Commit | Line | Data |
---|---|---|
5c2c6c95 ILT |
1 | // dwarf_reader.cc -- parse dwarf2/3 debug information |
2 | ||
4b95cf5c | 3 | // Copyright (C) 2007-2014 Free Software Foundation, Inc. |
5c2c6c95 ILT |
4 | // Written by Ian Lance Taylor <iant@google.com>. |
5 | ||
6 | // This file is part of gold. | |
7 | ||
8 | // This program is free software; you can redistribute it and/or modify | |
9 | // it under the terms of the GNU General Public License as published by | |
10 | // the Free Software Foundation; either version 3 of the License, or | |
11 | // (at your option) any later version. | |
12 | ||
13 | // This program is distributed in the hope that it will be useful, | |
14 | // but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | // GNU General Public License for more details. | |
17 | ||
18 | // You should have received a copy of the GNU General Public License | |
19 | // along with this program; if not, write to the Free Software | |
20 | // Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston, | |
21 | // MA 02110-1301, USA. | |
22 | ||
23 | #include "gold.h" | |
24 | ||
04bf7072 | 25 | #include <algorithm> |
ec673e64 | 26 | #include <utility> |
e4e5049b | 27 | #include <vector> |
04bf7072 | 28 | |
5c2c6c95 ILT |
29 | #include "elfcpp_swap.h" |
30 | #include "dwarf.h" | |
24badc65 | 31 | #include "object.h" |
4c50553d | 32 | #include "reloc.h" |
5c2c6c95 | 33 | #include "dwarf_reader.h" |
4f787271 | 34 | #include "int_encoding.h" |
a2e47362 | 35 | #include "compressed_output.h" |
5c2c6c95 | 36 | |
62b01cb5 | 37 | namespace gold { |
5c2c6c95 | 38 | |
c1027032 CC |
39 | // Class Sized_elf_reloc_mapper |
40 | ||
41 | // Initialize the relocation tracker for section RELOC_SHNDX. | |
42 | ||
43 | template<int size, bool big_endian> | |
44 | bool | |
45 | Sized_elf_reloc_mapper<size, big_endian>::do_initialize( | |
46 | unsigned int reloc_shndx, unsigned int reloc_type) | |
47 | { | |
48 | this->reloc_type_ = reloc_type; | |
49 | return this->track_relocs_.initialize(this->object_, reloc_shndx, | |
50 | reloc_type); | |
51 | } | |
52 | ||
53 | // Looks in the symtab to see what section a symbol is in. | |
54 | ||
55 | template<int size, bool big_endian> | |
56 | unsigned int | |
57 | Sized_elf_reloc_mapper<size, big_endian>::symbol_section( | |
58 | unsigned int symndx, Address* value, bool* is_ordinary) | |
59 | { | |
60 | const int symsize = elfcpp::Elf_sizes<size>::sym_size; | |
50ed5eb1 | 61 | gold_assert(static_cast<off_t>((symndx + 1) * symsize) <= this->symtab_size_); |
c1027032 CC |
62 | elfcpp::Sym<size, big_endian> elfsym(this->symtab_ + symndx * symsize); |
63 | *value = elfsym.get_st_value(); | |
64 | return this->object_->adjust_sym_shndx(symndx, elfsym.get_st_shndx(), | |
65 | is_ordinary); | |
66 | } | |
67 | ||
68 | // Return the section index and offset within the section of | |
69 | // the target of the relocation for RELOC_OFFSET. | |
70 | ||
71 | template<int size, bool big_endian> | |
72 | unsigned int | |
73 | Sized_elf_reloc_mapper<size, big_endian>::do_get_reloc_target( | |
74 | off_t reloc_offset, off_t* target_offset) | |
75 | { | |
76 | this->track_relocs_.advance(reloc_offset); | |
77 | if (reloc_offset != this->track_relocs_.next_offset()) | |
78 | return 0; | |
79 | unsigned int symndx = this->track_relocs_.next_symndx(); | |
80 | typename elfcpp::Elf_types<size>::Elf_Addr value; | |
81 | bool is_ordinary; | |
82 | unsigned int target_shndx = this->symbol_section(symndx, &value, | |
83 | &is_ordinary); | |
84 | if (!is_ordinary) | |
85 | return 0; | |
86 | if (this->reloc_type_ == elfcpp::SHT_RELA) | |
87 | value += this->track_relocs_.next_addend(); | |
88 | *target_offset = value; | |
89 | return target_shndx; | |
90 | } | |
91 | ||
92 | static inline Elf_reloc_mapper* | |
9fc236f3 | 93 | make_elf_reloc_mapper(Relobj* object, const unsigned char* symtab, |
c1027032 CC |
94 | off_t symtab_size) |
95 | { | |
9fc236f3 | 96 | if (object->elfsize() == 32) |
c1027032 | 97 | { |
9fc236f3 CC |
98 | if (object->is_big_endian()) |
99 | { | |
c1027032 | 100 | #ifdef HAVE_TARGET_32_BIG |
9fc236f3 CC |
101 | return new Sized_elf_reloc_mapper<32, true>(object, symtab, |
102 | symtab_size); | |
103 | #else | |
104 | gold_unreachable(); | |
c1027032 | 105 | #endif |
9fc236f3 CC |
106 | } |
107 | else | |
108 | { | |
109 | #ifdef HAVE_TARGET_32_LITTLE | |
110 | return new Sized_elf_reloc_mapper<32, false>(object, symtab, | |
111 | symtab_size); | |
112 | #else | |
113 | gold_unreachable(); | |
c1027032 | 114 | #endif |
9fc236f3 CC |
115 | } |
116 | } | |
117 | else if (object->elfsize() == 64) | |
118 | { | |
119 | if (object->is_big_endian()) | |
120 | { | |
c1027032 | 121 | #ifdef HAVE_TARGET_64_BIG |
9fc236f3 CC |
122 | return new Sized_elf_reloc_mapper<64, true>(object, symtab, |
123 | symtab_size); | |
124 | #else | |
125 | gold_unreachable(); | |
c1027032 | 126 | #endif |
9fc236f3 CC |
127 | } |
128 | else | |
129 | { | |
130 | #ifdef HAVE_TARGET_64_LITTLE | |
131 | return new Sized_elf_reloc_mapper<64, false>(object, symtab, | |
132 | symtab_size); | |
133 | #else | |
134 | gold_unreachable(); | |
135 | #endif | |
136 | } | |
c1027032 | 137 | } |
9fc236f3 CC |
138 | else |
139 | gold_unreachable(); | |
c1027032 CC |
140 | } |
141 | ||
142 | // class Dwarf_abbrev_table | |
143 | ||
144 | void | |
145 | Dwarf_abbrev_table::clear_abbrev_codes() | |
146 | { | |
147 | for (unsigned int code = 0; code < this->low_abbrev_code_max_; ++code) | |
148 | { | |
149 | if (this->low_abbrev_codes_[code] != NULL) | |
150 | { | |
151 | delete this->low_abbrev_codes_[code]; | |
152 | this->low_abbrev_codes_[code] = NULL; | |
153 | } | |
154 | } | |
155 | for (Abbrev_code_table::iterator it = this->high_abbrev_codes_.begin(); | |
156 | it != this->high_abbrev_codes_.end(); | |
157 | ++it) | |
158 | { | |
159 | if (it->second != NULL) | |
160 | delete it->second; | |
161 | } | |
162 | this->high_abbrev_codes_.clear(); | |
163 | } | |
164 | ||
165 | // Read the abbrev table from an object file. | |
166 | ||
167 | bool | |
168 | Dwarf_abbrev_table::do_read_abbrevs( | |
169 | Relobj* object, | |
170 | unsigned int abbrev_shndx, | |
171 | off_t abbrev_offset) | |
172 | { | |
173 | this->clear_abbrev_codes(); | |
174 | ||
175 | // If we don't have relocations, abbrev_shndx will be 0, and | |
176 | // we'll have to hunt for the .debug_abbrev section. | |
177 | if (abbrev_shndx == 0 && this->abbrev_shndx_ > 0) | |
178 | abbrev_shndx = this->abbrev_shndx_; | |
179 | else if (abbrev_shndx == 0) | |
180 | { | |
181 | for (unsigned int i = 1; i < object->shnum(); ++i) | |
182 | { | |
183 | std::string name = object->section_name(i); | |
ec673e64 | 184 | if (name == ".debug_abbrev" || name == ".zdebug_abbrev") |
c1027032 CC |
185 | { |
186 | abbrev_shndx = i; | |
187 | // Correct the offset. For incremental update links, we have a | |
188 | // relocated offset that is relative to the output section, but | |
189 | // here we need an offset relative to the input section. | |
190 | abbrev_offset -= object->output_section_offset(i); | |
191 | break; | |
192 | } | |
193 | } | |
194 | if (abbrev_shndx == 0) | |
195 | return false; | |
196 | } | |
197 | ||
198 | // Get the section contents and decompress if necessary. | |
199 | if (abbrev_shndx != this->abbrev_shndx_) | |
200 | { | |
201 | if (this->owns_buffer_ && this->buffer_ != NULL) | |
202 | { | |
203 | delete[] this->buffer_; | |
204 | this->owns_buffer_ = false; | |
205 | } | |
206 | ||
207 | section_size_type buffer_size; | |
208 | this->buffer_ = | |
209 | object->decompressed_section_contents(abbrev_shndx, | |
210 | &buffer_size, | |
211 | &this->owns_buffer_); | |
212 | this->buffer_end_ = this->buffer_ + buffer_size; | |
213 | this->abbrev_shndx_ = abbrev_shndx; | |
214 | } | |
215 | ||
216 | this->buffer_pos_ = this->buffer_ + abbrev_offset; | |
217 | return true; | |
218 | } | |
219 | ||
220 | // Lookup the abbrev code entry for CODE. This function is called | |
221 | // only when the abbrev code is not in the direct lookup table. | |
222 | // It may be in the hash table, it may not have been read yet, | |
223 | // or it may not exist in the abbrev table. | |
224 | ||
225 | const Dwarf_abbrev_table::Abbrev_code* | |
226 | Dwarf_abbrev_table::do_get_abbrev(unsigned int code) | |
227 | { | |
228 | // See if the abbrev code is already in the hash table. | |
229 | Abbrev_code_table::const_iterator it = this->high_abbrev_codes_.find(code); | |
230 | if (it != this->high_abbrev_codes_.end()) | |
231 | return it->second; | |
232 | ||
233 | // Read and store abbrev code definitions until we find the | |
234 | // one we're looking for. | |
235 | for (;;) | |
236 | { | |
237 | // Read the abbrev code. A zero here indicates the end of the | |
238 | // abbrev table. | |
239 | size_t len; | |
240 | if (this->buffer_pos_ >= this->buffer_end_) | |
241 | return NULL; | |
242 | uint64_t nextcode = read_unsigned_LEB_128(this->buffer_pos_, &len); | |
243 | if (nextcode == 0) | |
244 | { | |
245 | this->buffer_pos_ = this->buffer_end_; | |
246 | return NULL; | |
247 | } | |
248 | this->buffer_pos_ += len; | |
249 | ||
250 | // Read the tag. | |
251 | if (this->buffer_pos_ >= this->buffer_end_) | |
252 | return NULL; | |
253 | uint64_t tag = read_unsigned_LEB_128(this->buffer_pos_, &len); | |
254 | this->buffer_pos_ += len; | |
255 | ||
256 | // Read the has_children flag. | |
