Remove DW_ADDR
[deliverable/binutils-gdb.git] / gdb / dwarf2 / 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
82ca8957 22#include "dwarf2/index-write.h"
23baa4cc 23
cd4fb1b2
SM
24#include "addrmap.h"
25#include "cli/cli-decode.h"
268a13a5
TT
26#include "gdbsupport/byte-vector.h"
27#include "gdbsupport/filestuff.h"
28#include "gdbsupport/gdb_unlinker.h"
29#include "gdbsupport/pathstuff.h"
30#include "gdbsupport/scoped_fd.h"
cd4fb1b2 31#include "complaints.h"
82ca8957 32#include "dwarf2/index-common.h"
cd4fb1b2 33#include "dwarf2.h"
82ca8957 34#include "dwarf2/read.h"
9fda78b6 35#include "dwarf2/dwz.h"
cd4fb1b2
SM
36#include "gdb/gdb-index.h"
37#include "gdbcmd.h"
38#include "objfiles.h"
39#include "psympriv.h"
3b00ef10 40#include "ada-lang.h"
cd4fb1b2 41
4de283e4
TT
42#include <algorithm>
43#include <cmath>
159ed7d9 44#include <forward_list>
4de283e4
TT
45#include <set>
46#include <unordered_map>
47#include <unordered_set>
48
cd4fb1b2
SM
49/* Ensure only legit values are used. */
50#define DW2_GDB_INDEX_SYMBOL_STATIC_SET_VALUE(cu_index, value) \
51 do { \
52 gdb_assert ((unsigned int) (value) <= 1); \
53 GDB_INDEX_SYMBOL_STATIC_SET_VALUE((cu_index), (value)); \
54 } while (0)
55
56/* Ensure only legit values are used. */
57#define DW2_GDB_INDEX_SYMBOL_KIND_SET_VALUE(cu_index, value) \
58 do { \
59 gdb_assert ((value) >= GDB_INDEX_SYMBOL_KIND_TYPE \
60 && (value) <= GDB_INDEX_SYMBOL_KIND_OTHER); \
61 GDB_INDEX_SYMBOL_KIND_SET_VALUE((cu_index), (value)); \
62 } while (0)
63
85102364 64/* Ensure we don't use more than the allotted number of bits for the CU. */
cd4fb1b2
SM
65#define DW2_GDB_INDEX_CU_SET_VALUE(cu_index, value) \
66 do { \
67 gdb_assert (((value) & ~GDB_INDEX_CU_MASK) == 0); \
68 GDB_INDEX_CU_SET_VALUE((cu_index), (value)); \
69 } while (0)
70
71/* The "save gdb-index" command. */
72
73/* Write SIZE bytes from the buffer pointed to by DATA to FILE, with
74 error checking. */
75
76static void
77file_write (FILE *file, const void *data, size_t size)
78{
79 if (fwrite (data, 1, size, file) != size)
80 error (_("couldn't data write to file"));
81}
82
83/* Write the contents of VEC to FILE, with error checking. */
84
85template<typename Elem, typename Alloc>
86static void
87file_write (FILE *file, const std::vector<Elem, Alloc> &vec)
88{
1f88d0c8
SM
89 if (!vec.empty ())
90 file_write (file, vec.data (), vec.size () * sizeof (vec[0]));
cd4fb1b2
SM
91}
92
93/* In-memory buffer to prepare data to be written later to a file. */
94class data_buf
95{
96public:
97 /* Copy DATA to the end of the buffer. */
98 template<typename T>
99 void append_data (const T &data)
100 {
101 std::copy (reinterpret_cast<const gdb_byte *> (&data),
102 reinterpret_cast<const gdb_byte *> (&data + 1),
103 grow (sizeof (data)));
104 }
105
106 /* Copy CSTR (a zero-terminated string) to the end of buffer. The
107 terminating zero is appended too. */
108 void append_cstr0 (const char *cstr)
109 {
110 const size_t size = strlen (cstr) + 1;
111 std::copy (cstr, cstr + size, grow (size));
112 }
113
114 /* Store INPUT as ULEB128 to the end of buffer. */
115 void append_unsigned_leb128 (ULONGEST input)
116 {
117 for (;;)
118 {
119 gdb_byte output = input & 0x7f;
120 input >>= 7;
121 if (input)
122 output |= 0x80;
123 append_data (output);
124 if (input == 0)
125 break;
126 }
127 }
128
129 /* Accept a host-format integer in VAL and append it to the buffer
130 as a target-format integer which is LEN bytes long. */
131 void append_uint (size_t len, bfd_endian byte_order, ULONGEST val)
132 {
133 ::store_unsigned_integer (grow (len), len, byte_order, val);
134 }
135
136 /* Return the size of the buffer. */
137 size_t size () const
138 {
139 return m_vec.size ();
140 }
141
142 /* Return true iff the buffer is empty. */
143 bool empty () const
144 {
145 return m_vec.empty ();
146 }
147
148 /* Write the buffer to FILE. */
149 void file_write (FILE *file) const
150 {
151 ::file_write (file, m_vec);
152 }
153
154private:
155 /* Grow SIZE bytes at the end of the buffer. Returns a pointer to
156 the start of the new block. */
157 gdb_byte *grow (size_t size)
158 {
159 m_vec.resize (m_vec.size () + size);
b4be9bfd 160 return &*(m_vec.end () - size);
cd4fb1b2
SM
161 }
162
163 gdb::byte_vector m_vec;
164};
165
166/* An entry in the symbol table. */
167struct symtab_index_entry
168{
169 /* The name of the symbol. */
170 const char *name;
171 /* The offset of the name in the constant pool. */
172 offset_type index_offset;
173 /* A sorted vector of the indices of all the CUs that hold an object
174 of this name. */
175 std::vector<offset_type> cu_indices;
176};
177
178/* The symbol table. This is a power-of-2-sized hash table. */
179struct mapped_symtab
180{
181 mapped_symtab ()
182 {
183 data.resize (1024);
184 }
185
186 offset_type n_elements = 0;
187 std::vector<symtab_index_entry> data;
7ab96794
TV
188
189 /* Temporary storage for Ada names. */
190 auto_obstack m_string_obstack;
cd4fb1b2
SM
191};
192
193/* Find a slot in SYMTAB for the symbol NAME. Returns a reference to
194 the slot.
195
196 Function is used only during write_hash_table so no index format backward
197 compatibility is needed. */
198
199static symtab_index_entry &
200find_slot (struct mapped_symtab *symtab, const char *name)
201{
202 offset_type index, step, hash = mapped_index_string_hash (INT_MAX, name);
203
204 index = hash & (symtab->data.size () - 1);
205 step = ((hash * 17) & (symtab->data.size () - 1)) | 1;
206
207 for (;;)
208 {
209 if (symtab->data[index].name == NULL
210 || strcmp (name, symtab->data[index].name) == 0)
211 return symtab->data[index];
212 index = (index + step) & (symtab->data.size () - 1);
213 }
214}
215
216/* Expand SYMTAB's hash table. */
217
218static void
219hash_expand (struct mapped_symtab *symtab)
220{
221 auto old_entries = std::move (symtab->data);
222
223 symtab->data.clear ();
224 symtab->data.resize (old_entries.size () * 2);
225
226 for (auto &it : old_entries)
227 if (it.name != NULL)
228 {
229 auto &ref = find_slot (symtab, it.name);
230 ref = std::move (it);
231 }
232}
233
234/* Add an entry to SYMTAB. NAME is the name of the symbol.
235 CU_INDEX is the index of the CU in which the symbol appears.
236 IS_STATIC is one if the symbol is static, otherwise zero (global). */
237
238static void
239add_index_entry (struct mapped_symtab *symtab, const char *name,
240 int is_static, gdb_index_symbol_kind kind,
241 offset_type cu_index)
242{
243 offset_type cu_index_and_attrs;
244
245 ++symtab->n_elements;
246 if (4 * symtab->n_elements / 3 >= symtab->data.size ())
247 hash_expand (symtab);
248
249 symtab_index_entry &slot = find_slot (symtab, name);
250 if (slot.name == NULL)
251 {
252 slot.name = name;
253 /* index_offset is set later. */
254 }
255
256 cu_index_and_attrs = 0;
257 DW2_GDB_INDEX_CU_SET_VALUE (cu_index_and_attrs, cu_index);
258 DW2_GDB_INDEX_SYMBOL_STATIC_SET_VALUE (cu_index_and_attrs, is_static);
259 DW2_GDB_INDEX_SYMBOL_KIND_SET_VALUE (cu_index_and_attrs, kind);
260
261 /* We don't want to record an index value twice as we want to avoid the
262 duplication.
263 We process all global symbols and then all static symbols
264 (which would allow us to avoid the duplication by only having to check
265 the last entry pushed), but a symbol could have multiple kinds in one CU.
