Disambiguate info_print_options
[deliverable/binutils-gdb.git] / gdb / dwarf-index-write.c
CommitLineData
cd4fb1b2
SM
1/* DWARF index writing support for GDB.
2
b811d2c2 3 Copyright (C) 1994-2020 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
23baa4cc
SM
22#include "dwarf-index-write.h"
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
SM
31#include "complaints.h"
32#include "dwarf-index-common.h"
33#include "dwarf2.h"
34#include "dwarf2read.h"
35#include "gdb/gdb-index.h"
36#include "gdbcmd.h"
37#include "objfiles.h"
38#include "psympriv.h"
3b00ef10 39#include "ada-lang.h"
cd4fb1b2 40
4de283e4
TT
41#include <algorithm>
42#include <cmath>
43#include <set>
44#include <unordered_map>
45#include <unordered_set>
46
cd4fb1b2
SM
47/* Ensure only legit values are used. */
48#define DW2_GDB_INDEX_SYMBOL_STATIC_SET_VALUE(cu_index, value) \
49 do { \
50 gdb_assert ((unsigned int) (value) <= 1); \
51 GDB_INDEX_SYMBOL_STATIC_SET_VALUE((cu_index), (value)); \
52 } while (0)
53
54/* Ensure only legit values are used. */
55#define DW2_GDB_INDEX_SYMBOL_KIND_SET_VALUE(cu_index, value) \
56 do { \
57 gdb_assert ((value) >= GDB_INDEX_SYMBOL_KIND_TYPE \
58 && (value) <= GDB_INDEX_SYMBOL_KIND_OTHER); \
59 GDB_INDEX_SYMBOL_KIND_SET_VALUE((cu_index), (value)); \
60 } while (0)
61
85102364 62/* Ensure we don't use more than the allotted number of bits for the CU. */
cd4fb1b2
SM
63#define DW2_GDB_INDEX_CU_SET_VALUE(cu_index, value) \
64 do { \
65 gdb_assert (((value) & ~GDB_INDEX_CU_MASK) == 0); \
66 GDB_INDEX_CU_SET_VALUE((cu_index), (value)); \
67 } while (0)
68
69/* The "save gdb-index" command. */
70
71/* Write SIZE bytes from the buffer pointed to by DATA to FILE, with
72 error checking. */
73
74static void
75file_write (FILE *file, const void *data, size_t size)
76{
77 if (fwrite (data, 1, size, file) != size)
78 error (_("couldn't data write to file"));
79}
80
81/* Write the contents of VEC to FILE, with error checking. */
82
83template<typename Elem, typename Alloc>
84static void
85file_write (FILE *file, const std::vector<Elem, Alloc> &vec)
86{
1f88d0c8
SM
87 if (!vec.empty ())
88 file_write (file, vec.data (), vec.size () * sizeof (vec[0]));
cd4fb1b2
SM
89}
90
91/* In-memory buffer to prepare data to be written later to a file. */
92class data_buf
93{
94public:
95 /* Copy DATA to the end of the buffer. */
96 template<typename T>
97 void append_data (const T &data)
98 {
99 std::copy (reinterpret_cast<const gdb_byte *> (&data),
100 reinterpret_cast<const gdb_byte *> (&data + 1),
101 grow (sizeof (data)));
102 }
103
104 /* Copy CSTR (a zero-terminated string) to the end of buffer. The
105 terminating zero is appended too. */
106 void append_cstr0 (const char *cstr)
107 {
108 const size_t size = strlen (cstr) + 1;
109 std::copy (cstr, cstr + size, grow (size));
110 }
111
112 /* Store INPUT as ULEB128 to the end of buffer. */
113 void append_unsigned_leb128 (ULONGEST input)
114 {
115 for (;;)
116 {
117 gdb_byte output = input & 0x7f;
118 input >>= 7;
119 if (input)
120 output |= 0x80;
121 append_data (output);
122 if (input == 0)
123 break;
124 }
125 }
126
127 /* Accept a host-format integer in VAL and append it to the buffer
128 as a target-format integer which is LEN bytes long. */
129 void append_uint (size_t len, bfd_endian byte_order, ULONGEST val)
130 {
131 ::store_unsigned_integer (grow (len), len, byte_order, val);
132 }
133
134 /* Return the size of the buffer. */
135 size_t size () const
136 {
137 return m_vec.size ();
138 }
139
140 /* Return true iff the buffer is empty. */
141 bool empty () const
142 {
143 return m_vec.empty ();
144 }
145
146 /* Write the buffer to FILE. */
147 void file_write (FILE *file) const
148 {
149 ::file_write (file, m_vec);
150 }
151
152private:
153 /* Grow SIZE bytes at the end of the buffer. Returns a pointer to
154 the start of the new block. */
155 gdb_byte *grow (size_t size)
156 {
157 m_vec.resize (m_vec.size () + size);
b4be9bfd 158 return &*(m_vec.end () - size);
cd4fb1b2
SM
159 }
160
161 gdb::byte_vector m_vec;
162};
163
164/* An entry in the symbol table. */
165struct symtab_index_entry
166{
167 /* The name of the symbol. */
168 const char *name;
169 /* The offset of the name in the constant pool. */
170 offset_type index_offset;
171 /* A sorted vector of the indices of all the CUs that hold an object
172 of this name. */
173 std::vector<offset_type> cu_indices;
174};
175
176/* The symbol table. This is a power-of-2-sized hash table. */
177struct mapped_symtab
178{
179 mapped_symtab ()
180 {
181 data.resize (1024);
182 }
183
184 offset_type n_elements = 0;
185 std::vector<symtab_index_entry> data;
186};
187
188/* Find a slot in SYMTAB for the symbol NAME. Returns a reference to
189 the slot.
190
191 Function is used only during write_hash_table so no index format backward
192 compatibility is needed. */
193
194static symtab_index_entry &
195find_slot (struct mapped_symtab *symtab, const char *name)
196{
197 offset_type index, step, hash = mapped_index_string_hash (INT_MAX, name);
198
199 index = hash & (symtab->data.size () - 1);
200 step = ((hash * 17) & (symtab->data.size () - 1)) | 1;
201
202 for (;;)
203 {
204 if (symtab->data[index].name == NULL
205 || strcmp (name, symtab->data[index].name) == 0)
206 return symtab->data[index];
207 index = (index + step) & (symtab->data.size () - 1);
208 }
209}
210
211/* Expand SYMTAB's hash table. */
212
213static void
214hash_expand (struct mapped_symtab *symtab)
215{
216 auto old_entries = std::move (symtab->data);
217
218 symtab->data.clear ();
219 symtab->data.resize (old_entries.size () * 2);
220
221 for (auto &it : old_entries)
222 if (it.name != NULL)
223 {
224 auto &ref = find_slot (symtab, it.name);
225 ref = std::move (it);
226 }
227}
228
229/* Add an entry to SYMTAB. NAME is the name of the symbol.
230 CU_INDEX is the index of the CU in which the symbol appears.
231 IS_STATIC is one if the symbol is static, otherwise zero (global). */
232
233static void
234add_index_entry (struct mapped_symtab *symtab, const char *name,
235 int is_static, gdb_index_symbol_kind kind,
236 offset_type cu_index)
237{
238 offset_type cu_index_and_attrs;
239
240 ++symtab->n_elements;
241 if (4 * symtab->n_elements / 3 >= symtab->data.size ())
242 hash_expand (symtab);
243
244 symtab_index_entry &slot = find_slot (symtab, name);
245 if (slot.name == NULL)
246 {
247 slot.name = name;
248 /* index_offset is set later. */
249 }
250
251 cu_index_and_attrs = 0;
252 DW2_GDB_INDEX_CU_SET_VALUE (cu_index_and_attrs, cu_index);
253 DW2_GDB_INDEX_SYMBOL_STATIC_SET_VALUE (cu_index_and_attrs, is_static);
254 DW2_GDB_INDEX_SYMBOL_KIND_SET_VALUE (cu_index_and_attrs, kind);
255
256 /* We don't want to record an index value twice as we want to avoid the
257 duplication.
258 We process all global symbols and then all static symbols
259 (which would allow us to avoid the duplication by only having to check
260 the last entry pushed), but a symbol could have multiple kinds in one CU.
