Commit | Line | Data |
---|---|---|
cd4fb1b2 SM |
1 | /* DWARF index writing support for GDB. |
2 | ||
3666a048 | 3 | Copyright (C) 1994-2021 Free Software Foundation, Inc. |
cd4fb1b2 SM |
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 | ||
82ca8957 | 22 | #include "dwarf2/index-write.h" |
23baa4cc | 23 | |
cd4fb1b2 SM |
24 | #include "addrmap.h" |
25 | #include "cli/cli-decode.h" | |
268a13a5 TT |
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" | |
cd4fb1b2 | 31 | #include "complaints.h" |
82ca8957 | 32 | #include "dwarf2/index-common.h" |
cd4fb1b2 | 33 | #include "dwarf2.h" |
82ca8957 | 34 | #include "dwarf2/read.h" |
9fda78b6 | 35 | #include "dwarf2/dwz.h" |
cd4fb1b2 SM |
36 | #include "gdb/gdb-index.h" |
37 | #include "gdbcmd.h" | |
38 | #include "objfiles.h" | |
39 | #include "psympriv.h" | |
3b00ef10 | 40 | #include "ada-lang.h" |
cd4fb1b2 | 41 | |
4de283e4 TT |
42 | #include <algorithm> |
43 | #include <cmath> | |
159ed7d9 | 44 | #include <forward_list> |
4de283e4 TT |
45 | #include <set> |
46 | #include <unordered_map> | |
47 | #include <unordered_set> | |
48 | ||
cd4fb1b2 SM |
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 \ | |
dda83cd7 | 60 | && (value) <= GDB_INDEX_SYMBOL_KIND_OTHER); \ |
cd4fb1b2 SM |
61 | GDB_INDEX_SYMBOL_KIND_SET_VALUE((cu_index), (value)); \ |
62 | } while (0) | |
63 | ||
85102364 | 64 | /* Ensure we don't use more than the allotted number of bits for the CU. */ |
cd4fb1b2 SM |
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 | { | |
1f88d0c8 SM |
89 | if (!vec.empty ()) |
90 | file_write (file, vec.data (), vec.size () * sizeof (vec[0])); | |
cd4fb1b2 SM |
91 | } |
92 | ||
93 | /* In-memory buffer to prepare data to be written later to a file. */ | |
94 | class data_buf | |
95 | { | |
96 | public: | |
42c2c694 TT |
97 | /* Copy ARRAY to the end of the buffer. */ |
98 | void append_array (gdb::array_view<const gdb_byte> array) | |
cd4fb1b2 | 99 | { |
42c2c694 | 100 | std::copy (array.begin (), array.end (), grow (array.size ())); |
cd4fb1b2 SM |
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; | |
42c2c694 | 120 | m_vec.push_back (output); |
cd4fb1b2 SM |
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 | ||
42c2c694 TT |
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 | ||
cd4fb1b2 SM |
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); | |
b4be9bfd | 163 | return &*(m_vec.end () - size); |
cd4fb1b2 SM |
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 | ||
181 | /* The symbol table. This is a power-of-2-sized hash table. */ | |
182 | struct mapped_symtab | |
183 | { | |
184 | mapped_symtab () | |
185 | { | |
186 | data.resize (1024); | |
187 | } | |
188 | ||
189 | offset_type n_elements = 0; | |
190 | std::vector<symtab_index_entry> data; | |
7ab96794 TV |
191 | |
192 | /* Temporary storage for Ada names. */ | |
193 | auto_obstack m_string_obstack; | |
cd4fb1b2 SM |
194 | }; |
195 | ||
196 | /* Find a slot in SYMTAB for the symbol NAME. Returns a reference to | |
197 | the slot. | |
198 | ||
199 | Function is used only during write_hash_table so no index format backward | |
200 | compatibility is needed. */ | |
201 | ||
202 | static symtab_index_entry & | |
203 | find_slot (struct mapped_symtab *symtab, const char *name) | |
204 | { | |
205 | offset_type index, step, hash = mapped_index_string_hash (INT_MAX, name); | |
206 | ||
207 | index = hash & (symtab->data.size () - 1); | |
208 | step = ((hash * 17) & (symtab->data.size () - 1)) | 1; | |
209 | ||
210 | for (;;) | |
211 | { | |
212 | if (symtab->data[index].name == NULL | |
213 | || strcmp (name, symtab->data[index].name) == 0) | |
214 | return symtab->data[index]; | |
215 | index = (index + step) & (symtab->data.size () - 1); | |
216 | } | |
217 | } | |
218 | ||
219 | /* Expand SYMTAB's hash table. */ | |
220 | ||
221 | static void | |
222 | hash_expand (struct mapped_symtab *symtab) | |
223 | { | |
224 | auto old_entries = std::move (symtab->data); | |
225 | ||
226 | symtab->data.clear (); | |
227 | symtab->data.resize (old_entries.size () * 2); | |
228 | ||
229 | for (auto &it : old_entries) | |
230 | if (it.name != NULL) | |
231 | { | |
232 | auto &ref = find_slot (symtab, it.name); | |
233 | ref = std::move (it); | |
234 | } | |
235 | } | |
236 | ||
237 | /* Add an entry to SYMTAB. NAME is the name of the symbol. | |
238 | CU_INDEX is the index of the CU in which the symbol appears. | |
239 | IS_STATIC is one if the symbol is static, otherwise zero (global). */ | |
240 | ||
241 | static void | |
242 | add_index_entry (struct mapped_symtab *symtab, const char *name, | |
243 | int is_static, gdb_index_symbol_kind kind, | |
244 | offset_type cu_index) | |
245 | { | |
246 | offset_type cu_index_and_attrs; | |
247 | ||
248 | ++symtab->n_elements; | |
249 | if (4 * symtab->n_elements / 3 >= symtab->data.size ()) | |
250 | hash_expand (symtab); | |
251 | ||
252 | symtab_index_entry &slot = find_slot (symtab, name); | |
253 | if (slot.name == NULL) | |
254 | { | |
255 | slot.name = name; | |
256 | /* index_offset is set later. */ | |
257 | } | |
258 | ||
259 | cu_index_and_attrs = 0; | |
260 | DW2_GDB_INDEX_CU_SET_VALUE (cu_index_and_attrs, cu_index); | |
261 | DW2_GDB_INDEX_SYMBOL_STATIC_SET_VALUE (cu_index_and_attrs, is_static); | |
262 | DW2_GDB_INDEX_SYMBOL_KIND_SET_VALUE (cu_index_and_attrs, kind); | |
263 | ||
264 | /* We don't want to record an index value twice as we want to avoid the | |
265 | duplication. | |
266 | We process all global symbols and then all static symbols | |
267 | (which would allow us to avoid the duplication by only having to check | |
268 | the last entry pushed), but a symbol could have multiple kinds in one CU. | |
269 | To keep things simple we don't worry about the duplication here and | |
85102364 | 270 | sort and uniquify the list after we've processed all symbols. */ |
cd4fb1b2 SM |
271 | slot.cu_indices.push_back (cu_index_and_attrs); |
272 | } | |
273 | ||
274 | /* Sort and remove duplicates of all symbols' cu_indices lists. */ | |
275 | ||
276 | static void | |
277 | uniquify_cu_indices (struct mapped_symtab *symtab) | |
278 | { | |
279 | for (auto &entry : symtab->data) | |
280 | { | |
281 | if (entry.name != NULL && !entry.cu_indices.empty ()) | |
282 | { | |
283 | auto &cu_indices = entry.cu_indices; | |
284 | std::sort (cu_indices.begin (), cu_indices.end ()); | |
285 | auto from = std::unique (cu_indices.begin (), cu_indices.end ()); | |
286 | cu_indices.erase (from, cu_indices.end ()); | |
287 | } | |
288 | } | |
289 | } | |
290 | ||
291 | /* A form of 'const char *' suitable for container keys. Only the | |
292 | pointer is stored. The strings themselves are compared, not the | |
293 | pointers. */ | |
294 | class c_str_view | |
295 | { | |
296 | public: | |
297 | c_str_view (const char *cstr) | |
298 | : m_cstr (cstr) | |
299 | {} | |
300 | ||
301 | bool operator== (const c_str_view &other) const | |
302 | { | |
303 | return strcmp (m_cstr, other.m_cstr) == 0; | |
304 | } | |
305 | ||
306 | /* Return the underlying C string. Note, the returned string is | |
307 | only a reference with lifetime of this object. */ | |
308 | const char *c_str () const | |
309 | { | |
310 | return m_cstr; | |
311 | } | |
312 | ||
313 | private: | |
314 | friend class c_str_view_hasher; | |
315 | const char *const m_cstr; | |
316 | }; | |
317 | ||
318 | /* A std::unordered_map::hasher for c_str_view that uses the right | |
319 | hash function for strings in a mapped index. */ | |
320 | class c_str_view_hasher | |
321 | { | |
322 | public: | |
323 | size_t operator () (const c_str_view &x) const | |
324 | { | |
325 | return mapped_index_string_hash (INT_MAX, x.m_cstr); | |
326 | } | |
327 | }; | |
328 | ||
329 | /* A std::unordered_map::hasher for std::vector<>. */ | |
330 | template<typename T> | |
331 | class vector_hasher | |
332 | { | |
333 | public: | |
334 | size_t operator () (const std::vector<T> &key) const | |
335 | { | |
336 | return iterative_hash (key.data (), | |
337 | sizeof (key.front ()) * key.size (), 0); | |
338 | } | |
339 | }; | |
340 | ||
341 | /* Write the mapped hash table SYMTAB to the data buffer OUTPUT, with | |
342 | constant pool entries going into the data buffer CPOOL. */ | |
343 | ||
344 | static void | |
345 | write_hash_table (mapped_symtab *symtab, data_buf &output, data_buf &cpool) | |
346 | { | |
347 | { | |
348 | /* Elements are sorted vectors of the indices of all the CUs that | |
349 | hold an object of this name. */ | |
350 | std::unordered_map<std::vector<offset_type>, offset_type, | |