266 To keep things simple we don't worry about the duplication here and
85102364 267 sort and uniquify the list after we've processed all symbols. */
cd4fb1b2
SM
268 slot.cu_indices.push_back (cu_index_and_attrs);
269}
270
271/* Sort and remove duplicates of all symbols' cu_indices lists. */
272
273static void
274uniquify_cu_indices (struct mapped_symtab *symtab)
275{
276 for (auto &entry : symtab->data)
277 {
278 if (entry.name != NULL && !entry.cu_indices.empty ())
279 {
280 auto &cu_indices = entry.cu_indices;
281 std::sort (cu_indices.begin (), cu_indices.end ());
282 auto from = std::unique (cu_indices.begin (), cu_indices.end ());
283 cu_indices.erase (from, cu_indices.end ());
284 }
285 }
286}
287
288/* A form of 'const char *' suitable for container keys. Only the
289 pointer is stored. The strings themselves are compared, not the
290 pointers. */
291class c_str_view
292{
293public:
294 c_str_view (const char *cstr)
295 : m_cstr (cstr)
296 {}
297
298 bool operator== (const c_str_view &other) const
299 {
300 return strcmp (m_cstr, other.m_cstr) == 0;
301 }
302
303 /* Return the underlying C string. Note, the returned string is
304 only a reference with lifetime of this object. */
305 const char *c_str () const
306 {
307 return m_cstr;
308 }
309
310private:
311 friend class c_str_view_hasher;
312 const char *const m_cstr;
313};
314
315/* A std::unordered_map::hasher for c_str_view that uses the right
316 hash function for strings in a mapped index. */
317class c_str_view_hasher
318{
319public:
320 size_t operator () (const c_str_view &x) const
321 {
322 return mapped_index_string_hash (INT_MAX, x.m_cstr);
323 }
324};
325
326/* A std::unordered_map::hasher for std::vector<>. */
327template<typename T>
328class vector_hasher
329{
330public:
331 size_t operator () (const std::vector<T> &key) const
332 {
333 return iterative_hash (key.data (),
334 sizeof (key.front ()) * key.size (), 0);
335 }
336};
337
338/* Write the mapped hash table SYMTAB to the data buffer OUTPUT, with
339 constant pool entries going into the data buffer CPOOL. */
340
341static void
342write_hash_table (mapped_symtab *symtab, data_buf &output, data_buf &cpool)
343{
344 {
345 /* Elements are sorted vectors of the indices of all the CUs that
346 hold an object of this name. */
347 std::unordered_map<std::vector<offset_type>, offset_type,
348 vector_hasher<offset_type>>
349 symbol_hash_table;
350
351 /* We add all the index vectors to the constant pool first, to
352 ensure alignment is ok. */
353 for (symtab_index_entry &entry : symtab->data)
354 {
355 if (entry.name == NULL)
356 continue;
357 gdb_assert (entry.index_offset == 0);
358
359 /* Finding before inserting is faster than always trying to
360 insert, because inserting always allocates a node, does the
361 lookup, and then destroys the new node if another node
362 already had the same key. C++17 try_emplace will avoid
363 this. */
364 const auto found
365 = symbol_hash_table.find (entry.cu_indices);
366 if (found != symbol_hash_table.end ())
367 {
368 entry.index_offset = found->second;
369 continue;
370 }
371
372 symbol_hash_table.emplace (entry.cu_indices, cpool.size ());
373 entry.index_offset = cpool.size ();
374 cpool.append_data (MAYBE_SWAP (entry.cu_indices.size ()));
375 for (const auto index : entry.cu_indices)
376 cpool.append_data (MAYBE_SWAP (index));
377 }
378 }
379
380 /* Now write out the hash table. */
381 std::unordered_map<c_str_view, offset_type, c_str_view_hasher> str_table;
382 for (const auto &entry : symtab->data)
383 {
384 offset_type str_off, vec_off;
385
386 if (entry.name != NULL)
387 {
388 const auto insertpair = str_table.emplace (entry.name, cpool.size ());
389 if (insertpair.second)
390 cpool.append_cstr0 (entry.name);
391 str_off = insertpair.first->second;
392 vec_off = entry.index_offset;
393 }
394 else
395 {
396 /* While 0 is a valid constant pool index, it is not valid
397 to have 0 for both offsets. */
398 str_off = 0;
399 vec_off = 0;
400 }
401
402 output.append_data (MAYBE_SWAP (str_off));
403 output.append_data (MAYBE_SWAP (vec_off));
404 }
405}
406
edfe0a0c 407typedef std::unordered_map<partial_symtab *, unsigned int> psym_index_map;
cd4fb1b2
SM
408
409/* Helper struct for building the address table. */
410struct addrmap_index_data
411{
412 addrmap_index_data (data_buf &addr_vec_, psym_index_map &cu_index_htab_)
413 : addr_vec (addr_vec_), cu_index_htab (cu_index_htab_)
414 {}
415
416 struct objfile *objfile;
417 data_buf &addr_vec;
418 psym_index_map &cu_index_htab;
419
420 /* Non-zero if the previous_* fields are valid.
421 We can't write an entry until we see the next entry (since it is only then
422 that we know the end of the entry). */
423 int previous_valid;
424 /* Index of the CU in the table of all CUs in the index file. */
425 unsigned int previous_cu_index;
426 /* Start address of the CU. */
427 CORE_ADDR previous_cu_start;
428};
429
430/* Write an address entry to ADDR_VEC. */
431
432static void
433add_address_entry (struct objfile *objfile, data_buf &addr_vec,
434 CORE_ADDR start, CORE_ADDR end, unsigned int cu_index)
435{
79748972
TT
436 addr_vec.append_uint (8, BFD_ENDIAN_LITTLE, start);
437 addr_vec.append_uint (8, BFD_ENDIAN_LITTLE, end);
cd4fb1b2
SM
438 addr_vec.append_data (MAYBE_SWAP (cu_index));
439}
440
441/* Worker function for traversing an addrmap to build the address table. */
442
443static int
444add_address_entry_worker (void *datap, CORE_ADDR start_addr, void *obj)
445{
446 struct addrmap_index_data *data = (struct addrmap_index_data *) datap;
edfe0a0c 447 partial_symtab *pst = (partial_symtab *) obj;
cd4fb1b2
SM
448
449 if (data->previous_valid)
450 add_address_entry (data->objfile, data->addr_vec,
451 data->previous_cu_start, start_addr,
452 data->previous_cu_index);
453
454 data->previous_cu_start = start_addr;
455 if (pst != NULL)
456 {
457 const auto it = data->cu_index_htab.find (pst);
458 gdb_assert (it != data->cu_index_htab.cend ());
459 data->previous_cu_index = it->second;
460 data->previous_valid = 1;
461 }
462 else
463 data->previous_valid = 0;
464
465 return 0;
466}
467
468/* Write OBJFILE's address map to ADDR_VEC.
469 CU_INDEX_HTAB is used to map addrmap entries to their CU indices
470 in the index file. */
471
472static void
473write_address_map (struct objfile *objfile, data_buf &addr_vec,
474 psym_index_map &cu_index_htab)
475{
476 struct addrmap_index_data addrmap_index_data (addr_vec, cu_index_htab);
477
478 /* When writing the address table, we have to cope with the fact that
479 the addrmap iterator only provides the start of a region; we have to
480 wait until the next invocation to get the start of the next region. */
481
482 addrmap_index_data.objfile = objfile;
483 addrmap_index_data.previous_valid = 0;
484
d320c2b5
TT
485 addrmap_foreach (objfile->partial_symtabs->psymtabs_addrmap,
486 add_address_entry_worker, &addrmap_index_data);
cd4fb1b2
SM
487
488 /* It's highly unlikely the last entry (end address = 0xff...ff)
489 is valid, but we should still handle it.
490 The end address is recorded as the start of the next region, but that
491 doesn't work here. To cope we pass 0xff...ff, this is a rare situation
492 anyway. */
493 if (addrmap_index_data.previous_valid)
494 add_address_entry (objfile, addr_vec,
495 addrmap_index_data.previous_cu_start, (CORE_ADDR) -1,
496 addrmap_index_data.previous_cu_index);
497}
498
499/* Return the symbol kind of PSYM. */
500
501static gdb_index_symbol_kind
502symbol_kind (struct partial_symbol *psym)
503{
8a6d4234
TT
504 domain_enum domain = psym->domain;
505 enum address_class aclass = psym->aclass;
cd4fb1b2
SM
506
507 switch (domain)
508 {
509 case VAR_DOMAIN:
510 switch (aclass)
511 {
512 case LOC_BLOCK:
513 return GDB_INDEX_SYMBOL_KIND_FUNCTION;
514 case LOC_TYPEDEF:
515 return GDB_INDEX_SYMBOL_KIND_TYPE;
516 case LOC_COMPUTED:
517 case LOC_CONST_BYTES:
518 case LOC_OPTIMIZED_OUT:
519 case LOC_STATIC:
520 return GDB_INDEX_SYMBOL_KIND_VARIABLE;
521 case LOC_CONST:
522 /* Note: It's currently impossible to recognize psyms as enum values
523 short of reading the type info. For now punt. */
524 return GDB_INDEX_SYMBOL_KIND_VARIABLE;
525 default:
526 /* There are other LOC_FOO values that one might want to classify
527 as variables, but dwarf2read.c doesn't currently use them. */
528 return GDB_INDEX_SYMBOL_KIND_OTHER;
529 }
530 case STRUCT_DOMAIN:
531 return GDB_INDEX_SYMBOL_KIND_TYPE;
532 default:
533 return GDB_INDEX_SYMBOL_KIND_OTHER;
534 }
535}
536
537/* Add a list of partial symbols to SYMTAB. */
538
539static void
540write_psymbols (struct mapped_symtab *symtab,
541 std::unordered_set<partial_symbol *> &psyms_seen,
542 struct partial_symbol **psymp,
543 int count,
544 offset_type cu_index,
545 int is_static)
546{
547 for (; count-- > 0; ++psymp)
548 {
549 struct partial_symbol *psym = *psymp;
7ab96794 550 const char *name = psym->ginfo.search_name ();
cd4fb1b2 551
c1b5c1eb 552 if (psym->ginfo.language () == language_ada)
7ab96794
TV
553 {
554 /* We want to ensure that the Ada main function's name appears
555 verbatim in the index. However, this name will be of the
556 form "_ada_mumble", and will be rewritten by ada_decode.
