Refresh regs window in display_registers_from
[deliverable/binutils-gdb.git] / gdb / dwarf2 / index-write.c
CommitLineData
cd4fb1b2
SM
1/* DWARF index writing support for GDB.
2
3666a048 3 Copyright (C) 1994-2021 Free Software Foundation, Inc.
cd4fb1b2
SM
4
5 This file is part of GDB.
6
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
11
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
16
17 You should have received a copy of the GNU General Public License
18 along with this program. If not, see <http://www.gnu.org/licenses/>. */
19
20#include "defs.h"
21
82ca8957 22#include "dwarf2/index-write.h"
23baa4cc 23
cd4fb1b2
SM
24#include "addrmap.h"
25#include "cli/cli-decode.h"
268a13a5
TT
26#include "gdbsupport/byte-vector.h"
27#include "gdbsupport/filestuff.h"
28#include "gdbsupport/gdb_unlinker.h"
29#include "gdbsupport/pathstuff.h"
30#include "gdbsupport/scoped_fd.h"
cd4fb1b2 31#include "complaints.h"
82ca8957 32#include "dwarf2/index-common.h"
cd4fb1b2 33#include "dwarf2.h"
82ca8957 34#include "dwarf2/read.h"
9fda78b6 35#include "dwarf2/dwz.h"
cd4fb1b2
SM
36#include "gdb/gdb-index.h"
37#include "gdbcmd.h"
38#include "objfiles.h"
39#include "psympriv.h"
3b00ef10 40#include "ada-lang.h"
cd4fb1b2 41
4de283e4
TT
42#include <algorithm>
43#include <cmath>
159ed7d9 44#include <forward_list>
4de283e4
TT
45#include <set>
46#include <unordered_map>
47#include <unordered_set>
48
cd4fb1b2
SM
49/* Ensure only legit values are used. */
50#define DW2_GDB_INDEX_SYMBOL_STATIC_SET_VALUE(cu_index, value) \
51 do { \
52 gdb_assert ((unsigned int) (value) <= 1); \
53 GDB_INDEX_SYMBOL_STATIC_SET_VALUE((cu_index), (value)); \
54 } while (0)
55
56/* Ensure only legit values are used. */
57#define DW2_GDB_INDEX_SYMBOL_KIND_SET_VALUE(cu_index, value) \
58 do { \
59 gdb_assert ((value) >= GDB_INDEX_SYMBOL_KIND_TYPE \
dda83cd7 60 && (value) <= GDB_INDEX_SYMBOL_KIND_OTHER); \
cd4fb1b2
SM
61 GDB_INDEX_SYMBOL_KIND_SET_VALUE((cu_index), (value)); \
62 } while (0)
63
85102364 64/* Ensure we don't use more than the allotted number of bits for the CU. */
cd4fb1b2
SM
65#define DW2_GDB_INDEX_CU_SET_VALUE(cu_index, value) \
66 do { \
67 gdb_assert (((value) & ~GDB_INDEX_CU_MASK) == 0); \
68 GDB_INDEX_CU_SET_VALUE((cu_index), (value)); \
69 } while (0)
70
71/* The "save gdb-index" command. */
72
73/* Write SIZE bytes from the buffer pointed to by DATA to FILE, with
74 error checking. */
75
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,
932539d7 542 const std::vector<partial_symbol *> &symbols,
cd4fb1b2
SM
543 offset_type cu_index,
544 int is_static)
545{
932539d7 546 for (partial_symbol *psym : symbols)
cd4fb1b2 547 {
7ab96794 548 const char *name = psym->ginfo.search_name ();
cd4fb1b2 549
c1b5c1eb 550 if (psym->ginfo.language () == language_ada)
7ab96794
TV
551 {
552 /* We want to ensure that the Ada main function's name appears
553 verbatim in the index. However, this name will be of the
554 form "_ada_mumble", and will be rewritten by ada_decode.
555 So, recognize it specially here and add it to the index by
556 hand. */
557 if (strcmp (main_name (), name) == 0)
558 {
559 gdb_index_symbol_kind kind = symbol_kind (psym);
560
561 add_index_entry (symtab, name, is_static, kind, cu_index);
562 }
563
564 /* In order for the index to work when read back into gdb, it
565 has to supply a funny form of the name: it should be the
566 encoded name, with any suffixes stripped. Using the
567 ordinary encoded name will not work properly with the
568 searching logic in find_name_components_bounds; nor will
569 using the decoded name. Furthermore, an Ada "verbatim"
570 name (of the form "<MumBle>") must be entered without the
571 angle brackets. Note that the current index is unusual,
572 see PR symtab/24820 for details. */
573 std::string decoded = ada_decode (name);
574 if (decoded[0] == '<')
575 name = (char *) obstack_copy0 (&symtab->m_string_obstack,
576 decoded.c_str () + 1,
577 decoded.length () - 2);
578 else
579 name = obstack_strdup (&symtab->m_string_obstack,
580 ada_encode (decoded.c_str ()));
581 }
cd4fb1b2
SM
582
583 /* Only add a given psymbol once. */
584 if (psyms_seen.insert (psym).second)
585 {
586 gdb_index_symbol_kind kind = symbol_kind (psym);
587
7ab96794 588 add_index_entry (symtab, name, is_static, kind, cu_index);
cd4fb1b2
SM
589 }
590 }
591}
592
593/* A helper struct used when iterating over debug_types. */
594struct signatured_type_index_data
595{
596 signatured_type_index_data (data_buf &types_list_,
dda83cd7 597 std::unordered_set<partial_symbol *> &psyms_seen_)
cd4fb1b2
SM
598 : types_list (types_list_), psyms_seen (psyms_seen_)
599 {}
600
601 struct objfile *objfile;
602 struct mapped_symtab *symtab;
603 data_buf &types_list;
604 std::unordered_set<partial_symbol *> &psyms_seen;
605 int cu_index;
606};
607
608/* A helper function that writes a single signatured_type to an
609 obstack. */
610
611static int
612write_one_signatured_type (void **slot, void *d)
613{
614 struct signatured_type_index_data *info
615 = (struct signatured_type_index_data *) d;
616 struct signatured_type *entry = (struct signatured_type *) *slot;
edfe0a0c 617 partial_symtab *psymtab = entry->per_cu.v.psymtab;
cd4fb1b2 618
2bd3e4b8
TV
619 if (psymtab == nullptr)
620 {
621 /* We can end up here when processing a skeleton CU referring to a
622 .dwo file that hasn't been found. There's not much we can do in
623 such a case, so skip this CU. */
624 return 1;
625 }
626
932539d7
TT
627 write_psymbols (info->symtab, info->psyms_seen,
628 psymtab->global_psymbols, info->cu_index,
cd4fb1b2 629 0);
932539d7
TT
630 write_psymbols (info->symtab, info->psyms_seen,