257 | if (this->buffer_pos_ >= this->buffer_end_) | |
258 | return NULL; | |
259 | bool has_children = *this->buffer_pos_ == elfcpp::DW_CHILDREN_yes; | |
260 | this->buffer_pos_ += 1; | |
261 | ||
262 | // Read the list of (attribute, form) pairs. | |
263 | Abbrev_code* entry = new Abbrev_code(tag, has_children); | |
264 | for (;;) | |
265 | { | |
266 | // Read the attribute. | |
267 | if (this->buffer_pos_ >= this->buffer_end_) | |
268 | return NULL; | |
269 | uint64_t attr = read_unsigned_LEB_128(this->buffer_pos_, &len); | |
270 | this->buffer_pos_ += len; | |
271 | ||
272 | // Read the form. | |
273 | if (this->buffer_pos_ >= this->buffer_end_) | |
274 | return NULL; | |
275 | uint64_t form = read_unsigned_LEB_128(this->buffer_pos_, &len); | |
276 | this->buffer_pos_ += len; | |
277 | ||
278 | // A (0,0) pair terminates the list. | |
279 | if (attr == 0 && form == 0) | |
280 | break; | |
281 | ||
282 | if (attr == elfcpp::DW_AT_sibling) | |
283 | entry->has_sibling_attribute = true; | |
284 | ||
285 | entry->add_attribute(attr, form); | |
286 | } | |
287 | ||
288 | this->store_abbrev(nextcode, entry); | |
289 | if (nextcode == code) | |
290 | return entry; | |
291 | } | |
292 | ||
293 | return NULL; | |
294 | } | |
295 | ||
296 | // class Dwarf_ranges_table | |
297 | ||
298 | // Read the ranges table from an object file. | |
299 | ||
300 | bool | |
301 | Dwarf_ranges_table::read_ranges_table( | |
302 | Relobj* object, | |
303 | const unsigned char* symtab, | |
304 | off_t symtab_size, | |
305 | unsigned int ranges_shndx) | |
306 | { | |
307 | // If we've already read this abbrev table, return immediately. | |
308 | if (this->ranges_shndx_ > 0 | |
309 | && this->ranges_shndx_ == ranges_shndx) | |
310 | return true; | |
311 | ||
312 | // If we don't have relocations, ranges_shndx will be 0, and | |
313 | // we'll have to hunt for the .debug_ranges section. | |
314 | if (ranges_shndx == 0 && this->ranges_shndx_ > 0) | |
315 | ranges_shndx = this->ranges_shndx_; | |
316 | else if (ranges_shndx == 0) | |
317 | { | |
318 | for (unsigned int i = 1; i < object->shnum(); ++i) | |
319 | { | |
320 | std::string name = object->section_name(i); | |
ec673e64 | 321 | if (name == ".debug_ranges" || name == ".zdebug_ranges") |
c1027032 CC |
322 | { |
323 | ranges_shndx = i; | |
324 | this->output_section_offset_ = object->output_section_offset(i); | |
325 | break; | |
326 | } | |
327 | } | |
328 | if (ranges_shndx == 0) | |
329 | return false; | |
330 | } | |
331 | ||
332 | // Get the section contents and decompress if necessary. | |
333 | if (ranges_shndx != this->ranges_shndx_) | |
334 | { | |
335 | if (this->owns_ranges_buffer_ && this->ranges_buffer_ != NULL) | |
336 | { | |
337 | delete[] this->ranges_buffer_; | |
338 | this->owns_ranges_buffer_ = false; | |
339 | } | |
340 | ||
341 | section_size_type buffer_size; | |
342 | this->ranges_buffer_ = | |
343 | object->decompressed_section_contents(ranges_shndx, | |
344 | &buffer_size, | |
345 | &this->owns_ranges_buffer_); | |
346 | this->ranges_buffer_end_ = this->ranges_buffer_ + buffer_size; | |
347 | this->ranges_shndx_ = ranges_shndx; | |
348 | } | |
349 | ||
350 | if (this->ranges_reloc_mapper_ != NULL) | |
351 | { | |
352 | delete this->ranges_reloc_mapper_; | |
353 | this->ranges_reloc_mapper_ = NULL; | |
354 | } | |
355 | ||
356 | // For incremental objects, we have no relocations. | |
357 | if (object->is_incremental()) | |
358 | return true; | |
359 | ||
360 | // Find the relocation section for ".debug_ranges". | |
361 | unsigned int reloc_shndx = 0; | |
362 | unsigned int reloc_type = 0; | |
363 | for (unsigned int i = 0; i < object->shnum(); ++i) | |
364 | { | |
365 | reloc_type = object->section_type(i); | |
366 | if ((reloc_type == elfcpp::SHT_REL | |
367 | || reloc_type == elfcpp::SHT_RELA) | |
368 | && object->section_info(i) == ranges_shndx) | |
369 | { | |
370 | reloc_shndx = i; | |
371 | break; | |
372 | } | |
373 | } | |
374 | ||
375 | this->ranges_reloc_mapper_ = make_elf_reloc_mapper(object, symtab, | |
376 | symtab_size); | |
377 | this->ranges_reloc_mapper_->initialize(reloc_shndx, reloc_type); | |
267257d2 | 378 | this->reloc_type_ = reloc_type; |
c1027032 CC |
379 | |
380 | return true; | |
381 | } | |
382 | ||
383 | // Read a range list from section RANGES_SHNDX at offset RANGES_OFFSET. | |
384 | ||
385 | Dwarf_range_list* | |
386 | Dwarf_ranges_table::read_range_list( | |
387 | Relobj* object, | |
388 | const unsigned char* symtab, | |
389 | off_t symtab_size, | |
390 | unsigned int addr_size, | |
391 | unsigned int ranges_shndx, | |
392 | off_t offset) | |
393 | { | |
394 | Dwarf_range_list* ranges; | |
395 | ||
396 | if (!this->read_ranges_table(object, symtab, symtab_size, ranges_shndx)) | |
397 | return NULL; | |
398 | ||
399 | // Correct the offset. For incremental update links, we have a | |
400 | // relocated offset that is relative to the output section, but | |
401 | // here we need an offset relative to the input section. | |
402 | offset -= this->output_section_offset_; | |
403 | ||
404 | // Read the range list at OFFSET. | |
405 | ranges = new Dwarf_range_list(); | |
406 | off_t base = 0; | |
407 | for (; | |
408 | this->ranges_buffer_ + offset < this->ranges_buffer_end_; | |
409 | offset += 2 * addr_size) | |
410 | { | |
411 | off_t start; | |
412 | off_t end; | |
413 | ||
414 | // Read the raw contents of the section. | |
415 | if (addr_size == 4) | |
416 | { | |
ed5d6712 CC |
417 | start = this->dwinfo_->read_from_pointer<32>(this->ranges_buffer_ |
418 | + offset); | |
419 | end = this->dwinfo_->read_from_pointer<32>(this->ranges_buffer_ | |
420 | + offset + 4); | |
c1027032 CC |
421 | } |
422 | else | |
423 | { | |
ed5d6712 CC |
424 | start = this->dwinfo_->read_from_pointer<64>(this->ranges_buffer_ |
425 | + offset); | |
426 | end = this->dwinfo_->read_from_pointer<64>(this->ranges_buffer_ | |
427 | + offset + 8); | |
c1027032 CC |
428 | } |
429 | ||
430 | // Check for relocations and adjust the values. | |
431 | unsigned int shndx1 = 0; | |
432 | unsigned int shndx2 = 0; | |
433 | if (this->ranges_reloc_mapper_ != NULL) | |
434 | { | |
267257d2 CC |
435 | shndx1 = this->lookup_reloc(offset, &start); |
436 | shndx2 = this->lookup_reloc(offset + addr_size, &end); | |
c1027032 CC |
437 | } |
438 | ||
439 | // End of list is marked by a pair of zeroes. | |
440 | if (shndx1 == 0 && start == 0 && end == 0) | |
441 | break; | |
442 | ||
443 | // A "base address selection entry" is identified by | |
444 | // 0xffffffff for the first value of the pair. The second | |
445 | // value is used as a base for subsequent range list entries. | |
446 | if (shndx1 == 0 && start == -1) | |
447 | base = end; | |
448 | else if (shndx1 == shndx2) | |
449 | { | |
450 | if (shndx1 == 0 || object->is_section_included(shndx1)) | |
451 | ranges->add(shndx1, base + start, base + end); | |
452 | } | |
453 | else | |
454 | gold_warning(_("%s: DWARF info may be corrupt; offsets in a " | |
455 | "range list entry are in different sections"), | |
456 | object->name().c_str()); | |
457 | } | |
458 | ||
459 | return ranges; | |
460 | } | |
461 | ||
267257d2 CC |
462 | // Look for a relocation at offset OFF in the range table, |
463 | // and return the section index and offset of the target. | |
464 | ||
465 | unsigned int | |
466 | Dwarf_ranges_table::lookup_reloc(off_t off, off_t* target_off) | |
467 | { | |
468 | off_t value; | |
469 | unsigned int shndx = | |
470 | this->ranges_reloc_mapper_->get_reloc_target(off, &value); | |
471 | if (shndx == 0) | |
472 | return 0; | |
473 | if (this->reloc_type_ == elfcpp::SHT_REL) | |
474 | *target_off += value; | |
475 | else | |
476 | *target_off = value; | |
477 | return shndx; | |
478 | } | |
479 | ||
c1027032 CC |
480 | // class Dwarf_pubnames_table |
481 | ||
234d4ab8 | 482 | // Read the pubnames section from the object file. |
c1027032 CC |
483 | |
484 | bool | |
234d4ab8 SA |
485 | Dwarf_pubnames_table::read_section(Relobj* object, const unsigned char* symtab, |
486 | off_t symtab_size) | |
c1027032 CC |
487 | { |
488 | section_size_type buffer_size; | |
234d4ab8 | 489 | unsigned int shndx = 0; |
ec673e64 CC |
490 | const char* name = this->is_pubtypes_ ? "pubtypes" : "pubnames"; |
491 | const char* gnu_name = (this->is_pubtypes_ | |
492 | ? "gnu_pubtypes" | |
493 | : "gnu_pubnames"); | |
c1027032 | 494 | |
234d4ab8 | 495 | for (unsigned int i = 1; i < object->shnum(); ++i) |
c1027032 | 496 | { |
ec673e64 CC |
497 | std::string section_name = object->section_name(i); |
498 | const char* section_name_suffix = section_name.c_str(); | |
499 | if (is_prefix_of(".debug_", section_name_suffix)) | |
500 | section_name_suffix += 7; | |
501 | else if (is_prefix_of(".zdebug_", section_name_suffix)) | |
502 | section_name_suffix += 8; | |
503 | else | |
504 | continue; | |
505 | if (strcmp(section_name_suffix, name) == 0) | |
506 | { | |
507 | shndx = i; | |
508 | this->output_section_offset_ = object->output_section_offset(i); | |
509 | break; | |
510 | } | |
511 | else if (strcmp(section_name_suffix, gnu_name) == 0) | |
234d4ab8 SA |
512 | { |
513 | shndx = i; | |
514 | this->output_section_offset_ = object->output_section_offset(i); | |
ec673e64 | 515 | this->is_gnu_style_ = true; |
234d4ab8 SA |
516 | break; |
517 | } | |
c1027032 | 518 | } |
234d4ab8 SA |
519 | if (shndx == 0) |
520 | return false; | |
521 | ||
c1027032 CC |
