| 1 | /* DWARF index writing support for GDB. |
| 2 | |
| 3 | Copyright (C) 1994-2019 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 "dwarf-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 "dwarf-index-common.h" |
| 33 | #include "dwarf2.h" |
| 34 | #include "dwarf2read.h" |
| 35 | #include "gdb/gdb-index.h" |
| 36 | #include "gdbcmd.h" |
| 37 | #include "objfiles.h" |
| 38 | #include "psympriv.h" |
| 39 | #include "ada-lang.h" |
| 40 | |
| 41 | #include <algorithm> |
| 42 | #include <cmath> |
| 43 | #include <set> |
| 44 | #include <unordered_map> |
| 45 | #include <unordered_set> |
| 46 | |
| 47 | /* Ensure only legit values are used. */ |
| 48 | #define DW2_GDB_INDEX_SYMBOL_STATIC_SET_VALUE(cu_index, value) \ |
| 49 | do { \ |
| 50 | gdb_assert ((unsigned int) (value) <= 1); \ |
| 51 | GDB_INDEX_SYMBOL_STATIC_SET_VALUE((cu_index), (value)); \ |
| 52 | } while (0) |
| 53 | |
| 54 | /* Ensure only legit values are used. */ |
| 55 | #define DW2_GDB_INDEX_SYMBOL_KIND_SET_VALUE(cu_index, value) \ |
| 56 | do { \ |
| 57 | gdb_assert ((value) >= GDB_INDEX_SYMBOL_KIND_TYPE \ |
| 58 | && (value) <= GDB_INDEX_SYMBOL_KIND_OTHER); \ |
| 59 | GDB_INDEX_SYMBOL_KIND_SET_VALUE((cu_index), (value)); \ |
| 60 | } while (0) |
| 61 | |
| 62 | /* Ensure we don't use more than the allotted number of bits for the CU. */ |
| 63 | #define DW2_GDB_INDEX_CU_SET_VALUE(cu_index, value) \ |
| 64 | do { \ |
| 65 | gdb_assert (((value) & ~GDB_INDEX_CU_MASK) == 0); \ |
| 66 | GDB_INDEX_CU_SET_VALUE((cu_index), (value)); \ |
| 67 | } while (0) |
| 68 | |
| 69 | /* The "save gdb-index" command. */ |
| 70 | |
| 71 | /* Write SIZE bytes from the buffer pointed to by DATA to FILE, with |
| 72 | error checking. */ |
| 73 | |
| 74 | static void |
| 75 | file_write (FILE *file, const void *data, size_t size) |
| 76 | { |
| 77 | if (fwrite (data, 1, size, file) != size) |
| 78 | error (_("couldn't data write to file")); |
| 79 | } |
| 80 | |
| 81 | /* Write the contents of VEC to FILE, with error checking. */ |
| 82 | |
| 83 | template<typename Elem, typename Alloc> |
| 84 | static void |
| 85 | file_write (FILE *file, const std::vector<Elem, Alloc> &vec) |
| 86 | { |
| 87 | if (!vec.empty ()) |
| 88 | file_write (file, vec.data (), vec.size () * sizeof (vec[0])); |
| 89 | } |
| 90 | |
| 91 | /* In-memory buffer to prepare data to be written later to a file. */ |
| 92 | class data_buf |
| 93 | { |
| 94 | public: |
| 95 | /* Copy DATA to the end of the buffer. */ |
| 96 | template<typename T> |
| 97 | void append_data (const T &data) |
| 98 | { |
| 99 | std::copy (reinterpret_cast<const gdb_byte *> (&data), |
| 100 | reinterpret_cast<const gdb_byte *> (&data + 1), |
| 101 | grow (sizeof (data))); |
| 102 | } |
| 103 | |
| 104 | /* Copy CSTR (a zero-terminated string) to the end of buffer. The |
| 105 | terminating zero is appended too. */ |
| 106 | void append_cstr0 (const char *cstr) |
| 107 | { |
| 108 | const size_t size = strlen (cstr) + 1; |
| 109 | std::copy (cstr, cstr + size, grow (size)); |
| 110 | } |
| 111 | |
| 112 | /* Store INPUT as ULEB128 to the end of buffer. */ |
| 113 | void append_unsigned_leb128 (ULONGEST input) |
| 114 | { |
| 115 | for (;;) |
| 116 | { |
| 117 | gdb_byte output = input & 0x7f; |
| 118 | input >>= 7; |
| 119 | if (input) |
| 120 | output |= 0x80; |
| 121 | append_data (output); |
| 122 | if (input == 0) |
| 123 | break; |
| 124 | } |
| 125 | } |
| 126 | |
| 127 | /* Accept a host-format integer in VAL and append it to the buffer |
| 128 | as a target-format integer which is LEN bytes long. */ |
| 129 | void append_uint (size_t len, bfd_endian byte_order, ULONGEST val) |
| 130 | { |
| 131 | ::store_unsigned_integer (grow (len), len, byte_order, val); |
| 132 | } |
| 133 | |
| 134 | /* Return the size of the buffer. */ |
| 135 | size_t size () const |
| 136 | { |
| 137 | return m_vec.size (); |
| 138 | } |
| 139 | |
| 140 | /* Return true iff the buffer is empty. */ |
| 141 | bool empty () const |
| 142 | { |
| 143 | return m_vec.empty (); |
| 144 | } |
| 145 | |
| 146 | /* Write the buffer to FILE. */ |
| 147 | void file_write (FILE *file) const |
| 148 | { |
| 149 | ::file_write (file, m_vec); |
| 150 | } |
| 151 | |
| 152 | private: |
| 153 | /* Grow SIZE bytes at the end of the buffer. Returns a pointer to |
| 154 | the start of the new block. */ |
| 155 | gdb_byte *grow (size_t size) |
| 156 | { |
| 157 | m_vec.resize (m_vec.size () + size); |
| 158 | return &*(m_vec.end () - size); |
| 159 | } |
| 160 | |
| 161 | gdb::byte_vector m_vec; |
| 162 | }; |
| 163 | |
| 164 | /* An entry in the symbol table. */ |
| 165 | struct symtab_index_entry |
| 166 | { |
| 167 | /* The name of the symbol. */ |
| 168 | const char *name; |
| 169 | /* The offset of the name in the constant pool. */ |
| 170 | offset_type index_offset; |
| 171 | /* A sorted vector of the indices of all the CUs that hold an object |
| 172 | of this name. */ |
| 173 | std::vector<offset_type> cu_indices; |
| 174 | }; |
| 175 | |
| 176 | /* The symbol table. This is a power-of-2-sized hash table. */ |
| 177 | struct mapped_symtab |
| 178 | { |
| 179 | mapped_symtab () |
| 180 | { |
| 181 | data.resize (1024); |
| 182 | } |
| 183 | |
| 184 | offset_type n_elements = 0; |
| 185 | std::vector<symtab_index_entry> data; |
| 186 | }; |
| 187 | |
| 188 | /* Find a slot in SYMTAB for the symbol NAME. Returns a reference to |
| 189 | the slot. |
| 190 | |
| 191 | Function is used only during write_hash_table so no index format backward |
| 192 | compatibility is needed. */ |
| 193 | |
| 194 | static symtab_index_entry & |
| 195 | find_slot (struct mapped_symtab *symtab, const char *name) |
| 196 | { |
| 197 | offset_type index, step, hash = mapped_index_string_hash (INT_MAX, name); |
| 198 | |
| 199 | index = hash & (symtab->data.size () - 1); |
| 200 | step = ((hash * 17) & (symtab->data.size () - 1)) | 1; |
| 201 | |
| 202 | for (;;) |
| 203 | { |
| 204 | if (symtab->data[index].name == NULL |
| 205 | || strcmp (name, symtab->data[index].name) == 0) |
| 206 | return symtab->data[index]; |
| 207 | index = (index + step) & (symtab->data.size () - 1); |
| 208 | } |
| 209 | } |
| 210 | |
| 211 | /* Expand SYMTAB's hash table. */ |
| 212 | |
| 213 | static void |
| 214 | hash_expand (struct mapped_symtab *symtab) |
| 215 | { |
| 216 | auto old_entries = std::move (symtab->data); |
| 217 | |
| 218 | symtab->data.clear (); |
| 219 | symtab->data.resize (old_entries.size () * 2); |
| 220 | |
| 221 | for (auto &it : old_entries) |
| 222 | if (it.name != NULL) |
| 223 | { |
| 224 | auto &ref = find_slot (symtab, it.name); |
| 225 | ref = std::move (it); |
| 226 | } |
| 227 | } |
| 228 | |
| 229 | /* Add an entry to SYMTAB. NAME is the name of the symbol. |
| 230 | CU_INDEX is the index of the CU in which the symbol appears. |
| 231 | IS_STATIC is one if the symbol is static, otherwise zero (global). */ |
| 232 | |
| 233 | static void |
| 234 | add_index_entry (struct mapped_symtab *symtab, const char *name, |
| 235 | int is_static, gdb_index_symbol_kind kind, |
| 236 | offset_type cu_index) |
| 237 | { |
| 238 | offset_type cu_index_and_attrs; |
| 239 | |
| 240 | ++symtab->n_elements; |
| 241 | if (4 * symtab->n_elements / 3 >= symtab->data.size ()) |
| 242 | hash_expand (symtab); |
| 243 | |
| 244 | symtab_index_entry &slot = find_slot (symtab, name); |
| 245 | if (slot.name == NULL) |
| 246 | { |
| 247 | slot.name = name; |
| 248 | /* index_offset is set later. */ |
| 249 | } |
| 250 | |
| 251 | cu_index_and_attrs = 0; |
| 252 | DW2_GDB_INDEX_CU_SET_VALUE (cu_index_and_attrs, cu_index); |
| 253 | DW2_GDB_INDEX_SYMBOL_STATIC_SET_VALUE (cu_index_and_attrs, is_static); |
| 254 | DW2_GDB_INDEX_SYMBOL_KIND_SET_VALUE (cu_index_and_attrs, kind); |
| 255 | |
| 256 | /* We don't want to record an index value twice as we want to avoid the |
| 257 | duplication. |
| 258 | We process all global symbols and then all static symbols |
| 259 | (which would allow us to avoid the duplication by only having to check |
| 260 | the last entry pushed), but a symbol could have multiple kinds in one CU. |
| 261 | To keep things simple we don't worry about the duplication here and |
| 262 | sort and uniquify the list after we've processed all symbols. */ |
| 263 | slot.cu_indices.push_back (cu_index_and_attrs); |
| 264 | } |
| 265 | |
| 266 | /* Sort and remove duplicates of all symbols' cu_indices lists. */ |
| 267 | |
| 268 | static void |
| 269 | uniquify_cu_indices (struct mapped_symtab *symtab) |