261 To keep things simple we don't worry about the duplication here and
85102364 262 sort and uniquify the list after we've processed all symbols. */
cd4fb1b2
SM
263 slot.cu_indices.push_back (cu_index_and_attrs);
264}
265
266/* Sort and remove duplicates of all symbols' cu_indices lists. */
267
268static void
269uniquify_cu_indices (struct mapped_symtab *symtab)
270{
271 for (auto &entry : symtab->data)
272 {
273 if (entry.name != NULL && !entry.cu_indices.empty ())
274 {
275 auto &cu_indices = entry.cu_indices;
276 std::sort (cu_indices.begin (), cu_indices.end ());
277 auto from = std::unique (cu_indices.begin (), cu_indices.end ());
278 cu_indices.erase (from, cu_indices.end ());
279 }
280 }
281}
282
283/* A form of 'const char *' suitable for container keys. Only the
284 pointer is stored. The strings themselves are compared, not the
285 pointers. */
286class c_str_view
287{
288public:
289 c_str_view (const char *cstr)
290 : m_cstr (cstr)
291 {}
292
293 bool operator== (const c_str_view &other) const
294 {
295 return strcmp (m_cstr, other.m_cstr) == 0;
296 }
297
298 /* Return the underlying C string. Note, the returned string is
299 only a reference with lifetime of this object. */
300 const char *c_str () const
301 {
302 return m_cstr;
303 }
304
305private:
306 friend class c_str_view_hasher;
307 const char *const m_cstr;
308};
309
310/* A std::unordered_map::hasher for c_str_view that uses the right
311 hash function for strings in a mapped index. */
312class c_str_view_hasher
313{
314public:
315 size_t operator () (const c_str_view &x) const
316 {
317 return mapped_index_string_hash (INT_MAX, x.m_cstr);
318 }
319};
320
321/* A std::unordered_map::hasher for std::vector<>. */
322template<typename T>
323class vector_hasher
324{
325public:
326 size_t operator () (const std::vector<T> &key) const
327 {
328 return iterative_hash (key.data (),
329 sizeof (key.front ()) * key.size (), 0);
330 }
331};
332
333/* Write the mapped hash table SYMTAB to the data buffer OUTPUT, with
334 constant pool entries going into the data buffer CPOOL. */
335
336static void
337write_hash_table (mapped_symtab *symtab, data_buf &output, data_buf &cpool)
338{
339 {
340 /* Elements are sorted vectors of the indices of all the CUs that
341 hold an object of this name. */
342 std::unordered_map<std::vector<offset_type>, offset_type,
343 vector_hasher<offset_type>>
344 symbol_hash_table;
345
346 /* We add all the index vectors to the constant pool first, to
347 ensure alignment is ok. */
348 for (symtab_index_entry &entry : symtab->data)
349 {
350 if (entry.name == NULL)
351 continue;
352 gdb_assert (entry.index_offset == 0);
353
354 /* Finding before inserting is faster than always trying to
355 insert, because inserting always allocates a node, does the
356 lookup, and then destroys the new node if another node
357 already had the same key. C++17 try_emplace will avoid
358 this. */
359 const auto found
360 = symbol_hash_table.find (entry.cu_indices);
361 if (found != symbol_hash_table.end ())
362 {
363 entry.index_offset = found->second;
364 continue;
365 }
366
367 symbol_hash_table.emplace (entry.cu_indices, cpool.size ());
368 entry.index_offset = cpool.size ();
369 cpool.append_data (MAYBE_SWAP (entry.cu_indices.size ()));
370 for (const auto index : entry.cu_indices)
371 cpool.append_data (MAYBE_SWAP (index));
372 }
373 }
374
375 /* Now write out the hash table. */
376 std::unordered_map<c_str_view, offset_type, c_str_view_hasher> str_table;
377 for (const auto &entry : symtab->data)
378 {
379 offset_type str_off, vec_off;
380
381 if (entry.name != NULL)
382 {
383 const auto insertpair = str_table.emplace (entry.name, cpool.size ());
384 if (insertpair.second)
385 cpool.append_cstr0 (entry.name);
386 str_off = insertpair.first->second;
387 vec_off = entry.index_offset;
388 }
389 else
390 {
391 /* While 0 is a valid constant pool index, it is not valid
392 to have 0 for both offsets. */
393 str_off = 0;
394 vec_off = 0;
395 }
396
397 output.append_data (MAYBE_SWAP (str_off));
398 output.append_data (MAYBE_SWAP (vec_off));
399 }
400}
401
891813be 402typedef std::unordered_map<dwarf2_psymtab *, unsigned int> psym_index_map;
cd4fb1b2
SM
403
404/* Helper struct for building the address table. */
405struct addrmap_index_data
406{
407 addrmap_index_data (data_buf &addr_vec_, psym_index_map &cu_index_htab_)
408 : addr_vec (addr_vec_), cu_index_htab (cu_index_htab_)
409 {}
410
411 struct objfile *objfile;
412 data_buf &addr_vec;
413 psym_index_map &cu_index_htab;
414
415 /* Non-zero if the previous_* fields are valid.
416 We can't write an entry until we see the next entry (since it is only then
417 that we know the end of the entry). */
418 int previous_valid;
419 /* Index of the CU in the table of all CUs in the index file. */
420 unsigned int previous_cu_index;
421 /* Start address of the CU. */
422 CORE_ADDR previous_cu_start;
423};
424
425/* Write an address entry to ADDR_VEC. */
426
427static void
428add_address_entry (struct objfile *objfile, data_buf &addr_vec,
429 CORE_ADDR start, CORE_ADDR end, unsigned int cu_index)
430{
79748972
TT
431 addr_vec.append_uint (8, BFD_ENDIAN_LITTLE, start);
432 addr_vec.append_uint (8, BFD_ENDIAN_LITTLE, end);
cd4fb1b2
SM
433 addr_vec.append_data (MAYBE_SWAP (cu_index));
434}
435
436/* Worker function for traversing an addrmap to build the address table. */
437
438static int
439add_address_entry_worker (void *datap, CORE_ADDR start_addr, void *obj)
440{
441 struct addrmap_index_data *data = (struct addrmap_index_data *) datap;
891813be 442 dwarf2_psymtab *pst = (dwarf2_psymtab *) obj;
cd4fb1b2
SM
443
444 if (data->previous_valid)
445 add_address_entry (data->objfile, data->addr_vec,
446 data->previous_cu_start, start_addr,
447 data->previous_cu_index);
448
449 data->previous_cu_start = start_addr;
450 if (pst != NULL)
451 {
452 const auto it = data->cu_index_htab.find (pst);
453 gdb_assert (it != data->cu_index_htab.cend ());
454 data->previous_cu_index = it->second;
455 data->previous_valid = 1;
456 }
457 else
458 data->previous_valid = 0;
459
460 return 0;
461}
462
463/* Write OBJFILE's address map to ADDR_VEC.
464 CU_INDEX_HTAB is used to map addrmap entries to their CU indices
465 in the index file. */
466
467static void
468write_address_map (struct objfile *objfile, data_buf &addr_vec,
469 psym_index_map &cu_index_htab)
470{
471 struct addrmap_index_data addrmap_index_data (addr_vec, cu_index_htab);
472
473 /* When writing the address table, we have to cope with the fact that
474 the addrmap iterator only provides the start of a region; we have to
475 wait until the next invocation to get the start of the next region. */
476
477 addrmap_index_data.objfile = objfile;
478 addrmap_index_data.previous_valid = 0;
479
d320c2b5
TT
480 addrmap_foreach (objfile->partial_symtabs->psymtabs_addrmap,
481 add_address_entry_worker, &addrmap_index_data);
cd4fb1b2
SM
482
483 /* It's highly unlikely the last entry (end address = 0xff...ff)
484 is valid, but we should still handle it.
