Remove most uses of ALL_OBJFILES
[deliverable/binutils-gdb.git] / gdb / dwarf-index-write.c
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 "addrmap.h"
23 #include "cli/cli-decode.h"
24 #include "common/byte-vector.h"
25 #include "common/filestuff.h"
26 #include "common/gdb_unlinker.h"
27 #include "common/pathstuff.h"
28 #include "common/scoped_fd.h"
29 #include "complaints.h"
30 #include "dwarf-index-common.h"
31 #include "dwarf2.h"
32 #include "dwarf2read.h"
33 #include "gdb/gdb-index.h"
34 #include "gdbcmd.h"
35 #include "objfiles.h"
36 #include "psympriv.h"
37
38 #include <algorithm>
39 #include <cmath>
40 #include <set>
41 #include <unordered_map>
42 #include <unordered_set>
43
44 /* Ensure only legit values are used. */
45 #define DW2_GDB_INDEX_SYMBOL_STATIC_SET_VALUE(cu_index, value) \
46 do { \
47 gdb_assert ((unsigned int) (value) <= 1); \
48 GDB_INDEX_SYMBOL_STATIC_SET_VALUE((cu_index), (value)); \
49 } while (0)
50
51 /* Ensure only legit values are used. */
52 #define DW2_GDB_INDEX_SYMBOL_KIND_SET_VALUE(cu_index, value) \
53 do { \
54 gdb_assert ((value) >= GDB_INDEX_SYMBOL_KIND_TYPE \
55 && (value) <= GDB_INDEX_SYMBOL_KIND_OTHER); \
56 GDB_INDEX_SYMBOL_KIND_SET_VALUE((cu_index), (value)); \
57 } while (0)
58
59 /* Ensure we don't use more than the alloted nuber of bits for the CU. */
60 #define DW2_GDB_INDEX_CU_SET_VALUE(cu_index, value) \
61 do { \
62 gdb_assert (((value) & ~GDB_INDEX_CU_MASK) == 0); \
63 GDB_INDEX_CU_SET_VALUE((cu_index), (value)); \
64 } while (0)
65
66 /* The "save gdb-index" command. */
67
68 /* Write SIZE bytes from the buffer pointed to by DATA to FILE, with
69 error checking. */
70
71 static void
72 file_write (FILE *file, const void *data, size_t size)
73 {
74 if (fwrite (data, 1, size, file) != size)
75 error (_("couldn't data write to file"));
76 }
77
78 /* Write the contents of VEC to FILE, with error checking. */
79
80 template<typename Elem, typename Alloc>
81 static void
82 file_write (FILE *file, const std::vector<Elem, Alloc> &vec)
83 {
84 if (!vec.empty ())
85 file_write (file, vec.data (), vec.size () * sizeof (vec[0]));
86 }
87
88 /* In-memory buffer to prepare data to be written later to a file. */
89 class data_buf
90 {
91 public:
92 /* Copy DATA to the end of the buffer. */
93 template<typename T>
94 void append_data (const T &data)
95 {
96 std::copy (reinterpret_cast<const gdb_byte *> (&data),
97 reinterpret_cast<const gdb_byte *> (&data + 1),
98 grow (sizeof (data)));
99 }
100
101 /* Copy CSTR (a zero-terminated string) to the end of buffer. The
102 terminating zero is appended too. */
103 void append_cstr0 (const char *cstr)
104 {
105 const size_t size = strlen (cstr) + 1;
106 std::copy (cstr, cstr + size, grow (size));
107 }
108
109 /* Store INPUT as ULEB128 to the end of buffer. */
110 void append_unsigned_leb128 (ULONGEST input)
111 {
112 for (;;)
113 {
114 gdb_byte output = input & 0x7f;
115 input >>= 7;
116 if (input)
117 output |= 0x80;
118 append_data (output);
119 if (input == 0)
120 break;
121 }
122 }
123
124 /* Accept a host-format integer in VAL and append it to the buffer
125 as a target-format integer which is LEN bytes long. */
126 void append_uint (size_t len, bfd_endian byte_order, ULONGEST val)
127 {
128 ::store_unsigned_integer (grow (len), len, byte_order, val);
129 }
130
131 /* Return the size of the buffer. */
132 size_t size () const
133 {
134 return m_vec.size ();
135 }
136
137 /* Return true iff the buffer is empty. */
138 bool empty () const
139 {
140 return m_vec.empty ();
141 }
142
143 /* Write the buffer to FILE. */
144 void file_write (FILE *file) const
145 {
146 ::file_write (file, m_vec);
147 }
148
149 private:
150 /* Grow SIZE bytes at the end of the buffer. Returns a pointer to
151 the start of the new block. */
152 gdb_byte *grow (size_t size)
153 {
154 m_vec.resize (m_vec.size () + size);
155 return &*(m_vec.end () - size);
156 }
157
158 gdb::byte_vector m_vec;
159 };
160
161 /* An entry in the symbol table. */
162 struct symtab_index_entry
163 {
164 /* The name of the symbol. */
165 const char *name;
166 /* The offset of the name in the constant pool. */
167 offset_type index_offset;
168 /* A sorted vector of the indices of all the CUs that hold an object
169 of this name. */
170 std::vector<offset_type> cu_indices;
171 };
172
173 /* The symbol table. This is a power-of-2-sized hash table. */
174 struct mapped_symtab
175 {
176 mapped_symtab ()
177 {
178 data.resize (1024);
179 }
180
181 offset_type n_elements = 0;
182 std::vector<symtab_index_entry> data;
183 };
184
185 /* Find a slot in SYMTAB for the symbol NAME. Returns a reference to
186 the slot.
187
188 Function is used only during write_hash_table so no index format backward
189 compatibility is needed. */
190
191 static symtab_index_entry &
192 find_slot (struct mapped_symtab *symtab, const char *name)
193 {
194 offset_type index, step, hash = mapped_index_string_hash (INT_MAX, name);
195
196 index = hash & (symtab->data.size () - 1);
197 step = ((hash * 17) & (symtab->data.size () - 1)) | 1;
198
199 for (;;)
200 {
201 if (symtab->data[index].name == NULL
202 || strcmp (name, symtab->data[index].name) == 0)
203 return symtab->data[index];
204 index = (index + step) & (symtab->data.size () - 1);
205 }
206 }
207
208 /* Expand SYMTAB's hash table. */
209
210 static void
211 hash_expand (struct mapped_symtab *symtab)
212 {
213 auto old_entries = std::move (symtab->data);
214
215 symtab->data.clear ();
216 symtab->data.resize (old_entries.size () * 2);
217
218 for (auto &it : old_entries)
219 if (it.name != NULL)
220 {
221 auto &ref = find_slot (symtab, it.name);
222 ref = std::move (it);
223 }
224 }
225
226 /* Add an entry to SYMTAB. NAME is the name of the symbol.
227 CU_INDEX is the index of the CU in which the symbol appears.
228 IS_STATIC is one if the symbol is static, otherwise zero (global). */
229
230 static void
231 add_index_entry (struct mapped_symtab *symtab, const char *name,
232 int is_static, gdb_index_symbol_kind kind,
233 offset_type cu_index)
234 {
235 offset_type cu_index_and_attrs;
236
237 ++symtab->n_elements;
238 if (4 * symtab->n_elements / 3 >= symtab->data.size ())
239 hash_expand (symtab);
240
241 symtab_index_entry &slot = find_slot (symtab, name);
242 if (slot.name == NULL)
243 {
244 slot.name = name;
245 /* index_offset is set later. */
246 }
247
248 cu_index_and_attrs = 0;
249 DW2_GDB_INDEX_CU_SET_VALUE (cu_index_and_attrs, cu_index);
250 DW2_GDB_INDEX_SYMBOL_STATIC_SET_VALUE (cu_index_and_attrs, is_static);
251 DW2_GDB_INDEX_SYMBOL_KIND_SET_VALUE (cu_index_and_attrs, kind);
252
253 /* We don't want to record an index value twice as we want to avoid the
254 duplication.