557 So, recognize it specially here and add it to the index by
558 hand. */
559 if (strcmp (main_name (), name) == 0)
560 {
561 gdb_index_symbol_kind kind = symbol_kind (psym);
562
563 add_index_entry (symtab, name, is_static, kind, cu_index);
564 }
565
566 /* In order for the index to work when read back into gdb, it
567 has to supply a funny form of the name: it should be the
568 encoded name, with any suffixes stripped. Using the
569 ordinary encoded name will not work properly with the
570 searching logic in find_name_components_bounds; nor will
571 using the decoded name. Furthermore, an Ada "verbatim"
572 name (of the form "<MumBle>") must be entered without the
573 angle brackets. Note that the current index is unusual,
574 see PR symtab/24820 for details. */
575 std::string decoded = ada_decode (name);
576 if (decoded[0] == '<')
577 name = (char *) obstack_copy0 (&symtab->m_string_obstack,
578 decoded.c_str () + 1,
579 decoded.length () - 2);
580 else
581 name = obstack_strdup (&symtab->m_string_obstack,
582 ada_encode (decoded.c_str ()));
583 }
cd4fb1b2
SM
584
585 /* Only add a given psymbol once. */
586 if (psyms_seen.insert (psym).second)
587 {
588 gdb_index_symbol_kind kind = symbol_kind (psym);
589
7ab96794 590 add_index_entry (symtab, name, is_static, kind, cu_index);
cd4fb1b2
SM
591 }
592 }
593}
594
595/* A helper struct used when iterating over debug_types. */
596struct signatured_type_index_data
597{
598 signatured_type_index_data (data_buf &types_list_,
599 std::unordered_set<partial_symbol *> &psyms_seen_)
600 : types_list (types_list_), psyms_seen (psyms_seen_)
601 {}
602
603 struct objfile *objfile;
604 struct mapped_symtab *symtab;
605 data_buf &types_list;
606 std::unordered_set<partial_symbol *> &psyms_seen;
607 int cu_index;
608};
609
610/* A helper function that writes a single signatured_type to an
611 obstack. */
612
613static int
614write_one_signatured_type (void **slot, void *d)
615{
616 struct signatured_type_index_data *info
617 = (struct signatured_type_index_data *) d;
618 struct signatured_type *entry = (struct signatured_type *) *slot;
edfe0a0c 619 partial_symtab *psymtab = entry->per_cu.v.psymtab;
cd4fb1b2
SM
620
621 write_psymbols (info->symtab,
622 info->psyms_seen,
d320c2b5 623 (info->objfile->partial_symtabs->global_psymbols.data ()
b4be9bfd 624 + psymtab->globals_offset),
cd4fb1b2
SM
625 psymtab->n_global_syms, info->cu_index,
626 0);
627 write_psymbols (info->symtab,
628 info->psyms_seen,
d320c2b5 629 (info->objfile->partial_symtabs->static_psymbols.data ()
b4be9bfd 630 + psymtab->statics_offset),
cd4fb1b2
SM
631 psymtab->n_static_syms, info->cu_index,
632 1);
633
634 info->types_list.append_uint (8, BFD_ENDIAN_LITTLE,
635 to_underlying (entry->per_cu.sect_off));
636 info->types_list.append_uint (8, BFD_ENDIAN_LITTLE,
637 to_underlying (entry->type_offset_in_tu));
638 info->types_list.append_uint (8, BFD_ENDIAN_LITTLE, entry->signature);
639
640 ++info->cu_index;
641
642 return 1;
643}
644
645/* Recurse into all "included" dependencies and count their symbols as
646 if they appeared in this psymtab. */
647
648static void
edfe0a0c 649recursively_count_psymbols (partial_symtab *psymtab,
cd4fb1b2
SM
650 size_t &psyms_seen)
651{
652 for (int i = 0; i < psymtab->number_of_dependencies; ++i)
653 if (psymtab->dependencies[i]->user != NULL)
edfe0a0c 654 recursively_count_psymbols (psymtab->dependencies[i],
cd4fb1b2
SM
655 psyms_seen);
656
657 psyms_seen += psymtab->n_global_syms;
658 psyms_seen += psymtab->n_static_syms;
659}
660
661/* Recurse into all "included" dependencies and write their symbols as
662 if they appeared in this psymtab. */
663
664static void
665recursively_write_psymbols (struct objfile *objfile,
edfe0a0c 666 partial_symtab *psymtab,
cd4fb1b2
SM
667 struct mapped_symtab *symtab,
668 std::unordered_set<partial_symbol *> &psyms_seen,
669 offset_type cu_index)
670{
671 int i;
672
673 for (i = 0; i < psymtab->number_of_dependencies; ++i)
674 if (psymtab->dependencies[i]->user != NULL)
891813be 675 recursively_write_psymbols (objfile,
edfe0a0c 676 psymtab->dependencies[i],
cd4fb1b2
SM
677 symtab, psyms_seen, cu_index);
678
679 write_psymbols (symtab,
680 psyms_seen,
d320c2b5
TT
681 (objfile->partial_symtabs->global_psymbols.data ()
682 + psymtab->globals_offset),
cd4fb1b2
SM
683 psymtab->n_global_syms, cu_index,
684 0);
685 write_psymbols (symtab,
686 psyms_seen,
d320c2b5
TT
687 (objfile->partial_symtabs->static_psymbols.data ()
688 + psymtab->statics_offset),
cd4fb1b2
SM
689 psymtab->n_static_syms, cu_index,
690 1);
691}
692
693/* DWARF-5 .debug_names builder. */
694class debug_names
695{
696public:
976ca316 697 debug_names (dwarf2_per_objfile *per_objfile, bool is_dwarf64,
cd4fb1b2
SM
698 bfd_endian dwarf5_byte_order)
699 : m_dwarf5_byte_order (dwarf5_byte_order),
700 m_dwarf32 (dwarf5_byte_order),
701 m_dwarf64 (dwarf5_byte_order),
702 m_dwarf (is_dwarf64
703 ? static_cast<dwarf &> (m_dwarf64)
704 : static_cast<dwarf &> (m_dwarf32)),
705 m_name_table_string_offs (m_dwarf.name_table_string_offs),
706 m_name_table_entry_offs (m_dwarf.name_table_entry_offs),
976ca316 707 m_debugstrlookup (per_objfile)
cd4fb1b2
SM
708 {}
709
710 int dwarf5_offset_size () const
711 {
712 const bool dwarf5_is_dwarf64 = &m_dwarf == &m_dwarf64;
713 return dwarf5_is_dwarf64 ? 8 : 4;
714 }
715
716 /* Is this symbol from DW_TAG_compile_unit or DW_TAG_type_unit? */
717 enum class unit_kind { cu, tu };
718
719 /* Insert one symbol. */
720 void insert (const partial_symbol *psym, int cu_index, bool is_static,
721 unit_kind kind)
722 {
723 const int dwarf_tag = psymbol_tag (psym);
724 if (dwarf_tag == 0)
725 return;
c9d95fa3 726 const char *name = psym->ginfo.search_name ();
3b00ef10 727
c1b5c1eb 728 if (psym->ginfo.language () == language_ada)
3b00ef10
TT
729 {
730 /* We want to ensure that the Ada main function's name appears
731 verbatim in the index. However, this name will be of the
732 form "_ada_mumble", and will be rewritten by ada_decode.
733 So, recognize it specially here and add it to the index by
734 hand. */
735 if (strcmp (main_name (), name) == 0)
736 {
737 const auto insertpair
738 = m_name_to_value_set.emplace (c_str_view (name),
739 std::set<symbol_value> ());
740 std::set<symbol_value> &value_set = insertpair.first->second;
741 value_set.emplace (symbol_value (dwarf_tag, cu_index, is_static,
742 kind));
743 }
744
745 /* In order for the index to work when read back into gdb, it
746 has to supply a funny form of the name: it should be the
747 encoded name, with any suffixes stripped. Using the
748 ordinary encoded name will not work properly with the
749 searching logic in find_name_components_bounds; nor will
750 using the decoded name. Furthermore, an Ada "verbatim"
751 name (of the form "<MumBle>") must be entered without the
752 angle brackets. Note that the current index is unusual,
753 see PR symtab/24820 for details. */
f945dedf 754 std::string decoded = ada_decode (name);
3b00ef10
TT
755 if (decoded[0] == '<')
756 name = (char *) obstack_copy0 (&m_string_obstack,
f945dedf
CB
757 decoded.c_str () + 1,
758 decoded.length () - 2);
3b00ef10 759 else
f945dedf
CB
760 name = obstack_strdup (&m_string_obstack,
761 ada_encode (decoded.c_str ()));
3b00ef10
TT
762 }
763
cd4fb1b2
SM
764 const auto insertpair
765 = m_name_to_value_set.emplace (c_str_view (name),
766 std::set<symbol_value> ());
767 std::set<symbol_value> &value_set = insertpair.first->second;
768 value_set.emplace (symbol_value (dwarf_tag, cu_index, is_static, kind));
769 }
770
771 /* Build all the tables. All symbols must be already inserted.
