2098  * Use open(O_RDONLY) to check readability directly instead of access(R_OK)
2099  * since access(R_OK) only checks with real UID/GID but open() use effective
2100  * UID/GID and actual capabilities (e.g. /proc/kcore requires CAP_SYS_RAWIO).
2101  */
2102 static bool filename__readable(const char *file)
2103 {
2104         int fd = open(file, O_RDONLY);
2105         if (fd < 0)
2106                 return false;
2107         close(fd);
2108         return true;
2109 }
2110
2111 static char *dso__find_kallsyms(struct dso *dso, struct map *map)
2112 {
2113         struct build_id bid;
2114         char sbuild_id[SBUILD_ID_SIZE];
2115         bool is_host = false;
2116         char path[PATH_MAX];
2117
2118         if (!dso__has_build_id(dso)) {
2119                 /*
2120                  * Last resort, if we don't have a build-id and couldn't find
2121                  * any vmlinux file, try the running kernel kallsyms table.
2122                  */
2123                 goto proc_kallsyms;
2124         }
2125
2126         if (sysfs__read_build_id("/sys/kernel/notes", &bid) == 0)
2127                 is_host = dso__build_id_equal(dso, &bid);
2128
2129         /* Try a fast path for /proc/kallsyms if possible */
2130         if (is_host) {
2131                 /*
2132                  * Do not check the build-id cache, unless we know we cannot use
2133                  * /proc/kcore or module maps don't match to /proc/kallsyms.
2134                  * To check readability of /proc/kcore, do not use access(R_OK)
2135                  * since /proc/kcore requires CAP_SYS_RAWIO to read and access
2136                  * can't check it.
2137                  */
2138                 if (filename__readable("/proc/kcore") &&
2139                     !validate_kcore_addresses("/proc/kallsyms", map))
2140                         goto proc_kallsyms;
2141         }
2142
2143         build_id__sprintf(dso__bid(dso), sbuild_id);
2144
2145         /* Find kallsyms in build-id cache with kcore */
2146         scnprintf(path, sizeof(path), "%s/%s/%s",
2147                   buildid_dir, DSO__NAME_KCORE, sbuild_id);
2148
2149         if (!find_matching_kcore(map, path, sizeof(path)))
2150                 return strdup(path);
2151
2152         /* Use current /proc/kallsyms if possible */
2153         if (is_host) {
2154 proc_kallsyms:
2155                 return strdup("/proc/kallsyms");
2156         }
2157
2158         /* Finally, find a cache of kallsyms */
2159         if (!build_id_cache__kallsyms_path(sbuild_id, path, sizeof(path))) {
2160                 pr_err("No kallsyms or vmlinux with build-id %s was found\n",
2161                        sbuild_id);
2162                 return NULL;
2163         }
2164
2165         return strdup(path);
2166 }
2167
2168 static int dso__load_kernel_sym(struct dso *dso, struct map *map)
2169 {
2170         int err;
2171         const char *kallsyms_filename = NULL;
2172         char *kallsyms_allocated_filename = NULL;
2173         char *filename = NULL;
2174
2175         /*
2176          * Step 1: if the user specified a kallsyms or vmlinux filename, use
2177          * it and only it, reporting errors to the user if it cannot be used.
2178          *
2179          * For instance, try to analyse an ARM perf.data file _without_ a
2180          * build-id, or if the user specifies the wrong path to the right
2181          * vmlinux file, obviously we can't fallback to another vmlinux (a
2182          * x86_86 one, on the machine where analysis is being performed, say),
2183          * or worse, /proc/kallsyms.
2184          *
2185          * If the specified file _has_ a build-id and there is a build-id
2186          * section in the perf.data file, we will still do the expected
2187          * validation in dso__load_vmlinux and will bail out if they don't
2188          * match.