255 We process all global symbols and then all static symbols
256 (which would allow us to avoid the duplication by only having to check
257 the last entry pushed), but a symbol could have multiple kinds in one CU.
258 To keep things simple we don't worry about the duplication here and
259 sort and uniqufy the list after we've processed all symbols. */
260 slot.cu_indices.push_back (cu_index_and_attrs);
261 }
262
263 /* Sort and remove duplicates of all symbols' cu_indices lists. */
264
265 static void
266 uniquify_cu_indices (struct mapped_symtab *symtab)
267 {
268 for (auto &entry : symtab->data)
269 {
270 if (entry.name != NULL && !entry.cu_indices.empty ())
271 {
272 auto &cu_indices = entry.cu_indices;
273 std::sort (cu_indices.begin (), cu_indices.end ());
274 auto from = std::unique (cu_indices.begin (), cu_indices.end ());
275 cu_indices.erase (from, cu_indices.end ());
276 }
277 }
278 }
279
280 /* A form of 'const char *' suitable for container keys. Only the
281 pointer is stored. The strings themselves are compared, not the
282 pointers. */
283 class c_str_view
284 {
285 public:
286 c_str_view (const char *cstr)
287 : m_cstr (cstr)
288 {}
289
290 bool operator== (const c_str_view &other) const
291 {
292 return strcmp (m_cstr, other.m_cstr) == 0;
293 }
294
295 /* Return the underlying C string. Note, the returned string is
296 only a reference with lifetime of this object. */
297 const char *c_str () const
298 {
299 return m_cstr;
300 }
301
302 private:
303 friend class c_str_view_hasher;
304 const char *const m_cstr;
305 };
306
307 /* A std::unordered_map::hasher for c_str_view that uses the right
308 hash function for strings in a mapped index. */
309 class c_str_view_hasher
310 {
311 public:
312 size_t operator () (const c_str_view &x) const
313 {
314 return mapped_index_string_hash (INT_MAX, x.m_cstr);
315 }
316 };
317
318 /* A std::unordered_map::hasher for std::vector<>. */
319 template<typename T>
320 class vector_hasher
321 {
322 public:
323 size_t operator () (const std::vector<T> &key) const
324 {
325 return iterative_hash (key.data (),
326 sizeof (key.front ()) * key.size (), 0);
327 }
328 };
329
330 /* Write the mapped hash table SYMTAB to the data buffer OUTPUT, with
331 constant pool entries going into the data buffer CPOOL. */
332
333 static void
334 write_hash_table (mapped_symtab *symtab, data_buf &output, data_buf &cpool)
335 {
336 {
337 /* Elements are sorted vectors of the indices of all the CUs that
338 hold an object of this name. */
339 std::unordered_map<std::vector<offset_type>, offset_type,
340 vector_hasher<offset_type>>
341 symbol_hash_table;
342
343 /* We add all the index vectors to the constant pool first, to
344 ensure alignment is ok. */
345 for (symtab_index_entry &entry : symtab->data)
346 {
347 if (entry.name == NULL)
348 continue;
349 gdb_assert (entry.index_offset == 0);
350
351 /* Finding before inserting is faster than always trying to
352 insert, because inserting always allocates a node, does the
353 lookup, and then destroys the new node if another node
354 already had the same key. C++17 try_emplace will avoid
355 this. */
356 const auto found
357 = symbol_hash_table.find (entry.cu_indices);
358 if (found != symbol_hash_table.end ())
359 {
360 entry.index_offset = found->second;
361 continue;
362 }
363
364 symbol_hash_table.emplace (entry.cu_indices, cpool.size ());
365 entry.index_offset = cpool.size ();
366 cpool.append_data (MAYBE_SWAP (entry.cu_indices.size ()));
367 for (const auto index : entry.cu_indices)
368 cpool.append_data (MAYBE_SWAP (index));
369 }
370 }
371
372 /* Now write out the hash table. */
373 std::unordered_map<c_str_view, offset_type, c_str_view_hasher> str_table;
374 for (const auto &entry : symtab->data)
375 {
376 offset_type str_off, vec_off;
377
378 if (entry.name != NULL)
379 {
380 const auto insertpair = str_table.emplace (entry.name, cpool.size ());
381 if (insertpair.second)
382 cpool.append_cstr0 (entry.name);
383 str_off = insertpair.first->second;
384 vec_off = entry.index_offset;
385 }
386 else
387 {
388 /* While 0 is a valid constant pool index, it is not valid
389 to have 0 for both offsets. */
390 str_off = 0;
391 vec_off = 0;
392 }
393
394 output.append_data (MAYBE_SWAP (str_off));
395 output.append_data (MAYBE_SWAP (vec_off));
396 }
397 }
398
399 typedef std::unordered_map<partial_symtab *, unsigned int> psym_index_map;
400
401 /* Helper struct for building the address table. */
402 struct addrmap_index_data
403 {
404 addrmap_index_data (data_buf &addr_vec_, psym_index_map &cu_index_htab_)
405 : addr_vec (addr_vec_), cu_index_htab (cu_index_htab_)
406 {}
407
408 struct objfile *objfile;
409 data_buf &addr_vec;
410 psym_index_map &cu_index_htab;
411
412 /* Non-zero if the previous_* fields are valid.
413 We can't write an entry until we see the next entry (since it is only then
414 that we know the end of the entry). */
415 int previous_valid;
416 /* Index of the CU in the table of all CUs in the index file. */
417 unsigned int previous_cu_index;
418 /* Start address of the CU. */
419 CORE_ADDR previous_cu_start;
420 };
421
422 /* Write an address entry to ADDR_VEC. */
423
424 static void
425 add_address_entry (struct objfile *objfile, data_buf &addr_vec,
426 CORE_ADDR start, CORE_ADDR end, unsigned int cu_index)
427 {
428 addr_vec.append_uint (8, BFD_ENDIAN_LITTLE, start);
429 addr_vec.append_uint (8, BFD_ENDIAN_LITTLE, end);
430 addr_vec.append_data (MAYBE_SWAP (cu_index));
431 }
432
433 /* Worker function for traversing an addrmap to build the address table. */
434
435 static int
436 add_address_entry_worker (void *datap, CORE_ADDR start_addr, void *obj)
437 {
438 struct addrmap_index_data *data = (struct addrmap_index_data *) datap;
439 struct partial_symtab *pst = (struct partial_symtab *) obj;
440
441 if (data->previous_valid)
442 add_address_entry (data->objfile, data->addr_vec,
443 data->previous_cu_start, start_addr,
444 data->previous_cu_index);
445
446 data->previous_cu_start = start_addr;
447 if (pst != NULL)
448 {
449 const auto it = data->cu_index_htab.find (pst);
450 gdb_assert (it != data->cu_index_htab.cend ());
451 data->previous_cu_index = it->second;
452 data->previous_valid = 1;
453 }
454 else
455 data->previous_valid = 0;
456
457 return 0;
458 }
459
460 /* Write OBJFILE's address map to ADDR_VEC.
461 CU_INDEX_HTAB is used to map addrmap entries to their CU indices
462 in the index file. */
463
464 static void
465 write_address_map (struct objfile *objfile, data_buf &addr_vec,
466 psym_index_map &cu_index_htab)
467 {
468 struct addrmap_index_data addrmap_index_data (addr_vec, cu_index_htab);
469
470 /* When writing the address table, we have to cope with the fact that
471 the addrmap iterator only provides the start of a region; we have to
472 wait until the next invocation to get the start of the next region. */
473
474 addrmap_index_data.objfile = objfile;
475 addrmap_index_data.previous_valid = 0;
476
477 addrmap_foreach (objfile->psymtabs_addrmap, add_address_entry_worker,
478 &addrmap_index_data);
479
480 /* It's highly unlikely the last entry (end address = 0xff...ff)
481 is valid, but we should still handle it.
