1 /* DWARF index writing support for GDB. 2 3 Copyright (C) 1994-2023 Free Software Foundation, Inc. 4 5 This file is part of GDB. 6 7 This program is free software; you can redistribute it and/or modify 8 it under the terms of the GNU General Public License as published by 9 the Free Software Foundation; either version 3 of the License, or 10 (at your option) any later version. 11 12 This program is distributed in the hope that it will be useful, 13 but WITHOUT ANY WARRANTY; without even the implied warranty of 14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 15 GNU General Public License for more details. 16 17 You should have received a copy of the GNU General Public License 18 along with this program. If not, see <http://www.gnu.org/licenses/>. */ 19 20 #include "defs.h" 21 22 #include "dwarf2/index-write.h" 23 24 #include "addrmap.h" 25 #include "cli/cli-decode.h" 26 #include "gdbsupport/byte-vector.h" 27 #include "gdbsupport/filestuff.h" 28 #include "gdbsupport/gdb_unlinker.h" 29 #include "gdbsupport/pathstuff.h" 30 #include "gdbsupport/scoped_fd.h" 31 #include "complaints.h" 32 #include "dwarf2/index-common.h" 33 #include "dwarf2.h" 34 #include "dwarf2/read.h" 35 #include "dwarf2/dwz.h" 36 #include "gdb/gdb-index.h" 37 #include "gdbcmd.h" 38 #include "objfiles.h" 39 #include "ada-lang.h" 40 #include "dwarf2/tag.h" 41 42 #include <algorithm> 43 #include <cmath> 44 #include <forward_list> 45 #include <set> 46 #include <unordered_map> 47 #include <unordered_set> 48 49 /* Ensure only legit values are used. */ 50 #define DW2_GDB_INDEX_SYMBOL_STATIC_SET_VALUE(cu_index, value) \ 51 do { \ 52 gdb_assert ((unsigned int) (value) <= 1); \ 53 GDB_INDEX_SYMBOL_STATIC_SET_VALUE((cu_index), (value)); \ 54 } while (0) 55 56 /* Ensure only legit values are used. */ 57 #define DW2_GDB_INDEX_SYMBOL_KIND_SET_VALUE(cu_index, value) \ 58 do { \ 59 gdb_assert ((value) >= GDB_INDEX_SYMBOL_KIND_TYPE \ 60 && (value) <= GDB_INDEX_SYMBOL_KIND_OTHER); \ 61 GDB_INDEX_SYMBOL_KIND_SET_VALUE((cu_index), (value)); \ 62 } while (0) 63 64 /* Ensure we don't use more than the allotted number of bits for the CU. */ 65 #define DW2_GDB_INDEX_CU_SET_VALUE(cu_index, value) \ 66 do { \ 67 gdb_assert (((value) & ~GDB_INDEX_CU_MASK) == 0); \ 68 GDB_INDEX_CU_SET_VALUE((cu_index), (value)); \ 69 } while (0) 70 71 /* The "save gdb-index" command. */ 72 73 /* Write SIZE bytes from the buffer pointed to by DATA to FILE, with 74 error checking. */ 75 76 static void 77 file_write (FILE *file, const void *data, size_t size) 78 { 79 if (fwrite (data, 1, size, file) != size) 80 error (_("couldn't data write to file")); 81 } 82 83 /* Write the contents of VEC to FILE, with error checking. */ 84 85 template<typename Elem, typename Alloc> 86 static void 87 file_write (FILE *file, const std::vector<Elem, Alloc> &vec) 88 { 89 if (!vec.empty ()) 90 file_write (file, vec.data (), vec.size () * sizeof (vec[0])); 91 } 92 93 /* In-memory buffer to prepare data to be written later to a file. */ 94 class data_buf 95 { 96 public: 97 /* Copy ARRAY to the end of the buffer. */ 98 void append_array (gdb::array_view<const gdb_byte> array) 99 { 100 std::copy (array.begin (), array.end (), grow (array.size ())); 101 } 102 103 /* Copy CSTR (a zero-terminated string) to the end of buffer. The 104 terminating zero is appended too. */ 105 void append_cstr0 (const char *cstr) 106 { 107 const size_t size = strlen (cstr) + 1; 108 std::copy (cstr, cstr + size, grow (size)); 109 } 110 111 /* Store INPUT as ULEB128 to the end of buffer. */ 112 void append_unsigned_leb128 (ULONGEST input) 113 { 114 for (;;) 115 { 116 gdb_byte output = input & 0x7f; 117 input >>= 7; 118 if (input) 119 output |= 0x80; 120 m_vec.push_back (output); 121 if (input == 0) 122 break; 123 } 124 } 125 126 /* Accept a host-format integer in VAL and append it to the buffer 127 as a target-format integer which is LEN bytes long. */ 128 void append_uint (size_t len, bfd_endian byte_order, ULONGEST val) 129 { 130 ::store_unsigned_integer (grow (len), len, byte_order, val); 131 } 132 133 /* Copy VALUE to the end of the buffer, little-endian. */ 134 void append_offset (offset_type value) 135 { 136 append_uint (sizeof (value), BFD_ENDIAN_LITTLE, value); 137 } 138 139 /* Return the size of the buffer. */ 140 size_t size () const 141 { 142 return m_vec.size (); 143 } 144 145 /* Return true iff the buffer is empty. */ 146 bool empty () const 147 { 148 return m_vec.empty (); 149 } 150 151 /* Write the buffer to FILE. */ 152 void file_write (FILE *file) const 153 { 154 ::file_write (file, m_vec); 155 } 156 157 private: 158 /* Grow SIZE bytes at the end of the buffer. Returns a pointer to 159 the start of the new block. */ 160 gdb_byte *grow (size_t size) 161 { 162 m_vec.resize (m_vec.size () + size); 163 return &*(m_vec.end () - size); 164 } 165 166 gdb::byte_vector m_vec; 167 }; 168 169 /* An entry in the symbol table. */ 170 struct symtab_index_entry 171 { 172 /* The name of the symbol. */ 173 const char *name; 174 /* The offset of the name in the constant pool. */ 175 offset_type index_offset; 176 /* A sorted vector of the indices of all the CUs that hold an object 177 of this name. */ 178 std::vector<offset_type> cu_indices; 179 180 /* Minimize CU_INDICES, sorting them and removing duplicates as 181 appropriate. */ 182 void minimize (); 183 }; 184 185 /* The symbol table. This is a power-of-2-sized hash table. */ 186 struct mapped_symtab 187 { 188 mapped_symtab () 189 { 190 data.resize (1024); 191 } 192 193 /* Minimize each entry in the symbol table, removing duplicates. */ 194 void minimize () 195 { 196 for (symtab_index_entry &item : data) 197 item.minimize (); 198 } 199 200 offset_type n_elements = 0; 201 std::vector<symtab_index_entry> data; 202 203 /* Temporary storage for names. */ 204 auto_obstack m_string_obstack; 205 }; 206 207 /* Find a slot in SYMTAB for the symbol NAME. Returns a reference to 208 the slot. 209 210 Function is used only during write_hash_table so no index format backward 211 compatibility is needed. */ 212 213 static symtab_index_entry & 214 find_slot (struct mapped_symtab *symtab, const char *name) 215 { 216 offset_type index, step, hash = mapped_index_string_hash (INT_MAX, name); 217 218 index = hash & (symtab->data.size () - 1); 219 step = ((hash * 17) & (symtab->data.size () - 1)) | 1; 220 221 for (;;) 222 { 223 if (symtab->data[index].name == NULL 224 || strcmp (name, symtab->data[index].name) == 0) 225 return symtab->data[index]; 226 index = (index + step) & (symtab->data.size () - 1); 227 } 228 } 229 230 /* Expand SYMTAB's hash table. */ 231 232 static void 233 hash_expand (struct mapped_symtab *symtab) 234 { 235 auto old_entries = std::move (symtab->data); 236 237 symtab->data.clear (); 238 symtab->data.resize (old_entries.size () * 2); 239 240 for (auto &it : old_entries) 241 if (it.name != NULL) 242 { 243 auto &ref = find_slot (symtab, it.name); 244 ref = std::move (it); 245 } 246 } 247 248 /* Add an entry to SYMTAB. NAME is the name of the symbol. 