631 psymtab->static_psymbols, info->cu_index,
cd4fb1b2
SM
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
932539d7
TT
657 psyms_seen += psymtab->global_psymbols.size ();
658 psyms_seen += psymtab->static_psymbols.size ();
cd4fb1b2
SM
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
932539d7
TT
679 write_psymbols (symtab, psyms_seen,
680 psymtab->global_psymbols, cu_index,
cd4fb1b2 681 0);
932539d7
TT
682 write_psymbols (symtab, psyms_seen,
683 psymtab->static_psymbols, cu_index,
cd4fb1b2
SM
684 1);
685}
686
687/* DWARF-5 .debug_names builder. */
688class debug_names
689{
690public:
976ca316 691 debug_names (dwarf2_per_objfile *per_objfile, bool is_dwarf64,
cd4fb1b2
SM
692 bfd_endian dwarf5_byte_order)
693 : m_dwarf5_byte_order (dwarf5_byte_order),
694 m_dwarf32 (dwarf5_byte_order),
695 m_dwarf64 (dwarf5_byte_order),
696 m_dwarf (is_dwarf64
697 ? static_cast<dwarf &> (m_dwarf64)
698 : static_cast<dwarf &> (m_dwarf32)),
699 m_name_table_string_offs (m_dwarf.name_table_string_offs),
700 m_name_table_entry_offs (m_dwarf.name_table_entry_offs),
976ca316 701 m_debugstrlookup (per_objfile)
cd4fb1b2
SM
702 {}
703
704 int dwarf5_offset_size () const
705 {
706 const bool dwarf5_is_dwarf64 = &m_dwarf == &m_dwarf64;
707 return dwarf5_is_dwarf64 ? 8 : 4;
708 }
709
710 /* Is this symbol from DW_TAG_compile_unit or DW_TAG_type_unit? */
711 enum class unit_kind { cu, tu };
712
713 /* Insert one symbol. */
714 void insert (const partial_symbol *psym, int cu_index, bool is_static,
715 unit_kind kind)
716 {
717 const int dwarf_tag = psymbol_tag (psym);
718 if (dwarf_tag == 0)
719 return;
c9d95fa3 720 const char *name = psym->ginfo.search_name ();
3b00ef10 721
c1b5c1eb 722 if (psym->ginfo.language () == language_ada)
3b00ef10
TT
723 {
724 /* We want to ensure that the Ada main function's name appears
725 verbatim in the index. However, this name will be of the
726 form "_ada_mumble", and will be rewritten by ada_decode.
727 So, recognize it specially here and add it to the index by
728 hand. */
729 if (strcmp (main_name (), name) == 0)
730 {
731 const auto insertpair
732 = m_name_to_value_set.emplace (c_str_view (name),
733 std::set<symbol_value> ());
734 std::set<symbol_value> &value_set = insertpair.first->second;
735 value_set.emplace (symbol_value (dwarf_tag, cu_index, is_static,
736 kind));
737 }
738
739 /* In order for the index to work when read back into gdb, it
740 has to supply a funny form of the name: it should be the
741 encoded name, with any suffixes stripped. Using the
742 ordinary encoded name will not work properly with the
743 searching logic in find_name_components_bounds; nor will
744 using the decoded name. Furthermore, an Ada "verbatim"
745 name (of the form "<MumBle>") must be entered without the
746 angle brackets. Note that the current index is unusual,
747 see PR symtab/24820 for details. */
f945dedf 748 std::string decoded = ada_decode (name);
3b00ef10
TT
749 if (decoded[0] == '<')
750 name = (char *) obstack_copy0 (&m_string_obstack,
f945dedf
CB
751 decoded.c_str () + 1,
752 decoded.length () - 2);
3b00ef10 753 else
f945dedf
CB
754 name = obstack_strdup (&m_string_obstack,
755 ada_encode (decoded.c_str ()));
3b00ef10
TT
756 }
757
cd4fb1b2
SM
758 const auto insertpair
759 = m_name_to_value_set.emplace (c_str_view (name),
760 std::set<symbol_value> ());
761 std::set<symbol_value> &value_set = insertpair.first->second;
762 value_set.emplace (symbol_value (dwarf_tag, cu_index, is_static, kind));
763 }
764
765 /* Build all the tables. All symbols must be already inserted.
766 This function does not call file_write, caller has to do it
767 afterwards. */
768 void build ()
769 {
770 /* Verify the build method has not be called twice. */
771 gdb_assert (m_abbrev_table.empty ());
772 const size_t name_count = m_name_to_value_set.size ();
773 m_bucket_table.resize
774 (std::pow (2, std::ceil (std::log2 (name_count * 4 / 3))));
775 m_hash_table.reserve (name_count);
776 m_name_table_string_offs.reserve (name_count);
777 m_name_table_entry_offs.reserve (name_count);
778
779 /* Map each hash of symbol to its name and value. */
780 struct hash_it_pair
781 {
782 uint32_t hash;
783 decltype (m_name_to_value_set)::const_iterator it;
784 };
785 std::vector<std::forward_list<hash_it_pair>> bucket_hash;
786 bucket_hash.resize (m_bucket_table.size ());
787 for (decltype (m_name_to_value_set)::const_iterator it
788 = m_name_to_value_set.cbegin ();
789 it != m_name_to_value_set.cend ();
790 ++it)
791 {
792 const char *const name = it->first.c_str ();
793 const uint32_t hash = dwarf5_djb_hash (name);
794 hash_it_pair hashitpair;
795 hashitpair.hash = hash;
796 hashitpair.it = it;
797 auto &slot = bucket_hash[hash % bucket_hash.size()];
798 slot.push_front (std::move (hashitpair));
799 }
800 for (size_t bucket_ix = 0; bucket_ix < bucket_hash.size (); ++bucket_ix)
801 {
802 const std::forward_list<hash_it_pair> &hashitlist
803 = bucket_hash[bucket_ix];
804 if (hashitlist.empty ())
805 continue;
806 uint32_t &bucket_slot = m_bucket_table[bucket_ix];
807 /* The hashes array is indexed starting at 1. */
808 store_unsigned_integer (reinterpret_cast<gdb_byte *> (&bucket_slot),
809 sizeof (bucket_slot), m_dwarf5_byte_order,
810 m_hash_table.size () + 1);
811 for (const hash_it_pair &hashitpair : hashitlist)
812 {
813 m_hash_table.push_back (0);
814 store_unsigned_integer (reinterpret_cast<gdb_byte *>
815 (&m_hash_table.back ()),
816 sizeof (m_hash_table.back ()),
817 m_dwarf5_byte_order, hashitpair.hash);
818 const c_str_view &name = hashitpair.it->first;
819 const std::set<symbol_value> &value_set = hashitpair.it->second;