351 | vector_hasher<offset_type>> | |
352 | symbol_hash_table; | |
353 | ||
354 | /* We add all the index vectors to the constant pool first, to | |
355 | ensure alignment is ok. */ | |
356 | for (symtab_index_entry &entry : symtab->data) | |
357 | { | |
358 | if (entry.name == NULL) | |
359 | continue; | |
360 | gdb_assert (entry.index_offset == 0); | |
361 | ||
362 | /* Finding before inserting is faster than always trying to | |
363 | insert, because inserting always allocates a node, does the | |
364 | lookup, and then destroys the new node if another node | |
365 | already had the same key. C++17 try_emplace will avoid | |
366 | this. */ | |
367 | const auto found | |
368 | = symbol_hash_table.find (entry.cu_indices); | |
369 | if (found != symbol_hash_table.end ()) | |
370 | { | |
371 | entry.index_offset = found->second; | |
372 | continue; | |
373 | } | |
374 | ||
375 | symbol_hash_table.emplace (entry.cu_indices, cpool.size ()); | |
376 | entry.index_offset = cpool.size (); | |
42c2c694 | 377 | cpool.append_offset (entry.cu_indices.size ()); |
cd4fb1b2 | 378 | for (const auto index : entry.cu_indices) |
42c2c694 | 379 | cpool.append_offset (index); |
cd4fb1b2 SM |
380 | } |
381 | } | |
382 | ||
383 | /* Now write out the hash table. */ | |
384 | std::unordered_map<c_str_view, offset_type, c_str_view_hasher> str_table; | |
385 | for (const auto &entry : symtab->data) | |
386 | { | |
387 | offset_type str_off, vec_off; | |
388 | ||
389 | if (entry.name != NULL) | |
390 | { | |
391 | const auto insertpair = str_table.emplace (entry.name, cpool.size ()); | |
392 | if (insertpair.second) | |
393 | cpool.append_cstr0 (entry.name); | |
394 | str_off = insertpair.first->second; | |
395 | vec_off = entry.index_offset; | |
396 | } | |
397 | else | |
398 | { | |
399 | /* While 0 is a valid constant pool index, it is not valid | |
400 | to have 0 for both offsets. */ | |
401 | str_off = 0; | |
402 | vec_off = 0; | |
403 | } | |
404 | ||
42c2c694 TT |
405 | output.append_offset (str_off); |
406 | output.append_offset (vec_off); | |
cd4fb1b2 SM |
407 | } |
408 | } | |
409 | ||
edfe0a0c | 410 | typedef std::unordered_map<partial_symtab *, unsigned int> psym_index_map; |
cd4fb1b2 SM |
411 | |
412 | /* Helper struct for building the address table. */ | |
413 | struct addrmap_index_data | |
414 | { | |
415 | addrmap_index_data (data_buf &addr_vec_, psym_index_map &cu_index_htab_) | |
416 | : addr_vec (addr_vec_), cu_index_htab (cu_index_htab_) | |
417 | {} | |
418 | ||
cd4fb1b2 SM |
419 | data_buf &addr_vec; |
420 | psym_index_map &cu_index_htab; | |
421 | ||
422 | /* Non-zero if the previous_* fields are valid. | |
423 | We can't write an entry until we see the next entry (since it is only then | |
424 | that we know the end of the entry). */ | |
425 | int previous_valid; | |
426 | /* Index of the CU in the table of all CUs in the index file. */ | |
427 | unsigned int previous_cu_index; | |
428 | /* Start address of the CU. */ | |
429 | CORE_ADDR previous_cu_start; | |
430 | }; | |
431 | ||
432 | /* Write an address entry to ADDR_VEC. */ | |
433 | ||
434 | static void | |
79cc99f6 | 435 | add_address_entry (data_buf &addr_vec, |
cd4fb1b2 SM |
436 | CORE_ADDR start, CORE_ADDR end, unsigned int cu_index) |
437 | { | |
79748972 TT |
438 | addr_vec.append_uint (8, BFD_ENDIAN_LITTLE, start); |
439 | addr_vec.append_uint (8, BFD_ENDIAN_LITTLE, end); | |
42c2c694 | 440 | addr_vec.append_offset (cu_index); |
cd4fb1b2 SM |
441 | } |
442 | ||
443 | /* Worker function for traversing an addrmap to build the address table. */ | |
444 | ||
445 | static int | |
446 | add_address_entry_worker (void *datap, CORE_ADDR start_addr, void *obj) | |
447 | { | |
448 | struct addrmap_index_data *data = (struct addrmap_index_data *) datap; | |
edfe0a0c | 449 | partial_symtab *pst = (partial_symtab *) obj; |
cd4fb1b2 SM |
450 | |
451 | if (data->previous_valid) | |
79cc99f6 | 452 | add_address_entry (data->addr_vec, |
cd4fb1b2 SM |
453 | data->previous_cu_start, start_addr, |
454 | data->previous_cu_index); | |
455 | ||
456 | data->previous_cu_start = start_addr; | |
457 | if (pst != NULL) | |
458 | { | |
459 | const auto it = data->cu_index_htab.find (pst); | |
460 | gdb_assert (it != data->cu_index_htab.cend ()); | |
461 | data->previous_cu_index = it->second; | |
462 | data->previous_valid = 1; | |
463 | } | |
464 | else | |
465 | data->previous_valid = 0; | |
466 | ||
467 | return 0; | |
468 | } | |
469 | ||
79cc99f6 | 470 | /* Write PER_BFD's address map to ADDR_VEC. |
cd4fb1b2 SM |
471 | CU_INDEX_HTAB is used to map addrmap entries to their CU indices |
472 | in the index file. */ | |
473 | ||
474 | static void | |
79cc99f6 | 475 | write_address_map (dwarf2_per_bfd *per_bfd, data_buf &addr_vec, |
cd4fb1b2 SM |
476 | psym_index_map &cu_index_htab) |
477 | { | |
478 | struct addrmap_index_data addrmap_index_data (addr_vec, cu_index_htab); | |
479 | ||
480 | /* When writing the address table, we have to cope with the fact that | |
481 | the addrmap iterator only provides the start of a region; we have to | |
482 | wait until the next invocation to get the start of the next region. */ | |
483 | ||
cd4fb1b2 SM |
484 | addrmap_index_data.previous_valid = 0; |
485 | ||
79cc99f6 | 486 | addrmap_foreach (per_bfd->partial_symtabs->psymtabs_addrmap, |
d320c2b5 | 487 | add_address_entry_worker, &addrmap_index_data); |
cd4fb1b2 SM |
488 | |
489 | /* It's highly unlikely the last entry (end address = 0xff...ff) | |
490 | is valid, but we should still handle it. | |
491 | The end address is recorded as the start of the next region, but that | |
492 | doesn't work here. To cope we pass 0xff...ff, this is a rare situation | |
493 | anyway. */ | |
494 | if (addrmap_index_data.previous_valid) | |
79cc99f6 | 495 | add_address_entry (addr_vec, |
cd4fb1b2 SM |
496 | addrmap_index_data.previous_cu_start, (CORE_ADDR) -1, |
497 | addrmap_index_data.previous_cu_index); | |
498 | } | |
499 | ||
500 | /* Return the symbol kind of PSYM. */ | |
501 | ||
502 | static gdb_index_symbol_kind | |
503 | symbol_kind (struct partial_symbol *psym) | |
504 | { | |
8a6d4234 TT |
505 | domain_enum domain = psym->domain; |
506 | enum address_class aclass = psym->aclass; | |
cd4fb1b2 SM |
507 | |
508 | switch (domain) | |
509 | { | |
510 | case VAR_DOMAIN: | |
511 | switch (aclass) | |
512 | { | |
513 | case LOC_BLOCK: | |
514 | return GDB_INDEX_SYMBOL_KIND_FUNCTION; | |
515 | case LOC_TYPEDEF: | |
516 | return GDB_INDEX_SYMBOL_KIND_TYPE; | |
517 | case LOC_COMPUTED: | |
518 | case LOC_CONST_BYTES: | |
519 | case LOC_OPTIMIZED_OUT: | |
520 | case LOC_STATIC: | |
521 | return GDB_INDEX_SYMBOL_KIND_VARIABLE; | |
522 | case LOC_CONST: | |
523 | /* Note: It's currently impossible to recognize psyms as enum values | |
524 | short of reading the type info. For now punt. */ | |
525 | return GDB_INDEX_SYMBOL_KIND_VARIABLE; | |
526 | default: | |
527 | /* There are other LOC_FOO values that one might want to classify | |
528 | as variables, but dwarf2read.c doesn't currently use them. */ | |
529 | return GDB_INDEX_SYMBOL_KIND_OTHER; | |
530 | } | |
531 | case STRUCT_DOMAIN: | |
532 | return GDB_INDEX_SYMBOL_KIND_TYPE; | |
533 | default: | |
534 | return GDB_INDEX_SYMBOL_KIND_OTHER; | |
535 | } | |
536 | } | |
537 | ||
538 | /* Add a list of partial symbols to SYMTAB. */ | |
539 | ||
540 | static void | |
541 | write_psymbols (struct mapped_symtab *symtab, | |
542 | std::unordered_set<partial_symbol *> &psyms_seen, | |
932539d7 | 543 | const std::vector<partial_symbol *> &symbols, |
cd4fb1b2 SM |
544 | offset_type cu_index, |
545 | int is_static) | |
546 | { | |
932539d7 | 547 | for (partial_symbol *psym : symbols) |
cd4fb1b2 | 548 | { |
7ab96794 | 549 | const char *name = psym->ginfo.search_name (); |
cd4fb1b2 | 550 | |
c1b5c1eb | 551 | if (psym->ginfo.language () == language_ada) |
7ab96794 TV |
552 | { |
553 | /* We want to ensure that the Ada main function's name appears | |
554 | verbatim in the index. However, this name will be of the | |
555 | form "_ada_mumble", and will be rewritten by ada_decode. | |
556 | So, recognize it specially here and add it to the index by | |
557 | hand. */ | |
558 | if (strcmp (main_name (), name) == 0) | |
559 | { | |
560 | gdb_index_symbol_kind kind = symbol_kind (psym); | |
561 | ||
562 | add_index_entry (symtab, name, is_static, kind, cu_index); | |
563 | } | |
564 | ||
565 | /* In order for the index to work when read back into gdb, it | |
566 | has to supply a funny form of the name: it should be the | |