249 CU_INDEX is the index of the CU in which the symbol appears. 250 IS_STATIC is one if the symbol is static, otherwise zero (global). */ 251 252 static void 253 add_index_entry (struct mapped_symtab *symtab, const char *name, 254 int is_static, gdb_index_symbol_kind kind, 255 offset_type cu_index) 256 { 257 offset_type cu_index_and_attrs; 258 259 ++symtab->n_elements; 260 if (4 * symtab->n_elements / 3 >= symtab->data.size ()) 261 hash_expand (symtab); 262 263 symtab_index_entry &slot = find_slot (symtab, name); 264 if (slot.name == NULL) 265 { 266 slot.name = name; 267 /* index_offset is set later. */ 268 } 269 270 cu_index_and_attrs = 0; 271 DW2_GDB_INDEX_CU_SET_VALUE (cu_index_and_attrs, cu_index); 272 DW2_GDB_INDEX_SYMBOL_STATIC_SET_VALUE (cu_index_and_attrs, is_static); 273 DW2_GDB_INDEX_SYMBOL_KIND_SET_VALUE (cu_index_and_attrs, kind); 274 275 /* We don't want to record an index value twice as we want to avoid the 276 duplication. 277 We process all global symbols and then all static symbols 278 (which would allow us to avoid the duplication by only having to check 279 the last entry pushed), but a symbol could have multiple kinds in one CU. 280 To keep things simple we don't worry about the duplication here and 281 sort and uniquify the list after we've processed all symbols. */ 282 slot.cu_indices.push_back (cu_index_and_attrs); 283 } 284 285 /* See symtab_index_entry. */ 286 287 void 288 symtab_index_entry::minimize () 289 { 290 if (name == nullptr || cu_indices.empty ()) 291 return; 292 293 std::sort (cu_indices.begin (), cu_indices.end ()); 294 auto from = std::unique (cu_indices.begin (), cu_indices.end ()); 295 cu_indices.erase (from, cu_indices.end ()); 296 297 /* We don't want to enter a variable or type more than once, so 298 remove any such duplicates from the list as well. When doing 299 this, we want to keep the entry from the first CU -- but this is 300 implicit due to the sort. This choice is done because it's 301 similar to what gdb historically did for partial symbols. */ 302 std::unordered_set<offset_type> seen; 303 from = std::remove_if (cu_indices.begin (), cu_indices.end (), 304 [&] (offset_type val) 305 { 306 gdb_index_symbol_kind kind = GDB_INDEX_SYMBOL_KIND_VALUE (val); 307 if (kind != GDB_INDEX_SYMBOL_KIND_TYPE 308 && kind != GDB_INDEX_SYMBOL_KIND_VARIABLE) 309 return false; 310 311 val &= ~GDB_INDEX_CU_MASK; 312 return !seen.insert (val).second; 313 }); 314 cu_indices.erase (from, cu_indices.end ()); 315 } 316 317 /* A form of 'const char *' suitable for container keys. Only the 318 pointer is stored. The strings themselves are compared, not the 319 pointers. */ 320 class c_str_view 321 { 322 public: 323 c_str_view (const char *cstr) 324 : m_cstr (cstr) 325 {} 326 327 bool operator== (const c_str_view &other) const 328 { 329 return strcmp (m_cstr, other.m_cstr) == 0; 330 } 331 332 /* Return the underlying C string. Note, the returned string is 333 only a reference with lifetime of this object. */ 334 const char *c_str () const 335 { 336 return m_cstr; 337 } 338 339 private: 340 friend class c_str_view_hasher; 341 const char *const m_cstr; 342 }; 343 344 /* A std::unordered_map::hasher for c_str_view that uses the right 345 hash function for strings in a mapped index. */ 346 class c_str_view_hasher 347 { 348 public: 349 size_t operator () (const c_str_view &x) const 350 { 351 return mapped_index_string_hash (INT_MAX, x.m_cstr); 352 } 353 }; 354 355 /* A std::unordered_map::hasher for std::vector<>. */ 356 template<typename T> 357 class vector_hasher 358 { 359 public: 360 size_t operator () (const std::vector<T> &key) const 361 { 362 return iterative_hash (key.data (), 363 sizeof (key.front ()) * key.size (), 0); 364 } 365 }; 366 367 /* Write the mapped hash table SYMTAB to the data buffer OUTPUT, with 368 constant pool entries going into the data buffer CPOOL. */ 369 370 static void 371 write_hash_table (mapped_symtab *symtab, data_buf &output, data_buf &cpool) 372 { 373 { 374 /* Elements are sorted vectors of the indices of all the CUs that 375 hold an object of this name. */ 376 std::unordered_map<std::vector<offset_type>, offset_type, 377 vector_hasher<offset_type>> 378 symbol_hash_table; 379 380 /* We add all the index vectors to the constant pool first, to 381 ensure alignment is ok. */ 382 for (symtab_index_entry &entry : symtab->data) 383 { 384 if (entry.name == NULL) 385 continue; 386 gdb_assert (entry.index_offset == 0); 387 388 /* Finding before inserting is faster than always trying to 389 insert, because inserting always allocates a node, does the 390 lookup, and then destroys the new node if another node 391 already had the same key. C++17 try_emplace will avoid 392 this. */ 393 const auto found 394 = symbol_hash_table.find (entry.cu_indices); 395 if (found != symbol_hash_table.end ()) 396 { 397 entry.index_offset = found->second; 398 continue; 399 } 400 401 symbol_hash_table.emplace (entry.cu_indices, cpool.size ()); 402 entry.index_offset = cpool.size (); 403 cpool.append_offset (entry.cu_indices.size ()); 404 for (const auto index : entry.cu_indices) 405 cpool.append_offset (index); 406 } 407 } 408 409 /* Now write out the hash table. */ 410 std::unordered_map<c_str_view, offset_type, c_str_view_hasher> str_table; 411 for (const auto &entry : symtab->data) 412 { 413 offset_type str_off, vec_off; 414 415 if (entry.name != NULL) 416 { 417 const auto insertpair = str_table.emplace (entry.name, cpool.size ()); 418 if (insertpair.second) 419 cpool.append_cstr0 (entry.name); 420 str_off = insertpair.first->second; 421 vec_off = entry.index_offset; 422 } 423 else 424 { 425 /* While 0 is a valid constant pool index, it is not valid 426 to have 0 for both offsets. */ 427 str_off = 0; 428 vec_off = 0; 429 } 430 431 output.append_offset (str_off); 432 output.append_offset (vec_off); 433 } 434 } 435 436 typedef std::unordered_map<dwarf2_per_cu_data *, unsigned int> cu_index_map; 437 438 /* Helper struct for building the address table. */ 439 struct addrmap_index_data 440 { 441 addrmap_index_data (data_buf &addr_vec_, cu_index_map &cu_index_htab_) 442 : addr_vec (addr_vec_), 443 cu_index_htab (cu_index_htab_) 444 {} 445 446 data_buf &addr_vec; 447 cu_index_map &cu_index_htab; 448 449 int operator() (CORE_ADDR start_addr, void *obj); 450 451 /* True if the previous_* fields are valid. 