820 m_name_table_string_offs.push_back_reorder
821 (m_debugstrlookup.lookup (name.c_str ()));
822 m_name_table_entry_offs.push_back_reorder (m_entry_pool.size ());
823 gdb_assert (!value_set.empty ());
824 for (const symbol_value &value : value_set)
825 {
826 int &idx = m_indexkey_to_idx[index_key (value.dwarf_tag,
827 value.is_static,
828 value.kind)];
829 if (idx == 0)
830 {
831 idx = m_idx_next++;
832 m_abbrev_table.append_unsigned_leb128 (idx);
833 m_abbrev_table.append_unsigned_leb128 (value.dwarf_tag);
834 m_abbrev_table.append_unsigned_leb128
835 (value.kind == unit_kind::cu ? DW_IDX_compile_unit
836 : DW_IDX_type_unit);
837 m_abbrev_table.append_unsigned_leb128 (DW_FORM_udata);
838 m_abbrev_table.append_unsigned_leb128 (value.is_static
839 ? DW_IDX_GNU_internal
840 : DW_IDX_GNU_external);
841 m_abbrev_table.append_unsigned_leb128 (DW_FORM_flag_present);
842
843 /* Terminate attributes list. */
844 m_abbrev_table.append_unsigned_leb128 (0);
845 m_abbrev_table.append_unsigned_leb128 (0);
846 }
847
848 m_entry_pool.append_unsigned_leb128 (idx);
849 m_entry_pool.append_unsigned_leb128 (value.cu_index);
850 }
851
852 /* Terminate the list of CUs. */
853 m_entry_pool.append_unsigned_leb128 (0);
854 }
855 }
856 gdb_assert (m_hash_table.size () == name_count);
857
858 /* Terminate tags list. */
859 m_abbrev_table.append_unsigned_leb128 (0);
860 }
861
862 /* Return .debug_names bucket count. This must be called only after
863 calling the build method. */
864 uint32_t bucket_count () const
865 {
866 /* Verify the build method has been already called. */
867 gdb_assert (!m_abbrev_table.empty ());
868 const uint32_t retval = m_bucket_table.size ();
869
870 /* Check for overflow. */
871 gdb_assert (retval == m_bucket_table.size ());
872 return retval;
873 }
874
875 /* Return .debug_names names count. This must be called only after
876 calling the build method. */
877 uint32_t name_count () const
878 {
879 /* Verify the build method has been already called. */
880 gdb_assert (!m_abbrev_table.empty ());
881 const uint32_t retval = m_hash_table.size ();
882
883 /* Check for overflow. */
884 gdb_assert (retval == m_hash_table.size ());
885 return retval;
886 }
887
888 /* Return number of bytes of .debug_names abbreviation table. This
889 must be called only after calling the build method. */
890 uint32_t abbrev_table_bytes () const
891 {
892 gdb_assert (!m_abbrev_table.empty ());
893 return m_abbrev_table.size ();
894 }
895
896 /* Recurse into all "included" dependencies and store their symbols
897 as if they appeared in this psymtab. */
898 void recursively_write_psymbols
899 (struct objfile *objfile,
edfe0a0c 900 partial_symtab *psymtab,
cd4fb1b2
SM
901 std::unordered_set<partial_symbol *> &psyms_seen,
902 int cu_index)
903 {
904 for (int i = 0; i < psymtab->number_of_dependencies; ++i)
905 if (psymtab->dependencies[i]->user != NULL)
891813be 906 recursively_write_psymbols
edfe0a0c 907 (objfile, psymtab->dependencies[i], psyms_seen, cu_index);
cd4fb1b2 908
932539d7
TT
909 write_psymbols (psyms_seen, psymtab->global_psymbols,
910 cu_index, false, unit_kind::cu);
911 write_psymbols (psyms_seen, psymtab->static_psymbols,
912 cu_index, true, unit_kind::cu);
cd4fb1b2
SM
913 }
914
915 /* Return number of bytes the .debug_names section will have. This
916 must be called only after calling the build method. */
917 size_t bytes () const
918 {
919 /* Verify the build method has been already called. */
920 gdb_assert (!m_abbrev_table.empty ());
921 size_t expected_bytes = 0;
922 expected_bytes += m_bucket_table.size () * sizeof (m_bucket_table[0]);
923 expected_bytes += m_hash_table.size () * sizeof (m_hash_table[0]);
924 expected_bytes += m_name_table_string_offs.bytes ();
925 expected_bytes += m_name_table_entry_offs.bytes ();
926 expected_bytes += m_abbrev_table.size ();
927 expected_bytes += m_entry_pool.size ();
928 return expected_bytes;
929 }
930
931 /* Write .debug_names to FILE_NAMES and .debug_str addition to
932 FILE_STR. This must be called only after calling the build
933 method. */
934 void file_write (FILE *file_names, FILE *file_str) const
935 {
936 /* Verify the build method has been already called. */
937 gdb_assert (!m_abbrev_table.empty ());
938 ::file_write (file_names, m_bucket_table);
939 ::file_write (file_names, m_hash_table);
940 m_name_table_string_offs.file_write (file_names);
941 m_name_table_entry_offs.file_write (file_names);
942 m_abbrev_table.file_write (file_names);
943 m_entry_pool.file_write (file_names);
944 m_debugstrlookup.file_write (file_str);
945 }
946
947 /* A helper user data for write_one_signatured_type. */
948 class write_one_signatured_type_data
949 {
950 public:
951 write_one_signatured_type_data (debug_names &nametable_,
dda83cd7 952 signatured_type_index_data &&info_)
cd4fb1b2
SM
953 : nametable (nametable_), info (std::move (info_))
954 {}
955 debug_names &nametable;
956 struct signatured_type_index_data info;
957 };
958
959 /* A helper function to pass write_one_signatured_type to
960 htab_traverse_noresize. */
961 static int
962 write_one_signatured_type (void **slot, void *d)
963 {
964 write_one_signatured_type_data *data = (write_one_signatured_type_data *) d;
965 struct signatured_type_index_data *info = &data->info;
966 struct signatured_type *entry = (struct signatured_type *) *slot;
967
968 data->nametable.write_one_signatured_type (entry, info);
969
970 return 1;
971 }
972
973private:
974
975 /* Storage for symbol names mapping them to their .debug_str section
976 offsets. */
977 class debug_str_lookup
978 {
979 public:
980
30baf67b 981 /* Object constructor to be called for current DWARF2_PER_OBJFILE.