772 This function does not call file_write, caller has to do it
773 afterwards. */
774 void build ()
775 {
776 /* Verify the build method has not be called twice. */
777 gdb_assert (m_abbrev_table.empty ());
778 const size_t name_count = m_name_to_value_set.size ();
779 m_bucket_table.resize
780 (std::pow (2, std::ceil (std::log2 (name_count * 4 / 3))));
781 m_hash_table.reserve (name_count);
782 m_name_table_string_offs.reserve (name_count);
783 m_name_table_entry_offs.reserve (name_count);
784
785 /* Map each hash of symbol to its name and value. */
786 struct hash_it_pair
787 {
788 uint32_t hash;
789 decltype (m_name_to_value_set)::const_iterator it;
790 };
791 std::vector<std::forward_list<hash_it_pair>> bucket_hash;
792 bucket_hash.resize (m_bucket_table.size ());
793 for (decltype (m_name_to_value_set)::const_iterator it
794 = m_name_to_value_set.cbegin ();
795 it != m_name_to_value_set.cend ();
796 ++it)
797 {
798 const char *const name = it->first.c_str ();
799 const uint32_t hash = dwarf5_djb_hash (name);
800 hash_it_pair hashitpair;
801 hashitpair.hash = hash;
802 hashitpair.it = it;
803 auto &slot = bucket_hash[hash % bucket_hash.size()];
804 slot.push_front (std::move (hashitpair));
805 }
806 for (size_t bucket_ix = 0; bucket_ix < bucket_hash.size (); ++bucket_ix)
807 {
808 const std::forward_list<hash_it_pair> &hashitlist
809 = bucket_hash[bucket_ix];
810 if (hashitlist.empty ())
811 continue;
812 uint32_t &bucket_slot = m_bucket_table[bucket_ix];
813 /* The hashes array is indexed starting at 1. */
814 store_unsigned_integer (reinterpret_cast<gdb_byte *> (&bucket_slot),
815 sizeof (bucket_slot), m_dwarf5_byte_order,
816 m_hash_table.size () + 1);
817 for (const hash_it_pair &hashitpair : hashitlist)
818 {
819 m_hash_table.push_back (0);
820 store_unsigned_integer (reinterpret_cast<gdb_byte *>
821 (&m_hash_table.back ()),
822 sizeof (m_hash_table.back ()),
823 m_dwarf5_byte_order, hashitpair.hash);
824 const c_str_view &name = hashitpair.it->first;
825 const std::set<symbol_value> &value_set = hashitpair.it->second;
826 m_name_table_string_offs.push_back_reorder
827 (m_debugstrlookup.lookup (name.c_str ()));
828 m_name_table_entry_offs.push_back_reorder (m_entry_pool.size ());
829 gdb_assert (!value_set.empty ());
830 for (const symbol_value &value : value_set)
831 {
832 int &idx = m_indexkey_to_idx[index_key (value.dwarf_tag,
833 value.is_static,
834 value.kind)];
835 if (idx == 0)
836 {
837 idx = m_idx_next++;
838 m_abbrev_table.append_unsigned_leb128 (idx);
839 m_abbrev_table.append_unsigned_leb128 (value.dwarf_tag);
840 m_abbrev_table.append_unsigned_leb128
841 (value.kind == unit_kind::cu ? DW_IDX_compile_unit
842 : DW_IDX_type_unit);
843 m_abbrev_table.append_unsigned_leb128 (DW_FORM_udata);
844 m_abbrev_table.append_unsigned_leb128 (value.is_static
845 ? DW_IDX_GNU_internal
846 : DW_IDX_GNU_external);
847 m_abbrev_table.append_unsigned_leb128 (DW_FORM_flag_present);
848
849 /* Terminate attributes list. */
850 m_abbrev_table.append_unsigned_leb128 (0);
851 m_abbrev_table.append_unsigned_leb128 (0);
852 }
853
854 m_entry_pool.append_unsigned_leb128 (idx);
855 m_entry_pool.append_unsigned_leb128 (value.cu_index);
856 }
857
858 /* Terminate the list of CUs. */
859 m_entry_pool.append_unsigned_leb128 (0);
860 }
861 }
862 gdb_assert (m_hash_table.size () == name_count);
863
864 /* Terminate tags list. */
865 m_abbrev_table.append_unsigned_leb128 (0);
866 }
867
868 /* Return .debug_names bucket count. This must be called only after
869 calling the build method. */
870 uint32_t bucket_count () const
871 {
872 /* Verify the build method has been already called. */
873 gdb_assert (!m_abbrev_table.empty ());
874 const uint32_t retval = m_bucket_table.size ();
875
876 /* Check for overflow. */
877 gdb_assert (retval == m_bucket_table.size ());
878 return retval;
879 }
880
881 /* Return .debug_names names count. This must be called only after
882 calling the build method. */
883 uint32_t name_count () const
884 {
885 /* Verify the build method has been already called. */
886 gdb_assert (!m_abbrev_table.empty ());
887 const uint32_t retval = m_hash_table.size ();
888
889 /* Check for overflow. */
890 gdb_assert (retval == m_hash_table.size ());
891 return retval;
892 }
893
894 /* Return number of bytes of .debug_names abbreviation table. This
895 must be called only after calling the build method. */
896 uint32_t abbrev_table_bytes () const
897 {
898 gdb_assert (!m_abbrev_table.empty ());
899 return m_abbrev_table.size ();
900 }
901
902 /* Recurse into all "included" dependencies and store their symbols
903 as if they appeared in this psymtab. */
904 void recursively_write_psymbols
905 (struct objfile *objfile,
edfe0a0c 906 partial_symtab *psymtab,
cd4fb1b2
SM
907 std::unordered_set<partial_symbol *> &psyms_seen,
908 int cu_index)
909 {
910 for (int i = 0; i < psymtab->number_of_dependencies; ++i)
911 if (psymtab->dependencies[i]->user != NULL)
891813be 912 recursively_write_psymbols
edfe0a0c 913 (objfile, psymtab->dependencies[i], psyms_seen, cu_index);
cd4fb1b2
SM
914
915 write_psymbols (psyms_seen,
d320c2b5
TT
916 (objfile->partial_symtabs->global_psymbols.data ()
917 + psymtab->globals_offset),
cd4fb1b2
SM
918 psymtab->n_global_syms, cu_index, false, unit_kind::cu);
919 write_psymbols (psyms_seen,
d320c2b5
TT
920 (objfile->partial_symtabs->static_psymbols.data ()
921 + psymtab->statics_offset),
cd4fb1b2
SM
922 psymtab->n_static_syms, cu_index, true, unit_kind::cu);
923 }
924
925 /* Return number of bytes the .debug_names section will have. This
926 must be called only after calling the build method. */
927 size_t bytes () const
928 {
929 /* Verify the build method has been already called. */
930 gdb_assert (!m_abbrev_table.empty ());
931 size_t expected_bytes = 0;
932 expected_bytes += m_bucket_table.size () * sizeof (m_bucket_table[0]);
933 expected_bytes += m_hash_table.size () * sizeof (m_hash_table[0]);
934 expected_bytes += m_name_table_string_offs.bytes ();
935 expected_bytes += m_name_table_entry_offs.bytes ();
936 expected_bytes += m_abbrev_table.size ();
937 expected_bytes += m_entry_pool.size ();
938 return expected_bytes;
939 }
940
941 /* Write .debug_names to FILE_NAMES and .debug_str addition to
942 FILE_STR. This must be called only after calling the build
943 method. */
944 void file_write (FILE *file_names, FILE *file_str) const
945 {
946 /* Verify the build method has been already called. */
947 gdb_assert (!m_abbrev_table.empty ());
948 ::file_write (file_names, m_bucket_table);
949 ::file_write (file_names, m_hash_table);
950 m_name_table_string_offs.file_write (file_names);
951 m_name_table_entry_offs.file_write (file_names);
952 m_abbrev_table.file_write (file_names);
953 m_entry_pool.file_write (file_names);
954 m_debugstrlookup.file_write (file_str);
955 }
956
957 /* A helper user data for write_one_signatured_type. */
958 class write_one_signatured_type_data
959 {
960 public:
961 write_one_signatured_type_data (debug_names &nametable_,
962 signatured_type_index_data &&info_)
963 : nametable (nametable_), info (std::move (info_))
964 {}
965 debug_names &nametable;
966 struct signatured_type_index_data info;
967 };
968
969 /* A helper function to pass write_one_signatured_type to
970 htab_traverse_noresize. */
971 static int
972 write_one_signatured_type (void **slot, void *d)
973 {
974 write_one_signatured_type_data *data = (write_one_signatured_type_data *) d;
975 struct signatured_type_index_data *info = &data->info;
976 struct signatured_type *entry = (struct signatured_type *) *slot;
977
978 data->nametable.write_one_signatured_type (entry, info);
979
980 return 1;
981 }
982
983private:
984
985 /* Storage for symbol names mapping them to their .debug_str section
986 offsets. */
987 class debug_str_lookup
988 {
989 public:
990
30baf67b 991 /* Object constructor to be called for current DWARF2_PER_OBJFILE.