| 270 | { |
| 271 | for (auto &entry : symtab->data) |
| 272 | { |
| 273 | if (entry.name != NULL && !entry.cu_indices.empty ()) |
| 274 | { |
| 275 | auto &cu_indices = entry.cu_indices; |
| 276 | std::sort (cu_indices.begin (), cu_indices.end ()); |
| 277 | auto from = std::unique (cu_indices.begin (), cu_indices.end ()); |
| 278 | cu_indices.erase (from, cu_indices.end ()); |
| 279 | } |
| 280 | } |
| 281 | } |
| 282 | |
| 283 | /* A form of 'const char *' suitable for container keys. Only the |
| 284 | pointer is stored. The strings themselves are compared, not the |
| 285 | pointers. */ |
| 286 | class c_str_view |
| 287 | { |
| 288 | public: |
| 289 | c_str_view (const char *cstr) |
| 290 | : m_cstr (cstr) |
| 291 | {} |
| 292 | |
| 293 | bool operator== (const c_str_view &other) const |
| 294 | { |
| 295 | return strcmp (m_cstr, other.m_cstr) == 0; |
| 296 | } |
| 297 | |
| 298 | /* Return the underlying C string. Note, the returned string is |
| 299 | only a reference with lifetime of this object. */ |
| 300 | const char *c_str () const |
| 301 | { |
| 302 | return m_cstr; |
| 303 | } |
| 304 | |
| 305 | private: |
| 306 | friend class c_str_view_hasher; |
| 307 | const char *const m_cstr; |
| 308 | }; |
| 309 | |
| 310 | /* A std::unordered_map::hasher for c_str_view that uses the right |
| 311 | hash function for strings in a mapped index. */ |
| 312 | class c_str_view_hasher |
| 313 | { |
| 314 | public: |
| 315 | size_t operator () (const c_str_view &x) const |
| 316 | { |
| 317 | return mapped_index_string_hash (INT_MAX, x.m_cstr); |
| 318 | } |
| 319 | }; |
| 320 | |
| 321 | /* A std::unordered_map::hasher for std::vector<>. */ |
| 322 | template<typename T> |
| 323 | class vector_hasher |
| 324 | { |
| 325 | public: |
| 326 | size_t operator () (const std::vector<T> &key) const |
| 327 | { |
| 328 | return iterative_hash (key.data (), |
| 329 | sizeof (key.front ()) * key.size (), 0); |
| 330 | } |
| 331 | }; |
| 332 | |
| 333 | /* Write the mapped hash table SYMTAB to the data buffer OUTPUT, with |
| 334 | constant pool entries going into the data buffer CPOOL. */ |
| 335 | |
| 336 | static void |
| 337 | write_hash_table (mapped_symtab *symtab, data_buf &output, data_buf &cpool) |
| 338 | { |
| 339 | { |
| 340 | /* Elements are sorted vectors of the indices of all the CUs that |
| 341 | hold an object of this name. */ |
| 342 | std::unordered_map<std::vector<offset_type>, offset_type, |
| 343 | vector_hasher<offset_type>> |
| 344 | symbol_hash_table; |
| 345 | |
| 346 | /* We add all the index vectors to the constant pool first, to |
| 347 | ensure alignment is ok. */ |
| 348 | for (symtab_index_entry &entry : symtab->data) |
| 349 | { |
| 350 | if (entry.name == NULL) |
| 351 | continue; |
| 352 | gdb_assert (entry.index_offset == 0); |
| 353 | |
| 354 | /* Finding before inserting is faster than always trying to |
| 355 | insert, because inserting always allocates a node, does the |
| 356 | lookup, and then destroys the new node if another node |
| 357 | already had the same key. C++17 try_emplace will avoid |
| 358 | this. */ |
| 359 | const auto found |
| 360 | = symbol_hash_table.find (entry.cu_indices); |
| 361 | if (found != symbol_hash_table.end ()) |
| 362 | { |
| 363 | entry.index_offset = found->second; |
| 364 | continue; |
| 365 | } |
| 366 | |
| 367 | symbol_hash_table.emplace (entry.cu_indices, cpool.size ()); |
| 368 | entry.index_offset = cpool.size (); |
| 369 | cpool.append_data (MAYBE_SWAP (entry.cu_indices.size ())); |
| 370 | for (const auto index : entry.cu_indices) |
| 371 | cpool.append_data (MAYBE_SWAP (index)); |
| 372 | } |
| 373 | } |
| 374 | |
| 375 | /* Now write out the hash table. */ |
| 376 | std::unordered_map<c_str_view, offset_type, c_str_view_hasher> str_table; |
| 377 | for (const auto &entry : symtab->data) |
| 378 | { |
| 379 | offset_type str_off, vec_off; |
| 380 | |
| 381 | if (entry.name != NULL) |
| 382 | { |
| 383 | const auto insertpair = str_table.emplace (entry.name, cpool.size ()); |
| 384 | if (insertpair.second) |
| 385 | cpool.append_cstr0 (entry.name); |
| 386 | str_off = insertpair.first->second; |
| 387 | vec_off = entry.index_offset; |
| 388 | } |
| 389 | else |
| 390 | { |
| 391 | /* While 0 is a valid constant pool index, it is not valid |
| 392 | to have 0 for both offsets. */ |
| 393 | str_off = 0; |
| 394 | vec_off = 0; |
| 395 | } |
| 396 | |
| 397 | output.append_data (MAYBE_SWAP (str_off)); |
| 398 | output.append_data (MAYBE_SWAP (vec_off)); |
| 399 | } |
| 400 | } |
| 401 | |
| 402 | typedef std::unordered_map<partial_symtab *, unsigned int> psym_index_map; |
| 403 | |
| 404 | /* Helper struct for building the address table. */ |
| 405 | struct addrmap_index_data |
| 406 | { |
| 407 | addrmap_index_data (data_buf &addr_vec_, psym_index_map &cu_index_htab_) |
| 408 | : addr_vec (addr_vec_), cu_index_htab (cu_index_htab_) |
| 409 | {} |
| 410 | |
| 411 | struct objfile *objfile; |
| 412 | data_buf &addr_vec; |
| 413 | psym_index_map &cu_index_htab; |
| 414 | |
| 415 | /* Non-zero if the previous_* fields are valid. |
| 416 | We can't write an entry until we see the next entry (since it is only then |
| 417 | that we know the end of the entry). */ |
| 418 | int previous_valid; |
| 419 | /* Index of the CU in the table of all CUs in the index file. */ |
| 420 | unsigned int previous_cu_index; |
| 421 | /* Start address of the CU. */ |
| 422 | CORE_ADDR previous_cu_start; |
| 423 | }; |
| 424 | |
| 425 | /* Write an address entry to ADDR_VEC. */ |
| 426 | |
| 427 | static void |
| 428 | add_address_entry (struct objfile *objfile, data_buf &addr_vec, |
| 429 | CORE_ADDR start, CORE_ADDR end, unsigned int cu_index) |
| 430 | { |
| 431 | addr_vec.append_uint (8, BFD_ENDIAN_LITTLE, start); |
| 432 | addr_vec.append_uint (8, BFD_ENDIAN_LITTLE, end); |
| 433 | addr_vec.append_data (MAYBE_SWAP (cu_index)); |
| 434 | } |
| 435 | |
| 436 | /* Worker function for traversing an addrmap to build the address table. */ |
| 437 | |
| 438 | static int |
| 439 | add_address_entry_worker (void *datap, CORE_ADDR start_addr, void *obj) |
| 440 | { |
| 441 | struct addrmap_index_data *data = (struct addrmap_index_data *) datap; |
| 442 | struct partial_symtab *pst = (struct partial_symtab *) obj; |
| 443 | |
| 444 | if (data->previous_valid) |
| 445 | add_address_entry (data->objfile, data->addr_vec, |
| 446 | data->previous_cu_start, start_addr, |
| 447 | data->previous_cu_index); |
| 448 | |
| 449 | data->previous_cu_start = start_addr; |
| 450 | if (pst != NULL) |
| 451 | { |
| 452 | const auto it = data->cu_index_htab.find (pst); |
| 453 | gdb_assert (it != data->cu_index_htab.cend ()); |
| 454 | data->previous_cu_index = it->second; |
| 455 | data->previous_valid = 1; |
| 456 | } |
| 457 | else |
| 458 | data->previous_valid = 0; |
| 459 | |
| 460 | return 0; |
| 461 | } |
| 462 | |
| 463 | /* Write OBJFILE's address map to ADDR_VEC. |
| 464 | CU_INDEX_HTAB is used to map addrmap entries to their CU indices |
| 465 | in the index file. */ |
| 466 | |
| 467 | static void |
| 468 | write_address_map (struct objfile *objfile, data_buf &addr_vec, |
| 469 | psym_index_map &cu_index_htab) |
| 470 | { |
| 471 | struct addrmap_index_data addrmap_index_data (addr_vec, cu_index_htab); |
| 472 | |
| 473 | /* When writing the address table, we have to cope with the fact that |
| 474 | the addrmap iterator only provides the start of a region; we have to |
| 475 | wait until the next invocation to get the start of the next region. */ |
| 476 | |
| 477 | addrmap_index_data.objfile = objfile; |
| 478 | addrmap_index_data.previous_valid = 0; |
| 479 | |
| 480 | addrmap_foreach (objfile->partial_symtabs->psymtabs_addrmap, |
| 481 | add_address_entry_worker, &addrmap_index_data); |
| 482 | |
| 483 | /* It's highly unlikely the last entry (end address = 0xff...ff) |
| 484 | is valid, but we should still handle it. |
| 485 | The end address is recorded as the start of the next region, but that |
| 486 | doesn't work here. To cope we pass 0xff...ff, this is a rare situation |
| 487 | anyway. */ |
| 488 | if (addrmap_index_data.previous_valid) |
| 489 | add_address_entry (objfile, addr_vec, |
| 490 | addrmap_index_data.previous_cu_start, (CORE_ADDR) -1, |
| 491 | addrmap_index_data.previous_cu_index); |
| 492 | } |
| 493 | |
| 494 | /* Return the symbol kind of PSYM. */ |
| 495 | |
| 496 | static gdb_index_symbol_kind |
| 497 | symbol_kind (struct partial_symbol *psym) |