cd4fb1b2 982 All .debug_str section strings are automatically stored. */
976ca316
SM
983 debug_str_lookup (dwarf2_per_objfile *per_objfile)
984 : m_abfd (per_objfile->objfile->obfd),
985 m_per_objfile (per_objfile)
cd4fb1b2 986 {
976ca316
SM
987 per_objfile->per_bfd->str.read (per_objfile->objfile);
988 if (per_objfile->per_bfd->str.buffer == NULL)
cd4fb1b2 989 return;
976ca316
SM
990 for (const gdb_byte *data = per_objfile->per_bfd->str.buffer;
991 data < (per_objfile->per_bfd->str.buffer
992 + per_objfile->per_bfd->str.size);)
cd4fb1b2
SM
993 {
994 const char *const s = reinterpret_cast<const char *> (data);
995 const auto insertpair
996 = m_str_table.emplace (c_str_view (s),
976ca316 997 data - per_objfile->per_bfd->str.buffer);
cd4fb1b2 998 if (!insertpair.second)
b98664d3 999 complaint (_("Duplicate string \"%s\" in "
cd4fb1b2
SM
1000 ".debug_str section [in module %s]"),
1001 s, bfd_get_filename (m_abfd));
1002 data += strlen (s) + 1;
1003 }
1004 }
1005
1006 /* Return offset of symbol name S in the .debug_str section. Add
1007 such symbol to the section's end if it does not exist there
1008 yet. */
1009 size_t lookup (const char *s)
1010 {
1011 const auto it = m_str_table.find (c_str_view (s));
1012 if (it != m_str_table.end ())
1013 return it->second;
976ca316 1014 const size_t offset = (m_per_objfile->per_bfd->str.size
cd4fb1b2
SM
1015 + m_str_add_buf.size ());
1016 m_str_table.emplace (c_str_view (s), offset);
1017 m_str_add_buf.append_cstr0 (s);
1018 return offset;
1019 }
1020
1021 /* Append the end of the .debug_str section to FILE. */
1022 void file_write (FILE *file) const
1023 {
1024 m_str_add_buf.file_write (file);
1025 }
1026
1027 private:
1028 std::unordered_map<c_str_view, size_t, c_str_view_hasher> m_str_table;
1029 bfd *const m_abfd;
976ca316 1030 dwarf2_per_objfile *m_per_objfile;
cd4fb1b2
SM
1031
1032 /* Data to add at the end of .debug_str for new needed symbol names. */
1033 data_buf m_str_add_buf;
1034 };
1035
1036 /* Container to map used DWARF tags to their .debug_names abbreviation
1037 tags. */
1038 class index_key
1039 {
1040 public:
1041 index_key (int dwarf_tag_, bool is_static_, unit_kind kind_)
1042 : dwarf_tag (dwarf_tag_), is_static (is_static_), kind (kind_)
1043 {
1044 }
1045
1046 bool
1047 operator== (const index_key &other) const
1048 {
1049 return (dwarf_tag == other.dwarf_tag && is_static == other.is_static
1050 && kind == other.kind);
1051 }
1052
1053 const int dwarf_tag;
1054 const bool is_static;
1055 const unit_kind kind;
1056 };
1057
1058 /* Provide std::unordered_map::hasher for index_key. */
1059 class index_key_hasher
1060 {
1061 public:
1062 size_t
1063 operator () (const index_key &key) const
1064 {
1065 return (std::hash<int>() (key.dwarf_tag) << 1) | key.is_static;
1066 }
1067 };
1068
1069 /* Parameters of one symbol entry. */
1070 class symbol_value
1071 {
1072 public:
1073 const int dwarf_tag, cu_index;
1074 const bool is_static;
1075 const unit_kind kind;
1076
1077 symbol_value (int dwarf_tag_, int cu_index_, bool is_static_,
1078 unit_kind kind_)
1079 : dwarf_tag (dwarf_tag_), cu_index (cu_index_), is_static (is_static_),
dda83cd7 1080 kind (kind_)
cd4fb1b2
SM
1081 {}
1082
1083 bool
1084 operator< (const symbol_value &other) const
1085 {
1086#define X(n) \
1087 do \
1088 { \
1089 if (n < other.n) \
1090 return true; \
1091 if (n > other.n) \
1092 return false; \
1093 } \
1094 while (0)
1095 X (dwarf_tag);
1096 X (is_static);
1097 X (kind);
1098 X (cu_index);
1099#undef X
1100 return false;
1101 }
1102 };
1103
1104 /* Abstract base class to unify DWARF-32 and DWARF-64 name table
1105 output. */
1106 class offset_vec
1107 {
1108 protected:
1109 const bfd_endian dwarf5_byte_order;
1110 public:
1111 explicit offset_vec (bfd_endian dwarf5_byte_order_)
1112 : dwarf5_byte_order (dwarf5_byte_order_)
1113 {}
1114
1115 /* Call std::vector::reserve for NELEM elements. */
1116 virtual void reserve (size_t nelem) = 0;
1117
1118 /* Call std::vector::push_back with store_unsigned_integer byte
1119 reordering for ELEM. */
1120 virtual void push_back_reorder (size_t elem) = 0;
1121
1122 /* Return expected output size in bytes. */
1123 virtual size_t bytes () const = 0;
1124
1125 /* Write name table to FILE. */
1126 virtual void file_write (FILE *file) const = 0;
1127 };
1128
1129 /* Template to unify DWARF-32 and DWARF-64 output. */
1130 template<typename OffsetSize>
1131 class offset_vec_tmpl : public offset_vec
1132 {
1133 public:
1134 explicit offset_vec_tmpl (bfd_endian dwarf5_byte_order_)
1135 : offset_vec (dwarf5_byte_order_)
1136 {}
1137
1138 /* Implement offset_vec::reserve. */
1139 void reserve (size_t nelem) override
1140 {
1141 m_vec.reserve (nelem);
1142 }
1143
1144 /* Implement offset_vec::push_back_reorder. */
1145 void push_back_reorder (size_t elem) override
1146 {
1147 m_vec.push_back (elem);
1148 /* Check for overflow. */
1149 gdb_assert (m_vec.back () == elem);
1150 store_unsigned_integer (reinterpret_cast<gdb_byte *> (&m_vec.back ()),
1151 sizeof (m_vec.back ()), dwarf5_byte_order, elem);
1152 }
1153
1154 /* Implement offset_vec::bytes. */
1155 size_t bytes () const override
1156 {
1157 return m_vec.size () * sizeof (m_vec[0]);
1158 }
1159
1160 /* Implement offset_vec::file_write. */
1161 void file_write (FILE *file) const override
1162 {
1163 ::file_write (file, m_vec);
1164 }
1165
1166 private:
1167 std::vector<OffsetSize> m_vec;
1168 };
1169
1170 /* Base class to unify DWARF-32 and DWARF-64 .debug_names output
1171 respecting name table width. */
1172 class dwarf
1173 {
1174 public:
1175 offset_vec &name_table_string_offs, &name_table_entry_offs;
1176
1177 dwarf (offset_vec &name_table_string_offs_,
1178 offset_vec &name_table_entry_offs_)
1179 : name_table_string_offs (name_table_string_offs_),
1180 name_table_entry_offs (name_table_entry_offs_)
1181 {
1182 }
1183 };
1184
1185 /* Template to unify DWARF-32 and DWARF-64 .debug_names output
1186 respecting name table width. */
1187 template<typename OffsetSize>
1188 class dwarf_tmpl : public dwarf
1189 {
1190 public:
1191 explicit dwarf_tmpl (bfd_endian dwarf5_byte_order_)
1192 : dwarf (m_name_table_string_offs, m_name_table_entry_offs),
1193 m_name_table_string_offs (dwarf5_byte_order_),
1194 m_name_table_entry_offs (dwarf5_byte_order_)
1195 {}
1196
1197 private:
1198 offset_vec_tmpl<OffsetSize> m_name_table_string_offs;
1199 offset_vec_tmpl<OffsetSize> m_name_table_entry_offs;
1200 };
1201
1202 /* Try to reconstruct original DWARF tag for given partial_symbol.