2189          */
2190         if (symbol_conf.kallsyms_name != NULL) {
2191                 kallsyms_filename = symbol_conf.kallsyms_name;
2192                 goto do_kallsyms;
2193         }
2194
2195         if (!symbol_conf.ignore_vmlinux && symbol_conf.vmlinux_name != NULL) {
2196                 return dso__load_vmlinux(dso, map, symbol_conf.vmlinux_name, false);
2197         }
2198
2199         /*
2200          * Before checking on common vmlinux locations, check if it's
2201          * stored as standard build id binary (not kallsyms) under
2202          * .debug cache.
2203          */
2204         if (!symbol_conf.ignore_vmlinux_buildid)
2205                 filename = __dso__build_id_filename(dso, NULL, 0, false, false);
2206         if (filename != NULL) {
2207                 err = dso__load_vmlinux(dso, map, filename, true);
2208                 if (err > 0)
2209                         return err;
2210         }
2211
2212         if (!symbol_conf.ignore_vmlinux && vmlinux_path != NULL) {
2213                 err = dso__load_vmlinux_path(dso, map);
2214                 if (err > 0)
2215                         return err;
2216         }
2217
2218         /* do not try local files if a symfs was given */
2219         if (symbol_conf.symfs[0] != 0)
2220                 return -1;
2221
2222         kallsyms_allocated_filename = dso__find_kallsyms(dso, map);
2223         if (!kallsyms_allocated_filename)
2224                 return -1;
2225
2226         kallsyms_filename = kallsyms_allocated_filename;
2227
2228 do_kallsyms:
2229         err = dso__load_kallsyms(dso, kallsyms_filename, map);
2230         if (err > 0)
2231                 pr_debug("Using %s for symbols\n", kallsyms_filename);
2232         free(kallsyms_allocated_filename);
2233
2234         if (err > 0 && !dso__is_kcore(dso)) {
2235                 dso__set_binary_type(dso, DSO_BINARY_TYPE__KALLSYMS);
2236                 dso__set_long_name(dso, DSO__NAME_KALLSYMS, false);
2237                 map__fixup_start(map);
2238                 map__fixup_end(map);
2239         }
2240
2241         return err;
2242 }
2243
2244 static int dso__load_guest_kernel_sym(struct dso *dso, struct map *map)
2245 {
2246         int err;
2247         const char *kallsyms_filename;
2248         struct machine *machine = maps__machine(map__kmaps(map));
2249         char path[PATH_MAX];
2250
2251         if (machine->kallsyms_filename) {
2252                 kallsyms_filename = machine->kallsyms_filename;
2253         } else if (machine__is_default_guest(machine)) {
2254                 /*
2255                  * if the user specified a vmlinux filename, use it and only
2256                  * it, reporting errors to the user if it cannot be used.
2257                  * Or use file guest_kallsyms inputted by user on commandline
2258                  */
2259                 if (symbol_conf.default_guest_vmlinux_name != NULL) {
2260                         err = dso__load_vmlinux(dso, map,
2261                                                 symbol_conf.default_guest_vmlinux_name,
2262                                                 false);
2263                         return err;
2264                 }
2265
2266                 kallsyms_filename = symbol_conf.default_guest_kallsyms;
2267                 if (!kallsyms_filename)
2268                         return -1;
2269         } else {
2270                 sprintf(path, "%s/proc/kallsyms", machine->root_dir);
2271                 kallsyms_filename = path;
2272         }
2273
2274         err = dso__load_kallsyms(dso, kallsyms_filename, map);
2275         if (err > 0)
2276                 pr_debug("Using %s for symbols\n", kallsyms_filename);
2277         if (err > 0 && !dso__is_kcore(dso)) {
2278                 dso__set_binary_type(dso, DSO_BINARY_TYPE__GUEST_KALLSYMS);
2279                 dso__set_long_name(dso, machine->mmap_name, false);
2280                 map__fixup_start(map);