567 | encoded name, with any suffixes stripped. Using the | |
568 | ordinary encoded name will not work properly with the | |
569 | searching logic in find_name_components_bounds; nor will | |
570 | using the decoded name. Furthermore, an Ada "verbatim" | |
571 | name (of the form "<MumBle>") must be entered without the | |
572 | angle brackets. Note that the current index is unusual, | |
573 | see PR symtab/24820 for details. */ | |
574 | std::string decoded = ada_decode (name); | |
575 | if (decoded[0] == '<') | |
576 | name = (char *) obstack_copy0 (&symtab->m_string_obstack, | |
577 | decoded.c_str () + 1, | |
578 | decoded.length () - 2); | |
579 | else | |
580 | name = obstack_strdup (&symtab->m_string_obstack, | |
581 | ada_encode (decoded.c_str ())); | |
582 | } | |
cd4fb1b2 SM |
583 | |
584 | /* Only add a given psymbol once. */ | |
585 | if (psyms_seen.insert (psym).second) | |
586 | { | |
587 | gdb_index_symbol_kind kind = symbol_kind (psym); | |
588 | ||
7ab96794 | 589 | add_index_entry (symtab, name, is_static, kind, cu_index); |
cd4fb1b2 SM |
590 | } |
591 | } | |
592 | } | |
593 | ||
594 | /* A helper struct used when iterating over debug_types. */ | |
595 | struct signatured_type_index_data | |
596 | { | |
597 | signatured_type_index_data (data_buf &types_list_, | |
dda83cd7 | 598 | std::unordered_set<partial_symbol *> &psyms_seen_) |
cd4fb1b2 SM |
599 | : types_list (types_list_), psyms_seen (psyms_seen_) |
600 | {} | |
601 | ||
602 | struct objfile *objfile; | |
603 | struct mapped_symtab *symtab; | |
604 | data_buf &types_list; | |
605 | std::unordered_set<partial_symbol *> &psyms_seen; | |
606 | int cu_index; | |
607 | }; | |
608 | ||
609 | /* A helper function that writes a single signatured_type to an | |
610 | obstack. */ | |
611 | ||
612 | static int | |
613 | write_one_signatured_type (void **slot, void *d) | |
614 | { | |
615 | struct signatured_type_index_data *info | |
616 | = (struct signatured_type_index_data *) d; | |
617 | struct signatured_type *entry = (struct signatured_type *) *slot; | |
edfe0a0c | 618 | partial_symtab *psymtab = entry->per_cu.v.psymtab; |
cd4fb1b2 | 619 | |
2bd3e4b8 TV |
620 | if (psymtab == nullptr) |
621 | { | |
622 | /* We can end up here when processing a skeleton CU referring to a | |
623 | .dwo file that hasn't been found. There's not much we can do in | |
624 | such a case, so skip this CU. */ | |
625 | return 1; | |
626 | } | |
627 | ||
932539d7 TT |
628 | write_psymbols (info->symtab, info->psyms_seen, |
629 | psymtab->global_psymbols, info->cu_index, | |
cd4fb1b2 | 630 | 0); |
932539d7 TT |
631 | write_psymbols (info->symtab, info->psyms_seen, |
632 | psymtab->static_psymbols, info->cu_index, | |
cd4fb1b2 SM |
633 | 1); |
634 | ||
635 | info->types_list.append_uint (8, BFD_ENDIAN_LITTLE, | |
636 | to_underlying (entry->per_cu.sect_off)); | |
637 | info->types_list.append_uint (8, BFD_ENDIAN_LITTLE, | |
638 | to_underlying (entry->type_offset_in_tu)); | |
639 | info->types_list.append_uint (8, BFD_ENDIAN_LITTLE, entry->signature); | |
640 | ||
641 | ++info->cu_index; | |
642 | ||
643 | return 1; | |
644 | } | |
645 | ||
646 | /* Recurse into all "included" dependencies and count their symbols as | |
647 | if they appeared in this psymtab. */ | |
648 | ||
649 | static void | |
edfe0a0c | 650 | recursively_count_psymbols (partial_symtab *psymtab, |
cd4fb1b2 SM |
651 | size_t &psyms_seen) |
652 | { | |
653 | for (int i = 0; i < psymtab->number_of_dependencies; ++i) | |
654 | if (psymtab->dependencies[i]->user != NULL) | |
edfe0a0c | 655 | recursively_count_psymbols (psymtab->dependencies[i], |
cd4fb1b2 SM |
656 | psyms_seen); |
657 | ||
932539d7 TT |
658 | psyms_seen += psymtab->global_psymbols.size (); |
659 | psyms_seen += psymtab->static_psymbols.size (); | |
cd4fb1b2 SM |
660 | } |
661 | ||
662 | /* Recurse into all "included" dependencies and write their symbols as | |
663 | if they appeared in this psymtab. */ | |
664 | ||
665 | static void | |
666 | recursively_write_psymbols (struct objfile *objfile, | |
edfe0a0c | 667 | partial_symtab *psymtab, |
cd4fb1b2 SM |
668 | struct mapped_symtab *symtab, |
669 | std::unordered_set<partial_symbol *> &psyms_seen, | |
670 | offset_type cu_index) | |
671 | { | |
672 | int i; | |
673 | ||
674 | for (i = 0; i < psymtab->number_of_dependencies; ++i) | |
675 | if (psymtab->dependencies[i]->user != NULL) | |
891813be | 676 | recursively_write_psymbols (objfile, |
edfe0a0c | 677 | psymtab->dependencies[i], |
cd4fb1b2 SM |
678 | symtab, psyms_seen, cu_index); |
679 | ||
932539d7 TT |
680 | write_psymbols (symtab, psyms_seen, |
681 | psymtab->global_psymbols, cu_index, | |
cd4fb1b2 | 682 | 0); |
932539d7 TT |
683 | write_psymbols (symtab, psyms_seen, |
684 | psymtab->static_psymbols, cu_index, | |
cd4fb1b2 SM |
685 | 1); |
686 | } | |
687 | ||
688 | /* DWARF-5 .debug_names builder. */ | |
689 | class debug_names | |
690 | { | |
691 | public: | |
976ca316 | 692 | debug_names (dwarf2_per_objfile *per_objfile, bool is_dwarf64, |
cd4fb1b2 SM |
693 | bfd_endian dwarf5_byte_order) |
694 | : m_dwarf5_byte_order (dwarf5_byte_order), | |
695 | m_dwarf32 (dwarf5_byte_order), | |
696 | m_dwarf64 (dwarf5_byte_order), | |
697 | m_dwarf (is_dwarf64 | |
698 | ? static_cast<dwarf &> (m_dwarf64) | |
699 | : static_cast<dwarf &> (m_dwarf32)), | |
700 | m_name_table_string_offs (m_dwarf.name_table_string_offs), | |
701 | m_name_table_entry_offs (m_dwarf.name_table_entry_offs), | |
976ca316 | 702 | m_debugstrlookup (per_objfile) |
cd4fb1b2 SM |
703 | {} |
704 | ||
705 | int dwarf5_offset_size () const | |
706 | { | |
707 | const bool dwarf5_is_dwarf64 = &m_dwarf == &m_dwarf64; | |
708 | return dwarf5_is_dwarf64 ? 8 : 4; | |
709 | } | |
710 | ||
711 | /* Is this symbol from DW_TAG_compile_unit or DW_TAG_type_unit? */ | |
712 | enum class unit_kind { cu, tu }; | |
713 | ||
714 | /* Insert one symbol. */ | |
715 | void insert (const partial_symbol *psym, int cu_index, bool is_static, | |
716 | unit_kind kind) | |
717 | { | |
718 | const int dwarf_tag = psymbol_tag (psym); | |
719 | if (dwarf_tag == 0) | |
720 | return; | |
c9d95fa3 | 721 | const char *name = psym->ginfo.search_name (); |
3b00ef10 | 722 | |
c1b5c1eb | 723 | if (psym->ginfo.language () == language_ada) |
3b00ef10 TT |
724 | { |
725 | /* We want to ensure that the Ada main function's name appears | |
726 | verbatim in the index. However, this name will be of the | |
727 | form "_ada_mumble", and will be rewritten by ada_decode. | |
728 | So, recognize it specially here and add it to the index by | |
729 | hand. */ | |
730 | if (strcmp (main_name (), name) == 0) | |
731 | { | |
732 | const auto insertpair | |
733 | = m_name_to_value_set.emplace (c_str_view (name), | |
734 | std::set<symbol_value> ()); | |
735 | std::set<symbol_value> &value_set = insertpair.first->second; | |
736 | value_set.emplace (symbol_value (dwarf_tag, cu_index, is_static, | |
737 | kind)); | |
738 | } | |
739 | ||
740 | /* In order for the index to work when read back into gdb, it | |
741 | has to supply a funny form of the name: it should be the | |
742 | encoded name, with any suffixes stripped. Using the | |
743 | ordinary encoded name will not work properly with the | |
744 | searching logic in find_name_components_bounds; nor will | |
745 | using the decoded name. Furthermore, an Ada "verbatim" | |
746 | name (of the form "<MumBle>") must be entered without the | |
747 | angle brackets. Note that the current index is unusual, | |
748 | see PR symtab/24820 for details. */ | |
f945dedf | 749 | std::string decoded = ada_decode (name); |
3b00ef10 TT |
750 | if (decoded[0] == '<') |
751 | name = (char *) obstack_copy0 (&m_string_obstack, | |
f945dedf CB |
752 | decoded.c_str () + 1, |
753 | decoded.length () - 2); | |
3b00ef10 | 754 | else |
f945dedf CB |
755 | name = obstack_strdup (&m_string_obstack, |
756 | ada_encode (decoded.c_str ())); | |
3b00ef10 TT |
757 | } |
758 | ||
cd4fb1b2 SM |
759 | const auto insertpair |
760 | = m_name_to_value_set.emplace (c_str_view (name), | |
761 | std::set<symbol_value> ()); | |
762 | std::set<symbol_value> &value_set = insertpair.first->second; | |
763 | value_set.emplace (symbol_value (dwarf_tag, cu_index, is_static, kind)); | |
764 | } | |
765 | ||
766 | /* Build all the tables. All symbols must be already inserted. | |
767 | This function does not call file_write, caller has to do it | |
768 | afterwards. */ | |
769 | void build () | |
770 | { | |
771 | /* Verify the build method has not be called twice. */ | |