1203 This function is not DWARF-5 compliant but it is sufficient for
1204 GDB as a DWARF-5 index consumer. */
1205 static int psymbol_tag (const struct partial_symbol *psym)
1206 {
8a6d4234
TT
1207 domain_enum domain = psym->domain;
1208 enum address_class aclass = psym->aclass;
cd4fb1b2
SM
1209
1210 switch (domain)
1211 {
1212 case VAR_DOMAIN:
1213 switch (aclass)
1214 {
1215 case LOC_BLOCK:
1216 return DW_TAG_subprogram;
1217 case LOC_TYPEDEF:
1218 return DW_TAG_typedef;
1219 case LOC_COMPUTED:
1220 case LOC_CONST_BYTES:
1221 case LOC_OPTIMIZED_OUT:
1222 case LOC_STATIC:
1223 return DW_TAG_variable;
1224 case LOC_CONST:
1225 /* Note: It's currently impossible to recognize psyms as enum values
1226 short of reading the type info. For now punt. */
1227 return DW_TAG_variable;
1228 default:
1229 /* There are other LOC_FOO values that one might want to classify
1230 as variables, but dwarf2read.c doesn't currently use them. */
1231 return DW_TAG_variable;
1232 }
1233 case STRUCT_DOMAIN:
1234 return DW_TAG_structure_type;
7666722f
TV
1235 case MODULE_DOMAIN:
1236 return DW_TAG_module;
cd4fb1b2
SM
1237 default:
1238 return 0;
1239 }
1240 }
1241
1242 /* Call insert for all partial symbols and mark them in PSYMS_SEEN. */
1243 void write_psymbols (std::unordered_set<partial_symbol *> &psyms_seen,
932539d7
TT
1244 const std::vector<partial_symbol *> &symbols,
1245 int cu_index, bool is_static, unit_kind kind)
cd4fb1b2 1246 {
932539d7 1247 for (partial_symbol *psym : symbols)
cd4fb1b2 1248 {
cd4fb1b2
SM
1249 /* Only add a given psymbol once. */
1250 if (psyms_seen.insert (psym).second)
1251 insert (psym, cu_index, is_static, kind);
1252 }
1253 }
1254
1255 /* A helper function that writes a single signatured_type
1256 to a debug_names. */
1257 void
1258 write_one_signatured_type (struct signatured_type *entry,
1259 struct signatured_type_index_data *info)
1260 {
edfe0a0c 1261 partial_symtab *psymtab = entry->per_cu.v.psymtab;
cd4fb1b2 1262
932539d7
TT
1263 write_psymbols (info->psyms_seen, psymtab->global_psymbols,
1264 info->cu_index, false, unit_kind::tu);
1265 write_psymbols (info->psyms_seen, psymtab->static_psymbols,
1266 info->cu_index, true, unit_kind::tu);
cd4fb1b2
SM
1267
1268 info->types_list.append_uint (dwarf5_offset_size (), m_dwarf5_byte_order,
1269 to_underlying (entry->per_cu.sect_off));
1270
1271 ++info->cu_index;
1272 }
1273
1274 /* Store value of each symbol. */
1275 std::unordered_map<c_str_view, std::set<symbol_value>, c_str_view_hasher>
1276 m_name_to_value_set;
1277
1278 /* Tables of DWARF-5 .debug_names. They are in object file byte
1279 order. */
1280 std::vector<uint32_t> m_bucket_table;
1281 std::vector<uint32_t> m_hash_table;
1282
1283 const bfd_endian m_dwarf5_byte_order;
1284 dwarf_tmpl<uint32_t> m_dwarf32;
1285 dwarf_tmpl<uint64_t> m_dwarf64;
1286 dwarf &m_dwarf;
1287 offset_vec &m_name_table_string_offs, &m_name_table_entry_offs;
1288 debug_str_lookup m_debugstrlookup;
1289
1290 /* Map each used .debug_names abbreviation tag parameter to its
1291 index value. */
1292 std::unordered_map<index_key, int, index_key_hasher> m_indexkey_to_idx;
1293
1294 /* Next unused .debug_names abbreviation tag for
1295 m_indexkey_to_idx. */
1296 int m_idx_next = 1;
1297
1298 /* .debug_names abbreviation table. */
1299 data_buf m_abbrev_table;
1300
1301 /* .debug_names entry pool. */
1302 data_buf m_entry_pool;
3b00ef10
TT
1303
1304 /* Temporary storage for Ada names. */
1305 auto_obstack m_string_obstack;
cd4fb1b2
SM
1306};
1307
1308/* Return iff any of the needed offsets does not fit into 32-bit
1309 .debug_names section. */
1310
1311static bool
976ca316 1312check_dwarf64_offsets (dwarf2_per_objfile *per_objfile)
cd4fb1b2 1313{
976ca316 1314 for (dwarf2_per_cu_data *per_cu : per_objfile->per_bfd->all_comp_units)
cd4fb1b2 1315 {
b76e467d 1316 if (to_underlying (per_cu->sect_off) >= (static_cast<uint64_t> (1) << 32))
cd4fb1b2
SM
1317 return true;
1318 }
976ca316 1319 for (const signatured_type *sigtype : per_objfile->per_bfd->all_type_units)
cd4fb1b2 1320 {
b2bdb8cf 1321 const dwarf2_per_cu_data &per_cu = sigtype->per_cu;
cd4fb1b2
SM
1322
1323 if (to_underlying (per_cu.sect_off) >= (static_cast<uint64_t> (1) << 32))
1324 return true;
1325 }
1326 return false;
1327}
1328
1329/* The psyms_seen set is potentially going to be largish (~40k
1330 elements when indexing a -g3 build of GDB itself). Estimate the
1331 number of elements in order to avoid too many rehashes, which
1332 require rebuilding buckets and thus many trips to
1333 malloc/free. */
1334
1335static size_t
976ca316 1336psyms_seen_size (dwarf2_per_objfile *per_objfile)
cd4fb1b2
SM
1337{
1338 size_t psyms_count = 0;
976ca316 1339 for (dwarf2_per_cu_data *per_cu : per_objfile->per_bfd->all_comp_units)
cd4fb1b2 1340 {
edfe0a0c 1341 partial_symtab *psymtab = per_cu->v.psymtab;
cd4fb1b2
SM
1342
1343 if (psymtab != NULL && psymtab->user == NULL)
1344 recursively_count_psymbols (psymtab, psyms_count);
1345 }
1346 /* Generating an index for gdb itself shows a ratio of
1347 TOTAL_SEEN_SYMS/UNIQUE_SYMS or ~5. 4 seems like a good bet. */
1348 return psyms_count / 4;
1349}
1350
c4973306
SM
1351/* Assert that FILE's size is EXPECTED_SIZE. Assumes file's seek
1352 position is at the end of the file. */
cd4fb1b2 1353
c4973306
SM
1354static void
1355assert_file_size (FILE *file, size_t expected_size)
1356{
1357 const auto file_size = ftell (file);