| 498 | { |
| 499 | domain_enum domain = psym->domain; |
| 500 | enum address_class aclass = psym->aclass; |
| 501 | |
| 502 | switch (domain) |
| 503 | { |
| 504 | case VAR_DOMAIN: |
| 505 | switch (aclass) |
| 506 | { |
| 507 | case LOC_BLOCK: |
| 508 | return GDB_INDEX_SYMBOL_KIND_FUNCTION; |
| 509 | case LOC_TYPEDEF: |
| 510 | return GDB_INDEX_SYMBOL_KIND_TYPE; |
| 511 | case LOC_COMPUTED: |
| 512 | case LOC_CONST_BYTES: |
| 513 | case LOC_OPTIMIZED_OUT: |
| 514 | case LOC_STATIC: |
| 515 | return GDB_INDEX_SYMBOL_KIND_VARIABLE; |
| 516 | case LOC_CONST: |
| 517 | /* Note: It's currently impossible to recognize psyms as enum values |
| 518 | short of reading the type info. For now punt. */ |
| 519 | return GDB_INDEX_SYMBOL_KIND_VARIABLE; |
| 520 | default: |
| 521 | /* There are other LOC_FOO values that one might want to classify |
| 522 | as variables, but dwarf2read.c doesn't currently use them. */ |
| 523 | return GDB_INDEX_SYMBOL_KIND_OTHER; |
| 524 | } |
| 525 | case STRUCT_DOMAIN: |
| 526 | return GDB_INDEX_SYMBOL_KIND_TYPE; |
| 527 | default: |
| 528 | return GDB_INDEX_SYMBOL_KIND_OTHER; |
| 529 | } |
| 530 | } |
| 531 | |
| 532 | /* Add a list of partial symbols to SYMTAB. */ |
| 533 | |
| 534 | static void |
| 535 | write_psymbols (struct mapped_symtab *symtab, |
| 536 | std::unordered_set<partial_symbol *> &psyms_seen, |
| 537 | struct partial_symbol **psymp, |
| 538 | int count, |
| 539 | offset_type cu_index, |
| 540 | int is_static) |
| 541 | { |
| 542 | for (; count-- > 0; ++psymp) |
| 543 | { |
| 544 | struct partial_symbol *psym = *psymp; |
| 545 | |
| 546 | if (psym->ginfo.language () == language_ada) |
| 547 | error (_("Ada is not currently supported by the index; " |
| 548 | "use the DWARF 5 index instead")); |
| 549 | |
| 550 | /* Only add a given psymbol once. */ |
| 551 | if (psyms_seen.insert (psym).second) |
| 552 | { |
| 553 | gdb_index_symbol_kind kind = symbol_kind (psym); |
| 554 | |
| 555 | add_index_entry (symtab, psym->ginfo.search_name (), |
| 556 | is_static, kind, cu_index); |
| 557 | } |
| 558 | } |
| 559 | } |
| 560 | |
| 561 | /* A helper struct used when iterating over debug_types. */ |
| 562 | struct signatured_type_index_data |
| 563 | { |
| 564 | signatured_type_index_data (data_buf &types_list_, |
| 565 | std::unordered_set<partial_symbol *> &psyms_seen_) |
| 566 | : types_list (types_list_), psyms_seen (psyms_seen_) |
| 567 | {} |
| 568 | |
| 569 | struct objfile *objfile; |
| 570 | struct mapped_symtab *symtab; |
| 571 | data_buf &types_list; |
| 572 | std::unordered_set<partial_symbol *> &psyms_seen; |
| 573 | int cu_index; |
| 574 | }; |
| 575 | |
| 576 | /* A helper function that writes a single signatured_type to an |
| 577 | obstack. */ |
| 578 | |
| 579 | static int |
| 580 | write_one_signatured_type (void **slot, void *d) |
| 581 | { |
| 582 | struct signatured_type_index_data *info |
| 583 | = (struct signatured_type_index_data *) d; |
| 584 | struct signatured_type *entry = (struct signatured_type *) *slot; |
| 585 | struct partial_symtab *psymtab = entry->per_cu.v.psymtab; |
| 586 | |
| 587 | write_psymbols (info->symtab, |
| 588 | info->psyms_seen, |
| 589 | (info->objfile->partial_symtabs->global_psymbols.data () |
| 590 | + psymtab->globals_offset), |
| 591 | psymtab->n_global_syms, info->cu_index, |
| 592 | 0); |
| 593 | write_psymbols (info->symtab, |
| 594 | info->psyms_seen, |
| 595 | (info->objfile->partial_symtabs->static_psymbols.data () |
| 596 | + psymtab->statics_offset), |
| 597 | psymtab->n_static_syms, info->cu_index, |
| 598 | 1); |
| 599 | |
| 600 | info->types_list.append_uint (8, BFD_ENDIAN_LITTLE, |
| 601 | to_underlying (entry->per_cu.sect_off)); |
| 602 | info->types_list.append_uint (8, BFD_ENDIAN_LITTLE, |
| 603 | to_underlying (entry->type_offset_in_tu)); |
| 604 | info->types_list.append_uint (8, BFD_ENDIAN_LITTLE, entry->signature); |
| 605 | |
| 606 | ++info->cu_index; |
| 607 | |
| 608 | return 1; |
| 609 | } |
| 610 | |
| 611 | /* Recurse into all "included" dependencies and count their symbols as |
| 612 | if they appeared in this psymtab. */ |
| 613 | |
| 614 | static void |
| 615 | recursively_count_psymbols (struct partial_symtab *psymtab, |
| 616 | size_t &psyms_seen) |
| 617 | { |
| 618 | for (int i = 0; i < psymtab->number_of_dependencies; ++i) |
| 619 | if (psymtab->dependencies[i]->user != NULL) |
| 620 | recursively_count_psymbols (psymtab->dependencies[i], |
| 621 | psyms_seen); |
| 622 | |
| 623 | psyms_seen += psymtab->n_global_syms; |
| 624 | psyms_seen += psymtab->n_static_syms; |
| 625 | } |
| 626 | |
| 627 | /* Recurse into all "included" dependencies and write their symbols as |
| 628 | if they appeared in this psymtab. */ |
| 629 | |
| 630 | static void |
| 631 | recursively_write_psymbols (struct objfile *objfile, |
| 632 | struct partial_symtab *psymtab, |
| 633 | struct mapped_symtab *symtab, |
| 634 | std::unordered_set<partial_symbol *> &psyms_seen, |
| 635 | offset_type cu_index) |
| 636 | { |
| 637 | int i; |
| 638 | |
| 639 | for (i = 0; i < psymtab->number_of_dependencies; ++i) |
| 640 | if (psymtab->dependencies[i]->user != NULL) |
| 641 | recursively_write_psymbols (objfile, psymtab->dependencies[i], |
| 642 | symtab, psyms_seen, cu_index); |
| 643 | |
| 644 | write_psymbols (symtab, |
| 645 | psyms_seen, |
| 646 | (objfile->partial_symtabs->global_psymbols.data () |
| 647 | + psymtab->globals_offset), |
| 648 | psymtab->n_global_syms, cu_index, |
| 649 | 0); |
| 650 | write_psymbols (symtab, |
| 651 | psyms_seen, |
| 652 | (objfile->partial_symtabs->static_psymbols.data () |
| 653 | + psymtab->statics_offset), |
| 654 | psymtab->n_static_syms, cu_index, |
| 655 | 1); |
| 656 | } |
| 657 | |
| 658 | /* DWARF-5 .debug_names builder. */ |
| 659 | class debug_names |
| 660 | { |
| 661 | public: |
| 662 | debug_names (struct dwarf2_per_objfile *dwarf2_per_objfile, bool is_dwarf64, |
| 663 | bfd_endian dwarf5_byte_order) |
| 664 | : m_dwarf5_byte_order (dwarf5_byte_order), |
| 665 | m_dwarf32 (dwarf5_byte_order), |
| 666 | m_dwarf64 (dwarf5_byte_order), |
| 667 | m_dwarf (is_dwarf64 |
| 668 | ? static_cast<dwarf &> (m_dwarf64) |
| 669 | : static_cast<dwarf &> (m_dwarf32)), |
| 670 | m_name_table_string_offs (m_dwarf.name_table_string_offs), |
| 671 | m_name_table_entry_offs (m_dwarf.name_table_entry_offs), |
| 672 | m_debugstrlookup (dwarf2_per_objfile) |
| 673 | {} |
| 674 | |
| 675 | int dwarf5_offset_size () const |
| 676 | { |
| 677 | const bool dwarf5_is_dwarf64 = &m_dwarf == &m_dwarf64; |
| 678 | return dwarf5_is_dwarf64 ? 8 : 4; |
| 679 | } |
| 680 | |
| 681 | /* Is this symbol from DW_TAG_compile_unit or DW_TAG_type_unit? */ |
| 682 | enum class unit_kind { cu, tu }; |
| 683 | |
| 684 | /* Insert one symbol. */ |
| 685 | void insert (const partial_symbol *psym, int cu_index, bool is_static, |
| 686 | unit_kind kind) |
| 687 | { |
| 688 | const int dwarf_tag = psymbol_tag (psym); |
| 689 | if (dwarf_tag == 0) |
| 690 | return; |
| 691 | const char *name = psym->ginfo.search_name (); |
| 692 | |
| 693 | if (psym->ginfo.language () == language_ada) |
| 694 | { |
| 695 | /* We want to ensure that the Ada main function's name appears |
| 696 | verbatim in the index. However, this name will be of the |
| 697 | form "_ada_mumble", and will be rewritten by ada_decode. |
| 698 | So, recognize it specially here and add it to the index by |
| 699 | hand. */ |
| 700 | if (strcmp (main_name (), name) == 0) |
| 701 | { |
| 702 | const auto insertpair |
| 703 | = m_name_to_value_set.emplace (c_str_view (name), |
| 704 | std::set<symbol_value> ()); |
| 705 | std::set<symbol_value> &value_set = insertpair.first->second; |
| 706 | value_set.emplace (symbol_value (dwarf_tag, cu_index, is_static, |
| 707 | kind)); |
| 708 | } |
| 709 | |
| 710 | /* In order for the index to work when read back into gdb, it |
| 711 | has to supply a funny form of the name: it should be the |
| 712 | encoded name, with any suffixes stripped. Using the |
| 713 | ordinary encoded name will not work properly with the |
| 714 | searching logic in find_name_components_bounds; nor will |
| 715 | using the decoded name. Furthermore, an Ada "verbatim" |
| 716 | name (of the form "<MumBle>") must be entered without the |
| 717 | angle brackets. Note that the current index is unusual, |
| 718 | see PR symtab/24820 for details. */ |
| 719 | std::string decoded = ada_decode (name); |
| 720 | if (decoded[0] == '<') |