522 | this->buffer_ = object->decompressed_section_contents(shndx, |
523 | &buffer_size, | |
524 | &this->owns_buffer_); | |
525 | if (this->buffer_ == NULL) | |
526 | return false; | |
527 | this->buffer_end_ = this->buffer_ + buffer_size; | |
234d4ab8 SA |
528 | |
529 | // For incremental objects, we have no relocations. | |
530 | if (object->is_incremental()) | |
531 | return true; | |
532 | ||
533 | // Find the relocation section | |
534 | unsigned int reloc_shndx = 0; | |
535 | unsigned int reloc_type = 0; | |
536 | for (unsigned int i = 0; i < object->shnum(); ++i) | |
537 | { | |
538 | reloc_type = object->section_type(i); | |
539 | if ((reloc_type == elfcpp::SHT_REL | |
540 | || reloc_type == elfcpp::SHT_RELA) | |
541 | && object->section_info(i) == shndx) | |
542 | { | |
543 | reloc_shndx = i; | |
544 | break; | |
545 | } | |
546 | } | |
547 | ||
548 | this->reloc_mapper_ = make_elf_reloc_mapper(object, symtab, symtab_size); | |
549 | this->reloc_mapper_->initialize(reloc_shndx, reloc_type); | |
550 | this->reloc_type_ = reloc_type; | |
551 | ||
c1027032 CC |
552 | return true; |
553 | } | |
554 | ||
555 | // Read the header for the set at OFFSET. | |
556 | ||
557 | bool | |
558 | Dwarf_pubnames_table::read_header(off_t offset) | |
559 | { | |
234d4ab8 SA |
560 | // Make sure we have actually read the section. |
561 | gold_assert(this->buffer_ != NULL); | |
562 | ||
c1027032 CC |
563 | // Correct the offset. For incremental update links, we have a |
564 | // relocated offset that is relative to the output section, but | |
565 | // here we need an offset relative to the input section. | |
566 | offset -= this->output_section_offset_; | |
567 | ||
568 | if (offset < 0 || offset + 14 >= this->buffer_end_ - this->buffer_) | |
569 | return false; | |
570 | ||
571 | const unsigned char* pinfo = this->buffer_ + offset; | |
572 | ||
573 | // Read the unit_length field. | |
234d4ab8 | 574 | uint64_t unit_length = this->dwinfo_->read_from_pointer<32>(pinfo); |
c1027032 CC |
575 | pinfo += 4; |
576 | if (unit_length == 0xffffffff) | |
577 | { | |
ed5d6712 | 578 | unit_length = this->dwinfo_->read_from_pointer<64>(pinfo); |
234d4ab8 | 579 | this->unit_length_ = unit_length + 12; |
c1027032 CC |
580 | pinfo += 8; |
581 | this->offset_size_ = 8; | |
582 | } | |
583 | else | |
234d4ab8 SA |
584 | { |
585 | this->unit_length_ = unit_length + 4; | |
586 | this->offset_size_ = 4; | |
587 | } | |
ec673e64 | 588 | this->end_of_table_ = pinfo + unit_length; |
c1027032 CC |
589 | |
590 | // Check the version. | |
ed5d6712 | 591 | unsigned int version = this->dwinfo_->read_from_pointer<16>(pinfo); |
c1027032 CC |
592 | pinfo += 2; |
593 | if (version != 2) | |
594 | return false; | |
50ed5eb1 | 595 | |
234d4ab8 SA |
596 | this->reloc_mapper_->get_reloc_target(pinfo - this->buffer_, |
597 | &this->cu_offset_); | |
598 | ||
c1027032 CC |
599 | // Skip the debug_info_offset and debug_info_size fields. |
600 | pinfo += 2 * this->offset_size_; | |
601 | ||
602 | if (pinfo >= this->buffer_end_) | |
603 | return false; | |
604 | ||
605 | this->pinfo_ = pinfo; | |
606 | return true; | |
607 | } | |
608 | ||
609 | // Read the next name from the set. | |
610 | ||
611 | const char* | |
ec673e64 | 612 | Dwarf_pubnames_table::next_name(uint8_t* flag_byte) |
c1027032 CC |
613 | { |
614 | const unsigned char* pinfo = this->pinfo_; | |
615 | ||
ec673e64 CC |
616 | // Check for end of list. The table should be terminated by an |
617 | // entry containing nothing but a DIE offset of 0. | |
618 | if (pinfo + this->offset_size_ >= this->end_of_table_) | |
c1027032 CC |
619 | return NULL; |
620 | ||
ec673e64 CC |
621 | // Skip the offset within the CU. If this is zero, but we're not |
622 | // at the end of the table, then we have a real pubnames entry | |
623 | // whose DIE offset is 0 (likely to be a GCC bug). Since we | |
624 | // don't actually use the DIE offset in building .gdb_index, | |
625 | // it's harmless. | |
626 | pinfo += this->offset_size_; | |
627 | ||
628 | if (this->is_gnu_style_) | |
629 | *flag_byte = *pinfo++; | |
630 | else | |
631 | *flag_byte = 0; | |
632 | ||
c1027032 CC |
633 | // Return a pointer to the string at the current location, |
634 | // and advance the pointer to the next entry. | |
635 | const char* ret = reinterpret_cast<const char*>(pinfo); | |
636 | while (pinfo < this->buffer_end_ && *pinfo != '\0') | |
637 | ++pinfo; | |
638 | if (pinfo < this->buffer_end_) | |
639 | ++pinfo; | |
640 | ||
641 | this->pinfo_ = pinfo; | |
642 | return ret; | |
643 | } | |
644 | ||
645 | // class Dwarf_die | |
646 | ||
647 | Dwarf_die::Dwarf_die( | |
648 | Dwarf_info_reader* dwinfo, | |
649 | off_t die_offset, | |
650 | Dwarf_die* parent) | |
651 | : dwinfo_(dwinfo), parent_(parent), die_offset_(die_offset), | |
652 | child_offset_(0), sibling_offset_(0), abbrev_code_(NULL), attributes_(), | |
653 | attributes_read_(false), name_(NULL), name_off_(-1), linkage_name_(NULL), | |
654 | linkage_name_off_(-1), string_shndx_(0), specification_(0), | |
655 | abstract_origin_(0) | |
656 | { | |
657 | size_t len; | |
658 | const unsigned char* pdie = dwinfo->buffer_at_offset(die_offset); | |
659 | if (pdie == NULL) | |
660 | return; | |
661 | unsigned int code = read_unsigned_LEB_128(pdie, &len); | |
662 | if (code == 0) | |
663 | { | |
664 | if (parent != NULL) | |
665 | parent->set_sibling_offset(die_offset + len); | |
666 | return; | |
667 | } | |
668 | this->attr_offset_ = len; | |
669 | ||
670 | // Lookup the abbrev code in the abbrev table. | |
671 | this->abbrev_code_ = dwinfo->get_abbrev(code); | |
672 | } | |
673 | ||
674 | // Read all the attributes of the DIE. | |
675 | ||
676 | bool | |
677 | Dwarf_die::read_attributes() | |
678 | { | |
679 | if (this->attributes_read_) | |
680 | return true; | |
681 | ||
682 | gold_assert(this->abbrev_code_ != NULL); | |
683 | ||
684 | const unsigned char* pdie = | |
685 | this->dwinfo_->buffer_at_offset(this->die_offset_); | |
686 | if (pdie == NULL) | |
687 | return false; | |
688 | const unsigned char* pattr = pdie + this->attr_offset_; | |
689 | ||
690 | unsigned int nattr = this->abbrev_code_->attributes.size(); | |
691 | this->attributes_.reserve(nattr); | |
692 | for (unsigned int i = 0; i < nattr; ++i) | |
693 | { | |
694 | size_t len; | |
695 | unsigned int attr = this->abbrev_code_->attributes[i].attr; | |
696 | unsigned int form = this->abbrev_code_->attributes[i].form; | |
697 | if (form == elfcpp::DW_FORM_indirect) | |
698 | { | |
699 | form = read_unsigned_LEB_128(pattr, &len); | |
700 | pattr += len; | |
701 | } | |
702 | off_t attr_off = this->die_offset_ + (pattr - pdie); | |
703 | bool ref_form = false; | |
704 | Attribute_value attr_value; | |
705 | attr_value.attr = attr; | |
706 | attr_value.form = form; | |
707 | attr_value.aux.shndx = 0; | |
708 | switch(form) | |
709 | { | |
c1027032 CC |
710 | case elfcpp::DW_FORM_flag_present: |
711 | attr_value.val.intval = 1; | |
712 | break; | |
713 | case elfcpp::DW_FORM_strp: | |
714 | { | |
715 | off_t str_off; | |
716 | if (this->dwinfo_->offset_size() == 4) | |
ed5d6712 | 717 | str_off = this->dwinfo_->read_from_pointer<32>(&pattr); |
c1027032 | 718 | else |
ed5d6712 | 719 | str_off = this->dwinfo_->read_from_pointer<64>(&pattr); |
c1027032 CC |
720 | unsigned int shndx = |
721 | this->dwinfo_->lookup_reloc(attr_off, &str_off); | |
722 | attr_value.aux.shndx = shndx; | |
723 | attr_value.val.refval = str_off; | |
724 | break; | |
725 | } | |
726 | case elfcpp::DW_FORM_sec_offset: | |
727 | { | |
728 | off_t sec_off; | |
729 | if (this->dwinfo_->offset_size() == 4) | |
ed5d6712 | 730 | sec_off = this->dwinfo_->read_from_pointer<32>(&pattr); |
c1027032 | 731 | else |
ed5d6712 | 732 | sec_off = this->dwinfo_->read_from_pointer<64>(&pattr); |
c1027032 CC |
733 | unsigned int shndx = |
734 | this->dwinfo_->lookup_reloc(attr_off, &sec_off); | |
735 | attr_value.aux.shndx = shndx; | |
736 | attr_value.val.refval = sec_off; | |
737 | ref_form = true; | |
738 | break; | |
739 | } | |
740 | case elfcpp::DW_FORM_addr: | |
741 | case elfcpp::DW_FORM_ref_addr: | |
742 | { | |
743 | off_t sec_off; | |
744 | if (this->dwinfo_->address_size() == 4) | |
ed5d6712 | 745 | sec_off = this->dwinfo_->read_from_pointer<32>(&pattr); |
c1027032 | 746 | else |
ed5d6712 | 747 | sec_off = this->dwinfo_->read_from_pointer<64>(&pattr); |
c1027032 CC |
748 | unsigned int shndx = |
749 | this->dwinfo_->lookup_reloc(attr_off, &sec_off); | |
750 | attr_value.aux.shndx = shndx; | |
751 | attr_value.val.refval = sec_off; | |
752 | ref_form = true; | |
753 | break; | |
754 | } | |
755 | case elfcpp::DW_FORM_block1: | |
756 | attr_value.aux.blocklen = *pattr++; | |
757 | attr_value.val.blockval = pattr; | |
758 | pattr += attr_value.aux.blocklen; | |
759 | break; | |
760 | case elfcpp::DW_FORM_block2: | |
ed5d6712 CC |
761 | attr_value.aux.blocklen = |
762 | this->dwinfo_->read_from_pointer<16>(&pattr); | |
c1027032 CC |
763 | attr_value.val.blockval = pattr; |
764 | pattr += attr_value.aux.blocklen; | |
765 | break; | |
766 | case elfcpp::DW_FORM_block4: | |
ed5d6712 CC |
767 | attr_value.aux.blocklen = |
768 | this->dwinfo_->read_from_pointer<32>(&pattr); | |
c1027032 CC |
769 | attr_value.val.blockval = pattr; |
770 | pattr += attr_value.aux.blocklen; | |
771 | break; | |
772 | case elfcpp::DW_FORM_block: | |
773 | case elfcpp::DW_FORM_exprloc: | |