1358 if (file_size == -1)
1359 perror_with_name (("ftell"));
1360 gdb_assert (file_size == expected_size);
1361}
1362
1363/* Write a gdb index file to OUT_FILE from all the sections passed as
1364 arguments. */
1365
1366static void
1367write_gdbindex_1 (FILE *out_file,
1368 const data_buf &cu_list,
1369 const data_buf &types_cu_list,
1370 const data_buf &addr_vec,
1371 const data_buf &symtab_vec,
1372 const data_buf &constant_pool)
1373{
1374 data_buf contents;
1375 const offset_type size_of_header = 6 * sizeof (offset_type);
1376 offset_type total_len = size_of_header;
1377
1378 /* The version number. */
1379 contents.append_data (MAYBE_SWAP (8));
1380
1381 /* The offset of the CU list from the start of the file. */
1382 contents.append_data (MAYBE_SWAP (total_len));
1383 total_len += cu_list.size ();
1384
1385 /* The offset of the types CU list from the start of the file. */
1386 contents.append_data (MAYBE_SWAP (total_len));
1387 total_len += types_cu_list.size ();
1388
1389 /* The offset of the address table from the start of the file. */
1390 contents.append_data (MAYBE_SWAP (total_len));
1391 total_len += addr_vec.size ();
1392
1393 /* The offset of the symbol table from the start of the file. */
1394 contents.append_data (MAYBE_SWAP (total_len));
1395 total_len += symtab_vec.size ();
1396
1397 /* The offset of the constant pool from the start of the file. */
1398 contents.append_data (MAYBE_SWAP (total_len));
1399 total_len += constant_pool.size ();
1400
1401 gdb_assert (contents.size () == size_of_header);
1402
1403 contents.file_write (out_file);
1404 cu_list.file_write (out_file);
1405 types_cu_list.file_write (out_file);
1406 addr_vec.file_write (out_file);
1407 symtab_vec.file_write (out_file);
1408 constant_pool.file_write (out_file);
1409
1410 assert_file_size (out_file, total_len);
1411}
1412
1413/* Write contents of a .gdb_index section for OBJFILE into OUT_FILE.
1414 If OBJFILE has an associated dwz file, write contents of a .gdb_index
1415 section for that dwz file into DWZ_OUT_FILE. If OBJFILE does not have an
1416 associated dwz file, DWZ_OUT_FILE must be NULL. */
1417
1418static void
976ca316 1419write_gdbindex (dwarf2_per_objfile *per_objfile, FILE *out_file,
c4973306 1420 FILE *dwz_out_file)
cd4fb1b2 1421{
976ca316 1422 struct objfile *objfile = per_objfile->objfile;
cd4fb1b2 1423 mapped_symtab symtab;
c4973306
SM
1424 data_buf objfile_cu_list;
1425 data_buf dwz_cu_list;
cd4fb1b2
SM
1426
1427 /* While we're scanning CU's create a table that maps a psymtab pointer
1428 (which is what addrmap records) to its index (which is what is recorded
1429 in the index file). This will later be needed to write the address
1430 table. */
1431 psym_index_map cu_index_htab;
976ca316 1432 cu_index_htab.reserve (per_objfile->per_bfd->all_comp_units.size ());
cd4fb1b2
SM
1433
1434 /* The CU list is already sorted, so we don't need to do additional
1435 work here. Also, the debug_types entries do not appear in
1436 all_comp_units, but only in their own hash table. */
1437
1438 std::unordered_set<partial_symbol *> psyms_seen
976ca316
SM
1439 (psyms_seen_size (per_objfile));
1440 for (int i = 0; i < per_objfile->per_bfd->all_comp_units.size (); ++i)
cd4fb1b2 1441 {
976ca316 1442 dwarf2_per_cu_data *per_cu = per_objfile->per_bfd->all_comp_units[i];
edfe0a0c 1443 partial_symtab *psymtab = per_cu->v.psymtab;
cd4fb1b2 1444
efba5c23
TV
1445 if (psymtab != NULL)
1446 {
1447 if (psymtab->user == NULL)
1448 recursively_write_psymbols (objfile, psymtab, &symtab,
1449 psyms_seen, i);
1450
1451 const auto insertpair = cu_index_htab.emplace (psymtab, i);
1452 gdb_assert (insertpair.second);
1453 }
cd4fb1b2 1454
c4973306
SM
1455 /* The all_comp_units list contains CUs read from the objfile as well as
1456 from the eventual dwz file. We need to place the entry in the
1457 corresponding index. */
1458 data_buf &cu_list = per_cu->is_dwz ? dwz_cu_list : objfile_cu_list;
cd4fb1b2
SM
1459 cu_list.append_uint (8, BFD_ENDIAN_LITTLE,
1460 to_underlying (per_cu->sect_off));
1461 cu_list.append_uint (8, BFD_ENDIAN_LITTLE, per_cu->length);
1462 }
1463
1464 /* Dump the address map. */
1465 data_buf addr_vec;
1466 write_address_map (objfile, addr_vec, cu_index_htab);
1467
1468 /* Write out the .debug_type entries, if any. */
1469 data_buf types_cu_list;
976ca316 1470 if (per_objfile->per_bfd->signatured_types)
cd4fb1b2
SM
1471 {
1472 signatured_type_index_data sig_data (types_cu_list,
1473 psyms_seen);
1474
1475 sig_data.objfile = objfile;
1476 sig_data.symtab = &symtab;
976ca316
SM
1477 sig_data.cu_index = per_objfile->per_bfd->all_comp_units.size ();
1478 htab_traverse_noresize (per_objfile->per_bfd->signatured_types.get (),
cd4fb1b2
SM
1479 write_one_signatured_type, &sig_data);
1480 }
1481
1482 /* Now that we've processed all symbols we can shrink their cu_indices
1483 lists. */
1484 uniquify_cu_indices (&symtab);
1485
1486 data_buf symtab_vec, constant_pool;
1487 write_hash_table (&symtab, symtab_vec, constant_pool);
1488
c4973306
SM
1489 write_gdbindex_1(out_file, objfile_cu_list, types_cu_list, addr_vec,
1490 symtab_vec, constant_pool);
cd4fb1b2 1491
c4973306
SM
1492 if (dwz_out_file != NULL)
1493 write_gdbindex_1 (dwz_out_file, dwz_cu_list, {}, {}, {}, {});
1494 else
1495 gdb_assert (dwz_cu_list.empty ());
cd4fb1b2
SM
1496}
1497
1498/* DWARF-5 augmentation string for GDB's DW_IDX_GNU_* extension. */