485 The end address is recorded as the start of the next region, but that
486 doesn't work here. To cope we pass 0xff...ff, this is a rare situation
487 anyway. */
488 if (addrmap_index_data.previous_valid)
489 add_address_entry (objfile, addr_vec,
490 addrmap_index_data.previous_cu_start, (CORE_ADDR) -1,
491 addrmap_index_data.previous_cu_index);
492}
493
494/* Return the symbol kind of PSYM. */
495
496static gdb_index_symbol_kind
497symbol_kind (struct partial_symbol *psym)
498{
8a6d4234
TT
499 domain_enum domain = psym->domain;
500 enum address_class aclass = psym->aclass;
cd4fb1b2
SM
501
502 switch (domain)
503 {
504 case VAR_DOMAIN:
505 switch (aclass)
506 {
507 case LOC_BLOCK:
508 return GDB_INDEX_SYMBOL_KIND_FUNCTION;
509 case LOC_TYPEDEF:
510 return GDB_INDEX_SYMBOL_KIND_TYPE;
511 case LOC_COMPUTED:
512 case LOC_CONST_BYTES:
513 case LOC_OPTIMIZED_OUT:
514 case LOC_STATIC:
515 return GDB_INDEX_SYMBOL_KIND_VARIABLE;
516 case LOC_CONST:
517 /* Note: It's currently impossible to recognize psyms as enum values
518 short of reading the type info. For now punt. */
519 return GDB_INDEX_SYMBOL_KIND_VARIABLE;
520 default:
521 /* There are other LOC_FOO values that one might want to classify
522 as variables, but dwarf2read.c doesn't currently use them. */
523 return GDB_INDEX_SYMBOL_KIND_OTHER;
524 }
525 case STRUCT_DOMAIN:
526 return GDB_INDEX_SYMBOL_KIND_TYPE;
527 default:
528 return GDB_INDEX_SYMBOL_KIND_OTHER;
529 }
530}
531
532/* Add a list of partial symbols to SYMTAB. */
533
534static void
535write_psymbols (struct mapped_symtab *symtab,
536 std::unordered_set<partial_symbol *> &psyms_seen,
537 struct partial_symbol **psymp,
538 int count,
539 offset_type cu_index,
540 int is_static)
541{
542 for (; count-- > 0; ++psymp)
543 {
544 struct partial_symbol *psym = *psymp;
545
c1b5c1eb 546 if (psym->ginfo.language () == language_ada)
3b00ef10
TT
547 error (_("Ada is not currently supported by the index; "
548 "use the DWARF 5 index instead"));
cd4fb1b2
SM
549
550 /* Only add a given psymbol once. */
551 if (psyms_seen.insert (psym).second)
552 {
553 gdb_index_symbol_kind kind = symbol_kind (psym);
554
c9d95fa3 555 add_index_entry (symtab, psym->ginfo.search_name (),
cd4fb1b2
SM
556 is_static, kind, cu_index);
557 }
558 }
559}
560
561/* A helper struct used when iterating over debug_types. */
562struct signatured_type_index_data
563{
564 signatured_type_index_data (data_buf &types_list_,
565 std::unordered_set<partial_symbol *> &psyms_seen_)
566 : types_list (types_list_), psyms_seen (psyms_seen_)
567 {}
568
569 struct objfile *objfile;
570 struct mapped_symtab *symtab;
571 data_buf &types_list;
572 std::unordered_set<partial_symbol *> &psyms_seen;
573 int cu_index;
574};
575
576/* A helper function that writes a single signatured_type to an
577 obstack. */
578
579static int
580write_one_signatured_type (void **slot, void *d)
581{
582 struct signatured_type_index_data *info
583 = (struct signatured_type_index_data *) d;
584 struct signatured_type *entry = (struct signatured_type *) *slot;
891813be 585 dwarf2_psymtab *psymtab = entry->per_cu.v.psymtab;
cd4fb1b2
SM
586
587 write_psymbols (info->symtab,
588 info->psyms_seen,
d320c2b5 589 (info->objfile->partial_symtabs->global_psymbols.data ()
b4be9bfd 590 + psymtab->globals_offset),
cd4fb1b2
SM
591 psymtab->n_global_syms, info->cu_index,
592 0);
593 write_psymbols (info->symtab,
594 info->psyms_seen,
d320c2b5 595 (info->objfile->partial_symtabs->static_psymbols.data ()
b4be9bfd 596 + psymtab->statics_offset),
cd4fb1b2
SM
597 psymtab->n_static_syms, info->cu_index,
598 1);
599
600 info->types_list.append_uint (8, BFD_ENDIAN_LITTLE,
601 to_underlying (entry->per_cu.sect_off));
602 info->types_list.append_uint (8, BFD_ENDIAN_LITTLE,
603 to_underlying (entry->type_offset_in_tu));
604 info->types_list.append_uint (8, BFD_ENDIAN_LITTLE, entry->signature);
605
606 ++info->cu_index;
607
608 return 1;
609}
610
611/* Recurse into all "included" dependencies and count their symbols as
612 if they appeared in this psymtab. */
613
614static void
891813be 615recursively_count_psymbols (dwarf2_psymtab *psymtab,
cd4fb1b2
SM
616 size_t &psyms_seen)
617{
618 for (int i = 0; i < psymtab->number_of_dependencies; ++i)
619 if (psymtab->dependencies[i]->user != NULL)
891813be 620 recursively_count_psymbols ((dwarf2_psymtab *) psymtab->dependencies[i],
cd4fb1b2
SM
621 psyms_seen);
622
623 psyms_seen += psymtab->n_global_syms;
624 psyms_seen += psymtab->n_static_syms;
625}
626
627/* Recurse into all "included" dependencies and write their symbols as
628 if they appeared in this psymtab. */
629
630static void
631recursively_write_psymbols (struct objfile *objfile,
891813be 632 dwarf2_psymtab *psymtab,
cd4fb1b2
SM
633 struct mapped_symtab *symtab,
634 std::unordered_set<partial_symbol *> &psyms_seen,
635 offset_type cu_index)
636{
637 int i;
638
639 for (i = 0; i < psymtab->number_of_dependencies; ++i)
640 if (psymtab->dependencies[i]->user != NULL)
891813be
TT
641 recursively_write_psymbols (objfile,
642 (dwarf2_psymtab *) psymtab->dependencies[i],
cd4fb1b2
SM
643 symtab, psyms_seen, cu_index);
644
645 write_psymbols (symtab,
646 psyms_seen,
d320c2b5
TT
647 (objfile->partial_symtabs->global_psymbols.data ()
648 + psymtab->globals_offset),
cd4fb1b2
SM
649 psymtab->n_global_syms, cu_index,
650 0);
651 write_psymbols (symtab,
652 psyms_seen,
d320c2b5
TT
653 (objfile->partial_symtabs->static_psymbols.data ()
654 + psymtab->statics_offset),
cd4fb1b2
SM
655 psymtab->n_static_syms, cu_index,
656 1);
657}
658
659/* DWARF-5 .debug_names builder. */
660class debug_names
661{
662public:
663 debug_names (struct dwarf2_per_objfile *dwarf2_per_objfile, bool is_dwarf64,
664 bfd_endian dwarf5_byte_order)
665 : m_dwarf5_byte_order (dwarf5_byte_order),
666 m_dwarf32 (dwarf5_byte_order),
667 m_dwarf64 (dwarf5_byte_order),
668 m_dwarf (is_dwarf64
669 ? static_cast<dwarf &> (m_dwarf64)
670 : static_cast<dwarf &> (m_dwarf32)),
671 m_name_table_string_offs (m_dwarf.name_table_string_offs),
672 m_name_table_entry_offs (m_dwarf.name_table_entry_offs),
673 m_debugstrlookup (dwarf2_per_objfile)
674 {}
675
676 int dwarf5_offset_size () const
677 {
678 const bool dwarf5_is_dwarf64 = &m_dwarf == &m_dwarf64;
679 return dwarf5_is_dwarf64 ? 8 : 4;
680 }
681
682 /* Is this symbol from DW_TAG_compile_unit or DW_TAG_type_unit? */
683 enum class unit_kind { cu, tu };
684
685 /* Insert one symbol. */
686 void insert (const partial_symbol *psym, int cu_index, bool is_static,
687 unit_kind kind)
688 {
689 const int dwarf_tag = psymbol_tag (psym);
690 if (dwarf_tag == 0)
691 return;
c9d95fa3 692 const char *name = psym->ginfo.search_name ();
3b00ef10 693
c1b5c1eb 694 if (psym->ginfo.language () == language_ada)
3b00ef10
TT
695 {
696 /* We want to ensure that the Ada main function's name appears
697 verbatim in the index. However, this name will be of the
698 form "_ada_mumble", and will be rewritten by ada_decode.
699 So, recognize it specially here and add it to the index by
700 hand. */
701 if (strcmp (main_name (), name) == 0)
702 {
703 const auto insertpair
704 = m_name_to_value_set.emplace (c_str_view (name),
705 std::set<symbol_value> ());
706 std::set<symbol_value> &value_set = insertpair.first->second;
707 value_set.emplace (symbol_value (dwarf_tag, cu_index, is_static,
708 kind));
709 }
710
711 /* In order for the index to work when read back into gdb, it
712 has to supply a funny form of the name: it should be the
713 encoded name, with any suffixes stripped. Using the
714 ordinary encoded name will not work properly with the
715 searching logic in find_name_components_bounds; nor will
716 using the decoded name. Furthermore, an Ada "verbatim"
717 name (of the form "<MumBle>") must be entered without the
718 angle brackets. Note that the current index is unusual,
719 see PR symtab/24820 for details. */
f945dedf 720 std::string decoded = ada_decode (name);
3b00ef10
TT
721 if (decoded[0] == '<')
722 name = (char *) obstack_copy0 (&m_string_obstack,
f945dedf
CB
723 decoded.c_str () + 1,
724 decoded.length () - 2);
3b00ef10 725 else
f945dedf
CB
726 name = obstack_strdup (&m_string_obstack,
727 ada_encode (decoded.c_str ()));
3b00ef10
TT
728 }
729
cd4fb1b2
SM
730 const auto insertpair
731 = m_name_to_value_set.emplace (c_str_view (name),
732 std::set<symbol_value> ());
733 std::set<symbol_value> &value_set = insertpair.first->second;
734 value_set.emplace (symbol_value (dwarf_tag, cu_index, is_static, kind));
735 }
736
737 /* Build all the tables. All symbols must be already inserted.