774 | attr_value.aux.blocklen = read_unsigned_LEB_128(pattr, &len); | |
775 | attr_value.val.blockval = pattr + len; | |
776 | pattr += len + attr_value.aux.blocklen; | |
777 | break; | |
778 | case elfcpp::DW_FORM_data1: | |
779 | case elfcpp::DW_FORM_flag: | |
780 | attr_value.val.intval = *pattr++; | |
781 | break; | |
782 | case elfcpp::DW_FORM_ref1: | |
783 | attr_value.val.refval = *pattr++; | |
784 | ref_form = true; | |
785 | break; | |
786 | case elfcpp::DW_FORM_data2: | |
ed5d6712 CC |
787 | attr_value.val.intval = |
788 | this->dwinfo_->read_from_pointer<16>(&pattr); | |
c1027032 CC |
789 | break; |
790 | case elfcpp::DW_FORM_ref2: | |
ed5d6712 CC |
791 | attr_value.val.refval = |
792 | this->dwinfo_->read_from_pointer<16>(&pattr); | |
c1027032 CC |
793 | ref_form = true; |
794 | break; | |
795 | case elfcpp::DW_FORM_data4: | |
796 | { | |
797 | off_t sec_off; | |
ed5d6712 | 798 | sec_off = this->dwinfo_->read_from_pointer<32>(&pattr); |
c1027032 CC |
799 | unsigned int shndx = |
800 | this->dwinfo_->lookup_reloc(attr_off, &sec_off); | |
801 | attr_value.aux.shndx = shndx; | |
802 | attr_value.val.intval = sec_off; | |
803 | break; | |
804 | } | |
805 | case elfcpp::DW_FORM_ref4: | |
806 | { | |
807 | off_t sec_off; | |
ed5d6712 | 808 | sec_off = this->dwinfo_->read_from_pointer<32>(&pattr); |
c1027032 CC |
809 | unsigned int shndx = |
810 | this->dwinfo_->lookup_reloc(attr_off, &sec_off); | |
811 | attr_value.aux.shndx = shndx; | |
812 | attr_value.val.refval = sec_off; | |
813 | ref_form = true; | |
814 | break; | |
815 | } | |
816 | case elfcpp::DW_FORM_data8: | |
817 | { | |
818 | off_t sec_off; | |
ed5d6712 | 819 | sec_off = this->dwinfo_->read_from_pointer<64>(&pattr); |
c1027032 CC |
820 | unsigned int shndx = |
821 | this->dwinfo_->lookup_reloc(attr_off, &sec_off); | |
822 | attr_value.aux.shndx = shndx; | |
823 | attr_value.val.intval = sec_off; | |
824 | break; | |
825 | } | |
826 | case elfcpp::DW_FORM_ref_sig8: | |
ed5d6712 CC |
827 | attr_value.val.uintval = |
828 | this->dwinfo_->read_from_pointer<64>(&pattr); | |
c1027032 CC |
829 | break; |
830 | case elfcpp::DW_FORM_ref8: | |
831 | { | |
832 | off_t sec_off; | |
ed5d6712 | 833 | sec_off = this->dwinfo_->read_from_pointer<64>(&pattr); |
c1027032 CC |
834 | unsigned int shndx = |
835 | this->dwinfo_->lookup_reloc(attr_off, &sec_off); | |
836 | attr_value.aux.shndx = shndx; | |
837 | attr_value.val.refval = sec_off; | |
838 | ref_form = true; | |
839 | break; | |
840 | } | |
841 | case elfcpp::DW_FORM_ref_udata: | |
842 | attr_value.val.refval = read_unsigned_LEB_128(pattr, &len); | |
843 | ref_form = true; | |
844 | pattr += len; | |
845 | break; | |
846 | case elfcpp::DW_FORM_udata: | |
d2d60eef CC |
847 | case elfcpp::DW_FORM_GNU_addr_index: |
848 | case elfcpp::DW_FORM_GNU_str_index: | |
c1027032 CC |
849 | attr_value.val.uintval = read_unsigned_LEB_128(pattr, &len); |
850 | pattr += len; | |
851 | break; | |
852 | case elfcpp::DW_FORM_sdata: | |
853 | attr_value.val.intval = read_signed_LEB_128(pattr, &len); | |
854 | pattr += len; | |
855 | break; | |
856 | case elfcpp::DW_FORM_string: | |
857 | attr_value.val.stringval = reinterpret_cast<const char*>(pattr); | |
858 | len = strlen(attr_value.val.stringval); | |
859 | pattr += len + 1; | |
860 | break; | |
861 | default: | |
862 | return false; | |
863 | } | |
864 | ||
865 | // Cache the most frequently-requested attributes. | |
866 | switch (attr) | |
867 | { | |
868 | case elfcpp::DW_AT_name: | |
869 | if (form == elfcpp::DW_FORM_string) | |
870 | this->name_ = attr_value.val.stringval; | |
871 | else if (form == elfcpp::DW_FORM_strp) | |
872 | { | |
873 | // All indirect strings should refer to the same | |
874 | // string section, so we just save the last one seen. | |
875 | this->string_shndx_ = attr_value.aux.shndx; | |
876 | this->name_off_ = attr_value.val.refval; | |
877 | } | |
878 | break; | |
879 | case elfcpp::DW_AT_linkage_name: | |
880 | case elfcpp::DW_AT_MIPS_linkage_name: | |
881 | if (form == elfcpp::DW_FORM_string) | |
882 | this->linkage_name_ = attr_value.val.stringval; | |
883 | else if (form == elfcpp::DW_FORM_strp) | |
884 | { | |
885 | // All indirect strings should refer to the same | |
886 | // string section, so we just save the last one seen. | |
887 | this->string_shndx_ = attr_value.aux.shndx; | |
888 | this->linkage_name_off_ = attr_value.val.refval; | |
889 | } | |
890 | break; | |
891 | case elfcpp::DW_AT_specification: | |
892 | if (ref_form) | |
893 | this->specification_ = attr_value.val.refval; | |
894 | break; | |
895 | case elfcpp::DW_AT_abstract_origin: | |
896 | if (ref_form) | |
897 | this->abstract_origin_ = attr_value.val.refval; | |
898 | break; | |
899 | case elfcpp::DW_AT_sibling: | |
900 | if (ref_form && attr_value.aux.shndx == 0) | |
901 | this->sibling_offset_ = attr_value.val.refval; | |
902 | default: | |
903 | break; | |
904 | } | |
905 | ||
906 | this->attributes_.push_back(attr_value); | |
907 | } | |
908 | ||
909 | // Now that we know where the next DIE begins, record the offset | |
910 | // to avoid later recalculation. | |
911 | if (this->has_children()) | |
912 | this->child_offset_ = this->die_offset_ + (pattr - pdie); | |
913 | else | |
914 | this->sibling_offset_ = this->die_offset_ + (pattr - pdie); | |
915 | ||
916 | this->attributes_read_ = true; | |
917 | return true; | |
918 | } | |
919 | ||
920 | // Skip all the attributes of the DIE and return the offset of the next DIE. | |
921 | ||
922 | off_t | |
923 | Dwarf_die::skip_attributes() | |
924 | { | |
c1027032 CC |
925 | gold_assert(this->abbrev_code_ != NULL); |
926 | ||
927 | const unsigned char* pdie = | |
928 | this->dwinfo_->buffer_at_offset(this->die_offset_); | |
929 | if (pdie == NULL) | |
930 | return 0; | |
931 | const unsigned char* pattr = pdie + this->attr_offset_; | |
932 | ||
933 | for (unsigned int i = 0; i < this->abbrev_code_->attributes.size(); ++i) | |
934 | { | |
935 | size_t len; | |
936 | unsigned int form = this->abbrev_code_->attributes[i].form; | |
937 | if (form == elfcpp::DW_FORM_indirect) | |
938 | { | |
939 | form = read_unsigned_LEB_128(pattr, &len); | |
940 | pattr += len; | |
941 | } | |
942 | switch(form) | |
943 | { | |
c1027032 CC |
944 | case elfcpp::DW_FORM_flag_present: |
945 | break; | |
946 | case elfcpp::DW_FORM_strp: | |
947 | case elfcpp::DW_FORM_sec_offset: | |
948 | pattr += this->dwinfo_->offset_size(); | |
949 | break; | |
950 | case elfcpp::DW_FORM_addr: | |
951 | case elfcpp::DW_FORM_ref_addr: | |
952 | pattr += this->dwinfo_->address_size(); | |
953 | break; | |
954 | case elfcpp::DW_FORM_block1: | |
955 | pattr += 1 + *pattr; | |
956 | break; | |
957 | case elfcpp::DW_FORM_block2: | |
958 | { | |
959 | uint16_t block_size; | |
ed5d6712 | 960 | block_size = this->dwinfo_->read_from_pointer<16>(&pattr); |
c1027032 CC |
961 | pattr += block_size; |
962 | break; | |
963 | } | |
964 | case elfcpp::DW_FORM_block4: | |
965 | { | |
966 | uint32_t block_size; | |
ed5d6712 | 967 | block_size = this->dwinfo_->read_from_pointer<32>(&pattr); |
c1027032 CC |
968 | pattr += block_size; |
969 | break; | |
970 | } | |
971 | case elfcpp::DW_FORM_block: | |
972 | case elfcpp::DW_FORM_exprloc: | |
973 | { | |
974 | uint64_t block_size; | |
975 | block_size = read_unsigned_LEB_128(pattr, &len); | |
976 | pattr += len + block_size; | |
977 | break; | |
978 | } | |
979 | case elfcpp::DW_FORM_data1: | |
980 | case elfcpp::DW_FORM_ref1: | |
981 | case elfcpp::DW_FORM_flag: | |
982 | pattr += 1; | |
983 | break; | |
984 | case elfcpp::DW_FORM_data2: | |
985 | case elfcpp::DW_FORM_ref2: | |
986 | pattr += 2; | |
987 | break; | |
988 | case elfcpp::DW_FORM_data4: | |
989 | case elfcpp::DW_FORM_ref4: | |
990 | pattr += 4; | |
991 | break; | |
992 | case elfcpp::DW_FORM_data8: | |
993 | case elfcpp::DW_FORM_ref8: | |
994 | case elfcpp::DW_FORM_ref_sig8: | |
995 | pattr += 8; | |
996 | break; | |
997 | case elfcpp::DW_FORM_ref_udata: | |
998 | case elfcpp::DW_FORM_udata: | |
d2d60eef CC |
999 | case elfcpp::DW_FORM_GNU_addr_index: |
1000 | case elfcpp::DW_FORM_GNU_str_index: | |
c1027032 CC |
1001 | read_unsigned_LEB_128(pattr, &len); |
1002 | pattr += len; | |
1003 | break; | |
1004 | case elfcpp::DW_FORM_sdata: | |
1005 | read_signed_LEB_128(pattr, &len); | |
1006 | pattr += len; | |
1007 | break; | |
1008 | case elfcpp::DW_FORM_string: | |
1009 | len = strlen(reinterpret_cast<const char*>(pattr)); | |
1010 | pattr += len + 1; | |
1011 | break; | |
1012 | default: | |
1013 | return 0; | |
1014 | } | |
1015 | } | |
1016 | ||
1017 | return this->die_offset_ + (pattr - pdie); | |
1018 | } | |
1019 | ||
1020 | // Get the name of the DIE and cache it. | |
1021 | ||
1022 | void | |
1023 | Dwarf_die::set_name() | |
1024 | { | |
1025 | if (this->name_ != NULL || !this->read_attributes()) | |
1026 | return; | |
1027 | if (this->name_off_ != -1) | |
1028 | this->name_ = this->dwinfo_->get_string(this->name_off_, | |
1029 | this->string_shndx_); | |
1030 | } | |
1031 | ||
1032 | // Get the linkage name of the DIE and cache it. | |
1033 | ||
1034 | void | |
1035 | Dwarf_die::set_linkage_name() | |
1036 | { | |
1037 | if (this->linkage_name_ != NULL || !this->read_attributes()) | |
1038 | return; | |
1039 | if (this->linkage_name_off_ != -1) | |
1040 | this->linkage_name_ = this->dwinfo_->get_string(this->linkage_name_off_, | |
1041 | this->string_shndx_); | |
1042 | } | |
1043 | ||