772 | gdb_assert (m_abbrev_table.empty ()); | |
773 | const size_t name_count = m_name_to_value_set.size (); | |
774 | m_bucket_table.resize | |
775 | (std::pow (2, std::ceil (std::log2 (name_count * 4 / 3)))); | |
776 | m_hash_table.reserve (name_count); | |
777 | m_name_table_string_offs.reserve (name_count); | |
778 | m_name_table_entry_offs.reserve (name_count); | |
779 | ||
780 | /* Map each hash of symbol to its name and value. */ | |
781 | struct hash_it_pair | |
782 | { | |
783 | uint32_t hash; | |
784 | decltype (m_name_to_value_set)::const_iterator it; | |
785 | }; | |
786 | std::vector<std::forward_list<hash_it_pair>> bucket_hash; | |
787 | bucket_hash.resize (m_bucket_table.size ()); | |
788 | for (decltype (m_name_to_value_set)::const_iterator it | |
789 | = m_name_to_value_set.cbegin (); | |
790 | it != m_name_to_value_set.cend (); | |
791 | ++it) | |
792 | { | |
793 | const char *const name = it->first.c_str (); | |
794 | const uint32_t hash = dwarf5_djb_hash (name); | |
795 | hash_it_pair hashitpair; | |
796 | hashitpair.hash = hash; | |
797 | hashitpair.it = it; | |
798 | auto &slot = bucket_hash[hash % bucket_hash.size()]; | |
799 | slot.push_front (std::move (hashitpair)); | |
800 | } | |
801 | for (size_t bucket_ix = 0; bucket_ix < bucket_hash.size (); ++bucket_ix) | |
802 | { | |
803 | const std::forward_list<hash_it_pair> &hashitlist | |
804 | = bucket_hash[bucket_ix]; | |
805 | if (hashitlist.empty ()) | |
806 | continue; | |
807 | uint32_t &bucket_slot = m_bucket_table[bucket_ix]; | |
808 | /* The hashes array is indexed starting at 1. */ | |
809 | store_unsigned_integer (reinterpret_cast<gdb_byte *> (&bucket_slot), | |
810 | sizeof (bucket_slot), m_dwarf5_byte_order, | |
811 | m_hash_table.size () + 1); | |
812 | for (const hash_it_pair &hashitpair : hashitlist) | |
813 | { | |
814 | m_hash_table.push_back (0); | |
815 | store_unsigned_integer (reinterpret_cast<gdb_byte *> | |
816 | (&m_hash_table.back ()), | |
817 | sizeof (m_hash_table.back ()), | |
818 | m_dwarf5_byte_order, hashitpair.hash); | |
819 | const c_str_view &name = hashitpair.it->first; | |
820 | const std::set<symbol_value> &value_set = hashitpair.it->second; | |
821 | m_name_table_string_offs.push_back_reorder | |
822 | (m_debugstrlookup.lookup (name.c_str ())); | |
823 | m_name_table_entry_offs.push_back_reorder (m_entry_pool.size ()); | |
824 | gdb_assert (!value_set.empty ()); | |
825 | for (const symbol_value &value : value_set) | |
826 | { | |
827 | int &idx = m_indexkey_to_idx[index_key (value.dwarf_tag, | |
828 | value.is_static, | |
829 | value.kind)]; | |
830 | if (idx == 0) | |
831 | { | |
832 | idx = m_idx_next++; | |
833 | m_abbrev_table.append_unsigned_leb128 (idx); | |
834 | m_abbrev_table.append_unsigned_leb128 (value.dwarf_tag); | |
835 | m_abbrev_table.append_unsigned_leb128 | |
836 | (value.kind == unit_kind::cu ? DW_IDX_compile_unit | |
837 | : DW_IDX_type_unit); | |
838 | m_abbrev_table.append_unsigned_leb128 (DW_FORM_udata); | |
839 | m_abbrev_table.append_unsigned_leb128 (value.is_static | |
840 | ? DW_IDX_GNU_internal | |
841 | : DW_IDX_GNU_external); | |
842 | m_abbrev_table.append_unsigned_leb128 (DW_FORM_flag_present); | |
843 | ||
844 | /* Terminate attributes list. */ | |
845 | m_abbrev_table.append_unsigned_leb128 (0); | |
846 | m_abbrev_table.append_unsigned_leb128 (0); | |
847 | } | |
848 | ||
849 | m_entry_pool.append_unsigned_leb128 (idx); | |
850 | m_entry_pool.append_unsigned_leb128 (value.cu_index); | |
851 | } | |
852 | ||
853 | /* Terminate the list of CUs. */ | |
854 | m_entry_pool.append_unsigned_leb128 (0); | |
855 | } | |
856 | } | |
857 | gdb_assert (m_hash_table.size () == name_count); | |
858 | ||
859 | /* Terminate tags list. */ | |
860 | m_abbrev_table.append_unsigned_leb128 (0); | |
861 | } | |
862 | ||
863 | /* Return .debug_names bucket count. This must be called only after | |
864 | calling the build method. */ | |
865 | uint32_t bucket_count () const | |
866 | { | |
867 | /* Verify the build method has been already called. */ | |
868 | gdb_assert (!m_abbrev_table.empty ()); | |
869 | const uint32_t retval = m_bucket_table.size (); | |
870 | ||
871 | /* Check for overflow. */ | |
872 | gdb_assert (retval == m_bucket_table.size ()); | |
873 | return retval; | |
874 | } | |
875 | ||
876 | /* Return .debug_names names count. This must be called only after | |
877 | calling the build method. */ | |
878 | uint32_t name_count () const | |
879 | { | |
880 | /* Verify the build method has been already called. */ | |
881 | gdb_assert (!m_abbrev_table.empty ()); | |
882 | const uint32_t retval = m_hash_table.size (); | |
883 | ||
884 | /* Check for overflow. */ | |
885 | gdb_assert (retval == m_hash_table.size ()); | |
886 | return retval; | |
887 | } | |
888 | ||
889 | /* Return number of bytes of .debug_names abbreviation table. This | |
890 | must be called only after calling the build method. */ | |
891 | uint32_t abbrev_table_bytes () const | |
892 | { | |
893 | gdb_assert (!m_abbrev_table.empty ()); | |
894 | return m_abbrev_table.size (); | |
895 | } | |
896 | ||
897 | /* Recurse into all "included" dependencies and store their symbols | |
898 | as if they appeared in this psymtab. */ | |
899 | void recursively_write_psymbols | |
900 | (struct objfile *objfile, | |
edfe0a0c | 901 | partial_symtab *psymtab, |
cd4fb1b2 SM |
902 | std::unordered_set<partial_symbol *> &psyms_seen, |
903 | int cu_index) | |
904 | { | |
905 | for (int i = 0; i < psymtab->number_of_dependencies; ++i) | |
906 | if (psymtab->dependencies[i]->user != NULL) | |
891813be | 907 | recursively_write_psymbols |
edfe0a0c | 908 | (objfile, psymtab->dependencies[i], psyms_seen, cu_index); |
cd4fb1b2 | 909 | |
932539d7 TT |
910 | write_psymbols (psyms_seen, psymtab->global_psymbols, |
911 | cu_index, false, unit_kind::cu); | |
912 | write_psymbols (psyms_seen, psymtab->static_psymbols, | |
913 | cu_index, true, unit_kind::cu); | |
cd4fb1b2 SM |
914 | } |
915 | ||
916 | /* Return number of bytes the .debug_names section will have. This | |
917 | must be called only after calling the build method. */ | |
918 | size_t bytes () const | |
919 | { | |
920 | /* Verify the build method has been already called. */ | |
921 | gdb_assert (!m_abbrev_table.empty ()); | |
922 | size_t expected_bytes = 0; | |
923 | expected_bytes += m_bucket_table.size () * sizeof (m_bucket_table[0]); | |
924 | expected_bytes += m_hash_table.size () * sizeof (m_hash_table[0]); | |
925 | expected_bytes += m_name_table_string_offs.bytes (); | |
926 | expected_bytes += m_name_table_entry_offs.bytes (); | |
927 | expected_bytes += m_abbrev_table.size (); | |
928 | expected_bytes += m_entry_pool.size (); | |
929 | return expected_bytes; | |
930 | } | |
931 | ||
932 | /* Write .debug_names to FILE_NAMES and .debug_str addition to | |
933 | FILE_STR. This must be called only after calling the build | |
934 | method. */ | |
935 | void file_write (FILE *file_names, FILE *file_str) const | |
936 | { | |
937 | /* Verify the build method has been already called. */ | |
938 | gdb_assert (!m_abbrev_table.empty ()); | |
939 | ::file_write (file_names, m_bucket_table); | |
940 | ::file_write (file_names, m_hash_table); | |
941 | m_name_table_string_offs.file_write (file_names); | |
942 | m_name_table_entry_offs.file_write (file_names); | |
943 | m_abbrev_table.file_write (file_names); | |
944 | m_entry_pool.file_write (file_names); | |
945 | m_debugstrlookup.file_write (file_str); | |
946 | } | |
947 | ||
948 | /* A helper user data for write_one_signatured_type. */ | |
949 | class write_one_signatured_type_data | |
950 | { | |
951 | public: | |
952 | write_one_signatured_type_data (debug_names &nametable_, | |
dda83cd7 | 953 | signatured_type_index_data &&info_) |
cd4fb1b2 SM |
954 | : nametable (nametable_), info (std::move (info_)) |
955 | {} | |
956 | debug_names &nametable; | |
957 | struct signatured_type_index_data info; | |
958 | }; | |
959 | ||
960 | /* A helper function to pass write_one_signatured_type to | |
961 | htab_traverse_noresize. */ | |
962 | static int | |
963 | write_one_signatured_type (void **slot, void *d) | |
964 | { | |
965 | write_one_signatured_type_data *data = (write_one_signatured_type_data *) d; | |
966 | struct signatured_type_index_data *info = &data->info; | |
967 | struct signatured_type *entry = (struct signatured_type *) *slot; | |
968 | ||
969 | data->nametable.write_one_signatured_type (entry, info); | |
970 | ||
971 | return 1; | |
972 | } | |
973 | ||
974 | private: | |