452 We can't write an entry until we see the next entry (since it is only then 453 that we know the end of the entry). */ 454 bool previous_valid = false; 455 /* Index of the CU in the table of all CUs in the index file. */ 456 unsigned int previous_cu_index = 0; 457 /* Start address of the CU. */ 458 CORE_ADDR previous_cu_start = 0; 459 }; 460 461 /* Write an address entry to ADDR_VEC. */ 462 463 static void 464 add_address_entry (data_buf &addr_vec, 465 CORE_ADDR start, CORE_ADDR end, unsigned int cu_index) 466 { 467 addr_vec.append_uint (8, BFD_ENDIAN_LITTLE, start); 468 addr_vec.append_uint (8, BFD_ENDIAN_LITTLE, end); 469 addr_vec.append_offset (cu_index); 470 } 471 472 /* Worker function for traversing an addrmap to build the address table. */ 473 474 int 475 addrmap_index_data::operator() (CORE_ADDR start_addr, void *obj) 476 { 477 dwarf2_per_cu_data *per_cu = (dwarf2_per_cu_data *) obj; 478 479 if (previous_valid) 480 add_address_entry (addr_vec, 481 previous_cu_start, start_addr, 482 previous_cu_index); 483 484 previous_cu_start = start_addr; 485 if (per_cu != NULL) 486 { 487 const auto it = cu_index_htab.find (per_cu); 488 gdb_assert (it != cu_index_htab.cend ()); 489 previous_cu_index = it->second; 490 previous_valid = true; 491 } 492 else 493 previous_valid = false; 494 495 return 0; 496 } 497 498 /* Write PER_BFD's address map to ADDR_VEC. 499 CU_INDEX_HTAB is used to map addrmap entries to their CU indices 500 in the index file. */ 501 502 static void 503 write_address_map (struct addrmap *addrmap, data_buf &addr_vec, 504 cu_index_map &cu_index_htab) 505 { 506 struct addrmap_index_data addrmap_index_data (addr_vec, cu_index_htab); 507 508 addrmap->foreach (addrmap_index_data); 509 510 /* It's highly unlikely the last entry (end address = 0xff...ff) 511 is valid, but we should still handle it. 512 The end address is recorded as the start of the next region, but that 513 doesn't work here. To cope we pass 0xff...ff, this is a rare situation 514 anyway. */ 515 if (addrmap_index_data.previous_valid) 516 add_address_entry (addr_vec, 517 addrmap_index_data.previous_cu_start, (CORE_ADDR) -1, 518 addrmap_index_data.previous_cu_index); 519 } 520 521 /* DWARF-5 .debug_names builder. */ 522 class debug_names 523 { 524 public: 525 debug_names (dwarf2_per_objfile *per_objfile, bool is_dwarf64, 526 bfd_endian dwarf5_byte_order) 527 : m_dwarf5_byte_order (dwarf5_byte_order), 528 m_dwarf32 (dwarf5_byte_order), 529 m_dwarf64 (dwarf5_byte_order), 530 m_dwarf (is_dwarf64 531 ? static_cast<dwarf &> (m_dwarf64) 532 : static_cast<dwarf &> (m_dwarf32)), 533 m_name_table_string_offs (m_dwarf.name_table_string_offs), 534 m_name_table_entry_offs (m_dwarf.name_table_entry_offs), 535 m_debugstrlookup (per_objfile) 536 {} 537 538 int dwarf5_offset_size () const 539 { 540 const bool dwarf5_is_dwarf64 = &m_dwarf == &m_dwarf64; 541 return dwarf5_is_dwarf64 ? 8 : 4; 542 } 543 544 /* Is this symbol from DW_TAG_compile_unit or DW_TAG_type_unit? */ 545 enum class unit_kind { cu, tu }; 546 547 /* Insert one symbol. */ 548 void insert (const cooked_index_entry *entry) 549 { 550 const auto it = m_cu_index_htab.find (entry->per_cu); 551 gdb_assert (it != m_cu_index_htab.cend ()); 552 const char *name = entry->full_name (&m_string_obstack); 553 554 /* This is incorrect but it mirrors gdb's historical behavior; and 555 because the current .debug_names generation is also incorrect, 556 it seems better to follow what was done before, rather than 557 introduce a mismatch between the newer and older gdb. */ 558 dwarf_tag tag = entry->tag; 559 if (tag != DW_TAG_typedef && tag_is_type (tag)) 560 tag = DW_TAG_structure_type; 561 else if (tag == DW_TAG_enumerator || tag == DW_TAG_constant) 562 tag = DW_TAG_variable; 563 564 int cu_index = it->second; 565 bool is_static = (entry->flags & IS_STATIC) != 0; 566 unit_kind kind = (entry->per_cu->is_debug_types 567 ? unit_kind::tu 568 : unit_kind::cu); 569 570 if (entry->per_cu->lang () == language_ada) 571 { 572 /* We want to ensure that the Ada main function's name appears 573 verbatim in the index. However, this name will be of the 574 form "_ada_mumble", and will be rewritten by ada_decode. 575 So, recognize it specially here and add it to the index by 576 hand. */ 577 if (strcmp (main_name (), name) == 0) 578 { 579 const auto insertpair 580 = m_name_to_value_set.emplace (c_str_view (name), 581 std::set<symbol_value> ()); 582 std::set<symbol_value> &value_set = insertpair.first->second; 583 value_set.emplace (symbol_value (tag, cu_index, is_static, kind)); 584 } 585 586 /* In order for the index to work when read back into gdb, it 587 has to supply a funny form of the name: it should be the 588 encoded name, with any suffixes stripped. Using the 589 ordinary encoded name will not work properly with the 590 searching logic in find_name_components_bounds; nor will 591 using the decoded name. Furthermore, an Ada "verbatim" 592 name (of the form "<MumBle>") must be entered without the 593 angle brackets. Note that the current index is unusual, 594 see PR symtab/24820 for details. */ 595 std::string decoded = ada_decode (name); 596 if (decoded[0] == '<') 597 name = (char *) obstack_copy0 (&m_string_obstack, 598 decoded.c_str () + 1, 599 decoded.length () - 2); 600 else 601 name = obstack_strdup (&m_string_obstack, 602 ada_encode (decoded.c_str ())); 603 } 604 605 const auto insertpair 606 = m_name_to_value_set.emplace (c_str_view (name), 607 std::set<symbol_value> ()); 608 std::set<symbol_value> &value_set = insertpair.first->second; 609 value_set.emplace (symbol_value (tag, cu_index, is_static, kind)); 610 } 611 612 /* Build all the tables. All symbols must be already inserted. 