1044 | // Return the value of attribute ATTR. | |
1045 | ||
1046 | const Dwarf_die::Attribute_value* | |
1047 | Dwarf_die::attribute(unsigned int attr) | |
1048 | { | |
1049 | if (!this->read_attributes()) | |
1050 | return NULL; | |
1051 | for (unsigned int i = 0; i < this->attributes_.size(); ++i) | |
1052 | { | |
1053 | if (this->attributes_[i].attr == attr) | |
1054 | return &this->attributes_[i]; | |
1055 | } | |
1056 | return NULL; | |
1057 | } | |
1058 | ||
1059 | const char* | |
1060 | Dwarf_die::string_attribute(unsigned int attr) | |
1061 | { | |
1062 | const Attribute_value* attr_val = this->attribute(attr); | |
1063 | if (attr_val == NULL) | |
1064 | return NULL; | |
1065 | switch (attr_val->form) | |
1066 | { | |
1067 | case elfcpp::DW_FORM_string: | |
1068 | return attr_val->val.stringval; | |
1069 | case elfcpp::DW_FORM_strp: | |
1070 | return this->dwinfo_->get_string(attr_val->val.refval, | |
1071 | attr_val->aux.shndx); | |
1072 | default: | |
1073 | return NULL; | |
1074 | } | |
1075 | } | |
1076 | ||
1077 | int64_t | |
1078 | Dwarf_die::int_attribute(unsigned int attr) | |
1079 | { | |
1080 | const Attribute_value* attr_val = this->attribute(attr); | |
1081 | if (attr_val == NULL) | |
1082 | return 0; | |
1083 | switch (attr_val->form) | |
1084 | { | |
c1027032 CC |
1085 | case elfcpp::DW_FORM_flag_present: |
1086 | case elfcpp::DW_FORM_data1: | |
1087 | case elfcpp::DW_FORM_flag: | |
1088 | case elfcpp::DW_FORM_data2: | |
1089 | case elfcpp::DW_FORM_data4: | |
1090 | case elfcpp::DW_FORM_data8: | |
1091 | case elfcpp::DW_FORM_sdata: | |
1092 | return attr_val->val.intval; | |
1093 | default: | |
1094 | return 0; | |
1095 | } | |
1096 | } | |
1097 | ||
1098 | uint64_t | |
1099 | Dwarf_die::uint_attribute(unsigned int attr) | |
1100 | { | |
1101 | const Attribute_value* attr_val = this->attribute(attr); | |
1102 | if (attr_val == NULL) | |
1103 | return 0; | |
1104 | switch (attr_val->form) | |
1105 | { | |
c1027032 CC |
1106 | case elfcpp::DW_FORM_flag_present: |
1107 | case elfcpp::DW_FORM_data1: | |
1108 | case elfcpp::DW_FORM_flag: | |
1109 | case elfcpp::DW_FORM_data4: | |
1110 | case elfcpp::DW_FORM_data8: | |
1111 | case elfcpp::DW_FORM_ref_sig8: | |
1112 | case elfcpp::DW_FORM_udata: | |
1113 | return attr_val->val.uintval; | |
1114 | default: | |
1115 | return 0; | |
1116 | } | |
1117 | } | |
1118 | ||
1119 | off_t | |
1120 | Dwarf_die::ref_attribute(unsigned int attr, unsigned int* shndx) | |
1121 | { | |
1122 | const Attribute_value* attr_val = this->attribute(attr); | |
1123 | if (attr_val == NULL) | |
1124 | return -1; | |
1125 | switch (attr_val->form) | |
1126 | { | |
1127 | case elfcpp::DW_FORM_sec_offset: | |
1128 | case elfcpp::DW_FORM_addr: | |
1129 | case elfcpp::DW_FORM_ref_addr: | |
1130 | case elfcpp::DW_FORM_ref1: | |
1131 | case elfcpp::DW_FORM_ref2: | |
1132 | case elfcpp::DW_FORM_ref4: | |
1133 | case elfcpp::DW_FORM_ref8: | |
1134 | case elfcpp::DW_FORM_ref_udata: | |
1135 | *shndx = attr_val->aux.shndx; | |
1136 | return attr_val->val.refval; | |
1137 | case elfcpp::DW_FORM_ref_sig8: | |
1138 | *shndx = attr_val->aux.shndx; | |
1139 | return attr_val->val.uintval; | |
1140 | case elfcpp::DW_FORM_data4: | |
1141 | case elfcpp::DW_FORM_data8: | |
1142 | *shndx = attr_val->aux.shndx; | |
1143 | return attr_val->val.intval; | |
1144 | default: | |
1145 | return -1; | |
1146 | } | |
1147 | } | |
1148 | ||
57923f48 MW |
1149 | off_t |
1150 | Dwarf_die::address_attribute(unsigned int attr, unsigned int* shndx) | |
1151 | { | |
1152 | const Attribute_value* attr_val = this->attribute(attr); | |
1153 | if (attr_val == NULL || attr_val->form != elfcpp::DW_FORM_addr) | |
1154 | return -1; | |
1155 | ||
1156 | *shndx = attr_val->aux.shndx; | |
1157 | return attr_val->val.refval; | |
1158 | } | |
1159 | ||
c1027032 CC |
1160 | // Return the offset of this DIE's first child. |
1161 | ||
1162 | off_t | |
1163 | Dwarf_die::child_offset() | |
1164 | { | |
1165 | gold_assert(this->abbrev_code_ != NULL); | |
1166 | if (!this->has_children()) | |
1167 | return 0; | |
1168 | if (this->child_offset_ == 0) | |
1169 | this->child_offset_ = this->skip_attributes(); | |
1170 | return this->child_offset_; | |
1171 | } | |
1172 | ||
1173 | // Return the offset of this DIE's next sibling. | |
1174 | ||
1175 | off_t | |
1176 | Dwarf_die::sibling_offset() | |
1177 | { | |
1178 | gold_assert(this->abbrev_code_ != NULL); | |
1179 | ||
1180 | if (this->sibling_offset_ != 0) | |
1181 | return this->sibling_offset_; | |
1182 | ||
1183 | if (!this->has_children()) | |
1184 | { | |
1185 | this->sibling_offset_ = this->skip_attributes(); | |
1186 | return this->sibling_offset_; | |
1187 | } | |
1188 | ||
1189 | if (this->has_sibling_attribute()) | |
1190 | { | |
1191 | if (!this->read_attributes()) | |
1192 | return 0; | |
1193 | if (this->sibling_offset_ != 0) | |
1194 | return this->sibling_offset_; | |
1195 | } | |
1196 | ||
1197 | // Skip over the children. | |
1198 | off_t child_offset = this->child_offset(); | |
1199 | while (child_offset > 0) | |
1200 | { | |
1201 | Dwarf_die die(this->dwinfo_, child_offset, this); | |
1202 | // The Dwarf_die ctor will set this DIE's sibling offset | |
1203 | // when it reads a zero abbrev code. | |
1204 | if (die.tag() == 0) | |
1205 | break; | |
1206 | child_offset = die.sibling_offset(); | |
1207 | } | |
1208 | ||
1209 | // This should be set by now. If not, there was a problem reading | |
1210 | // the DWARF info, and we return 0. | |
1211 | return this->sibling_offset_; | |
1212 | } | |
1213 | ||
1214 | // class Dwarf_info_reader | |
1215 | ||
c1027032 CC |
1216 | // Begin parsing the debug info. This calls visit_compilation_unit() |
1217 | // or visit_type_unit() for each compilation or type unit found in the | |
1218 | // section, and visit_die() for each top-level DIE. | |
1219 | ||
1220 | void | |
1221 | Dwarf_info_reader::parse() | |
1222 | { | |
9fc236f3 | 1223 | if (this->object_->is_big_endian()) |
c1027032 | 1224 | { |
9fc236f3 CC |
1225 | #if defined(HAVE_TARGET_32_BIG) || defined(HAVE_TARGET_64_BIG) |
1226 | this->do_parse<true>(); | |
1227 | #else | |
1228 | gold_unreachable(); | |
c1027032 | 1229 | #endif |
9fc236f3 CC |
1230 | } |
1231 | else | |
1232 | { | |
1233 | #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_64_LITTLE) | |
1234 | this->do_parse<false>(); | |
1235 | #else | |
1236 | gold_unreachable(); | |
c1027032 | 1237 | #endif |
c1027032 CC |
1238 | } |
1239 | } | |
1240 | ||
1241 | template<bool big_endian> | |
1242 | void | |
1243 | Dwarf_info_reader::do_parse() | |
1244 | { | |
1245 | // Get the section contents and decompress if necessary. | |
1246 | section_size_type buffer_size; | |
1247 | bool buffer_is_new; | |
1248 | this->buffer_ = this->object_->decompressed_section_contents(this->shndx_, | |
1249 | &buffer_size, | |
1250 | &buffer_is_new); | |
1251 | if (this->buffer_ == NULL || buffer_size == 0) | |
1252 | return; | |
1253 | this->buffer_end_ = this->buffer_ + buffer_size; | |
1254 | ||
1255 | // The offset of this input section in the output section. | |
1256 | off_t section_offset = this->object_->output_section_offset(this->shndx_); | |
1257 | ||
1258 | // Start tracking relocations for this section. | |
1259 | this->reloc_mapper_ = make_elf_reloc_mapper(this->object_, this->symtab_, | |
1260 | this->symtab_size_); | |
1261 | this->reloc_mapper_->initialize(this->reloc_shndx_, this->reloc_type_); | |
1262 | ||
1263 | // Loop over compilation units (or type units). | |
8787852d | 1264 | unsigned int abbrev_shndx = this->abbrev_shndx_; |
c1027032 CC |
1265 | off_t abbrev_offset = 0; |
1266 | const unsigned char* pinfo = this->buffer_; | |
1267 | while (pinfo < this->buffer_end_) | |
1268 | { | |
1269 | // Read the compilation (or type) unit header. | |
1270 | const unsigned char* cu_start = pinfo; | |
1271 | this->cu_offset_ = cu_start - this->buffer_; | |
1272 | this->cu_length_ = this->buffer_end_ - cu_start; | |
1273 | ||
1274 | // Read unit_length (4 or 12 bytes). | |
1275 | if (!this->check_buffer(pinfo + 4)) | |
1276 | break; | |
1277 | uint32_t unit_length = | |
1278 | elfcpp::Swap_unaligned<32, big_endian>::readval(pinfo); | |
1279 | pinfo += 4; | |
1280 | if (unit_length == 0xffffffff) | |
1281 | { | |
1282 | if (!this->check_buffer(pinfo + 8)) | |
1283 | break; | |
1284 | unit_length = elfcpp::Swap_unaligned<64, big_endian>::readval(pinfo); | |
1285 | pinfo += 8; | |
1286 | this->offset_size_ = 8; | |
1287 | } | |
1288 | else | |
1289 | this->offset_size_ = 4; | |
1290 | if (!this->check_buffer(pinfo + unit_length)) | |
1291 | break; | |
1292 | const unsigned char* cu_end = pinfo + unit_length; | |
1293 | this->cu_length_ = cu_end - cu_start; | |
1294 | if (!this->check_buffer(pinfo + 2 + this->offset_size_ + 1)) | |
1295 | break; | |
1296 | ||
1297 | // Read version (2 bytes). | |
1298 | this->cu_version_ = | |
1299 | elfcpp::Swap_unaligned<16, big_endian>::readval(pinfo); | |
1300 | pinfo += 2; | |
1301 | ||
1302 | // Read debug_abbrev_offset (4 or 8 bytes). | |
1303 | if (this->offset_size_ == 4) | |
1304 | abbrev_offset = elfcpp::Swap_unaligned<32, big_endian>::readval(pinfo); | |
1305 | else | |
1306 | abbrev_offset = elfcpp::Swap_unaligned<64, big_endian>::readval(pinfo); | |
1307 | if (this->reloc_shndx_ > 0) | |
1308 | { | |
1309 | off_t reloc_offset = pinfo - this->buffer_; | |
1310 | off_t value; | |