| 721 | name = (char *) obstack_copy0 (&m_string_obstack, |
| 722 | decoded.c_str () + 1, |
| 723 | decoded.length () - 2); |
| 724 | else |
| 725 | name = obstack_strdup (&m_string_obstack, |
| 726 | ada_encode (decoded.c_str ())); |
| 727 | } |
| 728 | |
| 729 | const auto insertpair |
| 730 | = m_name_to_value_set.emplace (c_str_view (name), |
| 731 | std::set<symbol_value> ()); |
| 732 | std::set<symbol_value> &value_set = insertpair.first->second; |
| 733 | value_set.emplace (symbol_value (dwarf_tag, cu_index, is_static, kind)); |
| 734 | } |
| 735 | |
| 736 | /* Build all the tables. All symbols must be already inserted. |
| 737 | This function does not call file_write, caller has to do it |
| 738 | afterwards. */ |
| 739 | void build () |
| 740 | { |
| 741 | /* Verify the build method has not be called twice. */ |
| 742 | gdb_assert (m_abbrev_table.empty ()); |
| 743 | const size_t name_count = m_name_to_value_set.size (); |
| 744 | m_bucket_table.resize |
| 745 | (std::pow (2, std::ceil (std::log2 (name_count * 4 / 3)))); |
| 746 | m_hash_table.reserve (name_count); |
| 747 | m_name_table_string_offs.reserve (name_count); |
| 748 | m_name_table_entry_offs.reserve (name_count); |
| 749 | |
| 750 | /* Map each hash of symbol to its name and value. */ |
| 751 | struct hash_it_pair |
| 752 | { |
| 753 | uint32_t hash; |
| 754 | decltype (m_name_to_value_set)::const_iterator it; |
| 755 | }; |
| 756 | std::vector<std::forward_list<hash_it_pair>> bucket_hash; |
| 757 | bucket_hash.resize (m_bucket_table.size ()); |
| 758 | for (decltype (m_name_to_value_set)::const_iterator it |
| 759 | = m_name_to_value_set.cbegin (); |
| 760 | it != m_name_to_value_set.cend (); |
| 761 | ++it) |
| 762 | { |
| 763 | const char *const name = it->first.c_str (); |
| 764 | const uint32_t hash = dwarf5_djb_hash (name); |
| 765 | hash_it_pair hashitpair; |
| 766 | hashitpair.hash = hash; |
| 767 | hashitpair.it = it; |
| 768 | auto &slot = bucket_hash[hash % bucket_hash.size()]; |
| 769 | slot.push_front (std::move (hashitpair)); |
| 770 | } |
| 771 | for (size_t bucket_ix = 0; bucket_ix < bucket_hash.size (); ++bucket_ix) |
| 772 | { |
| 773 | const std::forward_list<hash_it_pair> &hashitlist |
| 774 | = bucket_hash[bucket_ix]; |
| 775 | if (hashitlist.empty ()) |
| 776 | continue; |
| 777 | uint32_t &bucket_slot = m_bucket_table[bucket_ix]; |
| 778 | /* The hashes array is indexed starting at 1. */ |
| 779 | store_unsigned_integer (reinterpret_cast<gdb_byte *> (&bucket_slot), |
| 780 | sizeof (bucket_slot), m_dwarf5_byte_order, |
| 781 | m_hash_table.size () + 1); |
| 782 | for (const hash_it_pair &hashitpair : hashitlist) |
| 783 | { |
| 784 | m_hash_table.push_back (0); |
| 785 | store_unsigned_integer (reinterpret_cast<gdb_byte *> |
| 786 | (&m_hash_table.back ()), |
| 787 | sizeof (m_hash_table.back ()), |
| 788 | m_dwarf5_byte_order, hashitpair.hash); |
| 789 | const c_str_view &name = hashitpair.it->first; |
| 790 | const std::set<symbol_value> &value_set = hashitpair.it->second; |
| 791 | m_name_table_string_offs.push_back_reorder |
| 792 | (m_debugstrlookup.lookup (name.c_str ())); |
| 793 | m_name_table_entry_offs.push_back_reorder (m_entry_pool.size ()); |
| 794 | gdb_assert (!value_set.empty ()); |
| 795 | for (const symbol_value &value : value_set) |
| 796 | { |
| 797 | int &idx = m_indexkey_to_idx[index_key (value.dwarf_tag, |
| 798 | value.is_static, |
| 799 | value.kind)]; |
| 800 | if (idx == 0) |
| 801 | { |
| 802 | idx = m_idx_next++; |
| 803 | m_abbrev_table.append_unsigned_leb128 (idx); |
| 804 | m_abbrev_table.append_unsigned_leb128 (value.dwarf_tag); |
| 805 | m_abbrev_table.append_unsigned_leb128 |
| 806 | (value.kind == unit_kind::cu ? DW_IDX_compile_unit |
| 807 | : DW_IDX_type_unit); |
| 808 | m_abbrev_table.append_unsigned_leb128 (DW_FORM_udata); |
| 809 | m_abbrev_table.append_unsigned_leb128 (value.is_static |
| 810 | ? DW_IDX_GNU_internal |
| 811 | : DW_IDX_GNU_external); |
| 812 | m_abbrev_table.append_unsigned_leb128 (DW_FORM_flag_present); |
| 813 | |
| 814 | /* Terminate attributes list. */ |
| 815 | m_abbrev_table.append_unsigned_leb128 (0); |
| 816 | m_abbrev_table.append_unsigned_leb128 (0); |
| 817 | } |
| 818 | |
| 819 | m_entry_pool.append_unsigned_leb128 (idx); |
| 820 | m_entry_pool.append_unsigned_leb128 (value.cu_index); |
| 821 | } |
| 822 | |
| 823 | /* Terminate the list of CUs. */ |
| 824 | m_entry_pool.append_unsigned_leb128 (0); |
| 825 | } |
| 826 | } |
| 827 | gdb_assert (m_hash_table.size () == name_count); |
| 828 | |
| 829 | /* Terminate tags list. */ |
| 830 | m_abbrev_table.append_unsigned_leb128 (0); |
| 831 | } |
| 832 | |
| 833 | /* Return .debug_names bucket count. This must be called only after |
| 834 | calling the build method. */ |
| 835 | uint32_t bucket_count () const |
| 836 | { |
| 837 | /* Verify the build method has been already called. */ |
| 838 | gdb_assert (!m_abbrev_table.empty ()); |
| 839 | const uint32_t retval = m_bucket_table.size (); |
| 840 | |
| 841 | /* Check for overflow. */ |
| 842 | gdb_assert (retval == m_bucket_table.size ()); |
| 843 | return retval; |
| 844 | } |
| 845 | |
| 846 | /* Return .debug_names names count. This must be called only after |
| 847 | calling the build method. */ |
| 848 | uint32_t name_count () const |
| 849 | { |
| 850 | /* Verify the build method has been already called. */ |
| 851 | gdb_assert (!m_abbrev_table.empty ()); |
| 852 | const uint32_t retval = m_hash_table.size (); |
| 853 | |
| 854 | /* Check for overflow. */ |
| 855 | gdb_assert (retval == m_hash_table.size ()); |
| 856 | return retval; |
| 857 | } |
| 858 | |
| 859 | /* Return number of bytes of .debug_names abbreviation table. This |
| 860 | must be called only after calling the build method. */ |
| 861 | uint32_t abbrev_table_bytes () const |
| 862 | { |
| 863 | gdb_assert (!m_abbrev_table.empty ()); |
| 864 | return m_abbrev_table.size (); |
| 865 | } |
| 866 | |
| 867 | /* Recurse into all "included" dependencies and store their symbols |
| 868 | as if they appeared in this psymtab. */ |
| 869 | void recursively_write_psymbols |
| 870 | (struct objfile *objfile, |
| 871 | struct partial_symtab *psymtab, |
| 872 | std::unordered_set<partial_symbol *> &psyms_seen, |
| 873 | int cu_index) |
| 874 | { |
| 875 | for (int i = 0; i < psymtab->number_of_dependencies; ++i) |
| 876 | if (psymtab->dependencies[i]->user != NULL) |
| 877 | recursively_write_psymbols (objfile, psymtab->dependencies[i], |
| 878 | psyms_seen, cu_index); |
| 879 | |
| 880 | write_psymbols (psyms_seen, |
| 881 | (objfile->partial_symtabs->global_psymbols.data () |
| 882 | + psymtab->globals_offset), |
| 883 | psymtab->n_global_syms, cu_index, false, unit_kind::cu); |
| 884 | write_psymbols (psyms_seen, |
| 885 | (objfile->partial_symtabs->static_psymbols.data () |
| 886 | + psymtab->statics_offset), |
| 887 | psymtab->n_static_syms, cu_index, true, unit_kind::cu); |
| 888 | } |
| 889 | |
| 890 | /* Return number of bytes the .debug_names section will have. This |
| 891 | must be called only after calling the build method. */ |
| 892 | size_t bytes () const |
| 893 | { |
| 894 | /* Verify the build method has been already called. */ |
| 895 | gdb_assert (!m_abbrev_table.empty ()); |
| 896 | size_t expected_bytes = 0; |
| 897 | expected_bytes += m_bucket_table.size () * sizeof (m_bucket_table[0]); |
| 898 | expected_bytes += m_hash_table.size () * sizeof (m_hash_table[0]); |
| 899 | expected_bytes += m_name_table_string_offs.bytes (); |
| 900 | expected_bytes += m_name_table_entry_offs.bytes (); |
| 901 | expected_bytes += m_abbrev_table.size (); |
| 902 | expected_bytes += m_entry_pool.size (); |
| 903 | return expected_bytes; |
| 904 | } |
| 905 | |
| 906 | /* Write .debug_names to FILE_NAMES and .debug_str addition to |
| 907 | FILE_STR. This must be called only after calling the build |
| 908 | method. */ |
| 909 | void file_write (FILE *file_names, FILE *file_str) const |
| 910 | { |
| 911 | /* Verify the build method has been already called. */ |
| 912 | gdb_assert (!m_abbrev_table.empty ()); |
| 913 | ::file_write (file_names, m_bucket_table); |
| 914 | ::file_write (file_names, m_hash_table); |
| 915 | m_name_table_string_offs.file_write (file_names); |
| 916 | m_name_table_entry_offs.file_write (file_names); |
| 917 | m_abbrev_table.file_write (file_names); |
| 918 | m_entry_pool.file_write (file_names); |
| 919 | m_debugstrlookup.file_write (file_str); |
| 920 | } |
| 921 | |