975 | ||
976 | /* Storage for symbol names mapping them to their .debug_str section | |
977 | offsets. */ | |
978 | class debug_str_lookup | |
979 | { | |
980 | public: | |
981 | ||
30baf67b | 982 | /* Object constructor to be called for current DWARF2_PER_OBJFILE. |
cd4fb1b2 | 983 | All .debug_str section strings are automatically stored. */ |
976ca316 SM |
984 | debug_str_lookup (dwarf2_per_objfile *per_objfile) |
985 | : m_abfd (per_objfile->objfile->obfd), | |
986 | m_per_objfile (per_objfile) | |
cd4fb1b2 | 987 | { |
976ca316 SM |
988 | per_objfile->per_bfd->str.read (per_objfile->objfile); |
989 | if (per_objfile->per_bfd->str.buffer == NULL) | |
cd4fb1b2 | 990 | return; |
976ca316 SM |
991 | for (const gdb_byte *data = per_objfile->per_bfd->str.buffer; |
992 | data < (per_objfile->per_bfd->str.buffer | |
993 | + per_objfile->per_bfd->str.size);) | |
cd4fb1b2 SM |
994 | { |
995 | const char *const s = reinterpret_cast<const char *> (data); | |
996 | const auto insertpair | |
997 | = m_str_table.emplace (c_str_view (s), | |
976ca316 | 998 | data - per_objfile->per_bfd->str.buffer); |
cd4fb1b2 | 999 | if (!insertpair.second) |
b98664d3 | 1000 | complaint (_("Duplicate string \"%s\" in " |
cd4fb1b2 SM |
1001 | ".debug_str section [in module %s]"), |
1002 | s, bfd_get_filename (m_abfd)); | |
1003 | data += strlen (s) + 1; | |
1004 | } | |
1005 | } | |
1006 | ||
1007 | /* Return offset of symbol name S in the .debug_str section. Add | |
1008 | such symbol to the section's end if it does not exist there | |
1009 | yet. */ | |
1010 | size_t lookup (const char *s) | |
1011 | { | |
1012 | const auto it = m_str_table.find (c_str_view (s)); | |
1013 | if (it != m_str_table.end ()) | |
1014 | return it->second; | |
976ca316 | 1015 | const size_t offset = (m_per_objfile->per_bfd->str.size |
cd4fb1b2 SM |
1016 | + m_str_add_buf.size ()); |
1017 | m_str_table.emplace (c_str_view (s), offset); | |
1018 | m_str_add_buf.append_cstr0 (s); | |
1019 | return offset; | |
1020 | } | |
1021 | ||
1022 | /* Append the end of the .debug_str section to FILE. */ | |
1023 | void file_write (FILE *file) const | |
1024 | { | |
1025 | m_str_add_buf.file_write (file); | |
1026 | } | |
1027 | ||
1028 | private: | |
1029 | std::unordered_map<c_str_view, size_t, c_str_view_hasher> m_str_table; | |
1030 | bfd *const m_abfd; | |
976ca316 | 1031 | dwarf2_per_objfile *m_per_objfile; |
cd4fb1b2 SM |
1032 | |
1033 | /* Data to add at the end of .debug_str for new needed symbol names. */ | |
1034 | data_buf m_str_add_buf; | |
1035 | }; | |
1036 | ||
1037 | /* Container to map used DWARF tags to their .debug_names abbreviation | |
1038 | tags. */ | |
1039 | class index_key | |
1040 | { | |
1041 | public: | |
1042 | index_key (int dwarf_tag_, bool is_static_, unit_kind kind_) | |
1043 | : dwarf_tag (dwarf_tag_), is_static (is_static_), kind (kind_) | |
1044 | { | |
1045 | } | |
1046 | ||
1047 | bool | |
1048 | operator== (const index_key &other) const | |
1049 | { | |
1050 | return (dwarf_tag == other.dwarf_tag && is_static == other.is_static | |
1051 | && kind == other.kind); | |
1052 | } | |
1053 | ||
1054 | const int dwarf_tag; | |
1055 | const bool is_static; | |
1056 | const unit_kind kind; | |
1057 | }; | |
1058 | ||
1059 | /* Provide std::unordered_map::hasher for index_key. */ | |
1060 | class index_key_hasher | |
1061 | { | |
1062 | public: | |
1063 | size_t | |
1064 | operator () (const index_key &key) const | |
1065 | { | |
1066 | return (std::hash<int>() (key.dwarf_tag) << 1) | key.is_static; | |
1067 | } | |
1068 | }; | |
1069 | ||
1070 | /* Parameters of one symbol entry. */ | |
1071 | class symbol_value | |
1072 | { | |
1073 | public: | |
1074 | const int dwarf_tag, cu_index; | |
1075 | const bool is_static; | |
1076 | const unit_kind kind; | |
1077 | ||
1078 | symbol_value (int dwarf_tag_, int cu_index_, bool is_static_, | |
1079 | unit_kind kind_) | |
1080 | : dwarf_tag (dwarf_tag_), cu_index (cu_index_), is_static (is_static_), | |
dda83cd7 | 1081 | kind (kind_) |
cd4fb1b2 SM |
1082 | {} |
1083 | ||
1084 | bool | |
1085 | operator< (const symbol_value &other) const | |
1086 | { | |
1087 | #define X(n) \ | |
1088 | do \ | |
1089 | { \ | |
1090 | if (n < other.n) \ | |
1091 | return true; \ | |
1092 | if (n > other.n) \ | |
1093 | return false; \ | |
1094 | } \ | |
1095 | while (0) | |
1096 | X (dwarf_tag); | |
1097 | X (is_static); | |
1098 | X (kind); | |
1099 | X (cu_index); | |
1100 | #undef X | |
1101 | return false; | |
1102 | } | |
1103 | }; | |
1104 | ||
1105 | /* Abstract base class to unify DWARF-32 and DWARF-64 name table | |
1106 | output. */ | |
1107 | class offset_vec | |
1108 | { | |
1109 | protected: | |
1110 | const bfd_endian dwarf5_byte_order; | |
1111 | public: | |
1112 | explicit offset_vec (bfd_endian dwarf5_byte_order_) | |
1113 | : dwarf5_byte_order (dwarf5_byte_order_) | |
1114 | {} | |
1115 | ||
1116 | /* Call std::vector::reserve for NELEM elements. */ | |
1117 | virtual void reserve (size_t nelem) = 0; | |
1118 | ||
1119 | /* Call std::vector::push_back with store_unsigned_integer byte | |
1120 | reordering for ELEM. */ | |
1121 | virtual void push_back_reorder (size_t elem) = 0; | |
1122 | ||
1123 | /* Return expected output size in bytes. */ | |
1124 | virtual size_t bytes () const = 0; | |
1125 | ||
1126 | /* Write name table to FILE. */ | |
1127 | virtual void file_write (FILE *file) const = 0; | |
1128 | }; | |
1129 | ||
1130 | /* Template to unify DWARF-32 and DWARF-64 output. */ | |
1131 | template<typename OffsetSize> | |
1132 | class offset_vec_tmpl : public offset_vec | |
1133 | { | |
1134 | public: | |
1135 | explicit offset_vec_tmpl (bfd_endian dwarf5_byte_order_) | |
1136 | : offset_vec (dwarf5_byte_order_) | |
1137 | {} | |
1138 | ||
1139 | /* Implement offset_vec::reserve. */ | |
1140 | void reserve (size_t nelem) override | |
1141 | { | |
1142 | m_vec.reserve (nelem); | |
1143 | } | |
1144 | ||
1145 | /* Implement offset_vec::push_back_reorder. */ | |
1146 | void push_back_reorder (size_t elem) override | |
1147 | { | |
1148 | m_vec.push_back (elem); | |
1149 | /* Check for overflow. */ | |
1150 | gdb_assert (m_vec.back () == elem); | |
1151 | store_unsigned_integer (reinterpret_cast<gdb_byte *> (&m_vec.back ()), | |
1152 | sizeof (m_vec.back ()), dwarf5_byte_order, elem); | |
1153 | } | |
1154 | ||
1155 | /* Implement offset_vec::bytes. */ | |
1156 | size_t bytes () const override | |
1157 | { | |
1158 | return m_vec.size () * sizeof (m_vec[0]); | |
1159 | } | |
1160 | ||
1161 | /* Implement offset_vec::file_write. */ | |
1162 | void file_write (FILE *file) const override | |
1163 | { | |
1164 | ::file_write (file, m_vec); | |
1165 | } | |
1166 | ||
1167 | private: | |
1168 | std::vector<OffsetSize> m_vec; | |
1169 | }; | |
1170 | ||
1171 | /* Base class to unify DWARF-32 and DWARF-64 .debug_names output | |
1172 | respecting name table width. */ | |
1173 | class dwarf | |
1174 | { | |
1175 | public: | |
1176 | offset_vec &name_table_string_offs, &name_table_entry_offs; | |
1177 | ||
1178 | dwarf (offset_vec &name_table_string_offs_, | |
1179 | offset_vec &name_table_entry_offs_) | |
1180 | : name_table_string_offs (name_table_string_offs_), | |
1181 | name_table_entry_offs (name_table_entry_offs_) | |
1182 | { | |
1183 | } | |
1184 | }; | |
1185 | ||
1186 | /* Template to unify DWARF-32 and DWARF-64 .debug_names output | |
1187 | respecting name table width. */ | |
1188 | template<typename OffsetSize> | |
1189 | class dwarf_tmpl : public dwarf | |
1190 | { | |
1191 | public: | |
1192 | explicit dwarf_tmpl (bfd_endian dwarf5_byte_order_) | |
1193 | : dwarf (m_name_table_string_offs, m_name_table_entry_offs), | |
1194 | m_name_table_string_offs (dwarf5_byte_order_), | |
1195 | m_name_table_entry_offs (dwarf5_byte_order_) | |
1196 | {} | |
1197 | ||
1198 | private: | |
1199 | offset_vec_tmpl<OffsetSize> m_name_table_string_offs; | |
1200 | offset_vec_tmpl<OffsetSize> m_name_table_entry_offs; | |
1201 | }; | |
1202 | ||
1203 | /* Try to reconstruct original DWARF tag for given partial_symbol. | |
1204 | This function is not DWARF-5 compliant but it is sufficient for | |
1205 | GDB as a DWARF-5 index consumer. */ | |
1206 | static int psymbol_tag (const struct partial_symbol *psym) | |
1207 | { | |
8a6d4234 TT |
1208 | domain_enum domain = psym->domain; |
1209 | enum address_class aclass = psym->aclass; | |
cd4fb1b2 SM |
1210 | |
1211 | switch (domain) | |
1212 | { | |
1213 | case VAR_DOMAIN: | |
1214 | switch (aclass) | |
1215 | { | |