738 This function does not call file_write, caller has to do it
739 afterwards. */
740 void build ()
741 {
742 /* Verify the build method has not be called twice. */
743 gdb_assert (m_abbrev_table.empty ());
744 const size_t name_count = m_name_to_value_set.size ();
745 m_bucket_table.resize
746 (std::pow (2, std::ceil (std::log2 (name_count * 4 / 3))));
747 m_hash_table.reserve (name_count);
748 m_name_table_string_offs.reserve (name_count);
749 m_name_table_entry_offs.reserve (name_count);
750
751 /* Map each hash of symbol to its name and value. */
752 struct hash_it_pair
753 {
754 uint32_t hash;
755 decltype (m_name_to_value_set)::const_iterator it;
756 };
757 std::vector<std::forward_list<hash_it_pair>> bucket_hash;
758 bucket_hash.resize (m_bucket_table.size ());
759 for (decltype (m_name_to_value_set)::const_iterator it
760 = m_name_to_value_set.cbegin ();
761 it != m_name_to_value_set.cend ();
762 ++it)
763 {
764 const char *const name = it->first.c_str ();
765 const uint32_t hash = dwarf5_djb_hash (name);
766 hash_it_pair hashitpair;
767 hashitpair.hash = hash;
768 hashitpair.it = it;
769 auto &slot = bucket_hash[hash % bucket_hash.size()];
770 slot.push_front (std::move (hashitpair));
771 }
772 for (size_t bucket_ix = 0; bucket_ix < bucket_hash.size (); ++bucket_ix)
773 {
774 const std::forward_list<hash_it_pair> &hashitlist
775 = bucket_hash[bucket_ix];
776 if (hashitlist.empty ())
777 continue;
778 uint32_t &bucket_slot = m_bucket_table[bucket_ix];
779 /* The hashes array is indexed starting at 1. */
780 store_unsigned_integer (reinterpret_cast<gdb_byte *> (&bucket_slot),
781 sizeof (bucket_slot), m_dwarf5_byte_order,
782 m_hash_table.size () + 1);
783 for (const hash_it_pair &hashitpair : hashitlist)
784 {
785 m_hash_table.push_back (0);
786 store_unsigned_integer (reinterpret_cast<gdb_byte *>
787 (&m_hash_table.back ()),
788 sizeof (m_hash_table.back ()),
789 m_dwarf5_byte_order, hashitpair.hash);
790 const c_str_view &name = hashitpair.it->first;
791 const std::set<symbol_value> &value_set = hashitpair.it->second;
792 m_name_table_string_offs.push_back_reorder
793 (m_debugstrlookup.lookup (name.c_str ()));
794 m_name_table_entry_offs.push_back_reorder (m_entry_pool.size ());
795 gdb_assert (!value_set.empty ());
796 for (const symbol_value &value : value_set)
797 {
798 int &idx = m_indexkey_to_idx[index_key (value.dwarf_tag,
799 value.is_static,
800 value.kind)];
801 if (idx == 0)
802 {
803 idx = m_idx_next++;
804 m_abbrev_table.append_unsigned_leb128 (idx);
805 m_abbrev_table.append_unsigned_leb128 (value.dwarf_tag);
806 m_abbrev_table.append_unsigned_leb128
807 (value.kind == unit_kind::cu ? DW_IDX_compile_unit
808 : DW_IDX_type_unit);
809 m_abbrev_table.append_unsigned_leb128 (DW_FORM_udata);
810 m_abbrev_table.append_unsigned_leb128 (value.is_static
811 ? DW_IDX_GNU_internal
812 : DW_IDX_GNU_external);
813 m_abbrev_table.append_unsigned_leb128 (DW_FORM_flag_present);
814
815 /* Terminate attributes list. */
816 m_abbrev_table.append_unsigned_leb128 (0);
817 m_abbrev_table.append_unsigned_leb128 (0);
818 }
819
820 m_entry_pool.append_unsigned_leb128 (idx);
821 m_entry_pool.append_unsigned_leb128 (value.cu_index);
822 }
823
824 /* Terminate the list of CUs. */
825 m_entry_pool.append_unsigned_leb128 (0);
826 }
827 }
828 gdb_assert (m_hash_table.size () == name_count);
829
830 /* Terminate tags list. */
831 m_abbrev_table.append_unsigned_leb128 (0);
832 }
833
834 /* Return .debug_names bucket count. This must be called only after
835 calling the build method. */
836 uint32_t bucket_count () const
837 {
838 /* Verify the build method has been already called. */
839 gdb_assert (!m_abbrev_table.empty ());
840 const uint32_t retval = m_bucket_table.size ();
841
842 /* Check for overflow. */
843 gdb_assert (retval == m_bucket_table.size ());
844 return retval;
845 }
846
847 /* Return .debug_names names count. This must be called only after
848 calling the build method. */
849 uint32_t name_count () const
850 {
851 /* Verify the build method has been already called. */
852 gdb_assert (!m_abbrev_table.empty ());
853 const uint32_t retval = m_hash_table.size ();
854
855 /* Check for overflow. */
856 gdb_assert (retval == m_hash_table.size ());
857 return retval;
858 }
859
860 /* Return number of bytes of .debug_names abbreviation table. This
861 must be called only after calling the build method. */
862 uint32_t abbrev_table_bytes () const
863 {
864 gdb_assert (!m_abbrev_table.empty ());
865 return m_abbrev_table.size ();
866 }
867
868 /* Recurse into all "included" dependencies and store their symbols
869 as if they appeared in this psymtab. */
870 void recursively_write_psymbols
871 (struct objfile *objfile,
891813be 872 dwarf2_psymtab *psymtab,
cd4fb1b2
SM
873 std::unordered_set<partial_symbol *> &psyms_seen,
874 int cu_index)
875 {
876 for (int i = 0; i < psymtab->number_of_dependencies; ++i)
877 if (psymtab->dependencies[i]->user != NULL)
891813be
TT
878 recursively_write_psymbols
879 (objfile, (dwarf2_psymtab *) psymtab->dependencies[i], psyms_seen, cu_index);
cd4fb1b2
SM
880
881 write_psymbols (psyms_seen,
d320c2b5
TT
882 (objfile->partial_symtabs->global_psymbols.data ()
883 + psymtab->globals_offset),
cd4fb1b2
SM
884 psymtab->n_global_syms, cu_index, false, unit_kind::cu);
885 write_psymbols (psyms_seen,
d320c2b5
TT
886 (objfile->partial_symtabs->static_psymbols.data ()
887 + psymtab->statics_offset),
cd4fb1b2
SM
888 psymtab->n_static_syms, cu_index, true, unit_kind::cu);
889 }
890
891 /* Return number of bytes the .debug_names section will have. This
892 must be called only after calling the build method. */
893 size_t bytes () const
894 {
895 /* Verify the build method has been already called. */
896 gdb_assert (!m_abbrev_table.empty ());
897 size_t expected_bytes = 0;
898 expected_bytes += m_bucket_table.size () * sizeof (m_bucket_table[0]);
899 expected_bytes += m_hash_table.size () * sizeof (m_hash_table[0]);
900 expected_bytes += m_name_table_string_offs.bytes ();
901 expected_bytes += m_name_table_entry_offs.bytes ();
902 expected_bytes += m_abbrev_table.size ();
903 expected_bytes += m_entry_pool.size ();
904 return expected_bytes;
905 }
906
907 /* Write .debug_names to FILE_NAMES and .debug_str addition to
908 FILE_STR. This must be called only after calling the build
909 method. */
910 void file_write (FILE *file_names, FILE *file_str) const
911 {
912 /* Verify the build method has been already called. */
913 gdb_assert (!m_abbrev_table.empty ());
914 ::file_write (file_names, m_bucket_table);
915 ::file_write (file_names, m_hash_table);
916 m_name_table_string_offs.file_write (file_names);
917 m_name_table_entry_offs.file_write (file_names);
918 m_abbrev_table.file_write (file_names);
919 m_entry_pool.file_write (file_names);
920 m_debugstrlookup.file_write (file_str);
921 }
922
923 /* A helper user data for write_one_signatured_type. */
924 class write_one_signatured_type_data
925 {
926 public:
927 write_one_signatured_type_data (debug_names &nametable_,
928 signatured_type_index_data &&info_)
929 : nametable (nametable_), info (std::move (info_))
930 {}
931 debug_names &nametable;
932 struct signatured_type_index_data info;
933 };
934
935 /* A helper function to pass write_one_signatured_type to
936 htab_traverse_noresize. */
937 static int
938 write_one_signatured_type (void **slot, void *d)
939 {
940 write_one_signatured_type_data *data = (write_one_signatured_type_data *) d;
941 struct signatured_type_index_data *info = &data->info;
942 struct signatured_type *entry = (struct signatured_type *) *slot;
943
944 data->nametable.write_one_signatured_type (entry, info);
945
946 return 1;
947 }
948
949private:
950
951 /* Storage for symbol names mapping them to their .debug_str section
952 offsets. */
953 class debug_str_lookup
954 {
955 public:
956
30baf67b 957 /* Object constructor to be called for current DWARF2_PER_OBJFILE.