cd4fb1b2 992 All .debug_str section strings are automatically stored. */
976ca316
SM
993 debug_str_lookup (dwarf2_per_objfile *per_objfile)
994 : m_abfd (per_objfile->objfile->obfd),
995 m_per_objfile (per_objfile)
cd4fb1b2 996 {
976ca316
SM
997 per_objfile->per_bfd->str.read (per_objfile->objfile);
998 if (per_objfile->per_bfd->str.buffer == NULL)
cd4fb1b2 999 return;
976ca316
SM
1000 for (const gdb_byte *data = per_objfile->per_bfd->str.buffer;
1001 data < (per_objfile->per_bfd->str.buffer
1002 + per_objfile->per_bfd->str.size);)
cd4fb1b2
SM
1003 {
1004 const char *const s = reinterpret_cast<const char *> (data);
1005 const auto insertpair
1006 = m_str_table.emplace (c_str_view (s),
976ca316 1007 data - per_objfile->per_bfd->str.buffer);
cd4fb1b2 1008 if (!insertpair.second)
b98664d3 1009 complaint (_("Duplicate string \"%s\" in "
cd4fb1b2
SM
1010 ".debug_str section [in module %s]"),
1011 s, bfd_get_filename (m_abfd));
1012 data += strlen (s) + 1;
1013 }
1014 }
1015
1016 /* Return offset of symbol name S in the .debug_str section. Add
1017 such symbol to the section's end if it does not exist there
1018 yet. */
1019 size_t lookup (const char *s)
1020 {
1021 const auto it = m_str_table.find (c_str_view (s));
1022 if (it != m_str_table.end ())
1023 return it->second;
976ca316 1024 const size_t offset = (m_per_objfile->per_bfd->str.size
cd4fb1b2
SM
1025 + m_str_add_buf.size ());
1026 m_str_table.emplace (c_str_view (s), offset);
1027 m_str_add_buf.append_cstr0 (s);
1028 return offset;
1029 }
1030
1031 /* Append the end of the .debug_str section to FILE. */
1032 void file_write (FILE *file) const
1033 {
1034 m_str_add_buf.file_write (file);
1035 }
1036
1037 private:
1038 std::unordered_map<c_str_view, size_t, c_str_view_hasher> m_str_table;
1039 bfd *const m_abfd;
976ca316 1040 dwarf2_per_objfile *m_per_objfile;
cd4fb1b2
SM
1041
1042 /* Data to add at the end of .debug_str for new needed symbol names. */
1043 data_buf m_str_add_buf;
1044 };
1045
1046 /* Container to map used DWARF tags to their .debug_names abbreviation
1047 tags. */
1048 class index_key
1049 {
1050 public:
1051 index_key (int dwarf_tag_, bool is_static_, unit_kind kind_)
1052 : dwarf_tag (dwarf_tag_), is_static (is_static_), kind (kind_)
1053 {
1054 }
1055
1056 bool
1057 operator== (const index_key &other) const
1058 {
1059 return (dwarf_tag == other.dwarf_tag && is_static == other.is_static
1060 && kind == other.kind);
1061 }
1062
1063 const int dwarf_tag;
1064 const bool is_static;
1065 const unit_kind kind;
1066 };
1067
1068 /* Provide std::unordered_map::hasher for index_key. */
1069 class index_key_hasher
1070 {
1071 public:
1072 size_t
1073 operator () (const index_key &key) const
1074 {
1075 return (std::hash<int>() (key.dwarf_tag) << 1) | key.is_static;
1076 }
1077 };
1078
1079 /* Parameters of one symbol entry. */
1080 class symbol_value
1081 {
1082 public:
1083 const int dwarf_tag, cu_index;
1084 const bool is_static;
1085 const unit_kind kind;
1086
1087 symbol_value (int dwarf_tag_, int cu_index_, bool is_static_,
1088 unit_kind kind_)
1089 : dwarf_tag (dwarf_tag_), cu_index (cu_index_), is_static (is_static_),
1090 kind (kind_)
1091 {}
1092
1093 bool
1094 operator< (const symbol_value &other) const
1095 {
1096#define X(n) \
1097 do \
1098 { \
1099 if (n < other.n) \
1100 return true; \
1101 if (n > other.n) \
1102 return false; \
1103 } \
1104 while (0)
1105 X (dwarf_tag);
1106 X (is_static);
1107 X (kind);
1108 X (cu_index);
1109#undef X
1110 return false;
1111 }
1112 };
1113
1114 /* Abstract base class to unify DWARF-32 and DWARF-64 name table
1115 output. */
1116 class offset_vec
1117 {
1118 protected:
1119 const bfd_endian dwarf5_byte_order;
1120 public:
1121 explicit offset_vec (bfd_endian dwarf5_byte_order_)
1122 : dwarf5_byte_order (dwarf5_byte_order_)
1123 {}
1124
1125 /* Call std::vector::reserve for NELEM elements. */
1126 virtual void reserve (size_t nelem) = 0;
1127
1128 /* Call std::vector::push_back with store_unsigned_integer byte
1129 reordering for ELEM. */
1130 virtual void push_back_reorder (size_t elem) = 0;
1131
1132 /* Return expected output size in bytes. */
1133 virtual size_t bytes () const = 0;
1134
1135 /* Write name table to FILE. */
1136 virtual void file_write (FILE *file) const = 0;
1137 };
1138
1139 /* Template to unify DWARF-32 and DWARF-64 output. */
1140 template<typename OffsetSize>
1141 class offset_vec_tmpl : public offset_vec
1142 {
1143 public:
1144 explicit offset_vec_tmpl (bfd_endian dwarf5_byte_order_)
1145 : offset_vec (dwarf5_byte_order_)
1146 {}
1147
1148 /* Implement offset_vec::reserve. */
1149 void reserve (size_t nelem) override
1150 {
1151 m_vec.reserve (nelem);
1152 }
1153
1154 /* Implement offset_vec::push_back_reorder. */
1155 void push_back_reorder (size_t elem) override
1156 {
1157 m_vec.push_back (elem);
1158 /* Check for overflow. */
1159 gdb_assert (m_vec.back () == elem);
1160 store_unsigned_integer (reinterpret_cast<gdb_byte *> (&m_vec.back ()),
1161 sizeof (m_vec.back ()), dwarf5_byte_order, elem);
1162 }
1163
1164 /* Implement offset_vec::bytes. */
1165 size_t bytes () const override
1166 {
1167 return m_vec.size () * sizeof (m_vec[0]);
1168 }
1169
1170 /* Implement offset_vec::file_write. */
1171 void file_write (FILE *file) const override
1172 {
1173 ::file_write (file, m_vec);
1174 }
1175
1176 private:
1177 std::vector<OffsetSize> m_vec;
1178 };
1179
1180 /* Base class to unify DWARF-32 and DWARF-64 .debug_names output
1181 respecting name table width. */
1182 class dwarf
1183 {
1184 public:
1185 offset_vec &name_table_string_offs, &name_table_entry_offs;
1186
1187 dwarf (offset_vec &name_table_string_offs_,
1188 offset_vec &name_table_entry_offs_)
1189 : name_table_string_offs (name_table_string_offs_),
1190 name_table_entry_offs (name_table_entry_offs_)
1191 {
1192 }
1193 };
1194
1195 /* Template to unify DWARF-32 and DWARF-64 .debug_names output
1196 respecting name table width. */
1197 template<typename OffsetSize>
1198 class dwarf_tmpl : public dwarf
1199 {
1200 public:
1201 explicit dwarf_tmpl (bfd_endian dwarf5_byte_order_)
1202 : dwarf (m_name_table_string_offs, m_name_table_entry_offs),
1203 m_name_table_string_offs (dwarf5_byte_order_),
1204 m_name_table_entry_offs (dwarf5_byte_order_)
1205 {}
1206
1207 private:
1208 offset_vec_tmpl<OffsetSize> m_name_table_string_offs;
1209 offset_vec_tmpl<OffsetSize> m_name_table_entry_offs;
1210 };
1211
1212 /* Try to reconstruct original DWARF tag for given partial_symbol.