482 The end address is recorded as the start of the next region, but that
483 doesn't work here. To cope we pass 0xff...ff, this is a rare situation
484 anyway. */
485 if (addrmap_index_data.previous_valid)
486 add_address_entry (objfile, addr_vec,
487 addrmap_index_data.previous_cu_start, (CORE_ADDR) -1,
488 addrmap_index_data.previous_cu_index);
489 }
490
491 /* Return the symbol kind of PSYM. */
492
493 static gdb_index_symbol_kind
494 symbol_kind (struct partial_symbol *psym)
495 {
496 domain_enum domain = psym->domain;
497 enum address_class aclass = psym->aclass;
498
499 switch (domain)
500 {
501 case VAR_DOMAIN:
502 switch (aclass)
503 {
504 case LOC_BLOCK:
505 return GDB_INDEX_SYMBOL_KIND_FUNCTION;
506 case LOC_TYPEDEF:
507 return GDB_INDEX_SYMBOL_KIND_TYPE;
508 case LOC_COMPUTED:
509 case LOC_CONST_BYTES:
510 case LOC_OPTIMIZED_OUT:
511 case LOC_STATIC:
512 return GDB_INDEX_SYMBOL_KIND_VARIABLE;
513 case LOC_CONST:
514 /* Note: It's currently impossible to recognize psyms as enum values
515 short of reading the type info. For now punt. */
516 return GDB_INDEX_SYMBOL_KIND_VARIABLE;
517 default:
518 /* There are other LOC_FOO values that one might want to classify
519 as variables, but dwarf2read.c doesn't currently use them. */
520 return GDB_INDEX_SYMBOL_KIND_OTHER;
521 }
522 case STRUCT_DOMAIN:
523 return GDB_INDEX_SYMBOL_KIND_TYPE;
524 default:
525 return GDB_INDEX_SYMBOL_KIND_OTHER;
526 }
527 }
528
529 /* Add a list of partial symbols to SYMTAB. */
530
531 static void
532 write_psymbols (struct mapped_symtab *symtab,
533 std::unordered_set<partial_symbol *> &psyms_seen,
534 struct partial_symbol **psymp,
535 int count,
536 offset_type cu_index,
537 int is_static)
538 {
539 for (; count-- > 0; ++psymp)
540 {
541 struct partial_symbol *psym = *psymp;
542
543 if (psym->language == language_ada)
544 error (_("Ada is not currently supported by the index"));
545
546 /* Only add a given psymbol once. */
547 if (psyms_seen.insert (psym).second)
548 {
549 gdb_index_symbol_kind kind = symbol_kind (psym);
550
551 add_index_entry (symtab, symbol_search_name (psym),
552 is_static, kind, cu_index);
553 }
554 }
555 }
556
557 /* A helper struct used when iterating over debug_types. */
558 struct signatured_type_index_data
559 {
560 signatured_type_index_data (data_buf &types_list_,
561 std::unordered_set<partial_symbol *> &psyms_seen_)
562 : types_list (types_list_), psyms_seen (psyms_seen_)
563 {}
564
565 struct objfile *objfile;
566 struct mapped_symtab *symtab;
567 data_buf &types_list;
568 std::unordered_set<partial_symbol *> &psyms_seen;
569 int cu_index;
570 };
571
572 /* A helper function that writes a single signatured_type to an
573 obstack. */
574
575 static int
576 write_one_signatured_type (void **slot, void *d)
577 {
578 struct signatured_type_index_data *info
579 = (struct signatured_type_index_data *) d;
580 struct signatured_type *entry = (struct signatured_type *) *slot;
581 struct partial_symtab *psymtab = entry->per_cu.v.psymtab;
582
583 write_psymbols (info->symtab,
584 info->psyms_seen,
585 (info->objfile->global_psymbols.data ()
586 + psymtab->globals_offset),
587 psymtab->n_global_syms, info->cu_index,
588 0);
589 write_psymbols (info->symtab,
590 info->psyms_seen,
591 (info->objfile->static_psymbols.data ()
592 + psymtab->statics_offset),
593 psymtab->n_static_syms, info->cu_index,
594 1);
595
596 info->types_list.append_uint (8, BFD_ENDIAN_LITTLE,
597 to_underlying (entry->per_cu.sect_off));
598 info->types_list.append_uint (8, BFD_ENDIAN_LITTLE,
599 to_underlying (entry->type_offset_in_tu));
600 info->types_list.append_uint (8, BFD_ENDIAN_LITTLE, entry->signature);
601
602 ++info->cu_index;
603
604 return 1;
605 }
606
607 /* Recurse into all "included" dependencies and count their symbols as
608 if they appeared in this psymtab. */
609
610 static void
611 recursively_count_psymbols (struct partial_symtab *psymtab,
612 size_t &psyms_seen)
613 {
614 for (int i = 0; i < psymtab->number_of_dependencies; ++i)
615 if (psymtab->dependencies[i]->user != NULL)
616 recursively_count_psymbols (psymtab->dependencies[i],
617 psyms_seen);
618
619 psyms_seen += psymtab->n_global_syms;
620 psyms_seen += psymtab->n_static_syms;
621 }
622
623 /* Recurse into all "included" dependencies and write their symbols as
624 if they appeared in this psymtab. */
625
626 static void
627 recursively_write_psymbols (struct objfile *objfile,
628 struct partial_symtab *psymtab,
629 struct mapped_symtab *symtab,
630 std::unordered_set<partial_symbol *> &psyms_seen,
631 offset_type cu_index)
632 {
633 int i;
634
635 for (i = 0; i < psymtab->number_of_dependencies; ++i)
636 if (psymtab->dependencies[i]->user != NULL)
637 recursively_write_psymbols (objfile, psymtab->dependencies[i],
638 symtab, psyms_seen, cu_index);
639
640 write_psymbols (symtab,
641 psyms_seen,
642 objfile->global_psymbols.data () + psymtab->globals_offset,
643 psymtab->n_global_syms, cu_index,
644 0);
645 write_psymbols (symtab,
646 psyms_seen,
647 objfile->static_psymbols.data () + psymtab->statics_offset,
648 psymtab->n_static_syms, cu_index,
649 1);
650 }
651
652 /* DWARF-5 .debug_names builder. */
653 class debug_names
654 {
655 public:
656 debug_names (struct dwarf2_per_objfile *dwarf2_per_objfile, bool is_dwarf64,
657 bfd_endian dwarf5_byte_order)
658 : m_dwarf5_byte_order (dwarf5_byte_order),
659 m_dwarf32 (dwarf5_byte_order),
660 m_dwarf64 (dwarf5_byte_order),
661 m_dwarf (is_dwarf64
662 ? static_cast<dwarf &> (m_dwarf64)
663 : static_cast<dwarf &> (m_dwarf32)),
664 m_name_table_string_offs (m_dwarf.name_table_string_offs),
665 m_name_table_entry_offs (m_dwarf.name_table_entry_offs),
666 m_debugstrlookup (dwarf2_per_objfile)
667 {}
668
669 int dwarf5_offset_size () const
670 {
671 const bool dwarf5_is_dwarf64 = &m_dwarf == &m_dwarf64;
672 return dwarf5_is_dwarf64 ? 8 : 4;
673 }
674
675 /* Is this symbol from DW_TAG_compile_unit or DW_TAG_type_unit? */
676 enum class unit_kind { cu, tu };
677
678 /* Insert one symbol. */
679 void insert (const partial_symbol *psym, int cu_index, bool is_static,
680 unit_kind kind)
681 {
682 const int dwarf_tag = psymbol_tag (psym);
683 if (dwarf_tag == 0)
684 return;
685 const char *const name = symbol_search_name (psym);
686 const auto insertpair
687 = m_name_to_value_set.emplace (c_str_view (name),
688 std::set<symbol_value> ());
689 std::set<symbol_value> &value_set = insertpair.first->second;
690 value_set.emplace (symbol_value (dwarf_tag, cu_index, is_static, kind));
691 }
692
693 /* Build all the tables. All symbols must be already inserted.