1311 | abbrev_shndx = | |
1312 | this->reloc_mapper_->get_reloc_target(reloc_offset, &value); | |
1313 | if (abbrev_shndx == 0) | |
1314 | return; | |
1315 | if (this->reloc_type_ == elfcpp::SHT_REL) | |
1316 | abbrev_offset += value; | |
1317 | else | |
1318 | abbrev_offset = value; | |
1319 | } | |
1320 | pinfo += this->offset_size_; | |
1321 | ||
1322 | // Read address_size (1 byte). | |
1323 | this->address_size_ = *pinfo++; | |
1324 | ||
1325 | // For type units, read the two extra fields. | |
1326 | uint64_t signature = 0; | |
1327 | off_t type_offset = 0; | |
1328 | if (this->is_type_unit_) | |
1329 | { | |
1330 | if (!this->check_buffer(pinfo + 8 + this->offset_size_)) | |
1331 | break; | |
1332 | ||
1333 | // Read type_signature (8 bytes). | |
1334 | signature = elfcpp::Swap_unaligned<64, big_endian>::readval(pinfo); | |
1335 | pinfo += 8; | |
1336 | ||
1337 | // Read type_offset (4 or 8 bytes). | |
1338 | if (this->offset_size_ == 4) | |
1339 | type_offset = | |
1340 | elfcpp::Swap_unaligned<32, big_endian>::readval(pinfo); | |
1341 | else | |
1342 | type_offset = | |
1343 | elfcpp::Swap_unaligned<64, big_endian>::readval(pinfo); | |
1344 | pinfo += this->offset_size_; | |
1345 | } | |
1346 | ||
1347 | // Read the .debug_abbrev table. | |
1348 | this->abbrev_table_.read_abbrevs(this->object_, abbrev_shndx, | |
1349 | abbrev_offset); | |
1350 | ||
1351 | // Visit the root DIE. | |
1352 | Dwarf_die root_die(this, | |
1353 | pinfo - (this->buffer_ + this->cu_offset_), | |
1354 | NULL); | |
1355 | if (root_die.tag() != 0) | |
1356 | { | |
1357 | // Visit the CU or TU. | |
1358 | if (this->is_type_unit_) | |
1359 | this->visit_type_unit(section_offset + this->cu_offset_, | |
908794a9 CC |
1360 | cu_end - cu_start, type_offset, signature, |
1361 | &root_die); | |
c1027032 CC |
1362 | else |
1363 | this->visit_compilation_unit(section_offset + this->cu_offset_, | |
1364 | cu_end - cu_start, &root_die); | |
1365 | } | |
1366 | ||
1367 | // Advance to the next CU. | |
1368 | pinfo = cu_end; | |
1369 | } | |
1370 | ||
1371 | if (buffer_is_new) | |
1372 | { | |
1373 | delete[] this->buffer_; | |
1374 | this->buffer_ = NULL; | |
1375 | } | |
1376 | } | |
1377 | ||
1378 | // Read the DWARF string table. | |
1379 | ||
1380 | bool | |
1381 | Dwarf_info_reader::do_read_string_table(unsigned int string_shndx) | |
1382 | { | |
1383 | Relobj* object = this->object_; | |
1384 | ||
1385 | // If we don't have relocations, string_shndx will be 0, and | |
1386 | // we'll have to hunt for the .debug_str section. | |
1387 | if (string_shndx == 0) | |
1388 | { | |
1389 | for (unsigned int i = 1; i < this->object_->shnum(); ++i) | |
1390 | { | |
1391 | std::string name = object->section_name(i); | |
ec673e64 | 1392 | if (name == ".debug_str" || name == ".zdebug_str") |
c1027032 CC |
1393 | { |
1394 | string_shndx = i; | |
1395 | this->string_output_section_offset_ = | |
1396 | object->output_section_offset(i); | |
1397 | break; | |
1398 | } | |
1399 | } | |
1400 | if (string_shndx == 0) | |
1401 | return false; | |
1402 | } | |
1403 | ||
1404 | if (this->owns_string_buffer_ && this->string_buffer_ != NULL) | |
1405 | { | |
1406 | delete[] this->string_buffer_; | |
1407 | this->owns_string_buffer_ = false; | |
1408 | } | |
1409 | ||
1410 | // Get the secton contents and decompress if necessary. | |
1411 | section_size_type buffer_size; | |
1412 | const unsigned char* buffer = | |
1413 | object->decompressed_section_contents(string_shndx, | |
1414 | &buffer_size, | |
1415 | &this->owns_string_buffer_); | |
1416 | this->string_buffer_ = reinterpret_cast<const char*>(buffer); | |
1417 | this->string_buffer_end_ = this->string_buffer_ + buffer_size; | |
1418 | this->string_shndx_ = string_shndx; | |
1419 | return true; | |
1420 | } | |
1421 | ||
ed5d6712 CC |
1422 | // Read a possibly unaligned integer of SIZE. |
1423 | template <int valsize> | |
1424 | inline typename elfcpp::Valtype_base<valsize>::Valtype | |
1425 | Dwarf_info_reader::read_from_pointer(const unsigned char* source) | |
1426 | { | |
1427 | typename elfcpp::Valtype_base<valsize>::Valtype return_value; | |
1428 | if (this->object_->is_big_endian()) | |
1429 | return_value = elfcpp::Swap_unaligned<valsize, true>::readval(source); | |
1430 | else | |
1431 | return_value = elfcpp::Swap_unaligned<valsize, false>::readval(source); | |
1432 | return return_value; | |
1433 | } | |
1434 | ||
1435 | // Read a possibly unaligned integer of SIZE. Update SOURCE after read. | |
1436 | template <int valsize> | |
1437 | inline typename elfcpp::Valtype_base<valsize>::Valtype | |
1438 | Dwarf_info_reader::read_from_pointer(const unsigned char** source) | |
1439 | { | |
1440 | typename elfcpp::Valtype_base<valsize>::Valtype return_value; | |
1441 | if (this->object_->is_big_endian()) | |
1442 | return_value = elfcpp::Swap_unaligned<valsize, true>::readval(*source); | |
1443 | else | |
1444 | return_value = elfcpp::Swap_unaligned<valsize, false>::readval(*source); | |
1445 | *source += valsize / 8; | |
1446 | return return_value; | |
1447 | } | |
1448 | ||
c1027032 CC |
1449 | // Look for a relocation at offset ATTR_OFF in the dwarf info, |
1450 | // and return the section index and offset of the target. | |
1451 | ||
1452 | unsigned int | |
1453 | Dwarf_info_reader::lookup_reloc(off_t attr_off, off_t* target_off) | |
1454 | { | |
1455 | off_t value; | |
1456 | attr_off += this->cu_offset_; | |
1457 | unsigned int shndx = this->reloc_mapper_->get_reloc_target(attr_off, &value); | |
1458 | if (shndx == 0) | |
1459 | return 0; | |
1460 | if (this->reloc_type_ == elfcpp::SHT_REL) | |
1461 | *target_off += value; | |
1462 | else | |
1463 | *target_off = value; | |
1464 | return shndx; | |
1465 | } | |
1466 | ||
1467 | // Return a string from the DWARF string table. | |
1468 | ||
1469 | const char* | |
1470 | Dwarf_info_reader::get_string(off_t str_off, unsigned int string_shndx) | |
1471 | { | |
1472 | if (!this->read_string_table(string_shndx)) | |
1473 | return NULL; | |
1474 | ||
1475 | // Correct the offset. For incremental update links, we have a | |
1476 | // relocated offset that is relative to the output section, but | |
1477 | // here we need an offset relative to the input section. | |
1478 | str_off -= this->string_output_section_offset_; | |
1479 | ||
1480 | const char* p = this->string_buffer_ + str_off; | |
1481 | ||
1482 | if (p < this->string_buffer_ || p >= this->string_buffer_end_) | |
1483 | return NULL; | |
1484 | ||
1485 | return p; | |
1486 | } | |
1487 | ||
1488 | // The following are default, do-nothing, implementations of the | |
1489 | // hook methods normally provided by a derived class. We provide | |
1490 | // default implementations rather than no implementation so that | |
1491 | // a derived class needs to implement only the hooks that it needs | |
1492 | // to use. | |
1493 | ||
1494 | // Process a compilation unit and parse its child DIE. | |
1495 | ||
1496 | void | |
1497 | Dwarf_info_reader::visit_compilation_unit(off_t, off_t, Dwarf_die*) | |
1498 | { | |
1499 | } | |
1500 | ||
1501 | // Process a type unit and parse its child DIE. | |
1502 | ||
1503 | void | |
908794a9 | 1504 | Dwarf_info_reader::visit_type_unit(off_t, off_t, off_t, uint64_t, Dwarf_die*) |
c1027032 CC |
1505 | { |
1506 | } | |
1507 | ||
a68a081d CC |
1508 | // Print a warning about a corrupt debug section. |
1509 | ||
1510 | void | |
1511 | Dwarf_info_reader::warn_corrupt_debug_section() const | |
1512 | { | |
1513 | gold_warning(_("%s: corrupt debug info in %s"), | |
1514 | this->object_->name().c_str(), | |
1515 | this->object_->section_name(this->shndx_).c_str()); | |
1516 | } | |
1517 | ||
c1027032 CC |
1518 | // class Sized_dwarf_line_info |
1519 | ||
5c2c6c95 ILT |
1520 | struct LineStateMachine |
1521 | { | |
1522 | int file_num; | |
1523 | uint64_t address; | |
1524 | int line_num; | |
1525 | int column_num; | |
1526 | unsigned int shndx; // the section address refers to | |
1527 | bool is_stmt; // stmt means statement. | |
1528 | bool basic_block; | |
1529 | bool end_sequence; | |
1530 | }; | |
1531 | ||
1532 | static void | |
1533 | ResetLineStateMachine(struct LineStateMachine* lsm, bool default_is_stmt) | |
1534 | { | |
1535 | lsm->file_num = 1; | |
1536 | lsm->address = 0; | |
1537 | lsm->line_num = 1; | |
1538 | lsm->column_num = 0; | |
338f2eba | 1539 | lsm->shndx = -1U; |
5c2c6c95 ILT |
1540 | lsm->is_stmt = default_is_stmt; |
1541 | lsm->basic_block = false; | |
1542 | lsm->end_sequence = false; | |
1543 | } | |
1544 | ||
24badc65 | 1545 | template<int size, bool big_endian> |
5dd8762a CC |
1546 | Sized_dwarf_line_info<size, big_endian>::Sized_dwarf_line_info( |
1547 | Object* object, | |
1548 | unsigned int read_shndx) | |
1549 | : data_valid_(false), buffer_(NULL), buffer_start_(NULL), | |
c1027032 CC |
1550 | reloc_mapper_(NULL), symtab_buffer_(NULL), directories_(), files_(), |
1551 | current_header_index_(-1) | |
24badc65 ILT |
1552 | { |
1553 | unsigned int debug_shndx; | |
5dd8762a | 1554 | |
ab8056e0 CC |
1555 | for (debug_shndx = 1; debug_shndx < object->shnum(); ++debug_shndx) |
1556 | { | |
1557 | // FIXME: do this more efficiently: section_name() isn't super-fast | |
1558 | std::string name = object->section_name(debug_shndx); | |
1559 | if (name == ".debug_line" || name == ".zdebug_line") | |
1560 | { | |
1561 | section_size_type buffer_size; | |
5dd8762a CC |
1562 | bool is_new = false; |