1499static const gdb_byte dwarf5_gdb_augmentation[] = { 'G', 'D', 'B', 0 };
1500
1501/* Write a new .debug_names section for OBJFILE into OUT_FILE, write
1502 needed addition to .debug_str section to OUT_FILE_STR. Return how
1503 many bytes were expected to be written into OUT_FILE. */
1504
c4973306 1505static void
976ca316 1506write_debug_names (dwarf2_per_objfile *per_objfile,
cd4fb1b2
SM
1507 FILE *out_file, FILE *out_file_str)
1508{
976ca316
SM
1509 const bool dwarf5_is_dwarf64 = check_dwarf64_offsets (per_objfile);
1510 struct objfile *objfile = per_objfile->objfile;
cd4fb1b2 1511 const enum bfd_endian dwarf5_byte_order
08feed99 1512 = gdbarch_byte_order (objfile->arch ());
cd4fb1b2
SM
1513
1514 /* The CU list is already sorted, so we don't need to do additional
1515 work here. Also, the debug_types entries do not appear in
1516 all_comp_units, but only in their own hash table. */
1517 data_buf cu_list;
976ca316 1518 debug_names nametable (per_objfile, dwarf5_is_dwarf64, dwarf5_byte_order);
cd4fb1b2 1519 std::unordered_set<partial_symbol *>
976ca316
SM
1520 psyms_seen (psyms_seen_size (per_objfile));
1521 for (int i = 0; i < per_objfile->per_bfd->all_comp_units.size (); ++i)
cd4fb1b2 1522 {
976ca316 1523 const dwarf2_per_cu_data *per_cu = per_objfile->per_bfd->all_comp_units[i];
edfe0a0c 1524 partial_symtab *psymtab = per_cu->v.psymtab;
cd4fb1b2
SM
1525
1526 /* CU of a shared file from 'dwz -m' may be unused by this main
1527 file. It may be referenced from a local scope but in such
1528 case it does not need to be present in .debug_names. */
1529 if (psymtab == NULL)
1530 continue;
1531
1532 if (psymtab->user == NULL)
1533 nametable.recursively_write_psymbols (objfile, psymtab, psyms_seen, i);
1534
1535 cu_list.append_uint (nametable.dwarf5_offset_size (), dwarf5_byte_order,
1536 to_underlying (per_cu->sect_off));
1537 }
1538
1539 /* Write out the .debug_type entries, if any. */
1540 data_buf types_cu_list;
976ca316 1541 if (per_objfile->per_bfd->signatured_types)
cd4fb1b2
SM
1542 {
1543 debug_names::write_one_signatured_type_data sig_data (nametable,
1544 signatured_type_index_data (types_cu_list, psyms_seen));
1545
1546 sig_data.info.objfile = objfile;
1547 /* It is used only for gdb_index. */
1548 sig_data.info.symtab = nullptr;
1549 sig_data.info.cu_index = 0;
976ca316 1550 htab_traverse_noresize (per_objfile->per_bfd->signatured_types.get (),
cd4fb1b2
SM
1551 debug_names::write_one_signatured_type,
1552 &sig_data);
1553 }
1554
1555 nametable.build ();
1556
1557 /* No addr_vec - DWARF-5 uses .debug_aranges generated by GCC. */
1558
1559 const offset_type bytes_of_header
1560 = ((dwarf5_is_dwarf64 ? 12 : 4)
1561 + 2 + 2 + 7 * 4
1562 + sizeof (dwarf5_gdb_augmentation));
1563 size_t expected_bytes = 0;
1564 expected_bytes += bytes_of_header;
1565 expected_bytes += cu_list.size ();
1566 expected_bytes += types_cu_list.size ();
1567 expected_bytes += nametable.bytes ();
1568 data_buf header;
1569
1570 if (!dwarf5_is_dwarf64)
1571 {
1572 const uint64_t size64 = expected_bytes - 4;
1573 gdb_assert (size64 < 0xfffffff0);
1574 header.append_uint (4, dwarf5_byte_order, size64);
1575 }
1576 else
1577 {
1578 header.append_uint (4, dwarf5_byte_order, 0xffffffff);
1579 header.append_uint (8, dwarf5_byte_order, expected_bytes - 12);
1580 }
1581
1582 /* The version number. */
1583 header.append_uint (2, dwarf5_byte_order, 5);
1584
1585 /* Padding. */
1586 header.append_uint (2, dwarf5_byte_order, 0);
1587
1588 /* comp_unit_count - The number of CUs in the CU list. */
b76e467d 1589 header.append_uint (4, dwarf5_byte_order,
976ca316 1590 per_objfile->per_bfd->all_comp_units.size ());
cd4fb1b2
SM
1591
1592 /* local_type_unit_count - The number of TUs in the local TU
1593 list. */
b2bdb8cf 1594 header.append_uint (4, dwarf5_byte_order,
976ca316 1595 per_objfile->per_bfd->all_type_units.size ());
cd4fb1b2
SM
1596
1597 /* foreign_type_unit_count - The number of TUs in the foreign TU
1598 list. */
1599 header.append_uint (4, dwarf5_byte_order, 0);
1600
1601 /* bucket_count - The number of hash buckets in the hash lookup
1602 table. */
1603 header.append_uint (4, dwarf5_byte_order, nametable.bucket_count ());
1604
1605 /* name_count - The number of unique names in the index. */
1606 header.append_uint (4, dwarf5_byte_order, nametable.name_count ());
1607
1608 /* abbrev_table_size - The size in bytes of the abbreviations
1609 table. */
1610 header.append_uint (4, dwarf5_byte_order, nametable.abbrev_table_bytes ());
1611
1612 /* augmentation_string_size - The size in bytes of the augmentation
1613 string. This value is rounded up to a multiple of 4. */
1614 static_assert (sizeof (dwarf5_gdb_augmentation) % 4 == 0, "");
1615 header.append_uint (4, dwarf5_byte_order, sizeof (dwarf5_gdb_augmentation));
1616 header.append_data (dwarf5_gdb_augmentation);
1617
1618 gdb_assert (header.size () == bytes_of_header);
1619
1620 header.file_write (out_file);
1621 cu_list.file_write (out_file);
1622 types_cu_list.file_write (out_file);
1623 nametable.file_write (out_file, out_file_str);
1624
c4973306 1625 assert_file_size (out_file, expected_bytes);
cd4fb1b2
SM
1626}
1627
c4973306 1628/* This represents an index file being written (work-in-progress).
cd4fb1b2 1629
c4973306
SM
1630 The data is initially written to a temporary file. When the finalize method
1631 is called, the file is closed and moved to its final location.