2281                 map__fixup_end(map);
2282         }
2283
2284         return err;
2285 }
2286
2287 static void vmlinux_path__exit(void)
2288 {
2289         while (--vmlinux_path__nr_entries >= 0)
2290                 zfree(&vmlinux_path[vmlinux_path__nr_entries]);
2291         vmlinux_path__nr_entries = 0;
2292
2293         zfree(&vmlinux_path);
2294 }
2295
2296 static const char * const vmlinux_paths[] = {
2297         "vmlinux",
2298         "/boot/vmlinux"
2299 };
2300
2301 static const char * const vmlinux_paths_upd[] = {
2302         "/boot/vmlinux-%s",
2303         "/usr/lib/debug/boot/vmlinux-%s",
2304         "/lib/modules/%s/build/vmlinux",
2305         "/usr/lib/debug/lib/modules/%s/vmlinux",
2306         "/usr/lib/debug/boot/vmlinux-%s.debug"
2307 };
2308
2309 static int vmlinux_path__add(const char *new_entry)
2310 {
2311         vmlinux_path[vmlinux_path__nr_entries] = strdup(new_entry);
2312         if (vmlinux_path[vmlinux_path__nr_entries] == NULL)
2313                 return -1;
2314         ++vmlinux_path__nr_entries;
2315
2316         return 0;
2317 }
2318
2319 static int vmlinux_path__init(struct perf_env *env)
2320 {
2321         struct utsname uts;
2322         char bf[PATH_MAX];
2323         char *kernel_version;
2324         unsigned int i;
2325
2326         vmlinux_path = malloc(sizeof(char *) * (ARRAY_SIZE(vmlinux_paths) +
2327                               ARRAY_SIZE(vmlinux_paths_upd)));
2328         if (vmlinux_path == NULL)
2329                 return -1;
2330
2331         for (i = 0; i < ARRAY_SIZE(vmlinux_paths); i++)
2332                 if (vmlinux_path__add(vmlinux_paths[i]) < 0)
2333                         goto out_fail;
2334
2335         /* only try kernel version if no symfs was given */
2336         if (symbol_conf.symfs[0] != 0)
2337                 return 0;
2338
2339         if (env) {
2340                 kernel_version = env->os_release;
2341         } else {
2342                 if (uname(&uts) < 0)
2343                         goto out_fail;
2344
2345                 kernel_version = uts.release;
2346         }
2347
2348         for (i = 0; i < ARRAY_SIZE(vmlinux_paths_upd); i++) {
2349                 snprintf(bf, sizeof(bf), vmlinux_paths_upd[i], kernel_version);
2350                 if (vmlinux_path__add(bf) < 0)
2351                         goto out_fail;
2352         }
2353
2354         return 0;
2355
2356 out_fail:
2357         vmlinux_path__exit();
2358         return -1;
2359 }
2360
2361 int setup_list(struct strlist **list, const char *list_str,
2362                       const char *list_name)
2363 {
2364         if (list_str == NULL)
2365                 return 0;
2366
2367         *list = strlist__new(list_str, NULL);
2368         if (!*list) {
2369                 pr_err("problems parsing %s list\n", list_name);
2370                 return -1;
2371         }
2372
2373         symbol_conf.has_filter = true;
2374         return 0;
2375 }
2376
2377 int setup_intlist(struct intlist **list, const char *list_str,
2378                   const char *list_name)
2379 {
2380         if (list_str == NULL)
2381                 return 0;
2382
2383         *list = intlist__new(list_str);
2384         if (!*list) {
2385                 pr_err("problems parsing %s list\n", list_name);
2386                 return -1;
2387         }
2388         return 0;
2389 }
2390
2391 static int setup_addrlist(struct intlist **addr_list, struct strlist *sym_list)
2392 {
2393         struct str_node *pos, *tmp;
2394         unsigned long val;
2395         char *sep;
2396         const char *end;
2397         int i = 0, err;
2398
2399         *addr_list = intlist__new(NULL);
2400         if (!*addr_list)
2401                 return -1;
2402
2403         strlist__for_each_entry_safe(pos, tmp, sym_list) {
2404                 errno = 0;
2405                 val = strtoul(pos->s, &sep, 16);
2406                 if (errno || (sep == pos->s))
2407                         continue;