| 922 | /* A helper user data for write_one_signatured_type. */ |
| 923 | class write_one_signatured_type_data |
| 924 | { |
| 925 | public: |
| 926 | write_one_signatured_type_data (debug_names &nametable_, |
| 927 | signatured_type_index_data &&info_) |
| 928 | : nametable (nametable_), info (std::move (info_)) |
| 929 | {} |
| 930 | debug_names &nametable; |
| 931 | struct signatured_type_index_data info; |
| 932 | }; |
| 933 | |
| 934 | /* A helper function to pass write_one_signatured_type to |
| 935 | htab_traverse_noresize. */ |
| 936 | static int |
| 937 | write_one_signatured_type (void **slot, void *d) |
| 938 | { |
| 939 | write_one_signatured_type_data *data = (write_one_signatured_type_data *) d; |
| 940 | struct signatured_type_index_data *info = &data->info; |
| 941 | struct signatured_type *entry = (struct signatured_type *) *slot; |
| 942 | |
| 943 | data->nametable.write_one_signatured_type (entry, info); |
| 944 | |
| 945 | return 1; |
| 946 | } |
| 947 | |
| 948 | private: |
| 949 | |
| 950 | /* Storage for symbol names mapping them to their .debug_str section |
| 951 | offsets. */ |
| 952 | class debug_str_lookup |
| 953 | { |
| 954 | public: |
| 955 | |
| 956 | /* Object constructor to be called for current DWARF2_PER_OBJFILE. |
| 957 | All .debug_str section strings are automatically stored. */ |
| 958 | debug_str_lookup (struct dwarf2_per_objfile *dwarf2_per_objfile) |
| 959 | : m_abfd (dwarf2_per_objfile->objfile->obfd), |
| 960 | m_dwarf2_per_objfile (dwarf2_per_objfile) |
| 961 | { |
| 962 | dwarf2_read_section (dwarf2_per_objfile->objfile, |
| 963 | &dwarf2_per_objfile->str); |
| 964 | if (dwarf2_per_objfile->str.buffer == NULL) |
| 965 | return; |
| 966 | for (const gdb_byte *data = dwarf2_per_objfile->str.buffer; |
| 967 | data < (dwarf2_per_objfile->str.buffer |
| 968 | + dwarf2_per_objfile->str.size);) |
| 969 | { |
| 970 | const char *const s = reinterpret_cast<const char *> (data); |
| 971 | const auto insertpair |
| 972 | = m_str_table.emplace (c_str_view (s), |
| 973 | data - dwarf2_per_objfile->str.buffer); |
| 974 | if (!insertpair.second) |
| 975 | complaint (_("Duplicate string \"%s\" in " |
| 976 | ".debug_str section [in module %s]"), |
| 977 | s, bfd_get_filename (m_abfd)); |
| 978 | data += strlen (s) + 1; |
| 979 | } |
| 980 | } |
| 981 | |
| 982 | /* Return offset of symbol name S in the .debug_str section. Add |
| 983 | such symbol to the section's end if it does not exist there |
| 984 | yet. */ |
| 985 | size_t lookup (const char *s) |
| 986 | { |
| 987 | const auto it = m_str_table.find (c_str_view (s)); |
| 988 | if (it != m_str_table.end ()) |
| 989 | return it->second; |
| 990 | const size_t offset = (m_dwarf2_per_objfile->str.size |
| 991 | + m_str_add_buf.size ()); |
| 992 | m_str_table.emplace (c_str_view (s), offset); |
| 993 | m_str_add_buf.append_cstr0 (s); |
| 994 | return offset; |
| 995 | } |
| 996 | |
| 997 | /* Append the end of the .debug_str section to FILE. */ |
| 998 | void file_write (FILE *file) const |
| 999 | { |
| 1000 | m_str_add_buf.file_write (file); |
| 1001 | } |
| 1002 | |
| 1003 | private: |
| 1004 | std::unordered_map<c_str_view, size_t, c_str_view_hasher> m_str_table; |
| 1005 | bfd *const m_abfd; |
| 1006 | struct dwarf2_per_objfile *m_dwarf2_per_objfile; |
| 1007 | |
| 1008 | /* Data to add at the end of .debug_str for new needed symbol names. */ |
| 1009 | data_buf m_str_add_buf; |
| 1010 | }; |
| 1011 | |
| 1012 | /* Container to map used DWARF tags to their .debug_names abbreviation |
| 1013 | tags. */ |
| 1014 | class index_key |
| 1015 | { |
| 1016 | public: |
| 1017 | index_key (int dwarf_tag_, bool is_static_, unit_kind kind_) |
| 1018 | : dwarf_tag (dwarf_tag_), is_static (is_static_), kind (kind_) |
| 1019 | { |
| 1020 | } |
| 1021 | |
| 1022 | bool |
| 1023 | operator== (const index_key &other) const |
| 1024 | { |
| 1025 | return (dwarf_tag == other.dwarf_tag && is_static == other.is_static |
| 1026 | && kind == other.kind); |
| 1027 | } |
| 1028 | |
| 1029 | const int dwarf_tag; |
| 1030 | const bool is_static; |
| 1031 | const unit_kind kind; |
| 1032 | }; |
| 1033 | |
| 1034 | /* Provide std::unordered_map::hasher for index_key. */ |
| 1035 | class index_key_hasher |
| 1036 | { |
| 1037 | public: |
| 1038 | size_t |
| 1039 | operator () (const index_key &key) const |
| 1040 | { |
| 1041 | return (std::hash<int>() (key.dwarf_tag) << 1) | key.is_static; |
| 1042 | } |
| 1043 | }; |
| 1044 | |
| 1045 | /* Parameters of one symbol entry. */ |
| 1046 | class symbol_value |
| 1047 | { |
| 1048 | public: |
| 1049 | const int dwarf_tag, cu_index; |
| 1050 | const bool is_static; |
| 1051 | const unit_kind kind; |
| 1052 | |
| 1053 | symbol_value (int dwarf_tag_, int cu_index_, bool is_static_, |
| 1054 | unit_kind kind_) |
| 1055 | : dwarf_tag (dwarf_tag_), cu_index (cu_index_), is_static (is_static_), |
| 1056 | kind (kind_) |
| 1057 | {} |
| 1058 | |
| 1059 | bool |
| 1060 | operator< (const symbol_value &other) const |
| 1061 | { |
| 1062 | #define X(n) \ |
| 1063 | do \ |
| 1064 | { \ |
| 1065 | if (n < other.n) \ |
| 1066 | return true; \ |
| 1067 | if (n > other.n) \ |
| 1068 | return false; \ |
| 1069 | } \ |
| 1070 | while (0) |
| 1071 | X (dwarf_tag); |
| 1072 | X (is_static); |
| 1073 | X (kind); |
| 1074 | X (cu_index); |
| 1075 | #undef X |
| 1076 | return false; |
| 1077 | } |
| 1078 | }; |
| 1079 | |
| 1080 | /* Abstract base class to unify DWARF-32 and DWARF-64 name table |
| 1081 | output. */ |
| 1082 | class offset_vec |
| 1083 | { |
| 1084 | protected: |
| 1085 | const bfd_endian dwarf5_byte_order; |
| 1086 | public: |
| 1087 | explicit offset_vec (bfd_endian dwarf5_byte_order_) |
| 1088 | : dwarf5_byte_order (dwarf5_byte_order_) |
| 1089 | {} |
| 1090 | |
| 1091 | /* Call std::vector::reserve for NELEM elements. */ |
| 1092 | virtual void reserve (size_t nelem) = 0; |
| 1093 | |
| 1094 | /* Call std::vector::push_back with store_unsigned_integer byte |
| 1095 | reordering for ELEM. */ |
| 1096 | virtual void push_back_reorder (size_t elem) = 0; |
| 1097 | |
| 1098 | /* Return expected output size in bytes. */ |
| 1099 | virtual size_t bytes () const = 0; |
| 1100 | |
| 1101 | /* Write name table to FILE. */ |
| 1102 | virtual void file_write (FILE *file) const = 0; |
| 1103 | }; |
| 1104 | |
| 1105 | /* Template to unify DWARF-32 and DWARF-64 output. */ |
| 1106 | template<typename OffsetSize> |
| 1107 | class offset_vec_tmpl : public offset_vec |
| 1108 | { |
| 1109 | public: |
| 1110 | explicit offset_vec_tmpl (bfd_endian dwarf5_byte_order_) |
| 1111 | : offset_vec (dwarf5_byte_order_) |
| 1112 | {} |
| 1113 | |
| 1114 | /* Implement offset_vec::reserve. */ |
| 1115 | void reserve (size_t nelem) override |
| 1116 | { |
| 1117 | m_vec.reserve (nelem); |
| 1118 | } |
| 1119 | |
| 1120 | /* Implement offset_vec::push_back_reorder. */ |
| 1121 | void push_back_reorder (size_t elem) override |
| 1122 | { |
| 1123 | m_vec.push_back (elem); |
| 1124 | /* Check for overflow. */ |
| 1125 | gdb_assert (m_vec.back () == elem); |
| 1126 | store_unsigned_integer (reinterpret_cast<gdb_byte *> (&m_vec.back ()), |
| 1127 | sizeof (m_vec.back ()), dwarf5_byte_order, elem); |
| 1128 | } |
| 1129 | |
| 1130 | /* Implement offset_vec::bytes. */ |
| 1131 | size_t bytes () const override |
| 1132 | { |
| 1133 | return m_vec.size () * sizeof (m_vec[0]); |
| 1134 | } |
| 1135 | |
| 1136 | /* Implement offset_vec::file_write. */ |
| 1137 | void file_write (FILE *file) const override |
| 1138 | { |
| 1139 | ::file_write (file, m_vec); |
| 1140 | } |
| 1141 | |
| 1142 | private: |
| 1143 | std::vector<OffsetSize> m_vec; |
| 1144 | }; |
| 1145 | |
| 1146 | /* Base class to unify DWARF-32 and DWARF-64 .debug_names output |
| 1147 | respecting name table width. */ |
| 1148 | class dwarf |
| 1149 | { |
| 1150 | public: |
| 1151 | offset_vec &name_table_string_offs, &name_table_entry_offs; |
| 1152 | |
| 1153 | dwarf (offset_vec &name_table_string_offs_, |
| 1154 | offset_vec &name_table_entry_offs_) |
| 1155 | : name_table_string_offs (name_table_string_offs_), |
| 1156 | name_table_entry_offs (name_table_entry_offs_) |
| 1157 | { |
| 1158 | } |
| 1159 | }; |
| 1160 | |
| 1161 | /* Template to unify DWARF-32 and DWARF-64 .debug_names output |
| 1162 | respecting name table width. */ |
| 1163 | template<typename OffsetSize> |
| 1164 | class dwarf_tmpl : public dwarf |
| 1165 | { |
| 1166 | public: |
| 1167 | explicit dwarf_tmpl (bfd_endian dwarf5_byte_order_) |