1632
1633 On failure (if this object is being destroyed with having called finalize),
1634 the temporary file is closed and deleted. */
1635
1636struct index_wip_file
cd4fb1b2 1637{
c4973306
SM
1638 index_wip_file (const char *dir, const char *basename,
1639 const char *suffix)
1640 {
1641 filename = (std::string (dir) + SLASH_STRING + basename
1642 + suffix);
1643
1644 filename_temp = make_temp_filename (filename);
1645
1646 scoped_fd out_file_fd (gdb_mkostemp_cloexec (filename_temp.data (),
1647 O_BINARY));
1648 if (out_file_fd.get () == -1)
1649 perror_with_name (("mkstemp"));
1650
1651 out_file = out_file_fd.to_file ("wb");
1652
1653 if (out_file == nullptr)
1654 error (_("Can't open `%s' for writing"), filename_temp.data ());
1655
1656 unlink_file.emplace (filename_temp.data ());
1657 }
1658
1659 void finalize ()
1660 {
1661 /* We want to keep the file. */
1662 unlink_file->keep ();
1663
1664 /* Close and move the str file in place. */
1665 unlink_file.reset ();
1666 if (rename (filename_temp.data (), filename.c_str ()) != 0)
1667 perror_with_name (("rename"));
1668 }
1669
1670 std::string filename;
1671 gdb::char_vector filename_temp;
1672
1673 /* Order matters here; we want FILE to be closed before
1674 FILENAME_TEMP is unlinked, because on MS-Windows one cannot
1675 delete a file that is still open. So, we wrap the unlinker in an
1676 optional and emplace it once we know the file name. */
1677 gdb::optional<gdb::unlinker> unlink_file;
1678
1679 gdb_file_up out_file;
1680};
cd4fb1b2 1681
87d6a7aa 1682/* See dwarf-index-write.h. */
cd4fb1b2 1683
87d6a7aa 1684void
976ca316
SM
1685write_psymtabs_to_index (dwarf2_per_objfile *per_objfile, const char *dir,
1686 const char *basename, const char *dwz_basename,
cd4fb1b2
SM
1687 dw_index_kind index_kind)
1688{
976ca316 1689 struct objfile *objfile = per_objfile->objfile;
cd4fb1b2 1690
976ca316 1691 if (per_objfile->per_bfd->using_index)
cd4fb1b2
SM
1692 error (_("Cannot use an index to create the index"));
1693
976ca316 1694 if (per_objfile->per_bfd->types.size () > 1)
cd4fb1b2
SM
1695 error (_("Cannot make an index when the file has multiple .debug_types sections"));
1696
d320c2b5
TT
1697 if (!objfile->partial_symtabs->psymtabs
1698 || !objfile->partial_symtabs->psymtabs_addrmap)
cd4fb1b2
SM
1699 return;
1700
1701 struct stat st;
1702 if (stat (objfile_name (objfile), &st) < 0)
1703 perror_with_name (objfile_name (objfile));
1704
c4973306
SM
1705 const char *index_suffix = (index_kind == dw_index_kind::DEBUG_NAMES
1706 ? INDEX5_SUFFIX : INDEX4_SUFFIX);
cd4fb1b2 1707
c4973306
SM
1708 index_wip_file objfile_index_wip (dir, basename, index_suffix);
1709 gdb::optional<index_wip_file> dwz_index_wip;
cd4fb1b2 1710
c4973306
SM
1711 if (dwz_basename != NULL)
1712 dwz_index_wip.emplace (dir, dwz_basename, index_suffix);
cd4fb1b2
SM
1713
1714 if (index_kind == dw_index_kind::DEBUG_NAMES)
1715 {
c4973306
SM
1716 index_wip_file str_wip_file (dir, basename, DEBUG_STR_SUFFIX);
1717
976ca316 1718 write_debug_names (per_objfile, objfile_index_wip.out_file.get (),
c4973306
SM
1719 str_wip_file.out_file.get ());
1720
1721 str_wip_file.finalize ();
cd4fb1b2
SM
1722 }
1723 else
976ca316 1724 write_gdbindex (per_objfile, objfile_index_wip.out_file.get (),
c4973306
SM
1725 (dwz_index_wip.has_value ()
1726 ? dwz_index_wip->out_file.get () : NULL));
cd4fb1b2 1727
c4973306 1728 objfile_index_wip.finalize ();
87d6a7aa 1729
c4973306
SM
1730 if (dwz_index_wip.has_value ())
1731 dwz_index_wip->finalize ();
cd4fb1b2
SM
1732}
1733
1734/* Implementation of the `save gdb-index' command.
1735
1736 Note that the .gdb_index file format used by this command is
1737 documented in the GDB manual. Any changes here must be documented
1738 there. */
1739
1740static void
1741save_gdb_index_command (const char *arg, int from_tty)
1742{
cd4fb1b2
SM
1743 const char dwarf5space[] = "-dwarf-5 ";
1744 dw_index_kind index_kind = dw_index_kind::GDB_INDEX;
1745
1746 if (!arg)
1747 arg = "";
1748
1749 arg = skip_spaces (arg);
1750 if (strncmp (arg, dwarf5space, strlen (dwarf5space)) == 0)
1751 {
1752 index_kind = dw_index_kind::DEBUG_NAMES;
1753 arg += strlen (dwarf5space);
1754 arg = skip_spaces (arg);
1755 }
1756
1757 if (!*arg)
1758 error (_("usage: save gdb-index [-dwarf-5] DIRECTORY"));
1759
2030c079 1760 for (objfile *objfile : current_program_space->objfiles ())
aed57c53
TT
1761 {
1762 struct stat st;
cd4fb1b2 1763
aed57c53
TT
1764 /* If the objfile does not correspond to an actual file, skip it. */
1765 if (stat (objfile_name (objfile), &st) < 0)
1766 continue;
cd4fb1b2 1767
976ca316 1768 dwarf2_per_objfile *per_objfile = get_dwarf2_per_objfile (objfile);
cd4fb1b2 1769
976ca316 1770 if (per_objfile != NULL)
aed57c53 1771 {
a70b8144 1772 try
aed57c53
TT
1773 {
1774 const char *basename = lbasename (objfile_name (objfile));
976ca316 1775 const dwz_file *dwz = dwarf2_get_dwz_file (per_objfile->per_bfd);
c4973306
SM
1776 const char *dwz_basename = NULL;
1777
1778 if (dwz != NULL)
1779 dwz_basename = lbasename (dwz->filename ());
1780
976ca316
SM
1781 write_psymtabs_to_index (per_objfile, arg, basename, dwz_basename,
1782 index_kind);
aed57c53 1783 }
230d2906 1784 catch (const gdb_exception_error &except)
aed57c53
TT
1785 {
1786 exception_fprintf (gdb_stderr, except,
1787 _("Error while writing index for `%s': "),
1788 objfile_name (objfile));
1789 }
aed57c53 1790 }
cd4fb1b2 1791
aed57c53 1792 }
cd4fb1b2
SM
1793}
1794
6c265988 1795void _initialize_dwarf_index_write ();
cd4fb1b2
SM
1796void
1797_initialize_dwarf_index_write ()
1798{
1799 cmd_list_element *c = add_cmd ("gdb-index", class_files,
1800 save_gdb_index_command, _("\
1801Save a gdb-index file.\n\
1802Usage: save gdb-index [-dwarf-5] DIRECTORY\n\
1803\n\
1804No options create one file with .gdb-index extension for pre-DWARF-5\n\
1805compatible .gdb_index section. With -dwarf-5 creates two files with\n\
1806extension .debug_names and .debug_str for DWARF-5 .debug_names section."),
1807 &save_cmdlist);
1808 set_cmd_completer (c, filename_completer);
1809}
This page took 0.471277 seconds and 4 git commands to generate.