cd4fb1b2
SM
958 All .debug_str section strings are automatically stored. */
959 debug_str_lookup (struct dwarf2_per_objfile *dwarf2_per_objfile)
960 : m_abfd (dwarf2_per_objfile->objfile->obfd),
961 m_dwarf2_per_objfile (dwarf2_per_objfile)
962 {
963 dwarf2_read_section (dwarf2_per_objfile->objfile,
964 &dwarf2_per_objfile->str);
965 if (dwarf2_per_objfile->str.buffer == NULL)
966 return;
967 for (const gdb_byte *data = dwarf2_per_objfile->str.buffer;
968 data < (dwarf2_per_objfile->str.buffer
969 + dwarf2_per_objfile->str.size);)
970 {
971 const char *const s = reinterpret_cast<const char *> (data);
972 const auto insertpair
973 = m_str_table.emplace (c_str_view (s),
974 data - dwarf2_per_objfile->str.buffer);
975 if (!insertpair.second)
b98664d3 976 complaint (_("Duplicate string \"%s\" in "
cd4fb1b2
SM
977 ".debug_str section [in module %s]"),
978 s, bfd_get_filename (m_abfd));
979 data += strlen (s) + 1;
980 }
981 }
982
983 /* Return offset of symbol name S in the .debug_str section. Add
984 such symbol to the section's end if it does not exist there
985 yet. */
986 size_t lookup (const char *s)
987 {
988 const auto it = m_str_table.find (c_str_view (s));
989 if (it != m_str_table.end ())
990 return it->second;
991 const size_t offset = (m_dwarf2_per_objfile->str.size
992 + m_str_add_buf.size ());
993 m_str_table.emplace (c_str_view (s), offset);
994 m_str_add_buf.append_cstr0 (s);
995 return offset;
996 }
997
998 /* Append the end of the .debug_str section to FILE. */
999 void file_write (FILE *file) const
1000 {
1001 m_str_add_buf.file_write (file);
1002 }
1003
1004 private:
1005 std::unordered_map<c_str_view, size_t, c_str_view_hasher> m_str_table;
1006 bfd *const m_abfd;
1007 struct dwarf2_per_objfile *m_dwarf2_per_objfile;
1008
1009 /* Data to add at the end of .debug_str for new needed symbol names. */
1010 data_buf m_str_add_buf;
1011 };
1012
1013 /* Container to map used DWARF tags to their .debug_names abbreviation
1014 tags. */
1015 class index_key
1016 {
1017 public:
1018 index_key (int dwarf_tag_, bool is_static_, unit_kind kind_)
1019 : dwarf_tag (dwarf_tag_), is_static (is_static_), kind (kind_)
1020 {
1021 }
1022
1023 bool
1024 operator== (const index_key &other) const
1025 {
1026 return (dwarf_tag == other.dwarf_tag && is_static == other.is_static
1027 && kind == other.kind);
1028 }
1029
1030 const int dwarf_tag;
1031 const bool is_static;
1032 const unit_kind kind;
1033 };
1034
1035 /* Provide std::unordered_map::hasher for index_key. */
1036 class index_key_hasher
1037 {
1038 public:
1039 size_t
1040 operator () (const index_key &key) const
1041 {
1042 return (std::hash<int>() (key.dwarf_tag) << 1) | key.is_static;
1043 }
1044 };
1045
1046 /* Parameters of one symbol entry. */
1047 class symbol_value
1048 {
1049 public:
1050 const int dwarf_tag, cu_index;
1051 const bool is_static;
1052 const unit_kind kind;
1053
1054 symbol_value (int dwarf_tag_, int cu_index_, bool is_static_,
1055 unit_kind kind_)
1056 : dwarf_tag (dwarf_tag_), cu_index (cu_index_), is_static (is_static_),
1057 kind (kind_)
1058 {}
1059
1060 bool
1061 operator< (const symbol_value &other) const
1062 {
1063#define X(n) \
1064 do \
1065 { \
1066 if (n < other.n) \
1067 return true; \
1068 if (n > other.n) \
1069 return false; \
1070 } \
1071 while (0)
1072 X (dwarf_tag);
1073 X (is_static);
1074 X (kind);
1075 X (cu_index);
1076#undef X
1077 return false;
1078 }
1079 };
1080
1081 /* Abstract base class to unify DWARF-32 and DWARF-64 name table
1082 output. */
1083 class offset_vec
1084 {
1085 protected:
1086 const bfd_endian dwarf5_byte_order;
1087 public:
1088 explicit offset_vec (bfd_endian dwarf5_byte_order_)
1089 : dwarf5_byte_order (dwarf5_byte_order_)
1090 {}
1091
1092 /* Call std::vector::reserve for NELEM elements. */
1093 virtual void reserve (size_t nelem) = 0;
1094
1095 /* Call std::vector::push_back with store_unsigned_integer byte
1096 reordering for ELEM. */
1097 virtual void push_back_reorder (size_t elem) = 0;
1098
1099 /* Return expected output size in bytes. */
1100 virtual size_t bytes () const = 0;
1101
1102 /* Write name table to FILE. */
1103 virtual void file_write (FILE *file) const = 0;
1104 };
1105
1106 /* Template to unify DWARF-32 and DWARF-64 output. */
1107 template<typename OffsetSize>
1108 class offset_vec_tmpl : public offset_vec
1109 {
1110 public:
1111 explicit offset_vec_tmpl (bfd_endian dwarf5_byte_order_)
1112 : offset_vec (dwarf5_byte_order_)
1113 {}
1114
1115 /* Implement offset_vec::reserve. */
1116 void reserve (size_t nelem) override
1117 {
1118 m_vec.reserve (nelem);
1119 }
1120
1121 /* Implement offset_vec::push_back_reorder. */
1122 void push_back_reorder (size_t elem) override
1123 {
1124 m_vec.push_back (elem);
1125 /* Check for overflow. */
1126 gdb_assert (m_vec.back () == elem);
1127 store_unsigned_integer (reinterpret_cast<gdb_byte *> (&m_vec.back ()),
1128 sizeof (m_vec.back ()), dwarf5_byte_order, elem);
1129 }
1130
1131 /* Implement offset_vec::bytes. */
1132 size_t bytes () const override
1133 {
1134 return m_vec.size () * sizeof (m_vec[0]);
1135 }
1136
1137 /* Implement offset_vec::file_write. */
1138 void file_write (FILE *file) const override
1139 {
1140 ::file_write (file, m_vec);
1141 }
1142
1143 private:
1144 std::vector<OffsetSize> m_vec;
1145 };
1146
1147 /* Base class to unify DWARF-32 and DWARF-64 .debug_names output
1148 respecting name table width. */
1149 class dwarf
1150 {
1151 public:
1152 offset_vec &name_table_string_offs, &name_table_entry_offs;
1153
1154 dwarf (offset_vec &name_table_string_offs_,
1155 offset_vec &name_table_entry_offs_)
1156 : name_table_string_offs (name_table_string_offs_),
1157 name_table_entry_offs (name_table_entry_offs_)
1158 {
1159 }
1160 };
1161
1162 /* Template to unify DWARF-32 and DWARF-64 .debug_names output
1163 respecting name table width. */
1164 template<typename OffsetSize>
1165 class dwarf_tmpl : public dwarf
1166 {
1167 public:
1168 explicit dwarf_tmpl (bfd_endian dwarf5_byte_order_)
1169 : dwarf (m_name_table_string_offs, m_name_table_entry_offs),
1170 m_name_table_string_offs (dwarf5_byte_order_),
1171 m_name_table_entry_offs (dwarf5_byte_order_)
1172 {}
1173
1174 private:
1175 offset_vec_tmpl<OffsetSize> m_name_table_string_offs;
1176 offset_vec_tmpl<OffsetSize> m_name_table_entry_offs;
1177 };
1178
1179 /* Try to reconstruct original DWARF tag for given partial_symbol.