| 1168 | : dwarf (m_name_table_string_offs, m_name_table_entry_offs), |
| 1169 | m_name_table_string_offs (dwarf5_byte_order_), |
| 1170 | m_name_table_entry_offs (dwarf5_byte_order_) |
| 1171 | {} |
| 1172 | |
| 1173 | private: |
| 1174 | offset_vec_tmpl<OffsetSize> m_name_table_string_offs; |
| 1175 | offset_vec_tmpl<OffsetSize> m_name_table_entry_offs; |
| 1176 | }; |
| 1177 | |
| 1178 | /* Try to reconstruct original DWARF tag for given partial_symbol. |
| 1179 | This function is not DWARF-5 compliant but it is sufficient for |
| 1180 | GDB as a DWARF-5 index consumer. */ |
| 1181 | static int psymbol_tag (const struct partial_symbol *psym) |
| 1182 | { |
| 1183 | domain_enum domain = psym->domain; |
| 1184 | enum address_class aclass = psym->aclass; |
| 1185 | |
| 1186 | switch (domain) |
| 1187 | { |
| 1188 | case VAR_DOMAIN: |
| 1189 | switch (aclass) |
| 1190 | { |
| 1191 | case LOC_BLOCK: |
| 1192 | return DW_TAG_subprogram; |
| 1193 | case LOC_TYPEDEF: |
| 1194 | return DW_TAG_typedef; |
| 1195 | case LOC_COMPUTED: |
| 1196 | case LOC_CONST_BYTES: |
| 1197 | case LOC_OPTIMIZED_OUT: |
| 1198 | case LOC_STATIC: |
| 1199 | return DW_TAG_variable; |
| 1200 | case LOC_CONST: |
| 1201 | /* Note: It's currently impossible to recognize psyms as enum values |
| 1202 | short of reading the type info. For now punt. */ |
| 1203 | return DW_TAG_variable; |
| 1204 | default: |
| 1205 | /* There are other LOC_FOO values that one might want to classify |
| 1206 | as variables, but dwarf2read.c doesn't currently use them. */ |
| 1207 | return DW_TAG_variable; |
| 1208 | } |
| 1209 | case STRUCT_DOMAIN: |
| 1210 | return DW_TAG_structure_type; |
| 1211 | default: |
| 1212 | return 0; |
| 1213 | } |
| 1214 | } |
| 1215 | |
| 1216 | /* Call insert for all partial symbols and mark them in PSYMS_SEEN. */ |
| 1217 | void write_psymbols (std::unordered_set<partial_symbol *> &psyms_seen, |
| 1218 | struct partial_symbol **psymp, int count, int cu_index, |
| 1219 | bool is_static, unit_kind kind) |
| 1220 | { |
| 1221 | for (; count-- > 0; ++psymp) |
| 1222 | { |
| 1223 | struct partial_symbol *psym = *psymp; |
| 1224 | |
| 1225 | /* Only add a given psymbol once. */ |
| 1226 | if (psyms_seen.insert (psym).second) |
| 1227 | insert (psym, cu_index, is_static, kind); |
| 1228 | } |
| 1229 | } |
| 1230 | |
| 1231 | /* A helper function that writes a single signatured_type |
| 1232 | to a debug_names. */ |
| 1233 | void |
| 1234 | write_one_signatured_type (struct signatured_type *entry, |
| 1235 | struct signatured_type_index_data *info) |
| 1236 | { |
| 1237 | struct partial_symtab *psymtab = entry->per_cu.v.psymtab; |
| 1238 | |
| 1239 | write_psymbols (info->psyms_seen, |
| 1240 | (info->objfile->partial_symtabs->global_psymbols.data () |
| 1241 | + psymtab->globals_offset), |
| 1242 | psymtab->n_global_syms, info->cu_index, false, |
| 1243 | unit_kind::tu); |
| 1244 | write_psymbols (info->psyms_seen, |
| 1245 | (info->objfile->partial_symtabs->static_psymbols.data () |
| 1246 | + psymtab->statics_offset), |
| 1247 | psymtab->n_static_syms, info->cu_index, true, |
| 1248 | unit_kind::tu); |
| 1249 | |
| 1250 | info->types_list.append_uint (dwarf5_offset_size (), m_dwarf5_byte_order, |
| 1251 | to_underlying (entry->per_cu.sect_off)); |
| 1252 | |
| 1253 | ++info->cu_index; |
| 1254 | } |
| 1255 | |
| 1256 | /* Store value of each symbol. */ |
| 1257 | std::unordered_map<c_str_view, std::set<symbol_value>, c_str_view_hasher> |
| 1258 | m_name_to_value_set; |
| 1259 | |
| 1260 | /* Tables of DWARF-5 .debug_names. They are in object file byte |
| 1261 | order. */ |
| 1262 | std::vector<uint32_t> m_bucket_table; |
| 1263 | std::vector<uint32_t> m_hash_table; |
| 1264 | |
| 1265 | const bfd_endian m_dwarf5_byte_order; |
| 1266 | dwarf_tmpl<uint32_t> m_dwarf32; |
| 1267 | dwarf_tmpl<uint64_t> m_dwarf64; |
| 1268 | dwarf &m_dwarf; |
| 1269 | offset_vec &m_name_table_string_offs, &m_name_table_entry_offs; |
| 1270 | debug_str_lookup m_debugstrlookup; |
| 1271 | |
| 1272 | /* Map each used .debug_names abbreviation tag parameter to its |
| 1273 | index value. */ |
| 1274 | std::unordered_map<index_key, int, index_key_hasher> m_indexkey_to_idx; |
| 1275 | |
| 1276 | /* Next unused .debug_names abbreviation tag for |
| 1277 | m_indexkey_to_idx. */ |
| 1278 | int m_idx_next = 1; |
| 1279 | |
| 1280 | /* .debug_names abbreviation table. */ |
| 1281 | data_buf m_abbrev_table; |
| 1282 | |
| 1283 | /* .debug_names entry pool. */ |
| 1284 | data_buf m_entry_pool; |
| 1285 | |
| 1286 | /* Temporary storage for Ada names. */ |
| 1287 | auto_obstack m_string_obstack; |
| 1288 | }; |
| 1289 | |
| 1290 | /* Return iff any of the needed offsets does not fit into 32-bit |
| 1291 | .debug_names section. */ |
| 1292 | |
| 1293 | static bool |
| 1294 | check_dwarf64_offsets (struct dwarf2_per_objfile *dwarf2_per_objfile) |
| 1295 | { |
| 1296 | for (dwarf2_per_cu_data *per_cu : dwarf2_per_objfile->all_comp_units) |
| 1297 | { |
| 1298 | if (to_underlying (per_cu->sect_off) >= (static_cast<uint64_t> (1) << 32)) |
| 1299 | return true; |
| 1300 | } |
| 1301 | for (const signatured_type *sigtype : dwarf2_per_objfile->all_type_units) |
| 1302 | { |
| 1303 | const dwarf2_per_cu_data &per_cu = sigtype->per_cu; |
| 1304 | |
| 1305 | if (to_underlying (per_cu.sect_off) >= (static_cast<uint64_t> (1) << 32)) |
| 1306 | return true; |
| 1307 | } |
| 1308 | return false; |
| 1309 | } |
| 1310 | |
| 1311 | /* The psyms_seen set is potentially going to be largish (~40k |
| 1312 | elements when indexing a -g3 build of GDB itself). Estimate the |
| 1313 | number of elements in order to avoid too many rehashes, which |
| 1314 | require rebuilding buckets and thus many trips to |
| 1315 | malloc/free. */ |
| 1316 | |
| 1317 | static size_t |
| 1318 | psyms_seen_size (struct dwarf2_per_objfile *dwarf2_per_objfile) |
| 1319 | { |
| 1320 | size_t psyms_count = 0; |
| 1321 | for (dwarf2_per_cu_data *per_cu : dwarf2_per_objfile->all_comp_units) |
| 1322 | { |
| 1323 | struct partial_symtab *psymtab = per_cu->v.psymtab; |
| 1324 | |
| 1325 | if (psymtab != NULL && psymtab->user == NULL) |
| 1326 | recursively_count_psymbols (psymtab, psyms_count); |
| 1327 | } |
| 1328 | /* Generating an index for gdb itself shows a ratio of |
| 1329 | TOTAL_SEEN_SYMS/UNIQUE_SYMS or ~5. 4 seems like a good bet. */ |
| 1330 | return psyms_count / 4; |
| 1331 | } |
| 1332 | |
| 1333 | /* Assert that FILE's size is EXPECTED_SIZE. Assumes file's seek |
| 1334 | position is at the end of the file. */ |
| 1335 | |
| 1336 | static void |
| 1337 | assert_file_size (FILE *file, size_t expected_size) |
| 1338 | { |
| 1339 | const auto file_size = ftell (file); |
| 1340 | if (file_size == -1) |
| 1341 | perror_with_name (("ftell")); |
| 1342 | gdb_assert (file_size == expected_size); |
| 1343 | } |
| 1344 | |
| 1345 | /* Write a gdb index file to OUT_FILE from all the sections passed as |
| 1346 | arguments. */ |
| 1347 | |
| 1348 | static void |
| 1349 | write_gdbindex_1 (FILE *out_file, |
| 1350 | const data_buf &cu_list, |
| 1351 | const data_buf &types_cu_list, |
| 1352 | const data_buf &addr_vec, |
| 1353 | const data_buf &symtab_vec, |
| 1354 | const data_buf &constant_pool) |
| 1355 | { |
| 1356 | data_buf contents; |
| 1357 | const offset_type size_of_header = 6 * sizeof (offset_type); |
| 1358 | offset_type total_len = size_of_header; |
| 1359 | |
| 1360 | /* The version number. */ |
| 1361 | contents.append_data (MAYBE_SWAP (8)); |
| 1362 | |
| 1363 | /* The offset of the CU list from the start of the file. */ |
| 1364 | contents.append_data (MAYBE_SWAP (total_len)); |
| 1365 | total_len += cu_list.size (); |
| 1366 | |
| 1367 | /* The offset of the types CU list from the start of the file. */ |
| 1368 | contents.append_data (MAYBE_SWAP (total_len)); |
| 1369 | total_len += types_cu_list.size (); |
| 1370 | |
| 1371 | /* The offset of the address table from the start of the file. */ |
| 1372 | contents.append_data (MAYBE_SWAP (total_len)); |
| 1373 | total_len += addr_vec.size (); |
| 1374 | |
| 1375 | /* The offset of the symbol table from the start of the file. */ |
| 1376 | contents.append_data (MAYBE_SWAP (total_len)); |
| 1377 | total_len += symtab_vec.size (); |
| 1378 | |
| 1379 | /* The offset of the constant pool from the start of the file. */ |
| 1380 | contents.append_data (MAYBE_SWAP (total_len)); |
| 1381 | total_len += constant_pool.size (); |
| 1382 | |