694 This function does not call file_write, caller has to do it
695 afterwards. */
696 void build ()
697 {
698 /* Verify the build method has not be called twice. */
699 gdb_assert (m_abbrev_table.empty ());
700 const size_t name_count = m_name_to_value_set.size ();
701 m_bucket_table.resize
702 (std::pow (2, std::ceil (std::log2 (name_count * 4 / 3))));
703 m_hash_table.reserve (name_count);
704 m_name_table_string_offs.reserve (name_count);
705 m_name_table_entry_offs.reserve (name_count);
706
707 /* Map each hash of symbol to its name and value. */
708 struct hash_it_pair
709 {
710 uint32_t hash;
711 decltype (m_name_to_value_set)::const_iterator it;
712 };
713 std::vector<std::forward_list<hash_it_pair>> bucket_hash;
714 bucket_hash.resize (m_bucket_table.size ());
715 for (decltype (m_name_to_value_set)::const_iterator it
716 = m_name_to_value_set.cbegin ();
717 it != m_name_to_value_set.cend ();
718 ++it)
719 {
720 const char *const name = it->first.c_str ();
721 const uint32_t hash = dwarf5_djb_hash (name);
722 hash_it_pair hashitpair;
723 hashitpair.hash = hash;
724 hashitpair.it = it;
725 auto &slot = bucket_hash[hash % bucket_hash.size()];
726 slot.push_front (std::move (hashitpair));
727 }
728 for (size_t bucket_ix = 0; bucket_ix < bucket_hash.size (); ++bucket_ix)
729 {
730 const std::forward_list<hash_it_pair> &hashitlist
731 = bucket_hash[bucket_ix];
732 if (hashitlist.empty ())
733 continue;
734 uint32_t &bucket_slot = m_bucket_table[bucket_ix];
735 /* The hashes array is indexed starting at 1. */
736 store_unsigned_integer (reinterpret_cast<gdb_byte *> (&bucket_slot),
737 sizeof (bucket_slot), m_dwarf5_byte_order,
738 m_hash_table.size () + 1);
739 for (const hash_it_pair &hashitpair : hashitlist)
740 {
741 m_hash_table.push_back (0);
742 store_unsigned_integer (reinterpret_cast<gdb_byte *>
743 (&m_hash_table.back ()),
744 sizeof (m_hash_table.back ()),
745 m_dwarf5_byte_order, hashitpair.hash);
746 const c_str_view &name = hashitpair.it->first;
747 const std::set<symbol_value> &value_set = hashitpair.it->second;
748 m_name_table_string_offs.push_back_reorder
749 (m_debugstrlookup.lookup (name.c_str ()));
750 m_name_table_entry_offs.push_back_reorder (m_entry_pool.size ());
751 gdb_assert (!value_set.empty ());
752 for (const symbol_value &value : value_set)
753 {
754 int &idx = m_indexkey_to_idx[index_key (value.dwarf_tag,
755 value.is_static,
756 value.kind)];
757 if (idx == 0)
758 {
759 idx = m_idx_next++;
760 m_abbrev_table.append_unsigned_leb128 (idx);
761 m_abbrev_table.append_unsigned_leb128 (value.dwarf_tag);
762 m_abbrev_table.append_unsigned_leb128
763 (value.kind == unit_kind::cu ? DW_IDX_compile_unit
764 : DW_IDX_type_unit);
765 m_abbrev_table.append_unsigned_leb128 (DW_FORM_udata);
766 m_abbrev_table.append_unsigned_leb128 (value.is_static
767 ? DW_IDX_GNU_internal
768 : DW_IDX_GNU_external);
769 m_abbrev_table.append_unsigned_leb128 (DW_FORM_flag_present);
770
771 /* Terminate attributes list. */
772 m_abbrev_table.append_unsigned_leb128 (0);
773 m_abbrev_table.append_unsigned_leb128 (0);
774 }
775
776 m_entry_pool.append_unsigned_leb128 (idx);
777 m_entry_pool.append_unsigned_leb128 (value.cu_index);
778 }
779
780 /* Terminate the list of CUs. */
781 m_entry_pool.append_unsigned_leb128 (0);
782 }
783 }
784 gdb_assert (m_hash_table.size () == name_count);
785
786 /* Terminate tags list. */
787 m_abbrev_table.append_unsigned_leb128 (0);
788 }
789
790 /* Return .debug_names bucket count. This must be called only after
791 calling the build method. */
792 uint32_t bucket_count () const
793 {
794 /* Verify the build method has been already called. */
795 gdb_assert (!m_abbrev_table.empty ());
796 const uint32_t retval = m_bucket_table.size ();
797
798 /* Check for overflow. */
799 gdb_assert (retval == m_bucket_table.size ());
800 return retval;
801 }
802
803 /* Return .debug_names names count. This must be called only after
804 calling the build method. */
805 uint32_t name_count () const
806 {
807 /* Verify the build method has been already called. */
808 gdb_assert (!m_abbrev_table.empty ());
809 const uint32_t retval = m_hash_table.size ();
810
811 /* Check for overflow. */
812 gdb_assert (retval == m_hash_table.size ());
813 return retval;
814 }
815
816 /* Return number of bytes of .debug_names abbreviation table. This
817 must be called only after calling the build method. */
818 uint32_t abbrev_table_bytes () const
819 {
820 gdb_assert (!m_abbrev_table.empty ());
821 return m_abbrev_table.size ();
822 }
823
824 /* Recurse into all "included" dependencies and store their symbols
825 as if they appeared in this psymtab. */
826 void recursively_write_psymbols
827 (struct objfile *objfile,
828 struct partial_symtab *psymtab,
829 std::unordered_set<partial_symbol *> &psyms_seen,
830 int cu_index)
831 {
832 for (int i = 0; i < psymtab->number_of_dependencies; ++i)
833 if (psymtab->dependencies[i]->user != NULL)
834 recursively_write_psymbols (objfile, psymtab->dependencies[i],
835 psyms_seen, cu_index);
836
837 write_psymbols (psyms_seen,
838 objfile->global_psymbols.data () + psymtab->globals_offset,
839 psymtab->n_global_syms, cu_index, false, unit_kind::cu);
840 write_psymbols (psyms_seen,
841 objfile->static_psymbols.data () + psymtab->statics_offset,
842 psymtab->n_static_syms, cu_index, true, unit_kind::cu);
843 }
844
845 /* Return number of bytes the .debug_names section will have. This
846 must be called only after calling the build method. */
847 size_t bytes () const
848 {
849 /* Verify the build method has been already called. */
850 gdb_assert (!m_abbrev_table.empty ());
851 size_t expected_bytes = 0;
852 expected_bytes += m_bucket_table.size () * sizeof (m_bucket_table[0]);
853 expected_bytes += m_hash_table.size () * sizeof (m_hash_table[0]);
854 expected_bytes += m_name_table_string_offs.bytes ();
855 expected_bytes += m_name_table_entry_offs.bytes ();
856 expected_bytes += m_abbrev_table.size ();
857 expected_bytes += m_entry_pool.size ();
858 return expected_bytes;
859 }
860
861 /* Write .debug_names to FILE_NAMES and .debug_str addition to
862 FILE_STR. This must be called only after calling the build
863 method. */
864 void file_write (FILE *file_names, FILE *file_str) const
865 {
866 /* Verify the build method has been already called. */
867 gdb_assert (!m_abbrev_table.empty ());
868 ::file_write (file_names, m_bucket_table);
869 ::file_write (file_names, m_hash_table);
870 m_name_table_string_offs.file_write (file_names);
871 m_name_table_entry_offs.file_write (file_names);
872 m_abbrev_table.file_write (file_names);
873 m_entry_pool.file_write (file_names);
874 m_debugstrlookup.file_write (file_str);
875 }
876
877 /* A helper user data for write_one_signatured_type. */
878 class write_one_signatured_type_data
879 {
880 public:
881 write_one_signatured_type_data (debug_names &nametable_,
882 signatured_type_index_data &&info_)
883 : nametable (nametable_), info (std::move (info_))
884 {}
885 debug_names &nametable;
886 struct signatured_type_index_data info;
887 };
888
889 /* A helper function to pass write_one_signatured_type to
890 htab_traverse_noresize. */
891 static int
892 write_one_signatured_type (void **slot, void *d)
893 {
894 write_one_signatured_type_data *data = (write_one_signatured_type_data *) d;
895 struct signatured_type_index_data *info = &data->info;
896 struct signatured_type *entry = (struct signatured_type *) *slot;
897
898 data->nametable.write_one_signatured_type (entry, info);
899
900 return 1;
901 }
902
903 private:
904
905 /* Storage for symbol names mapping them to their .debug_str section
906 offsets. */
907 class debug_str_lookup
908 {
909 public:
910
911 /* Object costructor to be called for current DWARF2_PER_OBJFILE.