2408
2409                 if (*sep != '\0') {
2410                         end = pos->s + strlen(pos->s) - 1;
2411                         while (end >= sep && isspace(*end))
2412                                 end--;
2413
2414                         if (end >= sep)
2415                                 continue;
2416                 }
2417
2418                 err = intlist__add(*addr_list, val);
2419                 if (err)
2420                         break;
2421
2422                 strlist__remove(sym_list, pos);
2423                 i++;
2424         }
2425
2426         if (i == 0) {
2427                 intlist__delete(*addr_list);
2428                 *addr_list = NULL;
2429         }
2430
2431         return 0;
2432 }
2433
2434 static bool symbol__read_kptr_restrict(void)
2435 {
2436         bool value = false;
2437         FILE *fp = fopen("/proc/sys/kernel/kptr_restrict", "r");
2438         bool used_root;
2439         bool cap_syslog = perf_cap__capable(CAP_SYSLOG, &used_root);
2440
2441         if (fp != NULL) {
2442                 char line[8];
2443
2444                 if (fgets(line, sizeof(line), fp) != NULL)
2445                         value = cap_syslog ? (atoi(line) >= 2) : (atoi(line) != 0);
2446
2447                 fclose(fp);
2448         }
2449
2450         /* Per kernel/kallsyms.c:
2451          * we also restrict when perf_event_paranoid > 1 w/o CAP_SYSLOG
2452          */
2453         if (perf_event_paranoid() > 1 && !cap_syslog)
2454                 value = true;
2455
2456         return value;
2457 }
2458
2459 int symbol__annotation_init(void)
2460 {
2461         if (symbol_conf.init_annotation)
2462                 return 0;
2463
2464         if (symbol_conf.initialized) {
2465                 pr_err("Annotation needs to be init before symbol__init()\n");
2466                 return -1;
2467         }
2468
2469         symbol_conf.priv_size += sizeof(struct annotation);
2470         symbol_conf.init_annotation = true;
2471         return 0;
2472 }
2473
2474 int symbol__init(struct perf_env *env)
2475 {
2476         const char *symfs;
2477
2478         if (symbol_conf.initialized)
2479                 return 0;
2480
2481         symbol_conf.priv_size = PERF_ALIGN(symbol_conf.priv_size, sizeof(u64));
2482
2483         symbol__elf_init();
2484
2485         if (symbol_conf.try_vmlinux_path && vmlinux_path__init(env) < 0)
2486                 return -1;
2487
2488         if (symbol_conf.field_sep && *symbol_conf.field_sep == '.') {
2489                 pr_err("'.' is the only non valid --field-separator argument\n");
2490                 return -1;
2491         }
2492
2493         if (setup_list(&symbol_conf.dso_list,
2494                        symbol_conf.dso_list_str, "dso") < 0)
2495                 return -1;
2496
2497         if (setup_list(&symbol_conf.comm_list,
2498                        symbol_conf.comm_list_str, "comm") < 0)
2499                 goto out_free_dso_list;
2500
2501         if (setup_intlist(&symbol_conf.pid_list,
2502                        symbol_conf.pid_list_str, "pid") < 0)
2503                 goto out_free_comm_list;
2504
2505         if (setup_intlist(&symbol_conf.tid_list,
2506                        symbol_conf.tid_list_str, "tid") < 0)
2507                 goto out_free_pid_list;
2508
2509         if (setup_list(&symbol_conf.sym_list,
2510                        symbol_conf.sym_list_str, "symbol") < 0)
2511                 goto out_free_tid_list;
2512
2513         if (symbol_conf.sym_list &&
2514             setup_addrlist(&symbol_conf.addr_list, symbol_conf.sym_list) < 0)
2515                 goto out_free_sym_list;
2516
2517         if (setup_list(&symbol_conf.bt_stop_list,
2518                        symbol_conf.bt_stop_list_str, "symbol") < 0)
2519                 goto out_free_sym_list;
2520
2521         /*
2522          * A path to symbols of "/" is identical to ""
2523          * reset here for simplicity.