1216 | case LOC_BLOCK: | |
1217 | return DW_TAG_subprogram; | |
1218 | case LOC_TYPEDEF: | |
1219 | return DW_TAG_typedef; | |
1220 | case LOC_COMPUTED: | |
1221 | case LOC_CONST_BYTES: | |
1222 | case LOC_OPTIMIZED_OUT: | |
1223 | case LOC_STATIC: | |
1224 | return DW_TAG_variable; | |
1225 | case LOC_CONST: | |
1226 | /* Note: It's currently impossible to recognize psyms as enum values | |
1227 | short of reading the type info. For now punt. */ | |
1228 | return DW_TAG_variable; | |
1229 | default: | |
1230 | /* There are other LOC_FOO values that one might want to classify | |
1231 | as variables, but dwarf2read.c doesn't currently use them. */ | |
1232 | return DW_TAG_variable; | |
1233 | } | |
1234 | case STRUCT_DOMAIN: | |
1235 | return DW_TAG_structure_type; | |
7666722f TV |
1236 | case MODULE_DOMAIN: |
1237 | return DW_TAG_module; | |
cd4fb1b2 SM |
1238 | default: |
1239 | return 0; | |
1240 | } | |
1241 | } | |
1242 | ||
1243 | /* Call insert for all partial symbols and mark them in PSYMS_SEEN. */ | |
1244 | void write_psymbols (std::unordered_set<partial_symbol *> &psyms_seen, | |
932539d7 TT |
1245 | const std::vector<partial_symbol *> &symbols, |
1246 | int cu_index, bool is_static, unit_kind kind) | |
cd4fb1b2 | 1247 | { |
932539d7 | 1248 | for (partial_symbol *psym : symbols) |
cd4fb1b2 | 1249 | { |
cd4fb1b2 SM |
1250 | /* Only add a given psymbol once. */ |
1251 | if (psyms_seen.insert (psym).second) | |
1252 | insert (psym, cu_index, is_static, kind); | |
1253 | } | |
1254 | } | |
1255 | ||
1256 | /* A helper function that writes a single signatured_type | |
1257 | to a debug_names. */ | |
1258 | void | |
1259 | write_one_signatured_type (struct signatured_type *entry, | |
1260 | struct signatured_type_index_data *info) | |
1261 | { | |
edfe0a0c | 1262 | partial_symtab *psymtab = entry->per_cu.v.psymtab; |
cd4fb1b2 | 1263 | |
932539d7 TT |
1264 | write_psymbols (info->psyms_seen, psymtab->global_psymbols, |
1265 | info->cu_index, false, unit_kind::tu); | |
1266 | write_psymbols (info->psyms_seen, psymtab->static_psymbols, | |
1267 | info->cu_index, true, unit_kind::tu); | |
cd4fb1b2 SM |
1268 | |
1269 | info->types_list.append_uint (dwarf5_offset_size (), m_dwarf5_byte_order, | |
1270 | to_underlying (entry->per_cu.sect_off)); | |
1271 | ||
1272 | ++info->cu_index; | |
1273 | } | |
1274 | ||
1275 | /* Store value of each symbol. */ | |
1276 | std::unordered_map<c_str_view, std::set<symbol_value>, c_str_view_hasher> | |
1277 | m_name_to_value_set; | |
1278 | ||
1279 | /* Tables of DWARF-5 .debug_names. They are in object file byte | |
1280 | order. */ | |
1281 | std::vector<uint32_t> m_bucket_table; | |
1282 | std::vector<uint32_t> m_hash_table; | |
1283 | ||
1284 | const bfd_endian m_dwarf5_byte_order; | |
1285 | dwarf_tmpl<uint32_t> m_dwarf32; | |
1286 | dwarf_tmpl<uint64_t> m_dwarf64; | |
1287 | dwarf &m_dwarf; | |
1288 | offset_vec &m_name_table_string_offs, &m_name_table_entry_offs; | |
1289 | debug_str_lookup m_debugstrlookup; | |
1290 | ||
1291 | /* Map each used .debug_names abbreviation tag parameter to its | |
1292 | index value. */ | |
1293 | std::unordered_map<index_key, int, index_key_hasher> m_indexkey_to_idx; | |
1294 | ||
1295 | /* Next unused .debug_names abbreviation tag for | |
1296 | m_indexkey_to_idx. */ | |
1297 | int m_idx_next = 1; | |
1298 | ||
1299 | /* .debug_names abbreviation table. */ | |
1300 | data_buf m_abbrev_table; | |
1301 | ||
1302 | /* .debug_names entry pool. */ | |
1303 | data_buf m_entry_pool; | |
3b00ef10 TT |
1304 | |
1305 | /* Temporary storage for Ada names. */ | |
1306 | auto_obstack m_string_obstack; | |
cd4fb1b2 SM |
1307 | }; |
1308 | ||
1309 | /* Return iff any of the needed offsets does not fit into 32-bit | |
1310 | .debug_names section. */ | |
1311 | ||
1312 | static bool | |
976ca316 | 1313 | check_dwarf64_offsets (dwarf2_per_objfile *per_objfile) |
cd4fb1b2 | 1314 | { |
976ca316 | 1315 | for (dwarf2_per_cu_data *per_cu : per_objfile->per_bfd->all_comp_units) |
cd4fb1b2 | 1316 | { |
b76e467d | 1317 | if (to_underlying (per_cu->sect_off) >= (static_cast<uint64_t> (1) << 32)) |
cd4fb1b2 SM |
1318 | return true; |
1319 | } | |
976ca316 | 1320 | for (const signatured_type *sigtype : per_objfile->per_bfd->all_type_units) |
cd4fb1b2 | 1321 | { |
b2bdb8cf | 1322 | const dwarf2_per_cu_data &per_cu = sigtype->per_cu; |
cd4fb1b2 SM |
1323 | |
1324 | if (to_underlying (per_cu.sect_off) >= (static_cast<uint64_t> (1) << 32)) | |
1325 | return true; | |
1326 | } | |
1327 | return false; | |
1328 | } | |
1329 | ||
1330 | /* The psyms_seen set is potentially going to be largish (~40k | |
1331 | elements when indexing a -g3 build of GDB itself). Estimate the | |
1332 | number of elements in order to avoid too many rehashes, which | |
1333 | require rebuilding buckets and thus many trips to | |
1334 | malloc/free. */ | |
1335 | ||
1336 | static size_t | |
976ca316 | 1337 | psyms_seen_size (dwarf2_per_objfile *per_objfile) |
cd4fb1b2 SM |
1338 | { |
1339 | size_t psyms_count = 0; | |
976ca316 | 1340 | for (dwarf2_per_cu_data *per_cu : per_objfile->per_bfd->all_comp_units) |
cd4fb1b2 | 1341 | { |
edfe0a0c | 1342 | partial_symtab *psymtab = per_cu->v.psymtab; |
cd4fb1b2 SM |
1343 | |
1344 | if (psymtab != NULL && psymtab->user == NULL) | |
1345 | recursively_count_psymbols (psymtab, psyms_count); | |
1346 | } | |
1347 | /* Generating an index for gdb itself shows a ratio of | |
1348 | TOTAL_SEEN_SYMS/UNIQUE_SYMS or ~5. 4 seems like a good bet. */ | |
1349 | return psyms_count / 4; | |
1350 | } | |
1351 | ||
c4973306 SM |
1352 | /* Assert that FILE's size is EXPECTED_SIZE. Assumes file's seek |
1353 | position is at the end of the file. */ | |
cd4fb1b2 | 1354 | |
c4973306 SM |
1355 | static void |
1356 | assert_file_size (FILE *file, size_t expected_size) | |
1357 | { | |
1358 | const auto file_size = ftell (file); | |
1359 | if (file_size == -1) | |
1360 | perror_with_name (("ftell")); | |
1361 | gdb_assert (file_size == expected_size); | |
1362 | } | |
1363 | ||
1364 | /* Write a gdb index file to OUT_FILE from all the sections passed as | |
1365 | arguments. */ | |
1366 | ||
1367 | static void | |
1368 | write_gdbindex_1 (FILE *out_file, | |
1369 | const data_buf &cu_list, | |
1370 | const data_buf &types_cu_list, | |
1371 | const data_buf &addr_vec, | |
1372 | const data_buf &symtab_vec, | |
1373 | const data_buf &constant_pool) | |
1374 | { | |
1375 | data_buf contents; | |
1376 | const offset_type size_of_header = 6 * sizeof (offset_type); | |
1377 | offset_type total_len = size_of_header; | |
1378 | ||
1379 | /* The version number. */ | |
42c2c694 | 1380 | contents.append_offset (8); |
c4973306 SM |
1381 | |
1382 | /* The offset of the CU list from the start of the file. */ | |
42c2c694 | 1383 | contents.append_offset (total_len); |
c4973306 SM |
1384 | total_len += cu_list.size (); |
1385 | ||
1386 | /* The offset of the types CU list from the start of the file. */ | |
42c2c694 | 1387 | contents.append_offset (total_len); |
c4973306 SM |
1388 | total_len += types_cu_list.size (); |
1389 | ||
1390 | /* The offset of the address table from the start of the file. */ | |
42c2c694 | 1391 | contents.append_offset (total_len); |
c4973306 SM |
1392 | total_len += addr_vec.size (); |
1393 | ||
1394 | /* The offset of the symbol table from the start of the file. */ | |
42c2c694 | 1395 | contents.append_offset (total_len); |
c4973306 SM |
1396 | total_len += symtab_vec.size (); |
1397 | ||
1398 | /* The offset of the constant pool from the start of the file. */ | |
42c2c694 | 1399 | contents.append_offset (total_len); |
c4973306 SM |
1400 | total_len += constant_pool.size (); |
1401 | ||
1402 | gdb_assert (contents.size () == size_of_header); | |
1403 | ||
1404 | contents.file_write (out_file); | |
1405 | cu_list.file_write (out_file); | |
1406 | types_cu_list.file_write (out_file); | |
1407 | addr_vec.file_write (out_file); | |
1408 | symtab_vec.file_write (out_file); | |
1409 | constant_pool.file_write (out_file); | |
1410 | ||
1411 | assert_file_size (out_file, total_len); | |
1412 | } | |
1413 | ||
1414 | /* Write contents of a .gdb_index section for OBJFILE into OUT_FILE. | |
1415 | If OBJFILE has an associated dwz file, write contents of a .gdb_index | |
1416 | section for that dwz file into DWZ_OUT_FILE. If OBJFILE does not have an | |
1417 | associated dwz file, DWZ_OUT_FILE must be NULL. */ | |
1418 | ||
1419 | static void | |
976ca316 | 1420 | write_gdbindex (dwarf2_per_objfile *per_objfile, FILE *out_file, |