912 All .debug_str section strings are automatically stored. */
913 debug_str_lookup (struct dwarf2_per_objfile *dwarf2_per_objfile)
914 : m_abfd (dwarf2_per_objfile->objfile->obfd),
915 m_dwarf2_per_objfile (dwarf2_per_objfile)
916 {
917 dwarf2_read_section (dwarf2_per_objfile->objfile,
918 &dwarf2_per_objfile->str);
919 if (dwarf2_per_objfile->str.buffer == NULL)
920 return;
921 for (const gdb_byte *data = dwarf2_per_objfile->str.buffer;
922 data < (dwarf2_per_objfile->str.buffer
923 + dwarf2_per_objfile->str.size);)
924 {
925 const char *const s = reinterpret_cast<const char *> (data);
926 const auto insertpair
927 = m_str_table.emplace (c_str_view (s),
928 data - dwarf2_per_objfile->str.buffer);
929 if (!insertpair.second)
930 complaint (_("Duplicate string \"%s\" in "
931 ".debug_str section [in module %s]"),
932 s, bfd_get_filename (m_abfd));
933 data += strlen (s) + 1;
934 }
935 }
936
937 /* Return offset of symbol name S in the .debug_str section. Add
938 such symbol to the section's end if it does not exist there
939 yet. */
940 size_t lookup (const char *s)
941 {
942 const auto it = m_str_table.find (c_str_view (s));
943 if (it != m_str_table.end ())
944 return it->second;
945 const size_t offset = (m_dwarf2_per_objfile->str.size
946 + m_str_add_buf.size ());
947 m_str_table.emplace (c_str_view (s), offset);
948 m_str_add_buf.append_cstr0 (s);
949 return offset;
950 }
951
952 /* Append the end of the .debug_str section to FILE. */
953 void file_write (FILE *file) const
954 {
955 m_str_add_buf.file_write (file);
956 }
957
958 private:
959 std::unordered_map<c_str_view, size_t, c_str_view_hasher> m_str_table;
960 bfd *const m_abfd;
961 struct dwarf2_per_objfile *m_dwarf2_per_objfile;
962
963 /* Data to add at the end of .debug_str for new needed symbol names. */
964 data_buf m_str_add_buf;
965 };
966
967 /* Container to map used DWARF tags to their .debug_names abbreviation
968 tags. */
969 class index_key
970 {
971 public:
972 index_key (int dwarf_tag_, bool is_static_, unit_kind kind_)
973 : dwarf_tag (dwarf_tag_), is_static (is_static_), kind (kind_)
974 {
975 }
976
977 bool
978 operator== (const index_key &other) const
979 {
980 return (dwarf_tag == other.dwarf_tag && is_static == other.is_static
981 && kind == other.kind);
982 }
983
984 const int dwarf_tag;
985 const bool is_static;
986 const unit_kind kind;
987 };
988
989 /* Provide std::unordered_map::hasher for index_key. */
990 class index_key_hasher
991 {
992 public:
993 size_t
994 operator () (const index_key &key) const
995 {
996 return (std::hash<int>() (key.dwarf_tag) << 1) | key.is_static;
997 }
998 };
999
1000 /* Parameters of one symbol entry. */
1001 class symbol_value
1002 {
1003 public:
1004 const int dwarf_tag, cu_index;
1005 const bool is_static;
1006 const unit_kind kind;
1007
1008 symbol_value (int dwarf_tag_, int cu_index_, bool is_static_,
1009 unit_kind kind_)
1010 : dwarf_tag (dwarf_tag_), cu_index (cu_index_), is_static (is_static_),
1011 kind (kind_)
1012 {}
1013
1014 bool
1015 operator< (const symbol_value &other) const
1016 {
1017 #define X(n) \
1018 do \
1019 { \
1020 if (n < other.n) \
1021 return true; \
1022 if (n > other.n) \
1023 return false; \
1024 } \
1025 while (0)
1026 X (dwarf_tag);
1027 X (is_static);
1028 X (kind);
1029 X (cu_index);
1030 #undef X
1031 return false;
1032 }
1033 };
1034
1035 /* Abstract base class to unify DWARF-32 and DWARF-64 name table
1036 output. */
1037 class offset_vec
1038 {
1039 protected:
1040 const bfd_endian dwarf5_byte_order;
1041 public:
1042 explicit offset_vec (bfd_endian dwarf5_byte_order_)
1043 : dwarf5_byte_order (dwarf5_byte_order_)
1044 {}
1045
1046 /* Call std::vector::reserve for NELEM elements. */
1047 virtual void reserve (size_t nelem) = 0;
1048
1049 /* Call std::vector::push_back with store_unsigned_integer byte
1050 reordering for ELEM. */
1051 virtual void push_back_reorder (size_t elem) = 0;
1052
1053 /* Return expected output size in bytes. */
1054 virtual size_t bytes () const = 0;
1055
1056 /* Write name table to FILE. */
1057 virtual void file_write (FILE *file) const = 0;
1058 };
1059
1060 /* Template to unify DWARF-32 and DWARF-64 output. */
1061 template<typename OffsetSize>
1062 class offset_vec_tmpl : public offset_vec
1063 {
1064 public:
1065 explicit offset_vec_tmpl (bfd_endian dwarf5_byte_order_)
1066 : offset_vec (dwarf5_byte_order_)
1067 {}
1068
1069 /* Implement offset_vec::reserve. */
1070 void reserve (size_t nelem) override
1071 {
1072 m_vec.reserve (nelem);
1073 }
1074
1075 /* Implement offset_vec::push_back_reorder. */
1076 void push_back_reorder (size_t elem) override
1077 {
1078 m_vec.push_back (elem);
1079 /* Check for overflow. */
1080 gdb_assert (m_vec.back () == elem);
1081 store_unsigned_integer (reinterpret_cast<gdb_byte *> (&m_vec.back ()),
1082 sizeof (m_vec.back ()), dwarf5_byte_order, elem);
1083 }
1084
1085 /* Implement offset_vec::bytes. */
1086 size_t bytes () const override
1087 {
1088 return m_vec.size () * sizeof (m_vec[0]);
1089 }
1090
1091 /* Implement offset_vec::file_write. */
1092 void file_write (FILE *file) const override
1093 {
1094 ::file_write (file, m_vec);
1095 }
1096
1097 private:
1098 std::vector<OffsetSize> m_vec;
1099 };
1100
1101 /* Base class to unify DWARF-32 and DWARF-64 .debug_names output
1102 respecting name table width. */
1103 class dwarf
1104 {
1105 public:
1106 offset_vec &name_table_string_offs, &name_table_entry_offs;
1107
1108 dwarf (offset_vec &name_table_string_offs_,
1109 offset_vec &name_table_entry_offs_)
1110 : name_table_string_offs (name_table_string_offs_),
1111 name_table_entry_offs (name_table_entry_offs_)
1112 {
1113 }
1114 };
1115
1116 /* Template to unify DWARF-32 and DWARF-64 .debug_names output
1117 respecting name table width. */
1118 template<typename OffsetSize>
1119 class dwarf_tmpl : public dwarf
1120 {
1121 public:
1122 explicit dwarf_tmpl (bfd_endian dwarf5_byte_order_)
1123 : dwarf (m_name_table_string_offs, m_name_table_entry_offs),
1124 m_name_table_string_offs (dwarf5_byte_order_),
1125 m_name_table_entry_offs (dwarf5_byte_order_)
1126 {}
1127
1128 private:
1129 offset_vec_tmpl<OffsetSize> m_name_table_string_offs;
1130 offset_vec_tmpl<OffsetSize> m_name_table_entry_offs;
1131 };
1132
1133 /* Try to reconstruct original DWARF tag for given partial_symbol.