1180 This function is not DWARF-5 compliant but it is sufficient for
1181 GDB as a DWARF-5 index consumer. */
1182 static int psymbol_tag (const struct partial_symbol *psym)
1183 {
8a6d4234
TT
1184 domain_enum domain = psym->domain;
1185 enum address_class aclass = psym->aclass;
cd4fb1b2
SM
1186
1187 switch (domain)
1188 {
1189 case VAR_DOMAIN:
1190 switch (aclass)
1191 {
1192 case LOC_BLOCK:
1193 return DW_TAG_subprogram;
1194 case LOC_TYPEDEF:
1195 return DW_TAG_typedef;
1196 case LOC_COMPUTED:
1197 case LOC_CONST_BYTES:
1198 case LOC_OPTIMIZED_OUT:
1199 case LOC_STATIC:
1200 return DW_TAG_variable;
1201 case LOC_CONST:
1202 /* Note: It's currently impossible to recognize psyms as enum values
1203 short of reading the type info. For now punt. */
1204 return DW_TAG_variable;
1205 default:
1206 /* There are other LOC_FOO values that one might want to classify
1207 as variables, but dwarf2read.c doesn't currently use them. */
1208 return DW_TAG_variable;
1209 }
1210 case STRUCT_DOMAIN:
1211 return DW_TAG_structure_type;
1212 default:
1213 return 0;
1214 }
1215 }
1216
1217 /* Call insert for all partial symbols and mark them in PSYMS_SEEN. */
1218 void write_psymbols (std::unordered_set<partial_symbol *> &psyms_seen,
1219 struct partial_symbol **psymp, int count, int cu_index,
1220 bool is_static, unit_kind kind)
1221 {
1222 for (; count-- > 0; ++psymp)
1223 {
1224 struct partial_symbol *psym = *psymp;
1225
cd4fb1b2
SM
1226 /* Only add a given psymbol once. */
1227 if (psyms_seen.insert (psym).second)
1228 insert (psym, cu_index, is_static, kind);
1229 }
1230 }
1231
1232 /* A helper function that writes a single signatured_type
1233 to a debug_names. */
1234 void
1235 write_one_signatured_type (struct signatured_type *entry,
1236 struct signatured_type_index_data *info)
1237 {
891813be 1238 dwarf2_psymtab *psymtab = entry->per_cu.v.psymtab;
cd4fb1b2
SM
1239
1240 write_psymbols (info->psyms_seen,
d320c2b5 1241 (info->objfile->partial_symtabs->global_psymbols.data ()
b4be9bfd 1242 + psymtab->globals_offset),
cd4fb1b2
SM
1243 psymtab->n_global_syms, info->cu_index, false,
1244 unit_kind::tu);
1245 write_psymbols (info->psyms_seen,
d320c2b5 1246 (info->objfile->partial_symtabs->static_psymbols.data ()
b4be9bfd 1247 + psymtab->statics_offset),
cd4fb1b2
SM
1248 psymtab->n_static_syms, info->cu_index, true,
1249 unit_kind::tu);
1250
1251 info->types_list.append_uint (dwarf5_offset_size (), m_dwarf5_byte_order,
1252 to_underlying (entry->per_cu.sect_off));
1253
1254 ++info->cu_index;
1255 }
1256
1257 /* Store value of each symbol. */
1258 std::unordered_map<c_str_view, std::set<symbol_value>, c_str_view_hasher>
1259 m_name_to_value_set;
1260
1261 /* Tables of DWARF-5 .debug_names. They are in object file byte
1262 order. */
1263 std::vector<uint32_t> m_bucket_table;
1264 std::vector<uint32_t> m_hash_table;
1265
1266 const bfd_endian m_dwarf5_byte_order;
1267 dwarf_tmpl<uint32_t> m_dwarf32;
1268 dwarf_tmpl<uint64_t> m_dwarf64;
1269 dwarf &m_dwarf;
1270 offset_vec &m_name_table_string_offs, &m_name_table_entry_offs;
1271 debug_str_lookup m_debugstrlookup;
1272
1273 /* Map each used .debug_names abbreviation tag parameter to its
1274 index value. */
1275 std::unordered_map<index_key, int, index_key_hasher> m_indexkey_to_idx;
1276
1277 /* Next unused .debug_names abbreviation tag for
1278 m_indexkey_to_idx. */
1279 int m_idx_next = 1;
1280
1281 /* .debug_names abbreviation table. */
1282 data_buf m_abbrev_table;
1283
1284 /* .debug_names entry pool. */
1285 data_buf m_entry_pool;
3b00ef10
TT
1286
1287 /* Temporary storage for Ada names. */
1288 auto_obstack m_string_obstack;
cd4fb1b2
SM
1289};
1290
1291/* Return iff any of the needed offsets does not fit into 32-bit
1292 .debug_names section. */
1293
1294static bool
1295check_dwarf64_offsets (struct dwarf2_per_objfile *dwarf2_per_objfile)
1296{
b76e467d 1297 for (dwarf2_per_cu_data *per_cu : dwarf2_per_objfile->all_comp_units)
cd4fb1b2 1298 {
b76e467d 1299 if (to_underlying (per_cu->sect_off) >= (static_cast<uint64_t> (1) << 32))
cd4fb1b2
SM
1300 return true;
1301 }
b2bdb8cf 1302 for (const signatured_type *sigtype : dwarf2_per_objfile->all_type_units)
cd4fb1b2 1303 {
b2bdb8cf 1304 const dwarf2_per_cu_data &per_cu = sigtype->per_cu;
cd4fb1b2
SM
1305
1306 if (to_underlying (per_cu.sect_off) >= (static_cast<uint64_t> (1) << 32))
1307 return true;
1308 }
1309 return false;
1310}
1311
1312/* The psyms_seen set is potentially going to be largish (~40k
1313 elements when indexing a -g3 build of GDB itself). Estimate the
1314 number of elements in order to avoid too many rehashes, which
1315 require rebuilding buckets and thus many trips to
1316 malloc/free. */
1317
1318static size_t
1319psyms_seen_size (struct dwarf2_per_objfile *dwarf2_per_objfile)
1320{
1321 size_t psyms_count = 0;
b76e467d 1322 for (dwarf2_per_cu_data *per_cu : dwarf2_per_objfile->all_comp_units)
cd4fb1b2 1323 {
891813be 1324 dwarf2_psymtab *psymtab = per_cu->v.psymtab;
cd4fb1b2
SM
1325
1326 if (psymtab != NULL && psymtab->user == NULL)
1327 recursively_count_psymbols (psymtab, psyms_count);
1328 }
1329 /* Generating an index for gdb itself shows a ratio of
1330 TOTAL_SEEN_SYMS/UNIQUE_SYMS or ~5. 4 seems like a good bet. */
1331 return psyms_count / 4;
1332}
1333
c4973306
SM
1334/* Assert that FILE's size is EXPECTED_SIZE. Assumes file's seek
1335 position is at the end of the file. */
cd4fb1b2 1336
c4973306
SM
1337static void
1338assert_file_size (FILE *file, size_t expected_size)
1339{
1340 const auto file_size = ftell (file);
1341 if (file_size == -1)
1342 perror_with_name (("ftell"));
1343 gdb_assert (file_size == expected_size);
1344}
1345
1346/* Write a gdb index file to OUT_FILE from all the sections passed as
1347 arguments. */
1348
1349static void
1350write_gdbindex_1 (FILE *out_file,
1351 const data_buf &cu_list,
1352 const data_buf &types_cu_list,
1353 const data_buf &addr_vec,
1354 const data_buf &symtab_vec,
1355 const data_buf &constant_pool)
1356{
1357 data_buf contents;
1358 const offset_type size_of_header = 6 * sizeof (offset_type);
1359 offset_type total_len = size_of_header;
1360
1361 /* The version number. */
1362 contents.append_data (MAYBE_SWAP (8));
1363
1364 /* The offset of the CU list from the start of the file. */
1365 contents.append_data (MAYBE_SWAP (total_len));
1366 total_len += cu_list.size ();
1367
1368 /* The offset of the types CU list from the start of the file. */
1369 contents.append_data (MAYBE_SWAP (total_len));
1370 total_len += types_cu_list.size ();
1371
1372 /* The offset of the address table from the start of the file. */
1373 contents.append_data (MAYBE_SWAP (total_len));
1374 total_len += addr_vec.size ();
1375
1376 /* The offset of the symbol table from the start of the file. */
1377 contents.append_data (MAYBE_SWAP (total_len));
1378 total_len += symtab_vec.size ();
1379
1380 /* The offset of the constant pool from the start of the file. */
1381 contents.append_data (MAYBE_SWAP (total_len));
1382 total_len += constant_pool.size ();
1383
1384 gdb_assert (contents.size () == size_of_header);
1385
1386 contents.file_write (out_file);
1387 cu_list.file_write (out_file);
1388 types_cu_list.file_write (out_file);
1389 addr_vec.file_write (out_file);
1390 symtab_vec.file_write (out_file);
1391 constant_pool.file_write (out_file);
1392
1393 assert_file_size (out_file, total_len);
1394}
1395
1396/* Write contents of a .gdb_index section for OBJFILE into OUT_FILE.