1213 This function is not DWARF-5 compliant but it is sufficient for
1214 GDB as a DWARF-5 index consumer. */
1215 static int psymbol_tag (const struct partial_symbol *psym)
1216 {
8a6d4234
TT
1217 domain_enum domain = psym->domain;
1218 enum address_class aclass = psym->aclass;
cd4fb1b2
SM
1219
1220 switch (domain)
1221 {
1222 case VAR_DOMAIN:
1223 switch (aclass)
1224 {
1225 case LOC_BLOCK:
1226 return DW_TAG_subprogram;
1227 case LOC_TYPEDEF:
1228 return DW_TAG_typedef;
1229 case LOC_COMPUTED:
1230 case LOC_CONST_BYTES:
1231 case LOC_OPTIMIZED_OUT:
1232 case LOC_STATIC:
1233 return DW_TAG_variable;
1234 case LOC_CONST:
1235 /* Note: It's currently impossible to recognize psyms as enum values
1236 short of reading the type info. For now punt. */
1237 return DW_TAG_variable;
1238 default:
1239 /* There are other LOC_FOO values that one might want to classify
1240 as variables, but dwarf2read.c doesn't currently use them. */
1241 return DW_TAG_variable;
1242 }
1243 case STRUCT_DOMAIN:
1244 return DW_TAG_structure_type;
7666722f
TV
1245 case MODULE_DOMAIN:
1246 return DW_TAG_module;
cd4fb1b2
SM
1247 default:
1248 return 0;
1249 }
1250 }
1251
1252 /* Call insert for all partial symbols and mark them in PSYMS_SEEN. */
1253 void write_psymbols (std::unordered_set<partial_symbol *> &psyms_seen,
1254 struct partial_symbol **psymp, int count, int cu_index,
1255 bool is_static, unit_kind kind)
1256 {
1257 for (; count-- > 0; ++psymp)
1258 {
1259 struct partial_symbol *psym = *psymp;
1260
cd4fb1b2
SM
1261 /* Only add a given psymbol once. */
1262 if (psyms_seen.insert (psym).second)
1263 insert (psym, cu_index, is_static, kind);
1264 }
1265 }
1266
1267 /* A helper function that writes a single signatured_type
1268 to a debug_names. */
1269 void
1270 write_one_signatured_type (struct signatured_type *entry,
1271 struct signatured_type_index_data *info)
1272 {
edfe0a0c 1273 partial_symtab *psymtab = entry->per_cu.v.psymtab;
cd4fb1b2
SM
1274
1275 write_psymbols (info->psyms_seen,
d320c2b5 1276 (info->objfile->partial_symtabs->global_psymbols.data ()
b4be9bfd 1277 + psymtab->globals_offset),
cd4fb1b2
SM
1278 psymtab->n_global_syms, info->cu_index, false,
1279 unit_kind::tu);
1280 write_psymbols (info->psyms_seen,
d320c2b5 1281 (info->objfile->partial_symtabs->static_psymbols.data ()
b4be9bfd 1282 + psymtab->statics_offset),
cd4fb1b2
SM
1283 psymtab->n_static_syms, info->cu_index, true,
1284 unit_kind::tu);
1285
1286 info->types_list.append_uint (dwarf5_offset_size (), m_dwarf5_byte_order,
1287 to_underlying (entry->per_cu.sect_off));
1288
1289 ++info->cu_index;
1290 }
1291
1292 /* Store value of each symbol. */
1293 std::unordered_map<c_str_view, std::set<symbol_value>, c_str_view_hasher>
1294 m_name_to_value_set;
1295
1296 /* Tables of DWARF-5 .debug_names. They are in object file byte
1297 order. */
1298 std::vector<uint32_t> m_bucket_table;
1299 std::vector<uint32_t> m_hash_table;
1300
1301 const bfd_endian m_dwarf5_byte_order;
1302 dwarf_tmpl<uint32_t> m_dwarf32;
1303 dwarf_tmpl<uint64_t> m_dwarf64;
1304 dwarf &m_dwarf;
1305 offset_vec &m_name_table_string_offs, &m_name_table_entry_offs;
1306 debug_str_lookup m_debugstrlookup;
1307
1308 /* Map each used .debug_names abbreviation tag parameter to its
1309 index value. */
1310 std::unordered_map<index_key, int, index_key_hasher> m_indexkey_to_idx;
1311
1312 /* Next unused .debug_names abbreviation tag for
1313 m_indexkey_to_idx. */
1314 int m_idx_next = 1;
1315
1316 /* .debug_names abbreviation table. */
1317 data_buf m_abbrev_table;
1318
1319 /* .debug_names entry pool. */
1320 data_buf m_entry_pool;
3b00ef10
TT
1321
1322 /* Temporary storage for Ada names. */
1323 auto_obstack m_string_obstack;
cd4fb1b2
SM
1324};
1325
1326/* Return iff any of the needed offsets does not fit into 32-bit
1327 .debug_names section. */
1328
1329static bool
976ca316 1330check_dwarf64_offsets (dwarf2_per_objfile *per_objfile)
cd4fb1b2 1331{
976ca316 1332 for (dwarf2_per_cu_data *per_cu : per_objfile->per_bfd->all_comp_units)
cd4fb1b2 1333 {
b76e467d 1334 if (to_underlying (per_cu->sect_off) >= (static_cast<uint64_t> (1) << 32))
cd4fb1b2
SM
1335 return true;
1336 }
976ca316 1337 for (const signatured_type *sigtype : per_objfile->per_bfd->all_type_units)
cd4fb1b2 1338 {
b2bdb8cf 1339 const dwarf2_per_cu_data &per_cu = sigtype->per_cu;
cd4fb1b2
SM
1340
1341 if (to_underlying (per_cu.sect_off) >= (static_cast<uint64_t> (1) << 32))
1342 return true;
1343 }
1344 return false;
1345}
1346
1347/* The psyms_seen set is potentially going to be largish (~40k
1348 elements when indexing a -g3 build of GDB itself). Estimate the
1349 number of elements in order to avoid too many rehashes, which
1350 require rebuilding buckets and thus many trips to
1351 malloc/free. */
1352
1353static size_t
976ca316 1354psyms_seen_size (dwarf2_per_objfile *per_objfile)
cd4fb1b2
SM
1355{
1356 size_t psyms_count = 0;
976ca316 1357 for (dwarf2_per_cu_data *per_cu : per_objfile->per_bfd->all_comp_units)
cd4fb1b2 1358 {
edfe0a0c 1359 partial_symtab *psymtab = per_cu->v.psymtab;
cd4fb1b2
SM
1360
1361 if (psymtab != NULL && psymtab->user == NULL)
1362 recursively_count_psymbols (psymtab, psyms_count);
1363 }
1364 /* Generating an index for gdb itself shows a ratio of
1365 TOTAL_SEEN_SYMS/UNIQUE_SYMS or ~5. 4 seems like a good bet. */
1366 return psyms_count / 4;
1367}
1368
c4973306
SM
1369/* Assert that FILE's size is EXPECTED_SIZE. Assumes file's seek
1370 position is at the end of the file. */
cd4fb1b2 1371
c4973306
SM
1372static void
1373assert_file_size (FILE *file, size_t expected_size)
1374{
1375 const auto file_size = ftell (file);
1376 if (file_size == -1)
1377 perror_with_name (("ftell"));
1378 gdb_assert (file_size == expected_size);
1379}
1380
1381/* Write a gdb index file to OUT_FILE from all the sections passed as
1382 arguments. */
1383
1384static void
1385write_gdbindex_1 (FILE *out_file,
1386 const data_buf &cu_list,
1387 const data_buf &types_cu_list,
1388 const data_buf &addr_vec,
1389 const data_buf &symtab_vec,
1390 const data_buf &constant_pool)
1391{
1392 data_buf contents;
1393 const offset_type size_of_header = 6 * sizeof (offset_type);
1394 offset_type total_len = size_of_header;
1395
1396 /* The version number. */
1397 contents.append_data (MAYBE_SWAP (8));
1398
1399 /* The offset of the CU list from the start of the file. */
1400 contents.append_data (MAYBE_SWAP (total_len));
1401 total_len += cu_list.size ();
1402
1403 /* The offset of the types CU list from the start of the file. */
1404 contents.append_data (MAYBE_SWAP (total_len));
1405 total_len += types_cu_list.size ();
1406
1407 /* The offset of the address table from the start of the file. */
1408 contents.append_data (MAYBE_SWAP (total_len));
1409 total_len += addr_vec.size ();
1410
1411 /* The offset of the symbol table from the start of the file. */
1412 contents.append_data (MAYBE_SWAP (total_len));
1413 total_len += symtab_vec.size ();
1414
1415 /* The offset of the constant pool from the start of the file. */
1416 contents.append_data (MAYBE_SWAP (total_len));
1417 total_len += constant_pool.size ();
1418
1419 gdb_assert (contents.size () == size_of_header);
1420
1421 contents.file_write (out_file);
1422 cu_list.file_write (out_file);
1423 types_cu_list.file_write (out_file);
1424 addr_vec.file_write (out_file);
1425 symtab_vec.file_write (out_file);
1426 constant_pool.file_write (out_file);
1427
1428 assert_file_size (out_file, total_len);
1429}
1430
1431/* Write contents of a .gdb_index section for OBJFILE into OUT_FILE.