1134 This function is not DWARF-5 compliant but it is sufficient for
1135 GDB as a DWARF-5 index consumer. */
1136 static int psymbol_tag (const struct partial_symbol *psym)
1137 {
1138 domain_enum domain = psym->domain;
1139 enum address_class aclass = psym->aclass;
1140
1141 switch (domain)
1142 {
1143 case VAR_DOMAIN:
1144 switch (aclass)
1145 {
1146 case LOC_BLOCK:
1147 return DW_TAG_subprogram;
1148 case LOC_TYPEDEF:
1149 return DW_TAG_typedef;
1150 case LOC_COMPUTED:
1151 case LOC_CONST_BYTES:
1152 case LOC_OPTIMIZED_OUT:
1153 case LOC_STATIC:
1154 return DW_TAG_variable;
1155 case LOC_CONST:
1156 /* Note: It's currently impossible to recognize psyms as enum values
1157 short of reading the type info. For now punt. */
1158 return DW_TAG_variable;
1159 default:
1160 /* There are other LOC_FOO values that one might want to classify
1161 as variables, but dwarf2read.c doesn't currently use them. */
1162 return DW_TAG_variable;
1163 }
1164 case STRUCT_DOMAIN:
1165 return DW_TAG_structure_type;
1166 default:
1167 return 0;
1168 }
1169 }
1170
1171 /* Call insert for all partial symbols and mark them in PSYMS_SEEN. */
1172 void write_psymbols (std::unordered_set<partial_symbol *> &psyms_seen,
1173 struct partial_symbol **psymp, int count, int cu_index,
1174 bool is_static, unit_kind kind)
1175 {
1176 for (; count-- > 0; ++psymp)
1177 {
1178 struct partial_symbol *psym = *psymp;
1179
1180 if (psym->language == language_ada)
1181 error (_("Ada is not currently supported by the index"));
1182
1183 /* Only add a given psymbol once. */
1184 if (psyms_seen.insert (psym).second)
1185 insert (psym, cu_index, is_static, kind);
1186 }
1187 }
1188
1189 /* A helper function that writes a single signatured_type
1190 to a debug_names. */
1191 void
1192 write_one_signatured_type (struct signatured_type *entry,
1193 struct signatured_type_index_data *info)
1194 {
1195 struct partial_symtab *psymtab = entry->per_cu.v.psymtab;
1196
1197 write_psymbols (info->psyms_seen,
1198 (info->objfile->global_psymbols.data ()
1199 + psymtab->globals_offset),
1200 psymtab->n_global_syms, info->cu_index, false,
1201 unit_kind::tu);
1202 write_psymbols (info->psyms_seen,
1203 (info->objfile->static_psymbols.data ()
1204 + psymtab->statics_offset),
1205 psymtab->n_static_syms, info->cu_index, true,
1206 unit_kind::tu);
1207
1208 info->types_list.append_uint (dwarf5_offset_size (), m_dwarf5_byte_order,
1209 to_underlying (entry->per_cu.sect_off));
1210
1211 ++info->cu_index;
1212 }
1213
1214 /* Store value of each symbol. */
1215 std::unordered_map<c_str_view, std::set<symbol_value>, c_str_view_hasher>
1216 m_name_to_value_set;
1217
1218 /* Tables of DWARF-5 .debug_names. They are in object file byte
1219 order. */
1220 std::vector<uint32_t> m_bucket_table;
1221 std::vector<uint32_t> m_hash_table;
1222
1223 const bfd_endian m_dwarf5_byte_order;
1224 dwarf_tmpl<uint32_t> m_dwarf32;
1225 dwarf_tmpl<uint64_t> m_dwarf64;
1226 dwarf &m_dwarf;
1227 offset_vec &m_name_table_string_offs, &m_name_table_entry_offs;
1228 debug_str_lookup m_debugstrlookup;
1229
1230 /* Map each used .debug_names abbreviation tag parameter to its
1231 index value. */
1232 std::unordered_map<index_key, int, index_key_hasher> m_indexkey_to_idx;
1233
1234 /* Next unused .debug_names abbreviation tag for
1235 m_indexkey_to_idx. */
1236 int m_idx_next = 1;
1237
1238 /* .debug_names abbreviation table. */
1239 data_buf m_abbrev_table;
1240
1241 /* .debug_names entry pool. */
1242 data_buf m_entry_pool;
1243 };
1244
1245 /* Return iff any of the needed offsets does not fit into 32-bit
1246 .debug_names section. */
1247
1248 static bool
1249 check_dwarf64_offsets (struct dwarf2_per_objfile *dwarf2_per_objfile)
1250 {
1251 for (dwarf2_per_cu_data *per_cu : dwarf2_per_objfile->all_comp_units)
1252 {
1253 if (to_underlying (per_cu->sect_off) >= (static_cast<uint64_t> (1) << 32))
1254 return true;
1255 }
1256 for (const signatured_type *sigtype : dwarf2_per_objfile->all_type_units)
1257 {
1258 const dwarf2_per_cu_data &per_cu = sigtype->per_cu;
1259
1260 if (to_underlying (per_cu.sect_off) >= (static_cast<uint64_t> (1) << 32))
1261 return true;
1262 }
1263 return false;
1264 }
1265
1266 /* The psyms_seen set is potentially going to be largish (~40k
1267 elements when indexing a -g3 build of GDB itself). Estimate the
1268 number of elements in order to avoid too many rehashes, which
1269 require rebuilding buckets and thus many trips to
1270 malloc/free. */
1271
1272 static size_t
1273 psyms_seen_size (struct dwarf2_per_objfile *dwarf2_per_objfile)
1274 {
1275 size_t psyms_count = 0;
1276 for (dwarf2_per_cu_data *per_cu : dwarf2_per_objfile->all_comp_units)
1277 {
1278 struct partial_symtab *psymtab = per_cu->v.psymtab;
1279
1280 if (psymtab != NULL && psymtab->user == NULL)
1281 recursively_count_psymbols (psymtab, psyms_count);
1282 }
1283 /* Generating an index for gdb itself shows a ratio of
1284 TOTAL_SEEN_SYMS/UNIQUE_SYMS or ~5. 4 seems like a good bet. */
1285 return psyms_count / 4;
1286 }
1287
1288 /* Write new .gdb_index section for OBJFILE into OUT_FILE.
1289 Return how many bytes were expected to be written into OUT_FILE. */
1290
1291 static size_t
1292 write_gdbindex (struct dwarf2_per_objfile *dwarf2_per_objfile, FILE *out_file)
1293 {
1294 struct objfile *objfile = dwarf2_per_objfile->objfile;
1295 mapped_symtab symtab;
1296 data_buf cu_list;
1297
1298 /* While we're scanning CU's create a table that maps a psymtab pointer
1299 (which is what addrmap records) to its index (which is what is recorded
1300 in the index file). This will later be needed to write the address
1301 table. */
1302 psym_index_map cu_index_htab;
1303 cu_index_htab.reserve (dwarf2_per_objfile->all_comp_units.size ());
1304
1305 /* The CU list is already sorted, so we don't need to do additional
1306 work here. Also, the debug_types entries do not appear in
1307 all_comp_units, but only in their own hash table. */
1308
1309 std::unordered_set<partial_symbol *> psyms_seen
1310 (psyms_seen_size (dwarf2_per_objfile));
1311 for (int i = 0; i < dwarf2_per_objfile->all_comp_units.size (); ++i)
1312 {
1313 struct dwarf2_per_cu_data *per_cu
1314 = dwarf2_per_objfile->all_comp_units[i];
1315 struct partial_symtab *psymtab = per_cu->v.psymtab;
1316
1317 /* CU of a shared file from 'dwz -m' may be unused by this main file.