1563 | this->buffer_ = object->decompressed_section_contents(debug_shndx, | |
1564 | &buffer_size, | |
1565 | &is_new); | |
1566 | if (is_new) | |
1567 | this->buffer_start_ = this->buffer_; | |
ab8056e0 CC |
1568 | this->buffer_end_ = this->buffer_ + buffer_size; |
1569 | break; | |
1570 | } | |
1571 | } | |
24badc65 | 1572 | if (this->buffer_ == NULL) |
c261a0be | 1573 | return; |
24badc65 ILT |
1574 | |
1575 | // Find the relocation section for ".debug_line". | |
af674d1d | 1576 | // We expect these for relobjs (.o's) but not dynobjs (.so's). |
c1027032 CC |
1577 | unsigned int reloc_shndx = 0; |
1578 | for (unsigned int i = 0; i < object->shnum(); ++i) | |
24badc65 | 1579 | { |
c1027032 | 1580 | unsigned int reloc_sh_type = object->section_type(i); |
24badc65 ILT |
1581 | if ((reloc_sh_type == elfcpp::SHT_REL |
1582 | || reloc_sh_type == elfcpp::SHT_RELA) | |
c1027032 | 1583 | && object->section_info(i) == debug_shndx) |
24badc65 | 1584 | { |
c1027032 | 1585 | reloc_shndx = i; |
4dbfafcc | 1586 | this->track_relocs_type_ = reloc_sh_type; |
24badc65 ILT |
1587 | break; |
1588 | } | |
1589 | } | |
24badc65 ILT |
1590 | |
1591 | // Finally, we need the symtab section to interpret the relocs. | |
c1027032 | 1592 | if (reloc_shndx != 0) |
af674d1d ILT |
1593 | { |
1594 | unsigned int symtab_shndx; | |
1595 | for (symtab_shndx = 0; symtab_shndx < object->shnum(); ++symtab_shndx) | |
1596 | if (object->section_type(symtab_shndx) == elfcpp::SHT_SYMTAB) | |
1597 | { | |
c1027032 CC |
1598 | this->symtab_buffer_ = object->section_contents( |
1599 | symtab_shndx, &this->symtab_buffer_size_, false); | |
af674d1d ILT |
1600 | break; |
1601 | } | |
1602 | if (this->symtab_buffer_ == NULL) | |
1603 | return; | |
1604 | } | |
24badc65 | 1605 | |
c1027032 CC |
1606 | this->reloc_mapper_ = |
1607 | new Sized_elf_reloc_mapper<size, big_endian>(object, | |
1608 | this->symtab_buffer_, | |
1609 | this->symtab_buffer_size_); | |
1610 | if (!this->reloc_mapper_->initialize(reloc_shndx, this->track_relocs_type_)) | |
1611 | return; | |
1612 | ||
24badc65 ILT |
1613 | // Now that we have successfully read all the data, parse the debug |
1614 | // info. | |
c261a0be | 1615 | this->data_valid_ = true; |
c1027032 | 1616 | this->read_line_mappings(read_shndx); |
24badc65 ILT |
1617 | } |
1618 | ||
5c2c6c95 ILT |
1619 | // Read the DWARF header. |
1620 | ||
1621 | template<int size, bool big_endian> | |
1622 | const unsigned char* | |
a55ce7fe | 1623 | Sized_dwarf_line_info<size, big_endian>::read_header_prolog( |
e43872e9 | 1624 | const unsigned char* lineptr) |
5c2c6c95 | 1625 | { |
deae2a14 | 1626 | uint32_t initial_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr); |
5c2c6c95 ILT |
1627 | lineptr += 4; |
1628 | ||
1629 | // In DWARF2/3, if the initial length is all 1 bits, then the offset | |
1630 | // size is 8 and we need to read the next 8 bytes for the real length. | |
1631 | if (initial_length == 0xffffffff) | |
1632 | { | |
1633 | header_.offset_size = 8; | |
deae2a14 | 1634 | initial_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr); |
5c2c6c95 ILT |
1635 | lineptr += 8; |
1636 | } | |
1637 | else | |
1638 | header_.offset_size = 4; | |
1639 | ||
1640 | header_.total_length = initial_length; | |
1641 | ||
1642 | gold_assert(lineptr + header_.total_length <= buffer_end_); | |
1643 | ||
deae2a14 | 1644 | header_.version = elfcpp::Swap_unaligned<16, big_endian>::readval(lineptr); |
5c2c6c95 ILT |
1645 | lineptr += 2; |
1646 | ||
1647 | if (header_.offset_size == 4) | |
deae2a14 | 1648 | header_.prologue_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr); |
5c2c6c95 | 1649 | else |
deae2a14 | 1650 | header_.prologue_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr); |
5c2c6c95 ILT |
1651 | lineptr += header_.offset_size; |
1652 | ||
1653 | header_.min_insn_length = *lineptr; | |
1654 | lineptr += 1; | |
1655 | ||
1656 | header_.default_is_stmt = *lineptr; | |
1657 | lineptr += 1; | |
1658 | ||
1659 | header_.line_base = *reinterpret_cast<const signed char*>(lineptr); | |
1660 | lineptr += 1; | |
1661 | ||
1662 | header_.line_range = *lineptr; | |
1663 | lineptr += 1; | |
1664 | ||
1665 | header_.opcode_base = *lineptr; | |
1666 | lineptr += 1; | |
1667 | ||
a869183f | 1668 | header_.std_opcode_lengths.resize(header_.opcode_base + 1); |
5c2c6c95 ILT |
1669 | header_.std_opcode_lengths[0] = 0; |
1670 | for (int i = 1; i < header_.opcode_base; i++) | |
1671 | { | |
1672 | header_.std_opcode_lengths[i] = *lineptr; | |
1673 | lineptr += 1; | |
1674 | } | |
1675 | ||
1676 | return lineptr; | |
1677 | } | |
1678 | ||
1679 | // The header for a debug_line section is mildly complicated, because | |
1680 | // the line info is very tightly encoded. | |
1681 | ||
e43872e9 | 1682 | template<int size, bool big_endian> |
5c2c6c95 | 1683 | const unsigned char* |
a55ce7fe | 1684 | Sized_dwarf_line_info<size, big_endian>::read_header_tables( |
e43872e9 | 1685 | const unsigned char* lineptr) |
5c2c6c95 | 1686 | { |
af674d1d ILT |
1687 | ++this->current_header_index_; |
1688 | ||
1689 | // Create a new directories_ entry and a new files_ entry for our new | |
1690 | // header. We initialize each with a single empty element, because | |
1691 | // dwarf indexes directory and filenames starting at 1. | |
1692 | gold_assert(static_cast<int>(this->directories_.size()) | |
1693 | == this->current_header_index_); | |
1694 | gold_assert(static_cast<int>(this->files_.size()) | |
1695 | == this->current_header_index_); | |
1696 | this->directories_.push_back(std::vector<std::string>(1)); | |
1697 | this->files_.push_back(std::vector<std::pair<int, std::string> >(1)); | |
1698 | ||
5c2c6c95 ILT |
1699 | // It is legal for the directory entry table to be empty. |
1700 | if (*lineptr) | |
1701 | { | |
1702 | int dirindex = 1; | |
1703 | while (*lineptr) | |
1704 | { | |
af674d1d ILT |
1705 | const char* dirname = reinterpret_cast<const char*>(lineptr); |
1706 | gold_assert(dirindex | |
1707 | == static_cast<int>(this->directories_.back().size())); | |
1708 | this->directories_.back().push_back(dirname); | |
1709 | lineptr += this->directories_.back().back().size() + 1; | |
5c2c6c95 ILT |
1710 | dirindex++; |
1711 | } | |
1712 | } | |
1713 | lineptr++; | |
1714 | ||
1715 | // It is also legal for the file entry table to be empty. | |
1716 | if (*lineptr) | |
1717 | { | |
1718 | int fileindex = 1; | |
1719 | size_t len; | |
1720 | while (*lineptr) | |
1721 | { | |
1722 | const char* filename = reinterpret_cast<const char*>(lineptr); | |
1723 | lineptr += strlen(filename) + 1; | |
1724 | ||
1725 | uint64_t dirindex = read_unsigned_LEB_128(lineptr, &len); | |
5c2c6c95 ILT |
1726 | lineptr += len; |
1727 | ||
af674d1d ILT |
1728 | if (dirindex >= this->directories_.back().size()) |
1729 | dirindex = 0; | |
1730 | int dirindexi = static_cast<int>(dirindex); | |
1731 | ||
5c2c6c95 ILT |
1732 | read_unsigned_LEB_128(lineptr, &len); // mod_time |
1733 | lineptr += len; | |
1734 | ||
1735 | read_unsigned_LEB_128(lineptr, &len); // filelength | |
1736 | lineptr += len; | |
1737 | ||
af674d1d ILT |
1738 | gold_assert(fileindex |
1739 | == static_cast<int>(this->files_.back().size())); | |
1740 | this->files_.back().push_back(std::make_pair(dirindexi, filename)); | |
5c2c6c95 ILT |
1741 | fileindex++; |
1742 | } | |
1743 | } | |
1744 | lineptr++; | |
1745 | ||
1746 | return lineptr; | |
1747 | } | |
1748 | ||
1749 | // Process a single opcode in the .debug.line structure. | |
1750 | ||
e43872e9 | 1751 | template<int size, bool big_endian> |
5c2c6c95 | 1752 | bool |
a55ce7fe | 1753 | Sized_dwarf_line_info<size, big_endian>::process_one_opcode( |
e43872e9 | 1754 | const unsigned char* start, struct LineStateMachine* lsm, size_t* len) |
5c2c6c95 ILT |
1755 | { |
1756 | size_t oplen = 0; | |
1757 | size_t templen; | |
1758 | unsigned char opcode = *start; | |
1759 | oplen++; | |
1760 | start++; | |
1761 | ||
1762 | // If the opcode is great than the opcode_base, it is a special | |
1763 | // opcode. Most line programs consist mainly of special opcodes. | |
1764 | if (opcode >= header_.opcode_base) | |
1765 | { | |
1766 | opcode -= header_.opcode_base; | |
1767 | const int advance_address = ((opcode / header_.line_range) | |
1768 | * header_.min_insn_length); | |
1769 | lsm->address += advance_address; | |
1770 | ||
1771 | const int advance_line = ((opcode % header_.line_range) | |
1772 | + header_.line_base); | |
1773 | lsm->line_num += advance_line; | |
1774 | lsm->basic_block = true; | |
1775 | *len = oplen; | |
1776 | return true; | |
1777 | } | |
1778 | ||
1779 | // Otherwise, we have the regular opcodes | |
1780 | switch (opcode) | |
1781 | { | |
1782 | case elfcpp::DW_LNS_copy: | |
1783 | lsm->basic_block = false; | |
1784 | *len = oplen; | |
1785 | return true; | |
1786 | ||
1787 | case elfcpp::DW_LNS_advance_pc: | |
1788 | { | |
1789 | const uint64_t advance_address | |
2ea97941 | 1790 | = read_unsigned_LEB_128(start, &templen); |
5c2c6c95 ILT |
1791 | oplen += templen; |
1792 | lsm->address += header_.min_insn_length * advance_address; | |
1793 | } | |
1794 | break; | |
1795 | ||
1796 | case elfcpp::DW_LNS_advance_line: | |
1797 | { | |
1798 | const uint64_t advance_line = read_signed_LEB_128(start, &templen); | |
1799 | oplen += templen; | |
1800 | lsm->line_num += advance_line; | |
1801 | } | |
1802 | break; | |
1803 | ||
1804 | case elfcpp::DW_LNS_set_file: | |