1432 If OBJFILE has an associated dwz file, write contents of a .gdb_index
1433 section for that dwz file into DWZ_OUT_FILE. If OBJFILE does not have an
1434 associated dwz file, DWZ_OUT_FILE must be NULL. */
1435
1436static void
976ca316 1437write_gdbindex (dwarf2_per_objfile *per_objfile, FILE *out_file,
c4973306 1438 FILE *dwz_out_file)
cd4fb1b2 1439{
976ca316 1440 struct objfile *objfile = per_objfile->objfile;
cd4fb1b2 1441 mapped_symtab symtab;
c4973306
SM
1442 data_buf objfile_cu_list;
1443 data_buf dwz_cu_list;
cd4fb1b2
SM
1444
1445 /* While we're scanning CU's create a table that maps a psymtab pointer
1446 (which is what addrmap records) to its index (which is what is recorded
1447 in the index file). This will later be needed to write the address
1448 table. */
1449 psym_index_map cu_index_htab;
976ca316 1450 cu_index_htab.reserve (per_objfile->per_bfd->all_comp_units.size ());
cd4fb1b2
SM
1451
1452 /* The CU list is already sorted, so we don't need to do additional
1453 work here. Also, the debug_types entries do not appear in
1454 all_comp_units, but only in their own hash table. */
1455
1456 std::unordered_set<partial_symbol *> psyms_seen
976ca316
SM
1457 (psyms_seen_size (per_objfile));
1458 for (int i = 0; i < per_objfile->per_bfd->all_comp_units.size (); ++i)
cd4fb1b2 1459 {
976ca316 1460 dwarf2_per_cu_data *per_cu = per_objfile->per_bfd->all_comp_units[i];
edfe0a0c 1461 partial_symtab *psymtab = per_cu->v.psymtab;
cd4fb1b2 1462
efba5c23
TV
1463 if (psymtab != NULL)
1464 {
1465 if (psymtab->user == NULL)
1466 recursively_write_psymbols (objfile, psymtab, &symtab,
1467 psyms_seen, i);
1468
1469 const auto insertpair = cu_index_htab.emplace (psymtab, i);
1470 gdb_assert (insertpair.second);
1471 }
cd4fb1b2 1472
c4973306
SM
1473 /* The all_comp_units list contains CUs read from the objfile as well as
1474 from the eventual dwz file. We need to place the entry in the
1475 corresponding index. */
1476 data_buf &cu_list = per_cu->is_dwz ? dwz_cu_list : objfile_cu_list;
cd4fb1b2
SM
1477 cu_list.append_uint (8, BFD_ENDIAN_LITTLE,
1478 to_underlying (per_cu->sect_off));
1479 cu_list.append_uint (8, BFD_ENDIAN_LITTLE, per_cu->length);
1480 }
1481
1482 /* Dump the address map. */
1483 data_buf addr_vec;
1484 write_address_map (objfile, addr_vec, cu_index_htab);
1485
1486 /* Write out the .debug_type entries, if any. */
1487 data_buf types_cu_list;
976ca316 1488 if (per_objfile->per_bfd->signatured_types)
cd4fb1b2
SM
1489 {
1490 signatured_type_index_data sig_data (types_cu_list,
1491 psyms_seen);
1492
1493 sig_data.objfile = objfile;
1494 sig_data.symtab = &symtab;
976ca316
SM
1495 sig_data.cu_index = per_objfile->per_bfd->all_comp_units.size ();
1496 htab_traverse_noresize (per_objfile->per_bfd->signatured_types.get (),
cd4fb1b2
SM
1497 write_one_signatured_type, &sig_data);
1498 }
1499
1500 /* Now that we've processed all symbols we can shrink their cu_indices
1501 lists. */
1502 uniquify_cu_indices (&symtab);
1503
1504 data_buf symtab_vec, constant_pool;
1505 write_hash_table (&symtab, symtab_vec, constant_pool);
1506
c4973306
SM
1507 write_gdbindex_1(out_file, objfile_cu_list, types_cu_list, addr_vec,
1508 symtab_vec, constant_pool);
cd4fb1b2 1509
c4973306
SM
1510 if (dwz_out_file != NULL)
1511 write_gdbindex_1 (dwz_out_file, dwz_cu_list, {}, {}, {}, {});
1512 else
1513 gdb_assert (dwz_cu_list.empty ());
cd4fb1b2
SM
1514}
1515
1516/* DWARF-5 augmentation string for GDB's DW_IDX_GNU_* extension. */
1517static const gdb_byte dwarf5_gdb_augmentation[] = { 'G', 'D', 'B', 0 };
1518
1519/* Write a new .debug_names section for OBJFILE into OUT_FILE, write
1520 needed addition to .debug_str section to OUT_FILE_STR. Return how
1521 many bytes were expected to be written into OUT_FILE. */
1522
c4973306 1523static void
976ca316 1524write_debug_names (dwarf2_per_objfile *per_objfile,
cd4fb1b2
SM
1525 FILE *out_file, FILE *out_file_str)
1526{
976ca316
SM
1527 const bool dwarf5_is_dwarf64 = check_dwarf64_offsets (per_objfile);
1528 struct objfile *objfile = per_objfile->objfile;
cd4fb1b2 1529 const enum bfd_endian dwarf5_byte_order
08feed99 1530 = gdbarch_byte_order (objfile->arch ());
cd4fb1b2
SM
1531
1532 /* The CU list is already sorted, so we don't need to do additional
1533 work here. Also, the debug_types entries do not appear in
1534 all_comp_units, but only in their own hash table. */
1535 data_buf cu_list;
976ca316 1536 debug_names nametable (per_objfile, dwarf5_is_dwarf64, dwarf5_byte_order);
cd4fb1b2 1537 std::unordered_set<partial_symbol *>
976ca316
SM
1538 psyms_seen (psyms_seen_size (per_objfile));
1539 for (int i = 0; i < per_objfile->per_bfd->all_comp_units.size (); ++i)
cd4fb1b2 1540 {
976ca316 1541 const dwarf2_per_cu_data *per_cu = per_objfile->per_bfd->all_comp_units[i];
edfe0a0c 1542 partial_symtab *psymtab = per_cu->v.psymtab;
cd4fb1b2
SM
1543
1544 /* CU of a shared file from 'dwz -m' may be unused by this main
1545 file. It may be referenced from a local scope but in such
1546 case it does not need to be present in .debug_names. */
1547 if (psymtab == NULL)
1548 continue;
1549
1550 if (psymtab->user == NULL)
1551 nametable.recursively_write_psymbols (objfile, psymtab, psyms_seen, i);
1552
1553 cu_list.append_uint (nametable.dwarf5_offset_size (), dwarf5_byte_order,
1554 to_underlying (per_cu->sect_off));
1555 }
1556
1557 /* Write out the .debug_type entries, if any. */
1558 data_buf types_cu_list;
976ca316 1559 if (per_objfile->per_bfd->signatured_types)
cd4fb1b2
SM
1560 {
1561 debug_names::write_one_signatured_type_data sig_data (nametable,
1562 signatured_type_index_data (types_cu_list, psyms_seen));
1563
1564 sig_data.info.objfile = objfile;
1565 /* It is used only for gdb_index. */
1566 sig_data.info.symtab = nullptr;
1567 sig_data.info.cu_index = 0;
976ca316 1568 htab_traverse_noresize (per_objfile->per_bfd->signatured_types.get (),
cd4fb1b2
SM
1569 debug_names::write_one_signatured_type,
1570 &sig_data);
1571 }
1572
1573 nametable.build ();
1574
1575 /* No addr_vec - DWARF-5 uses .debug_aranges generated by GCC. */
1576
1577 const offset_type bytes_of_header
1578 = ((dwarf5_is_dwarf64 ? 12 : 4)
1579 + 2 + 2 + 7 * 4
1580 + sizeof (dwarf5_gdb_augmentation));
1581 size_t expected_bytes = 0;
1582 expected_bytes += bytes_of_header;
1583 expected_bytes += cu_list.size ();
1584 expected_bytes += types_cu_list.size ();
1585 expected_bytes += nametable.bytes ();
1586 data_buf header;
1587
1588 if (!dwarf5_is_dwarf64)
1589 {
1590 const uint64_t size64 = expected_bytes - 4;
1591 gdb_assert (size64 < 0xfffffff0);
1592 header.append_uint (4, dwarf5_byte_order, size64);
1593 }
1594 else
1595 {
1596 header.append_uint (4, dwarf5_byte_order, 0xffffffff);
1597 header.append_uint (8, dwarf5_byte_order, expected_bytes - 12);
1598 }
1599
1600 /* The version number. */
1601 header.append_uint (2, dwarf5_byte_order, 5);
1602
1603 /* Padding. */
1604 header.append_uint (2, dwarf5_byte_order, 0);
1605
1606 /* comp_unit_count - The number of CUs in the CU list. */
b76e467d 1607 header.append_uint (4, dwarf5_byte_order,
976ca316 1608 per_objfile->per_bfd->all_comp_units.size ());
cd4fb1b2
SM
1609
1610 /* local_type_unit_count - The number of TUs in the local TU
1611 list. */
b2bdb8cf 1612 header.append_uint (4, dwarf5_byte_order,
976ca316 1613 per_objfile->per_bfd->all_type_units.size ());
cd4fb1b2
SM
1614
1615 /* foreign_type_unit_count - The number of TUs in the foreign TU
1616 list. */
1617 header.append_uint (4, dwarf5_byte_order, 0);
1618
1619 /* bucket_count - The number of hash buckets in the hash lookup
1620 table. */
1621 header.append_uint (4, dwarf5_byte_order, nametable.bucket_count ());
1622
1623 /* name_count - The number of unique names in the index. */
1624 header.append_uint (4, dwarf5_byte_order, nametable.name_count ());
1625
1626 /* abbrev_table_size - The size in bytes of the abbreviations
1627 table. */
1628 header.append_uint (4, dwarf5_byte_order, nametable.abbrev_table_bytes ());
1629
1630 /* augmentation_string_size - The size in bytes of the augmentation
1631 string. This value is rounded up to a multiple of 4. */
1632 static_assert (sizeof (dwarf5_gdb_augmentation) % 4 == 0, "");
1633 header.append_uint (4, dwarf5_byte_order, sizeof (dwarf5_gdb_augmentation));
1634 header.append_data (dwarf5_gdb_augmentation);
1635
1636 gdb_assert (header.size () == bytes_of_header);
1637
1638 header.file_write (out_file);
1639 cu_list.file_write (out_file);
1640 types_cu_list.file_write (out_file);
1641 nametable.file_write (out_file, out_file_str);
1642
c4973306 1643 assert_file_size (out_file, expected_bytes);
cd4fb1b2
SM
1644}
1645
c4973306 1646/* This represents an index file being written (work-in-progress).