c4973306 | 1421 | FILE *dwz_out_file) |
cd4fb1b2 | 1422 | { |
976ca316 | 1423 | struct objfile *objfile = per_objfile->objfile; |
cd4fb1b2 | 1424 | mapped_symtab symtab; |
c4973306 SM |
1425 | data_buf objfile_cu_list; |
1426 | data_buf dwz_cu_list; | |
cd4fb1b2 SM |
1427 | |
1428 | /* While we're scanning CU's create a table that maps a psymtab pointer | |
1429 | (which is what addrmap records) to its index (which is what is recorded | |
1430 | in the index file). This will later be needed to write the address | |
1431 | table. */ | |
1432 | psym_index_map cu_index_htab; | |
976ca316 | 1433 | cu_index_htab.reserve (per_objfile->per_bfd->all_comp_units.size ()); |
cd4fb1b2 SM |
1434 | |
1435 | /* The CU list is already sorted, so we don't need to do additional | |
1436 | work here. Also, the debug_types entries do not appear in | |
1437 | all_comp_units, but only in their own hash table. */ | |
1438 | ||
1439 | std::unordered_set<partial_symbol *> psyms_seen | |
976ca316 SM |
1440 | (psyms_seen_size (per_objfile)); |
1441 | for (int i = 0; i < per_objfile->per_bfd->all_comp_units.size (); ++i) | |
cd4fb1b2 | 1442 | { |
976ca316 | 1443 | dwarf2_per_cu_data *per_cu = per_objfile->per_bfd->all_comp_units[i]; |
edfe0a0c | 1444 | partial_symtab *psymtab = per_cu->v.psymtab; |
cd4fb1b2 | 1445 | |
efba5c23 TV |
1446 | if (psymtab != NULL) |
1447 | { | |
1448 | if (psymtab->user == NULL) | |
1449 | recursively_write_psymbols (objfile, psymtab, &symtab, | |
1450 | psyms_seen, i); | |
1451 | ||
1452 | const auto insertpair = cu_index_htab.emplace (psymtab, i); | |
1453 | gdb_assert (insertpair.second); | |
1454 | } | |
cd4fb1b2 | 1455 | |
c4973306 SM |
1456 | /* The all_comp_units list contains CUs read from the objfile as well as |
1457 | from the eventual dwz file. We need to place the entry in the | |
1458 | corresponding index. */ | |
1459 | data_buf &cu_list = per_cu->is_dwz ? dwz_cu_list : objfile_cu_list; | |
cd4fb1b2 SM |
1460 | cu_list.append_uint (8, BFD_ENDIAN_LITTLE, |
1461 | to_underlying (per_cu->sect_off)); | |
1462 | cu_list.append_uint (8, BFD_ENDIAN_LITTLE, per_cu->length); | |
1463 | } | |
1464 | ||
1465 | /* Dump the address map. */ | |
1466 | data_buf addr_vec; | |
79cc99f6 | 1467 | write_address_map (per_objfile->per_bfd, addr_vec, cu_index_htab); |
cd4fb1b2 SM |
1468 | |
1469 | /* Write out the .debug_type entries, if any. */ | |
1470 | data_buf types_cu_list; | |
976ca316 | 1471 | if (per_objfile->per_bfd->signatured_types) |
cd4fb1b2 SM |
1472 | { |
1473 | signatured_type_index_data sig_data (types_cu_list, | |
1474 | psyms_seen); | |
1475 | ||
1476 | sig_data.objfile = objfile; | |
1477 | sig_data.symtab = &symtab; | |
976ca316 SM |
1478 | sig_data.cu_index = per_objfile->per_bfd->all_comp_units.size (); |
1479 | htab_traverse_noresize (per_objfile->per_bfd->signatured_types.get (), | |
cd4fb1b2 SM |
1480 | write_one_signatured_type, &sig_data); |
1481 | } | |
1482 | ||
1483 | /* Now that we've processed all symbols we can shrink their cu_indices | |
1484 | lists. */ | |
1485 | uniquify_cu_indices (&symtab); | |
1486 | ||
1487 | data_buf symtab_vec, constant_pool; | |
1488 | write_hash_table (&symtab, symtab_vec, constant_pool); | |
1489 | ||
c4973306 SM |
1490 | write_gdbindex_1(out_file, objfile_cu_list, types_cu_list, addr_vec, |
1491 | symtab_vec, constant_pool); | |
cd4fb1b2 | 1492 | |
c4973306 SM |
1493 | if (dwz_out_file != NULL) |
1494 | write_gdbindex_1 (dwz_out_file, dwz_cu_list, {}, {}, {}, {}); | |
1495 | else | |
1496 | gdb_assert (dwz_cu_list.empty ()); | |
cd4fb1b2 SM |
1497 | } |
1498 | ||
1499 | /* DWARF-5 augmentation string for GDB's DW_IDX_GNU_* extension. */ | |
1500 | static const gdb_byte dwarf5_gdb_augmentation[] = { 'G', 'D', 'B', 0 }; | |
1501 | ||
1502 | /* Write a new .debug_names section for OBJFILE into OUT_FILE, write | |
1503 | needed addition to .debug_str section to OUT_FILE_STR. Return how | |
1504 | many bytes were expected to be written into OUT_FILE. */ | |
1505 | ||
c4973306 | 1506 | static void |
976ca316 | 1507 | write_debug_names (dwarf2_per_objfile *per_objfile, |
cd4fb1b2 SM |
1508 | FILE *out_file, FILE *out_file_str) |
1509 | { | |
976ca316 SM |
1510 | const bool dwarf5_is_dwarf64 = check_dwarf64_offsets (per_objfile); |
1511 | struct objfile *objfile = per_objfile->objfile; | |
cd4fb1b2 | 1512 | const enum bfd_endian dwarf5_byte_order |
08feed99 | 1513 | = gdbarch_byte_order (objfile->arch ()); |
cd4fb1b2 SM |
1514 | |
1515 | /* The CU list is already sorted, so we don't need to do additional | |
1516 | work here. Also, the debug_types entries do not appear in | |
1517 | all_comp_units, but only in their own hash table. */ | |
1518 | data_buf cu_list; | |
976ca316 | 1519 | debug_names nametable (per_objfile, dwarf5_is_dwarf64, dwarf5_byte_order); |
cd4fb1b2 | 1520 | std::unordered_set<partial_symbol *> |
976ca316 SM |
1521 | psyms_seen (psyms_seen_size (per_objfile)); |
1522 | for (int i = 0; i < per_objfile->per_bfd->all_comp_units.size (); ++i) | |
cd4fb1b2 | 1523 | { |
976ca316 | 1524 | const dwarf2_per_cu_data *per_cu = per_objfile->per_bfd->all_comp_units[i]; |
edfe0a0c | 1525 | partial_symtab *psymtab = per_cu->v.psymtab; |
cd4fb1b2 SM |
1526 | |
1527 | /* CU of a shared file from 'dwz -m' may be unused by this main | |
1528 | file. It may be referenced from a local scope but in such | |
1529 | case it does not need to be present in .debug_names. */ | |
1530 | if (psymtab == NULL) | |
1531 | continue; | |
1532 | ||
1533 | if (psymtab->user == NULL) | |
1534 | nametable.recursively_write_psymbols (objfile, psymtab, psyms_seen, i); | |
1535 | ||
1536 | cu_list.append_uint (nametable.dwarf5_offset_size (), dwarf5_byte_order, | |
1537 | to_underlying (per_cu->sect_off)); | |
1538 | } | |
1539 | ||
1540 | /* Write out the .debug_type entries, if any. */ | |
1541 | data_buf types_cu_list; | |
976ca316 | 1542 | if (per_objfile->per_bfd->signatured_types) |
cd4fb1b2 SM |
1543 | { |
1544 | debug_names::write_one_signatured_type_data sig_data (nametable, | |
1545 | signatured_type_index_data (types_cu_list, psyms_seen)); | |
1546 | ||
1547 | sig_data.info.objfile = objfile; | |
1548 | /* It is used only for gdb_index. */ | |
1549 | sig_data.info.symtab = nullptr; | |
1550 | sig_data.info.cu_index = 0; | |
976ca316 | 1551 | htab_traverse_noresize (per_objfile->per_bfd->signatured_types.get (), |
cd4fb1b2 SM |
1552 | debug_names::write_one_signatured_type, |
1553 | &sig_data); | |
1554 | } | |
1555 | ||
1556 | nametable.build (); | |
1557 | ||
1558 | /* No addr_vec - DWARF-5 uses .debug_aranges generated by GCC. */ | |
1559 | ||
1560 | const offset_type bytes_of_header | |
1561 | = ((dwarf5_is_dwarf64 ? 12 : 4) | |
1562 | + 2 + 2 + 7 * 4 | |
1563 | + sizeof (dwarf5_gdb_augmentation)); | |
1564 | size_t expected_bytes = 0; | |
1565 | expected_bytes += bytes_of_header; | |
1566 | expected_bytes += cu_list.size (); | |
1567 | expected_bytes += types_cu_list.size (); | |
1568 | expected_bytes += nametable.bytes (); | |
1569 | data_buf header; | |
1570 | ||
1571 | if (!dwarf5_is_dwarf64) | |
1572 | { | |
1573 | const uint64_t size64 = expected_bytes - 4; | |
1574 | gdb_assert (size64 < 0xfffffff0); | |
1575 | header.append_uint (4, dwarf5_byte_order, size64); | |
1576 | } | |
1577 | else | |
1578 | { | |
1579 | header.append_uint (4, dwarf5_byte_order, 0xffffffff); | |
1580 | header.append_uint (8, dwarf5_byte_order, expected_bytes - 12); | |
1581 | } | |
1582 | ||
1583 | /* The version number. */ | |
1584 | header.append_uint (2, dwarf5_byte_order, 5); | |
1585 | ||
1586 | /* Padding. */ | |
1587 | header.append_uint (2, dwarf5_byte_order, 0); | |
1588 | ||
1589 | /* comp_unit_count - The number of CUs in the CU list. */ | |
b76e467d | 1590 | header.append_uint (4, dwarf5_byte_order, |
976ca316 | 1591 | per_objfile->per_bfd->all_comp_units.size ()); |
cd4fb1b2 SM |
1592 | |
1593 | /* local_type_unit_count - The number of TUs in the local TU | |
1594 | list. */ | |
b2bdb8cf | 1595 | header.append_uint (4, dwarf5_byte_order, |
976ca316 | 1596 | per_objfile->per_bfd->all_type_units.size ()); |
cd4fb1b2 SM |
1597 | |
1598 | /* foreign_type_unit_count - The number of TUs in the foreign TU | |
1599 | list. */ | |
1600 | header.append_uint (4, dwarf5_byte_order, 0); | |
1601 | ||
1602 | /* bucket_count - The number of hash buckets in the hash lookup | |
1603 | table. */ | |
1604 | header.append_uint (4, dwarf5_byte_order, nametable.bucket_count ()); | |
1605 | ||
1606 | /* name_count - The number of unique names in the index. */ | |