2524          */
2525         symfs = realpath(symbol_conf.symfs, NULL);
2526         if (symfs == NULL)
2527                 symfs = symbol_conf.symfs;
2528         if (strcmp(symfs, "/") == 0)
2529                 symbol_conf.symfs = "";
2530         if (symfs != symbol_conf.symfs)
2531                 free((void *)symfs);
2532
2533         symbol_conf.kptr_restrict = symbol__read_kptr_restrict();
2534
2535         symbol_conf.initialized = true;
2536         return 0;
2537
2538 out_free_sym_list:
2539         strlist__delete(symbol_conf.sym_list);
2540         intlist__delete(symbol_conf.addr_list);
2541 out_free_tid_list:
2542         intlist__delete(symbol_conf.tid_list);
2543 out_free_pid_list:
2544         intlist__delete(symbol_conf.pid_list);
2545 out_free_comm_list:
2546         strlist__delete(symbol_conf.comm_list);
2547 out_free_dso_list:
2548         strlist__delete(symbol_conf.dso_list);
2549         return -1;
2550 }
2551
2552 void symbol__exit(void)
2553 {
2554         if (!symbol_conf.initialized)
2555                 return;
2556         strlist__delete(symbol_conf.bt_stop_list);
2557         strlist__delete(symbol_conf.sym_list);
2558         strlist__delete(symbol_conf.dso_list);
2559         strlist__delete(symbol_conf.comm_list);
2560         intlist__delete(symbol_conf.tid_list);
2561         intlist__delete(symbol_conf.pid_list);
2562         intlist__delete(symbol_conf.addr_list);
2563         vmlinux_path__exit();
2564         symbol_conf.sym_list = symbol_conf.dso_list = symbol_conf.comm_list = NULL;
2565         symbol_conf.bt_stop_list = NULL;
2566         symbol_conf.initialized = false;
2567 }
2568
2569 int symbol__config_symfs(const struct option *opt __maybe_unused,
2570                          const char *dir, int unset __maybe_unused)
2571 {
2572         char *bf = NULL;
2573         int ret;
2574
2575         symbol_conf.symfs = strdup(dir);
2576         if (symbol_conf.symfs == NULL)
2577                 return -ENOMEM;
2578
2579         /* skip the locally configured cache if a symfs is given, and
2580          * config buildid dir to symfs/.debug
2581          */
2582         ret = asprintf(&bf, "%s/%s", dir, ".debug");
2583         if (ret < 0)
2584                 return -ENOMEM;
2585
2586         set_buildid_dir(bf);
2587
2588         free(bf);
2589         return 0;
2590 }
2591
2592 /*
2593  * Checks that user supplied symbol kernel files are accessible because
2594  * the default mechanism for accessing elf files fails silently. i.e. if
2595  * debug syms for a build ID aren't found perf carries on normally. When
2596  * they are user supplied we should assume that the user doesn't want to
2597  * silently fail.
2598  */
2599 int symbol__validate_sym_arguments(void)
2600 {
2601         if (symbol_conf.vmlinux_name &&
2602             access(symbol_conf.vmlinux_name, R_OK)) {
2603                 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
2604                 return -EINVAL;
2605         }
2606         if (symbol_conf.kallsyms_name &&
2607             access(symbol_conf.kallsyms_name, R_OK)) {
2608                 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
2609                 return -EINVAL;
2610         }
2611         return 0;
2612 }