cd4fb1b2 1647
c4973306
SM
1648 The data is initially written to a temporary file. When the finalize method
1649 is called, the file is closed and moved to its final location.
1650
1651 On failure (if this object is being destroyed with having called finalize),
1652 the temporary file is closed and deleted. */
1653
1654struct index_wip_file
cd4fb1b2 1655{
c4973306
SM
1656 index_wip_file (const char *dir, const char *basename,
1657 const char *suffix)
1658 {
1659 filename = (std::string (dir) + SLASH_STRING + basename
1660 + suffix);
1661
1662 filename_temp = make_temp_filename (filename);
1663
1664 scoped_fd out_file_fd (gdb_mkostemp_cloexec (filename_temp.data (),
1665 O_BINARY));
1666 if (out_file_fd.get () == -1)
1667 perror_with_name (("mkstemp"));
1668
1669 out_file = out_file_fd.to_file ("wb");
1670
1671 if (out_file == nullptr)
1672 error (_("Can't open `%s' for writing"), filename_temp.data ());
1673
1674 unlink_file.emplace (filename_temp.data ());
1675 }
1676
1677 void finalize ()
1678 {
1679 /* We want to keep the file. */
1680 unlink_file->keep ();
1681
1682 /* Close and move the str file in place. */
1683 unlink_file.reset ();
1684 if (rename (filename_temp.data (), filename.c_str ()) != 0)
1685 perror_with_name (("rename"));
1686 }
1687
1688 std::string filename;
1689 gdb::char_vector filename_temp;
1690
1691 /* Order matters here; we want FILE to be closed before
1692 FILENAME_TEMP is unlinked, because on MS-Windows one cannot
1693 delete a file that is still open. So, we wrap the unlinker in an
1694 optional and emplace it once we know the file name. */
1695 gdb::optional<gdb::unlinker> unlink_file;
1696
1697 gdb_file_up out_file;
1698};
cd4fb1b2 1699
87d6a7aa 1700/* See dwarf-index-write.h. */
cd4fb1b2 1701
87d6a7aa 1702void
976ca316
SM
1703write_psymtabs_to_index (dwarf2_per_objfile *per_objfile, const char *dir,
1704 const char *basename, const char *dwz_basename,
cd4fb1b2
SM
1705 dw_index_kind index_kind)
1706{
976ca316 1707 struct objfile *objfile = per_objfile->objfile;
cd4fb1b2 1708
976ca316 1709 if (per_objfile->per_bfd->using_index)
cd4fb1b2
SM
1710 error (_("Cannot use an index to create the index"));
1711
976ca316 1712 if (per_objfile->per_bfd->types.size () > 1)
cd4fb1b2
SM
1713 error (_("Cannot make an index when the file has multiple .debug_types sections"));
1714
d320c2b5
TT
1715 if (!objfile->partial_symtabs->psymtabs
1716 || !objfile->partial_symtabs->psymtabs_addrmap)
cd4fb1b2
SM
1717 return;
1718
1719 struct stat st;
1720 if (stat (objfile_name (objfile), &st) < 0)
1721 perror_with_name (objfile_name (objfile));
1722
c4973306
SM
1723 const char *index_suffix = (index_kind == dw_index_kind::DEBUG_NAMES
1724 ? INDEX5_SUFFIX : INDEX4_SUFFIX);
cd4fb1b2 1725
c4973306
SM
1726 index_wip_file objfile_index_wip (dir, basename, index_suffix);
1727 gdb::optional<index_wip_file> dwz_index_wip;
cd4fb1b2 1728
c4973306
SM
1729 if (dwz_basename != NULL)
1730 dwz_index_wip.emplace (dir, dwz_basename, index_suffix);
cd4fb1b2
SM
1731
1732 if (index_kind == dw_index_kind::DEBUG_NAMES)
1733 {
c4973306
SM
1734 index_wip_file str_wip_file (dir, basename, DEBUG_STR_SUFFIX);
1735
976ca316 1736 write_debug_names (per_objfile, objfile_index_wip.out_file.get (),
c4973306
SM
1737 str_wip_file.out_file.get ());
1738
1739 str_wip_file.finalize ();
cd4fb1b2
SM
1740 }
1741 else
976ca316 1742 write_gdbindex (per_objfile, objfile_index_wip.out_file.get (),
c4973306
SM
1743 (dwz_index_wip.has_value ()
1744 ? dwz_index_wip->out_file.get () : NULL));
cd4fb1b2 1745
c4973306 1746 objfile_index_wip.finalize ();
87d6a7aa 1747
c4973306
SM
1748 if (dwz_index_wip.has_value ())
1749 dwz_index_wip->finalize ();
cd4fb1b2
SM
1750}
1751
1752/* Implementation of the `save gdb-index' command.
1753
1754 Note that the .gdb_index file format used by this command is
1755 documented in the GDB manual. Any changes here must be documented
1756 there. */
1757
1758static void
1759save_gdb_index_command (const char *arg, int from_tty)
1760{
cd4fb1b2
SM
1761 const char dwarf5space[] = "-dwarf-5 ";
1762 dw_index_kind index_kind = dw_index_kind::GDB_INDEX;
1763
1764 if (!arg)
1765 arg = "";
1766
1767 arg = skip_spaces (arg);
1768 if (strncmp (arg, dwarf5space, strlen (dwarf5space)) == 0)
1769 {
1770 index_kind = dw_index_kind::DEBUG_NAMES;
1771 arg += strlen (dwarf5space);
1772 arg = skip_spaces (arg);
1773 }
1774
1775 if (!*arg)
1776 error (_("usage: save gdb-index [-dwarf-5] DIRECTORY"));
1777
2030c079 1778 for (objfile *objfile : current_program_space->objfiles ())
aed57c53
TT
1779 {
1780 struct stat st;
cd4fb1b2 1781
aed57c53
TT
1782 /* If the objfile does not correspond to an actual file, skip it. */
1783 if (stat (objfile_name (objfile), &st) < 0)
1784 continue;
cd4fb1b2 1785
976ca316 1786 dwarf2_per_objfile *per_objfile = get_dwarf2_per_objfile (objfile);
cd4fb1b2 1787
976ca316 1788 if (per_objfile != NULL)
aed57c53 1789 {
a70b8144 1790 try
aed57c53
TT
1791 {
1792 const char *basename = lbasename (objfile_name (objfile));
976ca316 1793 const dwz_file *dwz = dwarf2_get_dwz_file (per_objfile->per_bfd);
c4973306
SM
1794 const char *dwz_basename = NULL;
1795
1796 if (dwz != NULL)
1797 dwz_basename = lbasename (dwz->filename ());
1798
976ca316
SM
1799 write_psymtabs_to_index (per_objfile, arg, basename, dwz_basename,
1800 index_kind);
aed57c53 1801 }
230d2906 1802 catch (const gdb_exception_error &except)
aed57c53
TT
1803 {
1804 exception_fprintf (gdb_stderr, except,
1805 _("Error while writing index for `%s': "),
1806 objfile_name (objfile));
1807 }
aed57c53 1808 }
cd4fb1b2 1809
aed57c53 1810 }
cd4fb1b2
SM
1811}
1812
6c265988 1813void _initialize_dwarf_index_write ();
cd4fb1b2
SM
1814void
1815_initialize_dwarf_index_write ()
1816{
1817 cmd_list_element *c = add_cmd ("gdb-index", class_files,
1818 save_gdb_index_command, _("\
1819Save a gdb-index file.\n\
1820Usage: save gdb-index [-dwarf-5] DIRECTORY\n\
1821\n\
1822No options create one file with .gdb-index extension for pre-DWARF-5\n\
1823compatible .gdb_index section. With -dwarf-5 creates two files with\n\
1824extension .debug_names and .debug_str for DWARF-5 .debug_names section."),
1825 &save_cmdlist);
1826 set_cmd_completer (c, filename_completer);
1827}
This page took 0.31791 seconds and 4 git commands to generate.