1607 | header.append_uint (4, dwarf5_byte_order, nametable.name_count ()); | |
1608 | ||
1609 | /* abbrev_table_size - The size in bytes of the abbreviations | |
1610 | table. */ | |
1611 | header.append_uint (4, dwarf5_byte_order, nametable.abbrev_table_bytes ()); | |
1612 | ||
1613 | /* augmentation_string_size - The size in bytes of the augmentation | |
1614 | string. This value is rounded up to a multiple of 4. */ | |
1615 | static_assert (sizeof (dwarf5_gdb_augmentation) % 4 == 0, ""); | |
1616 | header.append_uint (4, dwarf5_byte_order, sizeof (dwarf5_gdb_augmentation)); | |
42c2c694 | 1617 | header.append_array (dwarf5_gdb_augmentation); |
cd4fb1b2 SM |
1618 | |
1619 | gdb_assert (header.size () == bytes_of_header); | |
1620 | ||
1621 | header.file_write (out_file); | |
1622 | cu_list.file_write (out_file); | |
1623 | types_cu_list.file_write (out_file); | |
1624 | nametable.file_write (out_file, out_file_str); | |
1625 | ||
c4973306 | 1626 | assert_file_size (out_file, expected_bytes); |
cd4fb1b2 SM |
1627 | } |
1628 | ||
c4973306 | 1629 | /* This represents an index file being written (work-in-progress). |
cd4fb1b2 | 1630 | |
c4973306 SM |
1631 | The data is initially written to a temporary file. When the finalize method |
1632 | is called, the file is closed and moved to its final location. | |
1633 | ||
1634 | On failure (if this object is being destroyed with having called finalize), | |
1635 | the temporary file is closed and deleted. */ | |
1636 | ||
1637 | struct index_wip_file | |
cd4fb1b2 | 1638 | { |
c4973306 SM |
1639 | index_wip_file (const char *dir, const char *basename, |
1640 | const char *suffix) | |
1641 | { | |
1642 | filename = (std::string (dir) + SLASH_STRING + basename | |
1643 | + suffix); | |
1644 | ||
1645 | filename_temp = make_temp_filename (filename); | |
1646 | ||
1647 | scoped_fd out_file_fd (gdb_mkostemp_cloexec (filename_temp.data (), | |
1648 | O_BINARY)); | |
1649 | if (out_file_fd.get () == -1) | |
1650 | perror_with_name (("mkstemp")); | |
1651 | ||
1652 | out_file = out_file_fd.to_file ("wb"); | |
1653 | ||
1654 | if (out_file == nullptr) | |
1655 | error (_("Can't open `%s' for writing"), filename_temp.data ()); | |
1656 | ||
1657 | unlink_file.emplace (filename_temp.data ()); | |
1658 | } | |
1659 | ||
1660 | void finalize () | |
1661 | { | |
1662 | /* We want to keep the file. */ | |
1663 | unlink_file->keep (); | |
1664 | ||
1665 | /* Close and move the str file in place. */ | |
1666 | unlink_file.reset (); | |
1667 | if (rename (filename_temp.data (), filename.c_str ()) != 0) | |
1668 | perror_with_name (("rename")); | |
1669 | } | |
1670 | ||
1671 | std::string filename; | |
1672 | gdb::char_vector filename_temp; | |
1673 | ||
1674 | /* Order matters here; we want FILE to be closed before | |
1675 | FILENAME_TEMP is unlinked, because on MS-Windows one cannot | |
1676 | delete a file that is still open. So, we wrap the unlinker in an | |
1677 | optional and emplace it once we know the file name. */ | |
1678 | gdb::optional<gdb::unlinker> unlink_file; | |
1679 | ||
1680 | gdb_file_up out_file; | |
1681 | }; | |
cd4fb1b2 | 1682 | |
87d6a7aa | 1683 | /* See dwarf-index-write.h. */ |
cd4fb1b2 | 1684 | |
87d6a7aa | 1685 | void |
976ca316 SM |
1686 | write_psymtabs_to_index (dwarf2_per_objfile *per_objfile, const char *dir, |
1687 | const char *basename, const char *dwz_basename, | |
cd4fb1b2 SM |
1688 | dw_index_kind index_kind) |
1689 | { | |
79cc99f6 | 1690 | dwarf2_per_bfd *per_bfd = per_objfile->per_bfd; |
976ca316 | 1691 | struct objfile *objfile = per_objfile->objfile; |
cd4fb1b2 | 1692 | |
976ca316 | 1693 | if (per_objfile->per_bfd->using_index) |
cd4fb1b2 SM |
1694 | error (_("Cannot use an index to create the index")); |
1695 | ||
976ca316 | 1696 | if (per_objfile->per_bfd->types.size () > 1) |
cd4fb1b2 SM |
1697 | error (_("Cannot make an index when the file has multiple .debug_types sections")); |
1698 | ||
da314dd3 TT |
1699 | if (per_bfd->partial_symtabs == nullptr |
1700 | || !per_bfd->partial_symtabs->psymtabs | |
79cc99f6 | 1701 | || !per_bfd->partial_symtabs->psymtabs_addrmap) |
cd4fb1b2 SM |
1702 | return; |
1703 | ||
1704 | struct stat st; | |
1705 | if (stat (objfile_name (objfile), &st) < 0) | |
1706 | perror_with_name (objfile_name (objfile)); | |
1707 | ||
c4973306 SM |
1708 | const char *index_suffix = (index_kind == dw_index_kind::DEBUG_NAMES |
1709 | ? INDEX5_SUFFIX : INDEX4_SUFFIX); | |
cd4fb1b2 | 1710 | |
c4973306 SM |
1711 | index_wip_file objfile_index_wip (dir, basename, index_suffix); |
1712 | gdb::optional<index_wip_file> dwz_index_wip; | |
cd4fb1b2 | 1713 | |
c4973306 SM |
1714 | if (dwz_basename != NULL) |
1715 | dwz_index_wip.emplace (dir, dwz_basename, index_suffix); | |
cd4fb1b2 SM |
1716 | |
1717 | if (index_kind == dw_index_kind::DEBUG_NAMES) | |
1718 | { | |
c4973306 SM |
1719 | index_wip_file str_wip_file (dir, basename, DEBUG_STR_SUFFIX); |
1720 | ||
976ca316 | 1721 | write_debug_names (per_objfile, objfile_index_wip.out_file.get (), |
c4973306 SM |
1722 | str_wip_file.out_file.get ()); |
1723 | ||
1724 | str_wip_file.finalize (); | |
cd4fb1b2 SM |
1725 | } |
1726 | else | |
976ca316 | 1727 | write_gdbindex (per_objfile, objfile_index_wip.out_file.get (), |
c4973306 SM |
1728 | (dwz_index_wip.has_value () |
1729 | ? dwz_index_wip->out_file.get () : NULL)); | |
cd4fb1b2 | 1730 | |
c4973306 | 1731 | objfile_index_wip.finalize (); |
87d6a7aa | 1732 | |
c4973306 SM |
1733 | if (dwz_index_wip.has_value ()) |
1734 | dwz_index_wip->finalize (); | |
cd4fb1b2 SM |
1735 | } |
1736 | ||
1737 | /* Implementation of the `save gdb-index' command. | |
1738 | ||
1739 | Note that the .gdb_index file format used by this command is | |
1740 | documented in the GDB manual. Any changes here must be documented | |
1741 | there. */ | |
1742 | ||
1743 | static void | |
1744 | save_gdb_index_command (const char *arg, int from_tty) | |
1745 | { | |
cd4fb1b2 SM |
1746 | const char dwarf5space[] = "-dwarf-5 "; |
1747 | dw_index_kind index_kind = dw_index_kind::GDB_INDEX; | |
1748 | ||
1749 | if (!arg) | |
1750 | arg = ""; | |
1751 | ||
1752 | arg = skip_spaces (arg); | |
1753 | if (strncmp (arg, dwarf5space, strlen (dwarf5space)) == 0) | |
1754 | { | |
1755 | index_kind = dw_index_kind::DEBUG_NAMES; | |
1756 | arg += strlen (dwarf5space); | |
1757 | arg = skip_spaces (arg); | |
1758 | } | |
1759 | ||
1760 | if (!*arg) | |
1761 | error (_("usage: save gdb-index [-dwarf-5] DIRECTORY")); | |
1762 | ||
2030c079 | 1763 | for (objfile *objfile : current_program_space->objfiles ()) |
aed57c53 TT |
1764 | { |
1765 | struct stat st; | |
cd4fb1b2 | 1766 | |
aed57c53 TT |
1767 | /* If the objfile does not correspond to an actual file, skip it. */ |
1768 | if (stat (objfile_name (objfile), &st) < 0) | |
1769 | continue; | |
cd4fb1b2 | 1770 | |
976ca316 | 1771 | dwarf2_per_objfile *per_objfile = get_dwarf2_per_objfile (objfile); |
cd4fb1b2 | 1772 | |
976ca316 | 1773 | if (per_objfile != NULL) |
aed57c53 | 1774 | { |
a70b8144 | 1775 | try |
aed57c53 TT |
1776 | { |
1777 | const char *basename = lbasename (objfile_name (objfile)); | |
976ca316 | 1778 | const dwz_file *dwz = dwarf2_get_dwz_file (per_objfile->per_bfd); |
c4973306 SM |
1779 | const char *dwz_basename = NULL; |
1780 | ||
1781 | if (dwz != NULL) | |
1782 | dwz_basename = lbasename (dwz->filename ()); | |
1783 | ||
976ca316 SM |
1784 | write_psymtabs_to_index (per_objfile, arg, basename, dwz_basename, |
1785 | index_kind); | |
aed57c53 | 1786 | } |
230d2906 | 1787 | catch (const gdb_exception_error &except) |
aed57c53 TT |
1788 | { |
1789 | exception_fprintf (gdb_stderr, except, | |
1790 | _("Error while writing index for `%s': "), | |
1791 | objfile_name (objfile)); | |
1792 | } | |
aed57c53 | 1793 | } |
cd4fb1b2 | 1794 | |
aed57c53 | 1795 | } |
cd4fb1b2 SM |
1796 | } |
1797 | ||
6c265988 | 1798 | void _initialize_dwarf_index_write (); |
cd4fb1b2 SM |
1799 | void |
1800 | _initialize_dwarf_index_write () | |
1801 | { | |
1802 | cmd_list_element *c = add_cmd ("gdb-index", class_files, | |
1803 | save_gdb_index_command, _("\ | |
1804 | Save a gdb-index file.\n\ | |
1805 | Usage: save gdb-index [-dwarf-5] DIRECTORY\n\ | |
1806 | \n\ | |
1807 | No options create one file with .gdb-index extension for pre-DWARF-5\n\ | |
1808 | compatible .gdb_index section. With -dwarf-5 creates two files with\n\ | |
1809 | extension .debug_names and .debug_str for DWARF-5 .debug_names section."), | |
1810 | &save_cmdlist); | |
1811 | set_cmd_completer (c, filename_completer); | |
1812 | } |