613 This function does not call file_write, caller has to do it 614 afterwards. */ 615 void build () 616 { 617 /* Verify the build method has not be called twice. */ 618 gdb_assert (m_abbrev_table.empty ()); 619 const size_t name_count = m_name_to_value_set.size (); 620 m_bucket_table.resize 621 (std::pow (2, std::ceil (std::log2 (name_count * 4 / 3)))); 622 m_hash_table.reserve (name_count); 623 m_name_table_string_offs.reserve (name_count); 624 m_name_table_entry_offs.reserve (name_count); 625 626 /* Map each hash of symbol to its name and value. */ 627 struct hash_it_pair 628 { 629 uint32_t hash; 630 decltype (m_name_to_value_set)::const_iterator it; 631 }; 632 std::vector<std::forward_list<hash_it_pair>> bucket_hash; 633 bucket_hash.resize (m_bucket_table.size ()); 634 for (decltype (m_name_to_value_set)::const_iterator it 635 = m_name_to_value_set.cbegin (); 636 it != m_name_to_value_set.cend (); 637 ++it) 638 { 639 const char *const name = it->first.c_str (); 640 const uint32_t hash = dwarf5_djb_hash (name); 641 hash_it_pair hashitpair; 642 hashitpair.hash = hash; 643 hashitpair.it = it; 644 auto &slot = bucket_hash[hash % bucket_hash.size()]; 645 slot.push_front (std::move (hashitpair)); 646 } 647 for (size_t bucket_ix = 0; bucket_ix < bucket_hash.size (); ++bucket_ix) 648 { 649 const std::forward_list<hash_it_pair> &hashitlist 650 = bucket_hash[bucket_ix]; 651 if (hashitlist.empty ()) 652 continue; 653 uint32_t &bucket_slot = m_bucket_table[bucket_ix]; 654 /* The hashes array is indexed starting at 1. */ 655 store_unsigned_integer (reinterpret_cast<gdb_byte *> (&bucket_slot), 656 sizeof (bucket_slot), m_dwarf5_byte_order, 657 m_hash_table.size () + 1); 658 for (const hash_it_pair &hashitpair : hashitlist) 659 { 660 m_hash_table.push_back (0); 661 store_unsigned_integer (reinterpret_cast<gdb_byte *> 662 (&m_hash_table.back ()), 663 sizeof (m_hash_table.back ()), 664 m_dwarf5_byte_order, hashitpair.hash); 665 const c_str_view &name = hashitpair.it->first; 666 const std::set<symbol_value> &value_set = hashitpair.it->second; 667 m_name_table_string_offs.push_back_reorder 668 (m_debugstrlookup.lookup (name.c_str ())); 669 m_name_table_entry_offs.push_back_reorder (m_entry_pool.size ()); 670 gdb_assert (!value_set.empty ()); 671 for (const symbol_value &value : value_set) 672 { 673 int &idx = m_indexkey_to_idx[index_key (value.dwarf_tag, 674 value.is_static, 675 value.kind)]; 676 if (idx == 0) 677 { 678 idx = m_idx_next++; 679 m_abbrev_table.append_unsigned_leb128 (idx); 680 m_abbrev_table.append_unsigned_leb128 (value.dwarf_tag); 681 m_abbrev_table.append_unsigned_leb128 682 (value.kind == unit_kind::cu ? DW_IDX_compile_unit 683 : DW_IDX_type_unit); 684 m_abbrev_table.append_unsigned_leb128 (DW_FORM_udata); 685 m_abbrev_table.append_unsigned_leb128 (value.is_static 686 ? DW_IDX_GNU_internal 687 : DW_IDX_GNU_external); 688 m_abbrev_table.append_unsigned_leb128 (DW_FORM_flag_present); 689 690 /* Terminate attributes list. */ 691 m_abbrev_table.append_unsigned_leb128 (0); 692 m_abbrev_table.append_unsigned_leb128 (0); 693 } 694 695 m_entry_pool.append_unsigned_leb128 (idx); 696 m_entry_pool.append_unsigned_leb128 (value.cu_index); 697 } 698 699 /* Terminate the list of CUs. */ 700 m_entry_pool.append_unsigned_leb128 (0); 701 } 702 } 703 gdb_assert (m_hash_table.size () == name_count); 704 705 /* Terminate tags list. */ 706 m_abbrev_table.append_unsigned_leb128 (0); 707 } 708 709 /* Return .debug_names bucket count. This must be called only after 710 calling the build method. */ 711 uint32_t bucket_count () const 712 { 713 /* Verify the build method has been already called. */ 714 gdb_assert (!m_abbrev_table.empty ()); 715 const uint32_t retval = m_bucket_table.size (); 716 717 /* Check for overflow. */ 718 gdb_assert (retval == m_bucket_table.size ()); 719 return retval; 720 } 721 722 /* Return .debug_names names count. This must be called only after 723 calling the build method. */ 724 uint32_t name_count () const 725 { 726 /* Verify the build method has been already called. */ 727 gdb_assert (!m_abbrev_table.empty ()); 728 const uint32_t retval = m_hash_table.size (); 729 730 /* Check for overflow. */ 731 gdb_assert (retval == m_hash_table.size ()); 732 return retval; 733 } 734 735 /* Return number of bytes of .debug_names abbreviation table. This 736 must be called only after calling the build method. */ 737 uint32_t abbrev_table_bytes () const 738 { 739 gdb_assert (!m_abbrev_table.empty ()); 740 return m_abbrev_table.size (); 741 } 742 743 /* Return number of bytes the .debug_names section will have. This 744 must be called only after calling the build method. */ 745 size_t bytes () const 746 { 747 /* Verify the build method has been already called. */ 748 gdb_assert (!m_abbrev_table.empty ()); 749 size_t expected_bytes = 0; 750 expected_bytes += m_bucket_table.size () * sizeof (m_bucket_table[0]); 751 expected_bytes += m_hash_table.size () * sizeof (m_hash_table[0]); 752 expected_bytes += m_name_table_string_offs.bytes (); 753 expected_bytes += m_name_table_entry_offs.bytes (); 754 expected_bytes += m_abbrev_table.size (); 755 expected_bytes += m_entry_pool.size (); 756 return expected_bytes; 757 } 758 759 /* Write .debug_names to FILE_NAMES and .debug_str addition to 760 FILE_STR. This must be called only after calling the build 761 method. */ 762 void file_write (FILE *file_names, FILE *file_str) const 763 { 764 /* Verify the build method has been already called. */ 765 gdb_assert (!m_abbrev_table.empty ()); 766 ::file_write (file_names, m_bucket_table); 767 ::file_write (file_names, m_hash_table); 768 m_name_table_string_offs.file_write (file_names); 769 m_name_table_entry_offs.file_write (file_names); 770 m_abbrev_table.file_write (file_names); 771 m_entry_pool.file_write (file_names); 772 m_debugstrlookup.file_write (file_str); 773 } 774 775 void add_cu (dwarf2_per_cu_data *per_cu, offset_type index) 776 { 777 m_cu_index_htab.emplace (per_cu, index); 778 } 779 780 private: 781 782 /* Storage for symbol names mapping them to their .debug_str section 783 offsets. */ 784 class debug_str_lookup 785 { 786 public: 787 788 /* Object constructor to be called for current DWARF2_PER_OBJFILE. 