1397 If OBJFILE has an associated dwz file, write contents of a .gdb_index
1398 section for that dwz file into DWZ_OUT_FILE. If OBJFILE does not have an
1399 associated dwz file, DWZ_OUT_FILE must be NULL. */
1400
1401static void
1402write_gdbindex (struct dwarf2_per_objfile *dwarf2_per_objfile, FILE *out_file,
1403 FILE *dwz_out_file)
cd4fb1b2
SM
1404{
1405 struct objfile *objfile = dwarf2_per_objfile->objfile;
1406 mapped_symtab symtab;
c4973306
SM
1407 data_buf objfile_cu_list;
1408 data_buf dwz_cu_list;
cd4fb1b2
SM
1409
1410 /* While we're scanning CU's create a table that maps a psymtab pointer
1411 (which is what addrmap records) to its index (which is what is recorded
1412 in the index file). This will later be needed to write the address
1413 table. */
1414 psym_index_map cu_index_htab;
b76e467d 1415 cu_index_htab.reserve (dwarf2_per_objfile->all_comp_units.size ());
cd4fb1b2
SM
1416
1417 /* The CU list is already sorted, so we don't need to do additional
1418 work here. Also, the debug_types entries do not appear in
1419 all_comp_units, but only in their own hash table. */
1420
1421 std::unordered_set<partial_symbol *> psyms_seen
1422 (psyms_seen_size (dwarf2_per_objfile));
b76e467d 1423 for (int i = 0; i < dwarf2_per_objfile->all_comp_units.size (); ++i)
cd4fb1b2
SM
1424 {
1425 struct dwarf2_per_cu_data *per_cu
1426 = dwarf2_per_objfile->all_comp_units[i];
891813be 1427 dwarf2_psymtab *psymtab = per_cu->v.psymtab;
cd4fb1b2
SM
1428
1429 /* CU of a shared file from 'dwz -m' may be unused by this main file.
1430 It may be referenced from a local scope but in such case it does not
1431 need to be present in .gdb_index. */
1432 if (psymtab == NULL)
1433 continue;
1434
1435 if (psymtab->user == NULL)
1436 recursively_write_psymbols (objfile, psymtab, &symtab,
1437 psyms_seen, i);
1438
1439 const auto insertpair = cu_index_htab.emplace (psymtab, i);
1440 gdb_assert (insertpair.second);
1441
c4973306
SM
1442 /* The all_comp_units list contains CUs read from the objfile as well as
1443 from the eventual dwz file. We need to place the entry in the
1444 corresponding index. */
1445 data_buf &cu_list = per_cu->is_dwz ? dwz_cu_list : objfile_cu_list;
cd4fb1b2
SM
1446 cu_list.append_uint (8, BFD_ENDIAN_LITTLE,
1447 to_underlying (per_cu->sect_off));
1448 cu_list.append_uint (8, BFD_ENDIAN_LITTLE, per_cu->length);
1449 }
1450
1451 /* Dump the address map. */
1452 data_buf addr_vec;
1453 write_address_map (objfile, addr_vec, cu_index_htab);
1454
1455 /* Write out the .debug_type entries, if any. */
1456 data_buf types_cu_list;
1457 if (dwarf2_per_objfile->signatured_types)
1458 {
1459 signatured_type_index_data sig_data (types_cu_list,
1460 psyms_seen);
1461
1462 sig_data.objfile = objfile;
1463 sig_data.symtab = &symtab;
b76e467d 1464 sig_data.cu_index = dwarf2_per_objfile->all_comp_units.size ();
cd4fb1b2
SM
1465 htab_traverse_noresize (dwarf2_per_objfile->signatured_types,
1466 write_one_signatured_type, &sig_data);
1467 }
1468
1469 /* Now that we've processed all symbols we can shrink their cu_indices
1470 lists. */
1471 uniquify_cu_indices (&symtab);
1472
1473 data_buf symtab_vec, constant_pool;
1474 write_hash_table (&symtab, symtab_vec, constant_pool);
1475
c4973306
SM
1476 write_gdbindex_1(out_file, objfile_cu_list, types_cu_list, addr_vec,
1477 symtab_vec, constant_pool);
cd4fb1b2 1478
c4973306
SM
1479 if (dwz_out_file != NULL)
1480 write_gdbindex_1 (dwz_out_file, dwz_cu_list, {}, {}, {}, {});
1481 else
1482 gdb_assert (dwz_cu_list.empty ());
cd4fb1b2
SM
1483}
1484
1485/* DWARF-5 augmentation string for GDB's DW_IDX_GNU_* extension. */
1486static const gdb_byte dwarf5_gdb_augmentation[] = { 'G', 'D', 'B', 0 };
1487
1488/* Write a new .debug_names section for OBJFILE into OUT_FILE, write
1489 needed addition to .debug_str section to OUT_FILE_STR. Return how
1490 many bytes were expected to be written into OUT_FILE. */
1491
c4973306 1492static void
cd4fb1b2
SM
1493write_debug_names (struct dwarf2_per_objfile *dwarf2_per_objfile,
1494 FILE *out_file, FILE *out_file_str)
1495{
1496 const bool dwarf5_is_dwarf64 = check_dwarf64_offsets (dwarf2_per_objfile);
1497 struct objfile *objfile = dwarf2_per_objfile->objfile;
1498 const enum bfd_endian dwarf5_byte_order
1499 = gdbarch_byte_order (get_objfile_arch (objfile));
1500
1501 /* The CU list is already sorted, so we don't need to do additional
1502 work here. Also, the debug_types entries do not appear in
1503 all_comp_units, but only in their own hash table. */
1504 data_buf cu_list;
1505 debug_names nametable (dwarf2_per_objfile, dwarf5_is_dwarf64,
1506 dwarf5_byte_order);
1507 std::unordered_set<partial_symbol *>
1508 psyms_seen (psyms_seen_size (dwarf2_per_objfile));
b76e467d 1509 for (int i = 0; i < dwarf2_per_objfile->all_comp_units.size (); ++i)
cd4fb1b2
SM
1510 {
1511 const dwarf2_per_cu_data *per_cu = dwarf2_per_objfile->all_comp_units[i];
891813be 1512 dwarf2_psymtab *psymtab = per_cu->v.psymtab;
cd4fb1b2
SM
1513
1514 /* CU of a shared file from 'dwz -m' may be unused by this main
1515 file. It may be referenced from a local scope but in such
1516 case it does not need to be present in .debug_names. */
1517 if (psymtab == NULL)
1518 continue;
1519
1520 if (psymtab->user == NULL)
1521 nametable.recursively_write_psymbols (objfile, psymtab, psyms_seen, i);
1522
1523 cu_list.append_uint (nametable.dwarf5_offset_size (), dwarf5_byte_order,
1524 to_underlying (per_cu->sect_off));
1525 }
1526
1527 /* Write out the .debug_type entries, if any. */
1528 data_buf types_cu_list;
1529 if (dwarf2_per_objfile->signatured_types)
1530 {
1531 debug_names::write_one_signatured_type_data sig_data (nametable,
1532 signatured_type_index_data (types_cu_list, psyms_seen));
1533
1534 sig_data.info.objfile = objfile;
1535 /* It is used only for gdb_index. */
1536 sig_data.info.symtab = nullptr;
1537 sig_data.info.cu_index = 0;
1538 htab_traverse_noresize (dwarf2_per_objfile->signatured_types,
1539 debug_names::write_one_signatured_type,
1540 &sig_data);
1541 }
1542
1543 nametable.build ();
1544
1545 /* No addr_vec - DWARF-5 uses .debug_aranges generated by GCC. */
1546
1547 const offset_type bytes_of_header
1548 = ((dwarf5_is_dwarf64 ? 12 : 4)
1549 + 2 + 2 + 7 * 4
1550 + sizeof (dwarf5_gdb_augmentation));
1551 size_t expected_bytes = 0;
1552 expected_bytes += bytes_of_header;
1553 expected_bytes += cu_list.size ();
1554 expected_bytes += types_cu_list.size ();
1555 expected_bytes += nametable.bytes ();
1556 data_buf header;
1557
1558 if (!dwarf5_is_dwarf64)
1559 {
1560 const uint64_t size64 = expected_bytes - 4;
1561 gdb_assert (size64 < 0xfffffff0);
1562 header.append_uint (4, dwarf5_byte_order, size64);
1563 }
1564 else
1565 {
1566 header.append_uint (4, dwarf5_byte_order, 0xffffffff);
1567 header.append_uint (8, dwarf5_byte_order, expected_bytes - 12);
1568 }
1569
1570 /* The version number. */
1571 header.append_uint (2, dwarf5_byte_order, 5);
1572
1573 /* Padding. */
1574 header.append_uint (2, dwarf5_byte_order, 0);
1575
1576 /* comp_unit_count - The number of CUs in the CU list. */
b76e467d
SM
1577 header.append_uint (4, dwarf5_byte_order,
1578 dwarf2_per_objfile->all_comp_units.size ());
cd4fb1b2
SM
1579
1580 /* local_type_unit_count - The number of TUs in the local TU
1581 list. */
b2bdb8cf
SM
1582 header.append_uint (4, dwarf5_byte_order,
1583 dwarf2_per_objfile->all_type_units.size ());
cd4fb1b2
SM
1584
1585 /* foreign_type_unit_count - The number of TUs in the foreign TU
1586 list. */
1587 header.append_uint (4, dwarf5_byte_order, 0);
1588
1589 /* bucket_count - The number of hash buckets in the hash lookup
1590 table. */
1591 header.append_uint (4, dwarf5_byte_order, nametable.bucket_count ());
1592
1593 /* name_count - The number of unique names in the index. */
1594 header.append_uint (4, dwarf5_byte_order, nametable.name_count ());
1595
1596 /* abbrev_table_size - The size in bytes of the abbreviations
1597 table. */
1598 header.append_uint (4, dwarf5_byte_order, nametable.abbrev_table_bytes ());
1599
1600 /* augmentation_string_size - The size in bytes of the augmentation
1601 string. This value is rounded up to a multiple of 4. */
1602 static_assert (sizeof (dwarf5_gdb_augmentation) % 4 == 0, "");
1603 header.append_uint (4, dwarf5_byte_order, sizeof (dwarf5_gdb_augmentation));
1604 header.append_data (dwarf5_gdb_augmentation);
1605
1606 gdb_assert (header.size () == bytes_of_header);
1607
1608 header.file_write (out_file);
1609 cu_list.file_write (out_file);
1610 types_cu_list.file_write (out_file);
1611 nametable.file_write (out_file, out_file_str);
1612
c4973306 1613 assert_file_size (out_file, expected_bytes);
cd4fb1b2
SM
1614}
1615
c4973306 1616/* This represents an index file being written (work-in-progress).
cd4fb1b2 1617
c4973306
SM
1618 The data is initially written to a temporary file. When the finalize method
1619 is called, the file is closed and moved to its final location.