1318 It may be referenced from a local scope but in such case it does not
1319 need to be present in .gdb_index. */
1320 if (psymtab == NULL)
1321 continue;
1322
1323 if (psymtab->user == NULL)
1324 recursively_write_psymbols (objfile, psymtab, &symtab,
1325 psyms_seen, i);
1326
1327 const auto insertpair = cu_index_htab.emplace (psymtab, i);
1328 gdb_assert (insertpair.second);
1329
1330 cu_list.append_uint (8, BFD_ENDIAN_LITTLE,
1331 to_underlying (per_cu->sect_off));
1332 cu_list.append_uint (8, BFD_ENDIAN_LITTLE, per_cu->length);
1333 }
1334
1335 /* Dump the address map. */
1336 data_buf addr_vec;
1337 write_address_map (objfile, addr_vec, cu_index_htab);
1338
1339 /* Write out the .debug_type entries, if any. */
1340 data_buf types_cu_list;
1341 if (dwarf2_per_objfile->signatured_types)
1342 {
1343 signatured_type_index_data sig_data (types_cu_list,
1344 psyms_seen);
1345
1346 sig_data.objfile = objfile;
1347 sig_data.symtab = &symtab;
1348 sig_data.cu_index = dwarf2_per_objfile->all_comp_units.size ();
1349 htab_traverse_noresize (dwarf2_per_objfile->signatured_types,
1350 write_one_signatured_type, &sig_data);
1351 }
1352
1353 /* Now that we've processed all symbols we can shrink their cu_indices
1354 lists. */
1355 uniquify_cu_indices (&symtab);
1356
1357 data_buf symtab_vec, constant_pool;
1358 write_hash_table (&symtab, symtab_vec, constant_pool);
1359
1360 data_buf contents;
1361 const offset_type size_of_contents = 6 * sizeof (offset_type);
1362 offset_type total_len = size_of_contents;
1363
1364 /* The version number. */
1365 contents.append_data (MAYBE_SWAP (8));
1366
1367 /* The offset of the CU list from the start of the file. */
1368 contents.append_data (MAYBE_SWAP (total_len));
1369 total_len += cu_list.size ();
1370
1371 /* The offset of the types CU list from the start of the file. */
1372 contents.append_data (MAYBE_SWAP (total_len));
1373 total_len += types_cu_list.size ();
1374
1375 /* The offset of the address table from the start of the file. */
1376 contents.append_data (MAYBE_SWAP (total_len));
1377 total_len += addr_vec.size ();
1378
1379 /* The offset of the symbol table from the start of the file. */
1380 contents.append_data (MAYBE_SWAP (total_len));
1381 total_len += symtab_vec.size ();
1382
1383 /* The offset of the constant pool from the start of the file. */
1384 contents.append_data (MAYBE_SWAP (total_len));
1385 total_len += constant_pool.size ();
1386
1387 gdb_assert (contents.size () == size_of_contents);
1388
1389 contents.file_write (out_file);
1390 cu_list.file_write (out_file);
1391 types_cu_list.file_write (out_file);
1392 addr_vec.file_write (out_file);
1393 symtab_vec.file_write (out_file);
1394 constant_pool.file_write (out_file);
1395
1396 return total_len;
1397 }
1398
1399 /* DWARF-5 augmentation string for GDB's DW_IDX_GNU_* extension. */
1400 static const gdb_byte dwarf5_gdb_augmentation[] = { 'G', 'D', 'B', 0 };
1401
1402 /* Write a new .debug_names section for OBJFILE into OUT_FILE, write
1403 needed addition to .debug_str section to OUT_FILE_STR. Return how
1404 many bytes were expected to be written into OUT_FILE. */
1405
1406 static size_t
1407 write_debug_names (struct dwarf2_per_objfile *dwarf2_per_objfile,
1408 FILE *out_file, FILE *out_file_str)
1409 {
1410 const bool dwarf5_is_dwarf64 = check_dwarf64_offsets (dwarf2_per_objfile);
1411 struct objfile *objfile = dwarf2_per_objfile->objfile;
1412 const enum bfd_endian dwarf5_byte_order
1413 = gdbarch_byte_order (get_objfile_arch (objfile));
1414
1415 /* The CU list is already sorted, so we don't need to do additional
1416 work here. Also, the debug_types entries do not appear in
1417 all_comp_units, but only in their own hash table. */
1418 data_buf cu_list;
1419 debug_names nametable (dwarf2_per_objfile, dwarf5_is_dwarf64,
1420 dwarf5_byte_order);
1421 std::unordered_set<partial_symbol *>
1422 psyms_seen (psyms_seen_size (dwarf2_per_objfile));
1423 for (int i = 0; i < dwarf2_per_objfile->all_comp_units.size (); ++i)
1424 {
1425 const dwarf2_per_cu_data *per_cu = dwarf2_per_objfile->all_comp_units[i];
1426 partial_symtab *psymtab = per_cu->v.psymtab;
1427
1428 /* CU of a shared file from 'dwz -m' may be unused by this main
1429 file. It may be referenced from a local scope but in such
1430 case it does not need to be present in .debug_names. */
1431 if (psymtab == NULL)
1432 continue;
1433
1434 if (psymtab->user == NULL)
1435 nametable.recursively_write_psymbols (objfile, psymtab, psyms_seen, i);
1436
1437 cu_list.append_uint (nametable.dwarf5_offset_size (), dwarf5_byte_order,
1438 to_underlying (per_cu->sect_off));
1439 }
1440
1441 /* Write out the .debug_type entries, if any. */
1442 data_buf types_cu_list;
1443 if (dwarf2_per_objfile->signatured_types)
1444 {
1445 debug_names::write_one_signatured_type_data sig_data (nametable,
1446 signatured_type_index_data (types_cu_list, psyms_seen));
1447
1448 sig_data.info.objfile = objfile;
1449 /* It is used only for gdb_index. */
1450 sig_data.info.symtab = nullptr;
1451 sig_data.info.cu_index = 0;
1452 htab_traverse_noresize (dwarf2_per_objfile->signatured_types,
1453 debug_names::write_one_signatured_type,
1454 &sig_data);
1455 }
1456
1457 nametable.build ();
1458
1459 /* No addr_vec - DWARF-5 uses .debug_aranges generated by GCC. */
1460
1461 const offset_type bytes_of_header
1462 = ((dwarf5_is_dwarf64 ? 12 : 4)
1463 + 2 + 2 + 7 * 4
1464 + sizeof (dwarf5_gdb_augmentation));
1465 size_t expected_bytes = 0;
1466 expected_bytes += bytes_of_header;
1467 expected_bytes += cu_list.size ();
1468 expected_bytes += types_cu_list.size ();
1469 expected_bytes += nametable.bytes ();
1470 data_buf header;
1471
1472 if (!dwarf5_is_dwarf64)
1473 {
1474 const uint64_t size64 = expected_bytes - 4;
1475 gdb_assert (size64 < 0xfffffff0);
1476 header.append_uint (4, dwarf5_byte_order, size64);
1477 }
1478 else
1479 {
1480 header.append_uint (4, dwarf5_byte_order, 0xffffffff);
1481 header.append_uint (8, dwarf5_byte_order, expected_bytes - 12);
1482 }
1483
1484 /* The version number. */
1485 header.append_uint (2, dwarf5_byte_order, 5);
1486
1487 /* Padding. */
1488 header.append_uint (2, dwarf5_byte_order, 0);
1489
1490 /* comp_unit_count - The number of CUs in the CU list. */
1491 header.append_uint (4, dwarf5_byte_order,
1492 dwarf2_per_objfile->all_comp_units.size ());
1493
1494 /* local_type_unit_count - The number of TUs in the local TU
1495 list. */
1496 header.append_uint (4, dwarf5_byte_order,
1497 dwarf2_per_objfile->all_type_units.size ());
1498
1499 /* foreign_type_unit_count - The number of TUs in the foreign TU
1500 list. */
1501 header.append_uint (4, dwarf5_byte_order, 0);
1502
1503 /* bucket_count - The number of hash buckets in the hash lookup
1504 table. */
1505 header.append_uint (4, dwarf5_byte_order, nametable.bucket_count ());
1506