789 All .debug_str section strings are automatically stored. */ 790 debug_str_lookup (dwarf2_per_objfile *per_objfile) 791 : m_abfd (per_objfile->objfile->obfd.get ()), 792 m_per_objfile (per_objfile) 793 { 794 per_objfile->per_bfd->str.read (per_objfile->objfile); 795 if (per_objfile->per_bfd->str.buffer == NULL) 796 return; 797 for (const gdb_byte *data = per_objfile->per_bfd->str.buffer; 798 data < (per_objfile->per_bfd->str.buffer 799 + per_objfile->per_bfd->str.size);) 800 { 801 const char *const s = reinterpret_cast<const char *> (data); 802 const auto insertpair 803 = m_str_table.emplace (c_str_view (s), 804 data - per_objfile->per_bfd->str.buffer); 805 if (!insertpair.second) 806 complaint (_("Duplicate string \"%s\" in " 807 ".debug_str section [in module %s]"), 808 s, bfd_get_filename (m_abfd)); 809 data += strlen (s) + 1; 810 } 811 } 812 813 /* Return offset of symbol name S in the .debug_str section. Add 814 such symbol to the section's end if it does not exist there 815 yet. */ 816 size_t lookup (const char *s) 817 { 818 const auto it = m_str_table.find (c_str_view (s)); 819 if (it != m_str_table.end ()) 820 return it->second; 821 const size_t offset = (m_per_objfile->per_bfd->str.size 822 + m_str_add_buf.size ()); 823 m_str_table.emplace (c_str_view (s), offset); 824 m_str_add_buf.append_cstr0 (s); 825 return offset; 826 } 827 828 /* Append the end of the .debug_str section to FILE. */ 829 void file_write (FILE *file) const 830 { 831 m_str_add_buf.file_write (file); 832 } 833 834 private: 835 std::unordered_map<c_str_view, size_t, c_str_view_hasher> m_str_table; 836 bfd *const m_abfd; 837 dwarf2_per_objfile *m_per_objfile; 838 839 /* Data to add at the end of .debug_str for new needed symbol names. */ 840 data_buf m_str_add_buf; 841 }; 842 843 /* Container to map used DWARF tags to their .debug_names abbreviation 844 tags. */ 845 class index_key 846 { 847 public: 848 index_key (int dwarf_tag_, bool is_static_, unit_kind kind_) 849 : dwarf_tag (dwarf_tag_), is_static (is_static_), kind (kind_) 850 { 851 } 852 853 bool 854 operator== (const index_key &other) const 855 { 856 return (dwarf_tag == other.dwarf_tag && is_static == other.is_static 857 && kind == other.kind); 858 } 859 860 const int dwarf_tag; 861 const bool is_static; 862 const unit_kind kind; 863 }; 864 865 /* Provide std::unordered_map::hasher for index_key. */ 866 class index_key_hasher 867 { 868 public: 869 size_t 870 operator () (const index_key &key) const 871 { 872 return (std::hash<int>() (key.dwarf_tag) << 1) | key.is_static; 873 } 874 }; 875 876 /* Parameters of one symbol entry. */ 877 class symbol_value 878 { 879 public: 880 const int dwarf_tag, cu_index; 881 const bool is_static; 882 const unit_kind kind; 883 884 symbol_value (int dwarf_tag_, int cu_index_, bool is_static_, 885 unit_kind kind_) 886 : dwarf_tag (dwarf_tag_), cu_index (cu_index_), is_static (is_static_), 887 kind (kind_) 888 {} 889 890 bool 891 operator< (const symbol_value &other) const 892 { 893 #define X(n) \ 894 do \ 895 { \ 896 if (n < other.n) \ 897 return true; \ 898 if (n > other.n) \ 899 return false; \ 900 } \ 901 while (0) 902 X (dwarf_tag); 903 X (is_static); 904 X (kind); 905 X (cu_index); 906 #undef X 907 return false; 908 } 909 }; 910 911 /* Abstract base class to unify DWARF-32 and DWARF-64 name table 912 output. */ 913 class offset_vec 914 { 915 protected: 916 const bfd_endian dwarf5_byte_order; 917 public: 918 explicit offset_vec (bfd_endian dwarf5_byte_order_) 919 : dwarf5_byte_order (dwarf5_byte_order_) 920 {} 921 922 /* Call std::vector::reserve for NELEM elements. */ 923 virtual void reserve (size_t nelem) = 0; 924 925 /* Call std::vector::push_back with store_unsigned_integer byte 926 reordering for ELEM. */ 927 virtual void push_back_reorder (size_t elem) = 0; 928 929 /* Return expected output size in bytes. */ 930 virtual size_t bytes () const = 0; 931 932 /* Write name table to FILE. */ 933 virtual void file_write (FILE *file) const = 0; 934 }; 935 936 /* Template to unify DWARF-32 and DWARF-64 output. */ 937 template<typename OffsetSize> 938 class offset_vec_tmpl : public offset_vec 939 { 940 public: 941 explicit offset_vec_tmpl (bfd_endian dwarf5_byte_order_) 942 : offset_vec (dwarf5_byte_order_) 943 {} 944 945 /* Implement offset_vec::reserve. */ 946 void reserve (size_t nelem) override 947 { 948 m_vec.reserve (nelem); 949 } 950 951 /* Implement offset_vec::push_back_reorder. */ 952 void push_back_reorder (size_t elem) override 953 { 954 m_vec.push_back (elem); 955 /* Check for overflow. */ 956 gdb_assert (m_vec.back () == elem); 957 store_unsigned_integer (reinterpret_cast<gdb_byte *> (&m_vec.back ()), 958 sizeof (m_vec.back ()), dwarf5_byte_order, elem); 959 } 960 961 /* Implement offset_vec::bytes. */ 962 size_t bytes () const override 963 { 964 return m_vec.size () * sizeof (m_vec[0]); 965 } 966 967 /* Implement offset_vec::file_write. */ 968 void file_write (FILE *file) const override 969 { 970 ::file_write (file, m_vec); 971 } 972 973 private: 974 std::vector<OffsetSize> m_vec; 975 }; 976 977 /* Base class to unify DWARF-32 and DWARF-64 .debug_names output 978 respecting name table width. */ 979 class dwarf 980 { 981 public: 982 offset_vec &name_table_string_offs, &name_table_entry_offs; 983 984 dwarf (offset_vec &name_table_string_offs_, 985 offset_vec &name_table_entry_offs_) 986 : name_table_string_offs (name_table_string_offs_), 987 name_table_entry_offs (name_table_entry_offs_) 988 { 989 } 990 }; 991 992 /* Template to unify DWARF-32 and DWARF-64 .debug_names output 993 respecting name table width. */ 994 template<typename OffsetSize> 995 class dwarf_tmpl : public dwarf 996 { 997 public: 998 explicit dwarf_tmpl (bfd_endian dwarf5_byte_order_) 999 : dwarf (m_name_table_string_offs, m_name_table_entry_offs), 1000 m_name_table_string_offs (dwarf5_byte_order_), 1001 m_name_table_entry_offs (dwarf5_byte_order_) 1002 {} 1003 1004 private: 1005 offset_vec_tmpl<OffsetSize> m_name_table_string_offs; 1006 offset_vec_tmpl<OffsetSize> m_name_table_entry_offs; 1007 }; 1008 1009 /* Store value of each symbol. */ 1010 std::unordered_map<c_str_view, std::set<symbol_value>, c_str_view_hasher> 1011 m_name_to_value_set; 1012 1013 /* Tables of DWARF-5 .debug_names. They are in object file byte 1014 order. */ 1015 