| 1383 | gdb_assert (contents.size () == size_of_header); |
| 1384 | |
| 1385 | contents.file_write (out_file); |
| 1386 | cu_list.file_write (out_file); |
| 1387 | types_cu_list.file_write (out_file); |
| 1388 | addr_vec.file_write (out_file); |
| 1389 | symtab_vec.file_write (out_file); |
| 1390 | constant_pool.file_write (out_file); |
| 1391 | |
| 1392 | assert_file_size (out_file, total_len); |
| 1393 | } |
| 1394 | |
| 1395 | /* Write contents of a .gdb_index section for OBJFILE into OUT_FILE. |
| 1396 | If OBJFILE has an associated dwz file, write contents of a .gdb_index |
| 1397 | section for that dwz file into DWZ_OUT_FILE. If OBJFILE does not have an |
| 1398 | associated dwz file, DWZ_OUT_FILE must be NULL. */ |
| 1399 | |
| 1400 | static void |
| 1401 | write_gdbindex (struct dwarf2_per_objfile *dwarf2_per_objfile, FILE *out_file, |
| 1402 | FILE *dwz_out_file) |
| 1403 | { |
| 1404 | struct objfile *objfile = dwarf2_per_objfile->objfile; |
| 1405 | mapped_symtab symtab; |
| 1406 | data_buf objfile_cu_list; |
| 1407 | data_buf dwz_cu_list; |
| 1408 | |
| 1409 | /* While we're scanning CU's create a table that maps a psymtab pointer |
| 1410 | (which is what addrmap records) to its index (which is what is recorded |
| 1411 | in the index file). This will later be needed to write the address |
| 1412 | table. */ |
| 1413 | psym_index_map cu_index_htab; |
| 1414 | cu_index_htab.reserve (dwarf2_per_objfile->all_comp_units.size ()); |
| 1415 | |
| 1416 | /* The CU list is already sorted, so we don't need to do additional |
| 1417 | work here. Also, the debug_types entries do not appear in |
| 1418 | all_comp_units, but only in their own hash table. */ |
| 1419 | |
| 1420 | std::unordered_set<partial_symbol *> psyms_seen |
| 1421 | (psyms_seen_size (dwarf2_per_objfile)); |
| 1422 | for (int i = 0; i < dwarf2_per_objfile->all_comp_units.size (); ++i) |
| 1423 | { |
| 1424 | struct dwarf2_per_cu_data *per_cu |
| 1425 | = dwarf2_per_objfile->all_comp_units[i]; |
| 1426 | struct partial_symtab *psymtab = per_cu->v.psymtab; |
| 1427 | |
| 1428 | /* CU of a shared file from 'dwz -m' may be unused by this main file. |
| 1429 | It may be referenced from a local scope but in such case it does not |
| 1430 | need to be present in .gdb_index. */ |
| 1431 | if (psymtab == NULL) |
| 1432 | continue; |
| 1433 | |
| 1434 | if (psymtab->user == NULL) |
| 1435 | recursively_write_psymbols (objfile, psymtab, &symtab, |
| 1436 | psyms_seen, i); |
| 1437 | |
| 1438 | const auto insertpair = cu_index_htab.emplace (psymtab, i); |
| 1439 | gdb_assert (insertpair.second); |
| 1440 | |
| 1441 | /* The all_comp_units list contains CUs read from the objfile as well as |
| 1442 | from the eventual dwz file. We need to place the entry in the |
| 1443 | corresponding index. */ |
| 1444 | data_buf &cu_list = per_cu->is_dwz ? dwz_cu_list : objfile_cu_list; |
| 1445 | cu_list.append_uint (8, BFD_ENDIAN_LITTLE, |
| 1446 | to_underlying (per_cu->sect_off)); |
| 1447 | cu_list.append_uint (8, BFD_ENDIAN_LITTLE, per_cu->length); |
| 1448 | } |
| 1449 | |
| 1450 | /* Dump the address map. */ |
| 1451 | data_buf addr_vec; |
| 1452 | write_address_map (objfile, addr_vec, cu_index_htab); |
| 1453 | |
| 1454 | /* Write out the .debug_type entries, if any. */ |
| 1455 | data_buf types_cu_list; |
| 1456 | if (dwarf2_per_objfile->signatured_types) |
| 1457 | { |
| 1458 | signatured_type_index_data sig_data (types_cu_list, |
| 1459 | psyms_seen); |
| 1460 | |
| 1461 | sig_data.objfile = objfile; |
| 1462 | sig_data.symtab = &symtab; |
| 1463 | sig_data.cu_index = dwarf2_per_objfile->all_comp_units.size (); |
| 1464 | htab_traverse_noresize (dwarf2_per_objfile->signatured_types, |
| 1465 | write_one_signatured_type, &sig_data); |
| 1466 | } |
| 1467 | |
| 1468 | /* Now that we've processed all symbols we can shrink their cu_indices |
| 1469 | lists. */ |
| 1470 | uniquify_cu_indices (&symtab); |
| 1471 | |
| 1472 | data_buf symtab_vec, constant_pool; |
| 1473 | write_hash_table (&symtab, symtab_vec, constant_pool); |
| 1474 | |
| 1475 | write_gdbindex_1(out_file, objfile_cu_list, types_cu_list, addr_vec, |
| 1476 | symtab_vec, constant_pool); |
| 1477 | |
| 1478 | if (dwz_out_file != NULL) |
| 1479 | write_gdbindex_1 (dwz_out_file, dwz_cu_list, {}, {}, {}, {}); |
| 1480 | else |
| 1481 | gdb_assert (dwz_cu_list.empty ()); |
| 1482 | } |
| 1483 | |
| 1484 | /* DWARF-5 augmentation string for GDB's DW_IDX_GNU_* extension. */ |
| 1485 | static const gdb_byte dwarf5_gdb_augmentation[] = { 'G', 'D', 'B', 0 }; |
| 1486 | |
| 1487 | /* Write a new .debug_names section for OBJFILE into OUT_FILE, write |
| 1488 | needed addition to .debug_str section to OUT_FILE_STR. Return how |
| 1489 | many bytes were expected to be written into OUT_FILE. */ |
| 1490 | |
| 1491 | static void |
| 1492 | write_debug_names (struct dwarf2_per_objfile *dwarf2_per_objfile, |
| 1493 | FILE *out_file, FILE *out_file_str) |
| 1494 | { |
| 1495 | const bool dwarf5_is_dwarf64 = check_dwarf64_offsets (dwarf2_per_objfile); |
| 1496 | struct objfile *objfile = dwarf2_per_objfile->objfile; |
| 1497 | const enum bfd_endian dwarf5_byte_order |
| 1498 | = gdbarch_byte_order (get_objfile_arch (objfile)); |
| 1499 | |
| 1500 | /* The CU list is already sorted, so we don't need to do additional |
| 1501 | work here. Also, the debug_types entries do not appear in |
| 1502 | all_comp_units, but only in their own hash table. */ |
| 1503 | data_buf cu_list; |
| 1504 | debug_names nametable (dwarf2_per_objfile, dwarf5_is_dwarf64, |
| 1505 | dwarf5_byte_order); |
| 1506 | std::unordered_set<partial_symbol *> |
| 1507 | psyms_seen (psyms_seen_size (dwarf2_per_objfile)); |
| 1508 | for (int i = 0; i < dwarf2_per_objfile->all_comp_units.size (); ++i) |
| 1509 | { |
| 1510 | const dwarf2_per_cu_data *per_cu = dwarf2_per_objfile->all_comp_units[i]; |
| 1511 | partial_symtab *psymtab = per_cu->v.psymtab; |
| 1512 | |
| 1513 | /* CU of a shared file from 'dwz -m' may be unused by this main |
| 1514 | file. It may be referenced from a local scope but in such |
| 1515 | case it does not need to be present in .debug_names. */ |
| 1516 | if (psymtab == NULL) |
| 1517 | continue; |
| 1518 | |
| 1519 | if (psymtab->user == NULL) |
| 1520 | nametable.recursively_write_psymbols (objfile, psymtab, psyms_seen, i); |
| 1521 | |
| 1522 | cu_list.append_uint (nametable.dwarf5_offset_size (), dwarf5_byte_order, |
| 1523 | to_underlying (per_cu->sect_off)); |
| 1524 | } |
| 1525 | |
| 1526 | /* Write out the .debug_type entries, if any. */ |
| 1527 | data_buf types_cu_list; |
| 1528 | if (dwarf2_per_objfile->signatured_types) |
| 1529 | { |
| 1530 | debug_names::write_one_signatured_type_data sig_data (nametable, |
| 1531 | signatured_type_index_data (types_cu_list, psyms_seen)); |
| 1532 | |
| 1533 | sig_data.info.objfile = objfile; |
| 1534 | /* It is used only for gdb_index. */ |
| 1535 | sig_data.info.symtab = nullptr; |
| 1536 | sig_data.info.cu_index = 0; |
| 1537 | htab_traverse_noresize (dwarf2_per_objfile->signatured_types, |
| 1538 | debug_names::write_one_signatured_type, |
| 1539 | &sig_data); |
| 1540 | } |
| 1541 | |
| 1542 | nametable.build (); |
| 1543 | |
| 1544 | /* No addr_vec - DWARF-5 uses .debug_aranges generated by GCC. */ |
| 1545 | |
| 1546 | const offset_type bytes_of_header |
| 1547 | = ((dwarf5_is_dwarf64 ? 12 : 4) |
| 1548 | + 2 + 2 + 7 * 4 |
| 1549 | + sizeof (dwarf5_gdb_augmentation)); |
| 1550 | size_t expected_bytes = 0; |
| 1551 | expected_bytes += bytes_of_header; |
| 1552 | expected_bytes += cu_list.size (); |
| 1553 | expected_bytes += types_cu_list.size (); |
| 1554 | expected_bytes += nametable.bytes (); |
| 1555 | data_buf header; |
| 1556 | |
| 1557 | if (!dwarf5_is_dwarf64) |
| 1558 | { |
| 1559 | const uint64_t size64 = expected_bytes - 4; |
| 1560 | gdb_assert (size64 < 0xfffffff0); |
| 1561 | header.append_uint (4, dwarf5_byte_order, size64); |
| 1562 | } |
| 1563 | else |
| 1564 | { |
| 1565 | header.append_uint (4, dwarf5_byte_order, 0xffffffff); |
| 1566 | header.append_uint (8, dwarf5_byte_order, expected_bytes - 12); |
| 1567 | } |
| 1568 | |
| 1569 | /* The version number. */ |
| 1570 | header.append_uint (2, dwarf5_byte_order, 5); |
| 1571 | |
| 1572 | /* Padding. */ |
| 1573 | header.append_uint (2, dwarf5_byte_order, 0); |
| 1574 | |
| 1575 | /* comp_unit_count - The number of CUs in the CU list. */ |
| 1576 | header.append_uint (4, dwarf5_byte_order, |
| 1577 | dwarf2_per_objfile->all_comp_units.size ()); |