1620
1621 On failure (if this object is being destroyed with having called finalize),
1622 the temporary file is closed and deleted. */
1623
1624struct index_wip_file
cd4fb1b2 1625{
c4973306
SM
1626 index_wip_file (const char *dir, const char *basename,
1627 const char *suffix)
1628 {
1629 filename = (std::string (dir) + SLASH_STRING + basename
1630 + suffix);
1631
1632 filename_temp = make_temp_filename (filename);
1633
1634 scoped_fd out_file_fd (gdb_mkostemp_cloexec (filename_temp.data (),
1635 O_BINARY));
1636 if (out_file_fd.get () == -1)
1637 perror_with_name (("mkstemp"));
1638
1639 out_file = out_file_fd.to_file ("wb");
1640
1641 if (out_file == nullptr)
1642 error (_("Can't open `%s' for writing"), filename_temp.data ());
1643
1644 unlink_file.emplace (filename_temp.data ());
1645 }
1646
1647 void finalize ()
1648 {
1649 /* We want to keep the file. */
1650 unlink_file->keep ();
1651
1652 /* Close and move the str file in place. */
1653 unlink_file.reset ();
1654 if (rename (filename_temp.data (), filename.c_str ()) != 0)
1655 perror_with_name (("rename"));
1656 }
1657
1658 std::string filename;
1659 gdb::char_vector filename_temp;
1660
1661 /* Order matters here; we want FILE to be closed before
1662 FILENAME_TEMP is unlinked, because on MS-Windows one cannot
1663 delete a file that is still open. So, we wrap the unlinker in an
1664 optional and emplace it once we know the file name. */
1665 gdb::optional<gdb::unlinker> unlink_file;
1666
1667 gdb_file_up out_file;
1668};
cd4fb1b2 1669
87d6a7aa 1670/* See dwarf-index-write.h. */
cd4fb1b2 1671
87d6a7aa 1672void
cd4fb1b2 1673write_psymtabs_to_index (struct dwarf2_per_objfile *dwarf2_per_objfile,
87d6a7aa 1674 const char *dir, const char *basename,
c4973306 1675 const char *dwz_basename,
cd4fb1b2
SM
1676 dw_index_kind index_kind)
1677{
1678 struct objfile *objfile = dwarf2_per_objfile->objfile;
1679
1680 if (dwarf2_per_objfile->using_index)
1681 error (_("Cannot use an index to create the index"));
1682
fd5866f6 1683 if (dwarf2_per_objfile->types.size () > 1)
cd4fb1b2
SM
1684 error (_("Cannot make an index when the file has multiple .debug_types sections"));
1685
d320c2b5
TT
1686 if (!objfile->partial_symtabs->psymtabs
1687 || !objfile->partial_symtabs->psymtabs_addrmap)
cd4fb1b2
SM
1688 return;
1689
1690 struct stat st;
1691 if (stat (objfile_name (objfile), &st) < 0)
1692 perror_with_name (objfile_name (objfile));
1693
c4973306
SM
1694 const char *index_suffix = (index_kind == dw_index_kind::DEBUG_NAMES
1695 ? INDEX5_SUFFIX : INDEX4_SUFFIX);
cd4fb1b2 1696
c4973306
SM
1697 index_wip_file objfile_index_wip (dir, basename, index_suffix);
1698 gdb::optional<index_wip_file> dwz_index_wip;
cd4fb1b2 1699
c4973306
SM
1700 if (dwz_basename != NULL)
1701 dwz_index_wip.emplace (dir, dwz_basename, index_suffix);
cd4fb1b2
SM
1702
1703 if (index_kind == dw_index_kind::DEBUG_NAMES)
1704 {
c4973306
SM
1705 index_wip_file str_wip_file (dir, basename, DEBUG_STR_SUFFIX);
1706
1707 write_debug_names (dwarf2_per_objfile, objfile_index_wip.out_file.get (),
1708 str_wip_file.out_file.get ());
1709
1710 str_wip_file.finalize ();
cd4fb1b2
SM
1711 }
1712 else
c4973306
SM
1713 write_gdbindex (dwarf2_per_objfile, objfile_index_wip.out_file.get (),
1714 (dwz_index_wip.has_value ()
1715 ? dwz_index_wip->out_file.get () : NULL));
cd4fb1b2 1716
c4973306 1717 objfile_index_wip.finalize ();
87d6a7aa 1718
c4973306
SM
1719 if (dwz_index_wip.has_value ())
1720 dwz_index_wip->finalize ();
cd4fb1b2
SM
1721}
1722
1723/* Implementation of the `save gdb-index' command.
1724
1725 Note that the .gdb_index file format used by this command is
1726 documented in the GDB manual. Any changes here must be documented
1727 there. */
1728
1729static void
1730save_gdb_index_command (const char *arg, int from_tty)
1731{
cd4fb1b2
SM
1732 const char dwarf5space[] = "-dwarf-5 ";
1733 dw_index_kind index_kind = dw_index_kind::GDB_INDEX;
1734
1735 if (!arg)
1736 arg = "";
1737
1738 arg = skip_spaces (arg);
1739 if (strncmp (arg, dwarf5space, strlen (dwarf5space)) == 0)
1740 {
1741 index_kind = dw_index_kind::DEBUG_NAMES;
1742 arg += strlen (dwarf5space);
1743 arg = skip_spaces (arg);
1744 }
1745
1746 if (!*arg)
1747 error (_("usage: save gdb-index [-dwarf-5] DIRECTORY"));
1748
2030c079 1749 for (objfile *objfile : current_program_space->objfiles ())
aed57c53
TT
1750 {
1751 struct stat st;
cd4fb1b2 1752
aed57c53
TT
1753 /* If the objfile does not correspond to an actual file, skip it. */
1754 if (stat (objfile_name (objfile), &st) < 0)
1755 continue;
cd4fb1b2 1756
aed57c53
TT
1757 struct dwarf2_per_objfile *dwarf2_per_objfile
1758 = get_dwarf2_per_objfile (objfile);
cd4fb1b2 1759
aed57c53
TT
1760 if (dwarf2_per_objfile != NULL)
1761 {
a70b8144 1762 try
aed57c53
TT
1763 {
1764 const char *basename = lbasename (objfile_name (objfile));
c4973306
SM
1765 const dwz_file *dwz = dwarf2_get_dwz_file (dwarf2_per_objfile);
1766 const char *dwz_basename = NULL;
1767
1768 if (dwz != NULL)
1769 dwz_basename = lbasename (dwz->filename ());
1770
aed57c53 1771 write_psymtabs_to_index (dwarf2_per_objfile, arg, basename,
c4973306 1772 dwz_basename, index_kind);
aed57c53 1773 }
230d2906 1774 catch (const gdb_exception_error &except)
aed57c53
TT
1775 {
1776 exception_fprintf (gdb_stderr, except,
1777 _("Error while writing index for `%s': "),
1778 objfile_name (objfile));
1779 }
aed57c53 1780 }
cd4fb1b2 1781
aed57c53 1782 }
cd4fb1b2
SM
1783}
1784
6c265988 1785void _initialize_dwarf_index_write ();
cd4fb1b2
SM
1786void
1787_initialize_dwarf_index_write ()
1788{
1789 cmd_list_element *c = add_cmd ("gdb-index", class_files,
1790 save_gdb_index_command, _("\
1791Save a gdb-index file.\n\
1792Usage: save gdb-index [-dwarf-5] DIRECTORY\n\
1793\n\
1794No options create one file with .gdb-index extension for pre-DWARF-5\n\
1795compatible .gdb_index section. With -dwarf-5 creates two files with\n\
1796extension .debug_names and .debug_str for DWARF-5 .debug_names section."),
1797 &save_cmdlist);
1798 set_cmd_completer (c, filename_completer);
1799}
This page took 0.281673 seconds and 4 git commands to generate.