std::vector<uint32_t> m_bucket_table; 1016 std::vector<uint32_t> m_hash_table; 1017 1018 const bfd_endian m_dwarf5_byte_order; 1019 dwarf_tmpl<uint32_t> m_dwarf32; 1020 dwarf_tmpl<uint64_t> m_dwarf64; 1021 dwarf &m_dwarf; 1022 offset_vec &m_name_table_string_offs, &m_name_table_entry_offs; 1023 debug_str_lookup m_debugstrlookup; 1024 1025 /* Map each used .debug_names abbreviation tag parameter to its 1026 index value. */ 1027 std::unordered_map<index_key, int, index_key_hasher> m_indexkey_to_idx; 1028 1029 /* Next unused .debug_names abbreviation tag for 1030 m_indexkey_to_idx. */ 1031 int m_idx_next = 1; 1032 1033 /* .debug_names abbreviation table. */ 1034 data_buf m_abbrev_table; 1035 1036 /* .debug_names entry pool. */ 1037 data_buf m_entry_pool; 1038 1039 /* Temporary storage for Ada names. */ 1040 auto_obstack m_string_obstack; 1041 1042 cu_index_map m_cu_index_htab; 1043 }; 1044 1045 /* Return iff any of the needed offsets does not fit into 32-bit 1046 .debug_names section. */ 1047 1048 static bool 1049 check_dwarf64_offsets (dwarf2_per_objfile *per_objfile) 1050 { 1051 for (const auto &per_cu : per_objfile->per_bfd->all_units) 1052 { 1053 if (to_underlying (per_cu->sect_off) 1054 >= (static_cast<uint64_t> (1) << 32)) 1055 return true; 1056 } 1057 return false; 1058 } 1059 1060 /* Assert that FILE's size is EXPECTED_SIZE. Assumes file's seek 1061 position is at the end of the file. */ 1062 1063 static void 1064 assert_file_size (FILE *file, size_t expected_size) 1065 { 1066 const auto file_size = ftell (file); 1067 if (file_size == -1) 1068 perror_with_name (("ftell")); 1069 gdb_assert (file_size == expected_size); 1070 } 1071 1072 /* Write a gdb index file to OUT_FILE from all the sections passed as 1073 arguments. */ 1074 1075 static void 1076 write_gdbindex_1 (FILE *out_file, 1077 const data_buf &cu_list, 1078 const data_buf &types_cu_list, 1079 const data_buf &addr_vec, 1080 const data_buf &symtab_vec, 1081 const data_buf &constant_pool) 1082 { 1083 data_buf contents; 1084 const offset_type size_of_header = 6 * sizeof (offset_type); 1085 offset_type total_len = size_of_header; 1086 1087 /* The version number. */ 1088 contents.append_offset (8); 1089 1090 /* The offset of the CU list from the start of the file. */ 1091 contents.append_offset (total_len); 1092 total_len += cu_list.size (); 1093 1094 /* The offset of the types CU list from the start of the file. */ 1095 contents.append_offset (total_len); 1096 total_len += types_cu_list.size (); 1097 1098 /* The offset of the address table from the start of the file. */ 1099 contents.append_offset (total_len); 1100 total_len += addr_vec.size (); 1101 1102 /* The offset of the symbol table from the start of the file. */ 1103 contents.append_offset (total_len); 1104 total_len += symtab_vec.size (); 1105 1106 /* The offset of the constant pool from the start of the file. */ 1107 contents.append_offset (total_len); 1108 total_len += constant_pool.size (); 1109 1110 gdb_assert (contents.size () == size_of_header); 1111 1112 contents.file_write (out_file); 1113 cu_list.file_write (out_file); 1114 types_cu_list.file_write (out_file); 1115 addr_vec.file_write (out_file); 1116 symtab_vec.file_write (out_file); 1117 constant_pool.file_write (out_file); 1118 1119 assert_file_size (out_file, total_len); 1120 } 1121 1122 /* Write the contents of the internal "cooked" index. */ 1123 1124 static void 1125 write_cooked_index (cooked_index_vector *table, 1126 const cu_index_map &cu_index_htab, 1127 struct mapped_symtab *symtab) 1128 { 1129 const char *main_for_ada = main_name (); 1130 1131 for (const cooked_index_entry *entry : table->all_entries ()) 1132 { 1133 const auto it = cu_index_htab.find (entry->per_cu); 1134 gdb_assert (it != cu_index_htab.cend ()); 1135 1136 const char *name = entry->full_name (&symtab->m_string_obstack); 1137 1138 if (entry->per_cu->lang () == language_ada) 1139 { 1140 /* We want to ensure that the Ada main function's name 1141 appears verbatim in the index. However, this name will 1142 be of the form "_ada_mumble", and will be rewritten by 1143 ada_decode. So, recognize it specially here and add it 1144 to the index by hand. */ 1145 if (entry->tag == DW_TAG_subprogram 1146 && strcmp (main_for_ada, name) == 0) 1147 { 1148 /* Leave it alone. */ 1149 } 1150 else 1151 { 1152 /* In order for the index to work when read back into 1153 gdb, it has to use the encoded name, with any 1154 suffixes stripped. */ 1155 std::string encoded = ada_encode (name, false); 1156 name = obstack_strdup (&symtab->m_string_obstack, 1157 encoded.c_str ()); 1158 } 1159 } 1160 else if (entry->per_cu->lang () == language_cplus 1161 && (entry->flags & IS_LINKAGE) != 0) 1162 { 1163 /* GDB never put C++ linkage names into .gdb_index. The 1164 theory here is that a linkage name will normally be in 1165 the minimal symbols anyway, so including it in the index 1166 is usually redundant -- and the cases where it would not 1167 be redundant are rare and not worth supporting. */ 1168 continue; 1169 } 1170 else if ((entry->flags & IS_TYPE_DECLARATION) != 0) 1171 { 1172 /* Don't add type declarations to the index. */ 1173 continue; 1174 } 1175 1176 gdb_index_symbol_kind kind; 1177 if (entry->tag == DW_TAG_subprogram) 1178 kind = GDB_INDEX_SYMBOL_KIND_FUNCTION; 1179 else if (entry->tag == DW_TAG_variable 1180 || entry->tag == DW_TAG_constant 1181 || entry->tag == DW_TAG_enumerator) 1182 kind = GDB_INDEX_SYMBOL_KIND_VARIABLE; 1183 else if (entry->tag == DW_TAG_module 1184 || entry->tag == DW_TAG_common_block) 1185 kind = GDB_INDEX_SYMBOL_KIND_OTHER; 1186 else 1187 kind = GDB_INDEX_SYMBOL_KIND_TYPE; 1188 1189 add_index_entry (symtab, name, (entry->flags & IS_STATIC) != 0, 1190 kind, it->second); 1191 } 1192 } 1193 1194 /* Write contents of a .gdb_index section for OBJFILE into OUT_FILE. 1195 If OBJFILE has an associated dwz file, write contents of a .gdb_index 1196 section for that dwz file into DWZ_OUT_FILE. If OBJFILE does not have an 1197 associated dwz file, DWZ_OUT_FILE must be NULL. */ 1198 1199 static void 1200 write_gdbindex (dwarf2_per_objfile *per_objfile, 1201 cooked_index_vector *table, 1202 FILE *out_file, FILE *dwz_out_file) 1203 { 1204 mapped_symtab symtab; 1205 data_buf objfile_cu_list; 1206 data_buf dwz_cu_list; 1207 1208 /* While we're scanning CU's