| 1578 | |
| 1579 | /* local_type_unit_count - The number of TUs in the local TU |
| 1580 | list. */ |
| 1581 | header.append_uint (4, dwarf5_byte_order, |
| 1582 | dwarf2_per_objfile->all_type_units.size ()); |
| 1583 | |
| 1584 | /* foreign_type_unit_count - The number of TUs in the foreign TU |
| 1585 | list. */ |
| 1586 | header.append_uint (4, dwarf5_byte_order, 0); |
| 1587 | |
| 1588 | /* bucket_count - The number of hash buckets in the hash lookup |
| 1589 | table. */ |
| 1590 | header.append_uint (4, dwarf5_byte_order, nametable.bucket_count ()); |
| 1591 | |
| 1592 | /* name_count - The number of unique names in the index. */ |
| 1593 | header.append_uint (4, dwarf5_byte_order, nametable.name_count ()); |
| 1594 | |
| 1595 | /* abbrev_table_size - The size in bytes of the abbreviations |
| 1596 | table. */ |
| 1597 | header.append_uint (4, dwarf5_byte_order, nametable.abbrev_table_bytes ()); |
| 1598 | |
| 1599 | /* augmentation_string_size - The size in bytes of the augmentation |
| 1600 | string. This value is rounded up to a multiple of 4. */ |
| 1601 | static_assert (sizeof (dwarf5_gdb_augmentation) % 4 == 0, ""); |
| 1602 | header.append_uint (4, dwarf5_byte_order, sizeof (dwarf5_gdb_augmentation)); |
| 1603 | header.append_data (dwarf5_gdb_augmentation); |
| 1604 | |
| 1605 | gdb_assert (header.size () == bytes_of_header); |
| 1606 | |
| 1607 | header.file_write (out_file); |
| 1608 | cu_list.file_write (out_file); |
| 1609 | types_cu_list.file_write (out_file); |
| 1610 | nametable.file_write (out_file, out_file_str); |
| 1611 | |
| 1612 | assert_file_size (out_file, expected_bytes); |
| 1613 | } |
| 1614 | |
| 1615 | /* This represents an index file being written (work-in-progress). |
| 1616 | |
| 1617 | The data is initially written to a temporary file. When the finalize method |
| 1618 | is called, the file is closed and moved to its final location. |
| 1619 | |
| 1620 | On failure (if this object is being destroyed with having called finalize), |
| 1621 | the temporary file is closed and deleted. */ |
| 1622 | |
| 1623 | struct index_wip_file |
| 1624 | { |
| 1625 | index_wip_file (const char *dir, const char *basename, |
| 1626 | const char *suffix) |
| 1627 | { |
| 1628 | filename = (std::string (dir) + SLASH_STRING + basename |
| 1629 | + suffix); |
| 1630 | |
| 1631 | filename_temp = make_temp_filename (filename); |
| 1632 | |
| 1633 | scoped_fd out_file_fd (gdb_mkostemp_cloexec (filename_temp.data (), |
| 1634 | O_BINARY)); |
| 1635 | if (out_file_fd.get () == -1) |
| 1636 | perror_with_name (("mkstemp")); |
| 1637 | |
| 1638 | out_file = out_file_fd.to_file ("wb"); |
| 1639 | |
| 1640 | if (out_file == nullptr) |
| 1641 | error (_("Can't open `%s' for writing"), filename_temp.data ()); |
| 1642 | |
| 1643 | unlink_file.emplace (filename_temp.data ()); |
| 1644 | } |
| 1645 | |
| 1646 | void finalize () |
| 1647 | { |
| 1648 | /* We want to keep the file. */ |
| 1649 | unlink_file->keep (); |
| 1650 | |
| 1651 | /* Close and move the str file in place. */ |
| 1652 | unlink_file.reset (); |
| 1653 | if (rename (filename_temp.data (), filename.c_str ()) != 0) |
| 1654 | perror_with_name (("rename")); |
| 1655 | } |
| 1656 | |
| 1657 | std::string filename; |
| 1658 | gdb::char_vector filename_temp; |
| 1659 | |
| 1660 | /* Order matters here; we want FILE to be closed before |
| 1661 | FILENAME_TEMP is unlinked, because on MS-Windows one cannot |
| 1662 | delete a file that is still open. So, we wrap the unlinker in an |
| 1663 | optional and emplace it once we know the file name. */ |
| 1664 | gdb::optional<gdb::unlinker> unlink_file; |
| 1665 | |
| 1666 | gdb_file_up out_file; |
| 1667 | }; |
| 1668 | |
| 1669 | /* See dwarf-index-write.h. */ |
| 1670 | |
| 1671 | void |
| 1672 | write_psymtabs_to_index (struct dwarf2_per_objfile *dwarf2_per_objfile, |
| 1673 | const char *dir, const char *basename, |
| 1674 | const char *dwz_basename, |
| 1675 | dw_index_kind index_kind) |
| 1676 | { |
| 1677 | struct objfile *objfile = dwarf2_per_objfile->objfile; |
| 1678 | |
| 1679 | if (dwarf2_per_objfile->using_index) |
| 1680 | error (_("Cannot use an index to create the index")); |
| 1681 | |
| 1682 | if (dwarf2_per_objfile->types.size () > 1) |
| 1683 | error (_("Cannot make an index when the file has multiple .debug_types sections")); |
| 1684 | |
| 1685 | if (!objfile->partial_symtabs->psymtabs |
| 1686 | || !objfile->partial_symtabs->psymtabs_addrmap) |
| 1687 | return; |
| 1688 | |
| 1689 | struct stat st; |
| 1690 | if (stat (objfile_name (objfile), &st) < 0) |
| 1691 | perror_with_name (objfile_name (objfile)); |
| 1692 | |
| 1693 | const char *index_suffix = (index_kind == dw_index_kind::DEBUG_NAMES |
| 1694 | ? INDEX5_SUFFIX : INDEX4_SUFFIX); |
| 1695 | |
| 1696 | index_wip_file objfile_index_wip (dir, basename, index_suffix); |
| 1697 | gdb::optional<index_wip_file> dwz_index_wip; |
| 1698 | |
| 1699 | if (dwz_basename != NULL) |
| 1700 | dwz_index_wip.emplace (dir, dwz_basename, index_suffix); |
| 1701 | |
| 1702 | if (index_kind == dw_index_kind::DEBUG_NAMES) |
| 1703 | { |
| 1704 | index_wip_file str_wip_file (dir, basename, DEBUG_STR_SUFFIX); |
| 1705 | |
| 1706 | write_debug_names (dwarf2_per_objfile, objfile_index_wip.out_file.get (), |
| 1707 | str_wip_file.out_file.get ()); |
| 1708 | |
| 1709 | str_wip_file.finalize (); |
| 1710 | } |
| 1711 | else |
| 1712 | write_gdbindex (dwarf2_per_objfile, objfile_index_wip.out_file.get (), |
| 1713 | (dwz_index_wip.has_value () |
| 1714 | ? dwz_index_wip->out_file.get () : NULL)); |
| 1715 | |
| 1716 | objfile_index_wip.finalize (); |
| 1717 | |
| 1718 | if (dwz_index_wip.has_value ()) |
| 1719 | dwz_index_wip->finalize (); |
| 1720 | } |
| 1721 | |
| 1722 | /* Implementation of the `save gdb-index' command. |
| 1723 | |
| 1724 | Note that the .gdb_index file format used by this command is |
| 1725 | documented in the GDB manual. Any changes here must be documented |
| 1726 | there. */ |
| 1727 | |
| 1728 | static void |
| 1729 | save_gdb_index_command (const char *arg, int from_tty) |
| 1730 | { |
| 1731 | const char dwarf5space[] = "-dwarf-5 "; |
| 1732 | dw_index_kind index_kind = dw_index_kind::GDB_INDEX; |
| 1733 | |
| 1734 | if (!arg) |
| 1735 | arg = ""; |
| 1736 | |
| 1737 | arg = skip_spaces (arg); |
| 1738 | if (strncmp (arg, dwarf5space, strlen (dwarf5space)) == 0) |
| 1739 | { |
| 1740 | index_kind = dw_index_kind::DEBUG_NAMES; |
| 1741 | arg += strlen (dwarf5space); |
| 1742 | arg = skip_spaces (arg); |
| 1743 | } |
| 1744 | |
| 1745 | if (!*arg) |
| 1746 | error (_("usage: save gdb-index [-dwarf-5] DIRECTORY")); |
| 1747 | |
| 1748 | for (objfile *objfile : current_program_space->objfiles ()) |
| 1749 | { |
| 1750 | struct stat st; |
| 1751 | |
| 1752 | /* If the objfile does not correspond to an actual file, skip it. */ |
| 1753 | if (stat (objfile_name (objfile), &st) < 0) |
| 1754 | continue; |
| 1755 | |
| 1756 | struct dwarf2_per_objfile *dwarf2_per_objfile |
| 1757 | = get_dwarf2_per_objfile (objfile); |
| 1758 | |
| 1759 | if (dwarf2_per_objfile != NULL) |
| 1760 | { |
| 1761 | try |
| 1762 | { |
| 1763 | const char *basename = lbasename (objfile_name (objfile)); |
| 1764 | const dwz_file *dwz = dwarf2_get_dwz_file (dwarf2_per_objfile); |
| 1765 | const char *dwz_basename = NULL; |
| 1766 | |
| 1767 | if (dwz != NULL) |
| 1768 | dwz_basename = lbasename (dwz->filename ()); |
| 1769 | |
| 1770 | write_psymtabs_to_index (dwarf2_per_objfile, arg, basename, |
| 1771 | dwz_basename, index_kind); |
| 1772 | } |
| 1773 | catch (const gdb_exception_error &except) |
| 1774 | { |
| 1775 | exception_fprintf (gdb_stderr, except, |
| 1776 | _("Error while writing index for `%s': "), |
| 1777 | objfile_name (objfile)); |
| 1778 | } |
| 1779 | } |
| 1780 | |
| 1781 | } |
| 1782 | } |
| 1783 | |
| 1784 | void |
| 1785 | _initialize_dwarf_index_write () |
| 1786 | { |
| 1787 | cmd_list_element *c = add_cmd ("gdb-index", class_files, |
| 1788 | save_gdb_index_command, _("\ |
| 1789 | Save a gdb-index file.\n\ |
| 1790 | Usage: save gdb-index [-dwarf-5] DIRECTORY\n\ |
| 1791 | \n\ |
| 1792 | No options create one file with .gdb-index extension for pre-DWARF-5\n\ |
| 1793 | compatible .gdb_index section. With -dwarf-5 creates two files with\n\ |
| 1794 | extension .debug_names and .debug_str for DWARF-5 .debug_names section."), |
| 1795 | &save_cmdlist); |
| 1796 | set_cmd_completer (c, filename_completer); |
| 1797 | } |