1507 /* name_count - The number of unique names in the index. */
1508 header.append_uint (4, dwarf5_byte_order, nametable.name_count ());
1509
1510 /* abbrev_table_size - The size in bytes of the abbreviations
1511 table. */
1512 header.append_uint (4, dwarf5_byte_order, nametable.abbrev_table_bytes ());
1513
1514 /* augmentation_string_size - The size in bytes of the augmentation
1515 string. This value is rounded up to a multiple of 4. */
1516 static_assert (sizeof (dwarf5_gdb_augmentation) % 4 == 0, "");
1517 header.append_uint (4, dwarf5_byte_order, sizeof (dwarf5_gdb_augmentation));
1518 header.append_data (dwarf5_gdb_augmentation);
1519
1520 gdb_assert (header.size () == bytes_of_header);
1521
1522 header.file_write (out_file);
1523 cu_list.file_write (out_file);
1524 types_cu_list.file_write (out_file);
1525 nametable.file_write (out_file, out_file_str);
1526
1527 return expected_bytes;
1528 }
1529
1530 /* Assert that FILE's size is EXPECTED_SIZE. Assumes file's seek
1531 position is at the end of the file. */
1532
1533 static void
1534 assert_file_size (FILE *file, const char *filename, size_t expected_size)
1535 {
1536 const auto file_size = ftell (file);
1537 if (file_size == -1)
1538 error (_("Can't get `%s' size"), filename);
1539 gdb_assert (file_size == expected_size);
1540 }
1541
1542 /* See dwarf-index-write.h. */
1543
1544 void
1545 write_psymtabs_to_index (struct dwarf2_per_objfile *dwarf2_per_objfile,
1546 const char *dir, const char *basename,
1547 dw_index_kind index_kind)
1548 {
1549 struct objfile *objfile = dwarf2_per_objfile->objfile;
1550
1551 if (dwarf2_per_objfile->using_index)
1552 error (_("Cannot use an index to create the index"));
1553
1554 if (VEC_length (dwarf2_section_info_def, dwarf2_per_objfile->types) > 1)
1555 error (_("Cannot make an index when the file has multiple .debug_types sections"));
1556
1557 if (!objfile->psymtabs || !objfile->psymtabs_addrmap)
1558 return;
1559
1560 struct stat st;
1561 if (stat (objfile_name (objfile), &st) < 0)
1562 perror_with_name (objfile_name (objfile));
1563
1564 std::string filename (std::string (dir) + SLASH_STRING + basename
1565 + (index_kind == dw_index_kind::DEBUG_NAMES
1566 ? INDEX5_SUFFIX : INDEX4_SUFFIX));
1567 gdb::char_vector filename_temp = make_temp_filename (filename);
1568
1569 /* Order matters here; we want FILE to be closed before
1570 FILENAME_TEMP is unlinked, because on MS-Windows one cannot
1571 delete a file that is still open. So, we wrap the unlinker in an
1572 optional and emplace it once we know the file name. */
1573 gdb::optional<gdb::unlinker> unlink_file;
1574 scoped_fd out_file_fd (gdb_mkostemp_cloexec (filename_temp.data (),
1575 O_BINARY));
1576 if (out_file_fd.get () == -1)
1577 perror_with_name (("mkstemp"));
1578
1579 gdb_file_up out_file = out_file_fd.to_file ("wb");
1580 if (out_file == nullptr)
1581 error (_("Can't open `%s' for writing"), filename_temp.data ());
1582
1583 unlink_file.emplace (filename_temp.data ());
1584
1585 if (index_kind == dw_index_kind::DEBUG_NAMES)
1586 {
1587 std::string filename_str (std::string (dir) + SLASH_STRING
1588 + basename + DEBUG_STR_SUFFIX);
1589 gdb::char_vector filename_str_temp = make_temp_filename (filename_str);
1590
1591 /* As above, arrange to unlink the file only after the file
1592 descriptor has been closed. */
1593 gdb::optional<gdb::unlinker> unlink_file_str;
1594 scoped_fd out_file_str_fd
1595 (gdb_mkostemp_cloexec (filename_str_temp.data (), O_BINARY));
1596 if (out_file_str_fd.get () == -1)
1597 perror_with_name (("mkstemp"));
1598
1599 gdb_file_up out_file_str = out_file_str_fd.to_file ("wb");
1600 if (out_file_str == nullptr)
1601 error (_("Can't open `%s' for writing"), filename_str_temp.data ());
1602
1603 unlink_file_str.emplace (filename_str_temp.data ());
1604
1605 const size_t total_len
1606 = write_debug_names (dwarf2_per_objfile, out_file.get (),
1607 out_file_str.get ());
1608 assert_file_size (out_file.get (), filename_temp.data (), total_len);
1609
1610 /* We want to keep the file .debug_str file too. */
1611 unlink_file_str->keep ();
1612
1613 /* Close and move the str file in place. */
1614 out_file_str.reset ();
1615 if (rename (filename_str_temp.data (), filename_str.c_str ()) != 0)
1616 perror_with_name (("rename"));
1617 }
1618 else
1619 {
1620 const size_t total_len
1621 = write_gdbindex (dwarf2_per_objfile, out_file.get ());
1622 assert_file_size (out_file.get (), filename_temp.data (), total_len);
1623 }
1624
1625 /* We want to keep the file. */
1626 unlink_file->keep ();
1627
1628 /* Close and move the file in place. */
1629 out_file.reset ();
1630 if (rename (filename_temp.data (), filename.c_str ()) != 0)
1631 perror_with_name (("rename"));
1632 }
1633
1634 /* Implementation of the `save gdb-index' command.
1635
1636 Note that the .gdb_index file format used by this command is
1637 documented in the GDB manual. Any changes here must be documented
1638 there. */
1639
1640 static void
1641 save_gdb_index_command (const char *arg, int from_tty)
1642 {
1643 const char dwarf5space[] = "-dwarf-5 ";
1644 dw_index_kind index_kind = dw_index_kind::GDB_INDEX;
1645
1646 if (!arg)
1647 arg = "";
1648
1649 arg = skip_spaces (arg);
1650 if (strncmp (arg, dwarf5space, strlen (dwarf5space)) == 0)
1651 {
1652 index_kind = dw_index_kind::DEBUG_NAMES;
1653 arg += strlen (dwarf5space);
1654 arg = skip_spaces (arg);
1655 }
1656
1657 if (!*arg)
1658 error (_("usage: save gdb-index [-dwarf-5] DIRECTORY"));
1659
1660 for (objfile *objfile : all_objfiles (current_program_space))
1661 {
1662 struct stat st;
1663
1664 /* If the objfile does not correspond to an actual file, skip it. */
1665 if (stat (objfile_name (objfile), &st) < 0)
1666 continue;
1667
1668 struct dwarf2_per_objfile *dwarf2_per_objfile
1669 = get_dwarf2_per_objfile (objfile);
1670
1671 if (dwarf2_per_objfile != NULL)
1672 {
1673 TRY
1674 {
1675 const char *basename = lbasename (objfile_name (objfile));
1676 write_psymtabs_to_index (dwarf2_per_objfile, arg, basename,
1677 index_kind);
1678 }
1679 CATCH (except, RETURN_MASK_ERROR)
1680 {
1681 exception_fprintf (gdb_stderr, except,
1682 _("Error while writing index for `%s': "),
1683 objfile_name (objfile));
1684 }
1685 END_CATCH
1686 }
1687
1688 }
1689 }
1690
1691 void
1692 _initialize_dwarf_index_write ()
1693 {
1694 cmd_list_element *c = add_cmd ("gdb-index", class_files,
1695 save_gdb_index_command, _("\
1696 Save a gdb-index file.\n\
1697 Usage: save gdb-index [-dwarf-5] DIRECTORY\n\
1698 \n\
1699 No options create one file with .gdb-index extension for pre-DWARF-5\n\
1700 compatible .gdb_index section. With -dwarf-5 creates two files with\n\
1701 extension .debug_names and .debug_str for DWARF-5 .debug_names section."),
1702 &save_cmdlist);
1703 set_cmd_completer (c, filename_completer);
1704 }
This page took 0.088388 seconds and 5 git commands to generate.