create a table that maps a dwarf2_per_cu_data 1209 (which is what addrmap records) to its index (which is what is recorded 1210 in the index file). This will later be needed to write the address 1211 table. */ 1212 cu_index_map cu_index_htab; 1213 cu_index_htab.reserve (per_objfile->per_bfd->all_units.size ()); 1214 1215 /* Store out the .debug_type CUs, if any. */ 1216 data_buf types_cu_list; 1217 1218 /* The CU list is already sorted, so we don't need to do additional 1219 work here. Also, the debug_types entries do not appear in 1220 all_units, but only in their own hash table. */ 1221 1222 int counter = 0; 1223 int types_counter = 0; 1224 for (int i = 0; i < per_objfile->per_bfd->all_units.size (); ++i) 1225 { 1226 dwarf2_per_cu_data *per_cu 1227 = per_objfile->per_bfd->all_units[i].get (); 1228 1229 int &this_counter = per_cu->is_debug_types ? types_counter : counter; 1230 1231 const auto insertpair = cu_index_htab.emplace (per_cu, this_counter); 1232 gdb_assert (insertpair.second); 1233 1234 /* The all_units list contains CUs read from the objfile as well as 1235 from the eventual dwz file. We need to place the entry in the 1236 corresponding index. */ 1237 data_buf &cu_list = (per_cu->is_debug_types 1238 ? types_cu_list 1239 : per_cu->is_dwz ? dwz_cu_list : objfile_cu_list); 1240 cu_list.append_uint (8, BFD_ENDIAN_LITTLE, 1241 to_underlying (per_cu->sect_off)); 1242 if (per_cu->is_debug_types) 1243 { 1244 signatured_type *sig_type = (signatured_type *) per_cu; 1245 cu_list.append_uint (8, BFD_ENDIAN_LITTLE, 1246 to_underlying (sig_type->type_offset_in_tu)); 1247 cu_list.append_uint (8, BFD_ENDIAN_LITTLE, 1248 sig_type->signature); 1249 } 1250 else 1251 cu_list.append_uint (8, BFD_ENDIAN_LITTLE, per_cu->length ()); 1252 1253 ++this_counter; 1254 } 1255 1256 write_cooked_index (table, cu_index_htab, &symtab); 1257 1258 /* Dump the address map. */ 1259 data_buf addr_vec; 1260 for (auto map : table->get_addrmaps ()) 1261 write_address_map (map, addr_vec, cu_index_htab); 1262 1263 /* Now that we've processed all symbols we can shrink their cu_indices 1264 lists. */ 1265 symtab.minimize (); 1266 1267 data_buf symtab_vec, constant_pool; 1268 if (symtab.n_elements == 0) 1269 symtab.data.resize (0); 1270 1271 write_hash_table (&symtab, symtab_vec, constant_pool); 1272 1273 write_gdbindex_1(out_file, objfile_cu_list, types_cu_list, addr_vec, 1274 symtab_vec, constant_pool); 1275 1276 if (dwz_out_file != NULL) 1277 write_gdbindex_1 (dwz_out_file, dwz_cu_list, {}, {}, {}, {}); 1278 else 1279 gdb_assert (dwz_cu_list.empty ()); 1280 } 1281 1282 /* DWARF-5 augmentation string for GDB's DW_IDX_GNU_* extension. */ 1283 static const gdb_byte dwarf5_gdb_augmentation[] = { 'G', 'D', 'B', 0 }; 1284 1285 /* Write a new .debug_names section for OBJFILE into OUT_FILE, write 1286 needed addition to .debug_str section to OUT_FILE_STR. Return how 1287 many bytes were expected to be written into OUT_FILE. */ 1288 1289 static void 1290 write_debug_names (dwarf2_per_objfile *per_objfile, 1291 cooked_index_vector *table, 1292 FILE *out_file, FILE *out_file_str) 1293 { 1294 const bool dwarf5_is_dwarf64 = check_dwarf64_offsets (per_objfile); 1295 struct objfile *objfile = per_objfile->objfile; 1296 const enum bfd_endian dwarf5_byte_order 1297 = gdbarch_byte_order (objfile->arch ()); 1298 1299 /* The CU list is already sorted, so we don't need to do additional 1300 work here. Also, the debug_types entries do not appear in 1301 all_units, but only in their own hash table. */ 1302 data_buf cu_list; 1303 data_buf types_cu_list; 1304 debug_names nametable (per_objfile, dwarf5_is_dwarf64, dwarf5_byte_order); 1305 int counter = 0; 1306 int types_counter = 0; 1307 for (int i = 0; i < per_objfile->per_bfd->all_units.size (); ++i) 1308 { 1309 dwarf2_per_cu_data *per_cu 1310 = per_objfile->per_bfd->all_units[i].get (); 1311 1312 int &this_counter = per_cu->is_debug_types ? types_counter : counter; 1313 data_buf &this_list = per_cu->is_debug_types ? types_cu_list : cu_list; 1314 1315 nametable.add_cu (per_cu, this_counter); 1316 this_list.append_uint (nametable.dwarf5_offset_size (), 1317 dwarf5_byte_order, 1318 to_underlying (per_cu->sect_off)); 1319 ++this_counter; 1320 } 1321 1322 /* Verify that all units are represented. */ 1323 gdb_assert (counter == per_objfile->per_bfd->all_comp_units.size ()); 1324 gdb_assert (types_counter == per_objfile->per_bfd->all_type_units.size ()); 1325 1326 for (const cooked_index_entry *entry : table->all_entries ()) 1327 nametable.insert (entry); 1328 1329 nametable.build (); 1330 1331 /* No addr_vec - DWARF-5 uses .debug_aranges generated by GCC. */ 1332 1333 const offset_type bytes_of_header 1334 = ((dwarf5_is_dwarf64 ? 12 : 4) 1335 + 2 + 2 + 7 * 4 1336 + sizeof (dwarf5_gdb_augmentation)); 1337 size_t expected_bytes = 0; 1338 expected_bytes += bytes_of_header; 1339 expected_bytes += cu_list.size (); 1340 expected_bytes += types_cu_list.size (); 1341 expected_bytes += nametable.bytes (); 1342 data_buf header; 1343 1344 if (!dwarf5_is_dwarf64) 1345 { 1346 const uint64_t size64 = expected_bytes - 4; 1347 gdb_assert (size64 < 0xfffffff0); 1348 header.append_uint (4, dwarf5_byte_order, size64); 1349 } 1350 else 1351 { 1352 header.append_uint (4, dwarf5_byte_order, 0xffffffff); 1353 header.append_uint (8, dwarf5_byte_order, expected_bytes - 12); 1354 } 1355 1356 /* The version number. */ 1357 header.append_uint (2, dwarf5_byte_order, 5); 1358 1359 /* Padding. */ 1360 header.append_uint (2, dwarf5_byte_order, 0); 1361 1362 /* comp_unit_count - The number of CUs in the CU list. */ 1363 header.append_uint (4, dwarf5_byte_order, counter); 1364 1365 /* local_type_unit_count - The number of TUs in the local TU 1366 list. */ 1367 header.append_uint (4, dwarf5_byte_order, types_counter); 1368 1369 /* foreign_type_unit_count - The number of TUs in the foreign TU 1370 list. */ 1371 header.append_uint (4, dwarf5_byte_order, 0); 1372 1373 /* bucket_count - The number of hash buckets in the hash lookup 1374 table. */ 1375 header.append_uint (4, dwarf5_byte_order, nametable.bucket_count ()); 1376 1377 /* name_count - The number of unique names in the index. */ 1378 header.append_uint (4, dwarf5_byte_order, nametable.name_count ()); 1379 1380 /* abbrev_table_size - The size in bytes of the abbreviations 1381 table. */ 1382 header.append_uint (4, dwarf5_byte_order, nametable.abbrev_table_bytes ()); 1383 1384 /* augmentation_string_size - The size in bytes of the augmentation 1385 string. This value is rounded up to a multiple of 4. */ 1386 static_assert (sizeof (dwarf5_gdb_augmentation) % 4 == 0, ""); 1387 header.append_uint (4, dwarf5_byte_order, sizeof (dwarf5_gdb_augmentation)); 1388 header.append_array (dwarf5_gdb_augmentation); 1389 1390 gdb_assert (header.size () == bytes_of_header); 1391 1392 header.file_write (out_file); 1393 cu_list.file_write (out_file); 1394 types_cu_list.file_write (out_file); 1395 nametable.file_write (out_file, out_file_str); 1396 1397 assert_file_size (out_file, expected_bytes); 1398 } 1399 1400 /* This represents an index file being written (work-in-progress). 1401 1402 The data is initially written to a temporary file. When the finalize method 1403 is called, the file is closed and moved to its final location. 1404 1405 On failure (if this object is being destroyed with having called finalize), 1406 the temporary file is closed and deleted. */ 1407 1408 struct index_wip_file 1409 { 1410 index_wip_file (const char *dir, const char *basename, 1411 const char *suffix) 1412 { 1413 filename = (std::string (dir) + SLASH_STRING + basename 1414 + suffix); 1415 1416 filename_temp = make_temp_filename (filename); 1417 1418 scoped_fd out_file_fd = gdb_mkostemp_cloexec (filename_temp.data (), 1419 O_BINARY); 1420 if (out_file_fd.get () == -1) 1421 perror_with_name (("mkstemp")); 1422 1423 out_file = out_file_fd.to_file ("wb"); 1424 1425 if (out_file == nullptr) 1426 error (_("Can't open `%s' for writing"), filename_temp.data ()); 1427 1428 unlink_file.emplace (filename_temp.data ()); 1429 } 1430 1431 void finalize () 1432 { 1433 /* We want to keep the file. */ 1434 unlink_file->keep (); 1435 1436 /* Close and move the str file in place. */ 1437 unlink_file.reset (); 1438 if (rename (filename_temp.data (), filename.c_str ()) != 0) 1439 perror_with_name (("rename")); 1440 } 1441 1442 std::string filename; 1443 gdb::char_vector filename_temp; 1444 1445 /* Order matters here; we want FILE to be closed before 1446 FILENAME_TEMP is unlinked, because on MS-Windows one cannot 1447 delete a file that is still open. So, we wrap the unlinker in an 1448 optional and emplace it once we know the file name. */ 1449 gdb::optional<gdb::unlinker> unlink_file; 1450 1451 gdb_file_up out_file; 1452 }; 1453 1454 /* See dwarf-index-write.h. */ 1455 1456 void 1457 write_dwarf_index (dwarf2_per_objfile *per_objfile, const char *dir, 1458 const char *basename, const char *dwz_basename, 1459 dw_index_kind index_kind) 1460 { 1461 struct objfile *objfile = per_objfile->objfile; 1462 1463 if (per_objfile->per_bfd->index_table == nullptr) 1464 error (_("No debugging symbols")); 1465 cooked_index_vector *table 1466 = per_objfile->per_bfd->index_table->index_for_writing (); 1467 1468 if (per_objfile->per_bfd->types.size () > 1) 1469 error (_("Cannot make an index when the file has multiple .debug_types sections")); 1470 1471 1472 gdb_assert ((objfile->flags & OBJF_NOT_FILENAME) == 0); 1473 1474 const char *index_suffix = (index_kind == dw_index_kind::DEBUG_NAMES 1475 ? INDEX5_SUFFIX : INDEX4_SUFFIX); 1476 1477 index_wip_file objfile_index_wip (dir, basename, index_suffix); 1478 gdb::optional<index_wip_file> dwz_index_wip; 1479 1480 if (dwz_basename != NULL) 1481 dwz_index_wip.emplace (dir, dwz_basename, index_suffix); 1482 1483 if (index_kind == dw_index_kind::DEBUG_NAMES) 1484 { 1485 index_wip_file str_wip_file (dir, basename, DEBUG_STR_SUFFIX); 1486 1487 write_debug_names (per_objfile, table, objfile_index_wip.out_file.get (), 1488 str_wip_file.out_file.get ()); 1489 1490 str_wip_file.finalize (); 1491 } 1492 else 1493 write_gdbindex (per_objfile, table, objfile_index_wip.out_file.get (), 1494 (dwz_index_wip.has_value () 1495 ? dwz_index_wip->out_file.get () : NULL)); 1496 1497 objfile_index_wip.finalize (); 1498 1499 if (dwz_index_wip.has_value ()) 1500 dwz_index_wip->finalize (); 1501 } 1502 1503 /* Implementation of the `save gdb-index' command. 1504 1505 Note that the .gdb_index file format used by this command is 1506 documented in the GDB manual. Any changes here must be documented 1507 there. */ 1508 1509 static void 1510 save_gdb_index_command (const char *arg, int from_tty) 1511 { 1512 const char dwarf5space[] = "-dwarf-5 "; 1513 dw_index_kind index_kind = dw_index_kind::GDB_INDEX; 1514 1515 if (!arg) 1516 arg = ""; 1517 1518 arg = skip_spaces (arg); 1519 if (strncmp (arg, dwarf5space, strlen (dwarf5space)) == 0) 1520 { 1521 index_kind = dw_index_kind::DEBUG_NAMES; 1522 arg += strlen (dwarf5space); 1523 arg = skip_spaces (arg); 1524 } 1525 1526 if (!*arg) 1527 error (_("usage: save gdb-index [-dwarf-5] DIRECTORY")); 1528 1529 for (objfile *objfile : current_program_space->objfiles ()) 1530 { 1531 /* If the objfile does not correspond to an actual file, skip it. */ 1532 if ((objfile->flags & OBJF_NOT_FILENAME) != 0) 1533 continue; 1534 1535 dwarf2_per_objfile *per_objfile = get_dwarf2_per_objfile (objfile); 1536 1537 if (per_objfile != NULL) 1538 { 1539 try 1540 { 1541 const char *basename = lbasename (objfile_name (objfile)); 1542 const dwz_file *dwz = dwarf2_get_dwz_file (per_objfile->per_bfd); 1543 const char *dwz_basename = NULL; 1544 1545 if (dwz != NULL) 1546 dwz_basename = lbasename (dwz->filename ()); 1547 1548 write_dwarf_index (per_objfile, arg, basename, dwz_basename, 1549 index_kind); 1550 } 1551 catch (const gdb_exception_error &except) 1552 { 1553 exception_fprintf (gdb_stderr, except, 1554 _("Error while writing index for `%s': "), 1555 objfile_name (objfile)); 1556 } 1557 } 1558 1559 } 1560 } 1561 1562 void _initialize_dwarf_index_write (); 1563 void 1564 _initialize_dwarf_index_write () 1565 { 1566 cmd_list_element *c = add_cmd ("gdb-index", class_files, 1567 save_gdb_index_command, _("\ 1568 Save a gdb-index file.\n\ 1569 Usage: save gdb-index [-dwarf-5] DIRECTORY\n\ 1570 \n\ 1571 No options create one file with .gdb-index extension for pre-DWARF-5\n\ 1572 compatible .gdb_index section. With -dwarf-5 creates two files with\n\ 1573 extension .debug_names and .debug_str for DWARF-5 .debug_names section."), 1574 &save_cmdlist); 1575 set_cmd_completer (c, filename_completer); 1576 } 1577