1805 | { | |
1806 | const uint64_t fileno = read_unsigned_LEB_128(start, &templen); | |
1807 | oplen += templen; | |
1808 | lsm->file_num = fileno; | |
1809 | } | |
1810 | break; | |
1811 | ||
1812 | case elfcpp::DW_LNS_set_column: | |
1813 | { | |
1814 | const uint64_t colno = read_unsigned_LEB_128(start, &templen); | |
1815 | oplen += templen; | |
1816 | lsm->column_num = colno; | |
1817 | } | |
1818 | break; | |
1819 | ||
1820 | case elfcpp::DW_LNS_negate_stmt: | |
1821 | lsm->is_stmt = !lsm->is_stmt; | |
1822 | break; | |
1823 | ||
1824 | case elfcpp::DW_LNS_set_basic_block: | |
1825 | lsm->basic_block = true; | |
1826 | break; | |
1827 | ||
1828 | case elfcpp::DW_LNS_fixed_advance_pc: | |
1829 | { | |
1830 | int advance_address; | |
deae2a14 | 1831 | advance_address = elfcpp::Swap_unaligned<16, big_endian>::readval(start); |
5c2c6c95 ILT |
1832 | oplen += 2; |
1833 | lsm->address += advance_address; | |
1834 | } | |
1835 | break; | |
1836 | ||
1837 | case elfcpp::DW_LNS_const_add_pc: | |
1838 | { | |
1839 | const int advance_address = (header_.min_insn_length | |
1840 | * ((255 - header_.opcode_base) | |
1841 | / header_.line_range)); | |
1842 | lsm->address += advance_address; | |
1843 | } | |
1844 | break; | |
1845 | ||
1846 | case elfcpp::DW_LNS_extended_op: | |
1847 | { | |
1848 | const uint64_t extended_op_len | |
2ea97941 | 1849 | = read_unsigned_LEB_128(start, &templen); |
5c2c6c95 ILT |
1850 | start += templen; |
1851 | oplen += templen + extended_op_len; | |
1852 | ||
1853 | const unsigned char extended_op = *start; | |
1854 | start++; | |
1855 | ||
1856 | switch (extended_op) | |
1857 | { | |
1858 | case elfcpp::DW_LNE_end_sequence: | |
124dfc89 ILT |
1859 | // This means that the current byte is the one immediately |
1860 | // after a set of instructions. Record the current line | |
1861 | // for up to one less than the current address. | |
79e052ea | 1862 | lsm->line_num = -1; |
5c2c6c95 ILT |
1863 | lsm->end_sequence = true; |
1864 | *len = oplen; | |
1865 | return true; | |
1866 | ||
1867 | case elfcpp::DW_LNE_set_address: | |
4c50553d | 1868 | { |
4dbfafcc ILT |
1869 | lsm->address = |
1870 | elfcpp::Swap_unaligned<size, big_endian>::readval(start); | |
4c50553d | 1871 | typename Reloc_map::const_iterator it |
4dbfafcc | 1872 | = this->reloc_map_.find(start - this->buffer_); |
4c50553d ILT |
1873 | if (it != reloc_map_.end()) |
1874 | { | |
4dbfafcc ILT |
1875 | // If this is a SHT_RELA section, then ignore the |
1876 | // section contents. This assumes that this is a | |
1877 | // straight reloc which just uses the reloc addend. | |
1878 | // The reloc addend has already been included in the | |
1879 | // symbol value. | |
1880 | if (this->track_relocs_type_ == elfcpp::SHT_RELA) | |
1881 | lsm->address = 0; | |
1882 | // Add in the symbol value. | |
1883 | lsm->address += it->second.second; | |
4c50553d ILT |
1884 | lsm->shndx = it->second.first; |
1885 | } | |
1886 | else | |
1887 | { | |
af674d1d ILT |
1888 | // If we're a normal .o file, with relocs, every |
1889 | // set_address should have an associated relocation. | |
1890 | if (this->input_is_relobj()) | |
1891 | this->data_valid_ = false; | |
4c50553d ILT |
1892 | } |
1893 | break; | |
24badc65 | 1894 | } |
5c2c6c95 ILT |
1895 | case elfcpp::DW_LNE_define_file: |
1896 | { | |
1897 | const char* filename = reinterpret_cast<const char*>(start); | |
1898 | templen = strlen(filename) + 1; | |
1899 | start += templen; | |
1900 | ||
1901 | uint64_t dirindex = read_unsigned_LEB_128(start, &templen); | |
5c2c6c95 | 1902 | |
af674d1d ILT |
1903 | if (dirindex >= this->directories_.back().size()) |
1904 | dirindex = 0; | |
1905 | int dirindexi = static_cast<int>(dirindex); | |
1906 | ||
e8dd54e1 CC |
1907 | // This opcode takes two additional ULEB128 parameters |
1908 | // (mod_time and filelength), but we don't use those | |
1909 | // values. Because OPLEN already tells us how far to | |
1910 | // skip to the next opcode, we don't need to read | |
1911 | // them at all. | |
5c2c6c95 | 1912 | |
af674d1d | 1913 | this->files_.back().push_back(std::make_pair(dirindexi, |
5c2c6c95 ILT |
1914 | filename)); |
1915 | } | |
1916 | break; | |
1917 | } | |
1918 | } | |
1919 | break; | |
1920 | ||
1921 | default: | |
1922 | { | |
2ea97941 | 1923 | // Ignore unknown opcode silently |
5c2c6c95 ILT |
1924 | for (int i = 0; i < header_.std_opcode_lengths[opcode]; i++) |
1925 | { | |
2ea97941 | 1926 | size_t templen; |
5c2c6c95 ILT |
1927 | read_unsigned_LEB_128(start, &templen); |
1928 | start += templen; | |
1929 | oplen += templen; | |
1930 | } | |
1931 | } | |
1932 | break; | |
2ea97941 | 1933 | } |
5c2c6c95 ILT |
1934 | *len = oplen; |
1935 | return false; | |
1936 | } | |
1937 | ||
1938 | // Read the debug information at LINEPTR and store it in the line | |
1939 | // number map. | |
1940 | ||
e43872e9 | 1941 | template<int size, bool big_endian> |
5c2c6c95 | 1942 | unsigned const char* |
9430daf8 | 1943 | Sized_dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr, |
75aea3d0 | 1944 | unsigned int shndx) |
5c2c6c95 ILT |
1945 | { |
1946 | struct LineStateMachine lsm; | |
1947 | ||
1948 | // LENGTHSTART is the place the length field is based on. It is the | |
1949 | // point in the header after the initial length field. | |
1950 | const unsigned char* lengthstart = buffer_; | |
1951 | ||
1952 | // In 64 bit dwarf, the initial length is 12 bytes, because of the | |
1953 | // 0xffffffff at the start. | |
1954 | if (header_.offset_size == 8) | |
1955 | lengthstart += 12; | |
1956 | else | |
1957 | lengthstart += 4; | |
1958 | ||
1959 | while (lineptr < lengthstart + header_.total_length) | |
1960 | { | |
1961 | ResetLineStateMachine(&lsm, header_.default_is_stmt); | |
1962 | while (!lsm.end_sequence) | |
1963 | { | |
1964 | size_t oplength; | |
e43872e9 | 1965 | bool add_line = this->process_one_opcode(lineptr, &lsm, &oplength); |
9430daf8 ILT |
1966 | if (add_line |
1967 | && (shndx == -1U || lsm.shndx == -1U || shndx == lsm.shndx)) | |
5c2c6c95 ILT |
1968 | { |
1969 | Offset_to_lineno_entry entry | |
76677ad0 CC |
1970 | = { static_cast<off_t>(lsm.address), |
1971 | this->current_header_index_, | |
1972 | static_cast<unsigned int>(lsm.file_num), | |
1973 | true, lsm.line_num }; | |
7500420b ILT |
1974 | std::vector<Offset_to_lineno_entry>& |
1975 | map(this->line_number_map_[lsm.shndx]); | |
1976 | // If we see two consecutive entries with the same | |
71ff8986 ILT |
1977 | // offset and a real line number, then mark the first |
1978 | // one as non-canonical. | |
7500420b ILT |
1979 | if (!map.empty() |
1980 | && (map.back().offset == static_cast<off_t>(lsm.address)) | |
1981 | && lsm.line_num != -1 | |
1982 | && map.back().line_num != -1) | |
71ff8986 ILT |
1983 | map.back().last_line_for_offset = false; |
1984 | map.push_back(entry); | |
5c2c6c95 ILT |
1985 | } |
1986 | lineptr += oplength; | |
1987 | } | |
1988 | } | |
1989 | ||
1990 | return lengthstart + header_.total_length; | |
1991 | } | |
1992 | ||
4c50553d ILT |
1993 | // Read the relocations into a Reloc_map. |
1994 | ||
1995 | template<int size, bool big_endian> | |
1996 | void | |
c1027032 | 1997 | Sized_dwarf_line_info<size, big_endian>::read_relocs() |
4c50553d ILT |
1998 | { |
1999 | if (this->symtab_buffer_ == NULL) | |
2000 | return; | |
2001 | ||
c1027032 | 2002 | off_t value; |
4c50553d | 2003 | off_t reloc_offset; |
c1027032 | 2004 | while ((reloc_offset = this->reloc_mapper_->next_offset()) != -1) |
4c50553d | 2005 | { |
c1027032 CC |
2006 | const unsigned int shndx = |
2007 | this->reloc_mapper_->get_reloc_target(reloc_offset, &value); | |
d491d34e ILT |
2008 | |
2009 | // There is no reason to record non-ordinary section indexes, or | |
2010 | // SHN_UNDEF, because they will never match the real section. | |
c1027032 CC |
2011 | if (shndx != 0) |
2012 | this->reloc_map_[reloc_offset] = std::make_pair(shndx, value); | |
d491d34e | 2013 | |
c1027032 | 2014 | this->reloc_mapper_->advance(reloc_offset + 1); |
4c50553d ILT |
2015 | } |
2016 | } | |
2017 | ||
2018 | // Read the line number info. | |
2019 | ||
e43872e9 | 2020 | template<int size, bool big_endian> |
5c2c6c95 | 2021 | void |
c1027032 | 2022 | Sized_dwarf_line_info<size, big_endian>::read_line_mappings(unsigned int shndx) |
5c2c6c95 | 2023 | { |
c261a0be | 2024 | gold_assert(this->data_valid_ == true); |
24badc65 | 2025 | |
c1027032 | 2026 | this->read_relocs(); |
4c50553d | 2027 | while (this->buffer_ < this->buffer_end_) |
e43872e9 | 2028 | { |
4c50553d | 2029 | const unsigned char* lineptr = this->buffer_; |
e43872e9 ILT |
2030 | lineptr = this->read_header_prolog(lineptr); |
2031 | lineptr = this->read_header_tables(lineptr); | |
9430daf8 | 2032 | lineptr = this->read_lines(lineptr, shndx); |
4c50553d | 2033 | this->buffer_ = lineptr; |
e43872e9 ILT |
2034 | } |
2035 | ||
2036 | // Sort the lines numbers, so addr2line can use binary search. | |
2037 | for (typename Lineno_map::iterator it = line_number_map_.begin(); | |
5c2c6c95 ILT |
2038 | it != line_number_map_.end(); |
2039 | ++it) | |
2040 | // Each vector needs to be sorted by offset. | |
4c50553d | 2041 | std::sort(it->second.begin(), it->second.end()); |
5c2c6c95 ILT |
2042 | } |
2043 | ||
af674d1d ILT |
2044 | // Some processing depends on whether the input is a .o file or not. |
2045 | // For instance, .o files have relocs, and have .debug_lines | |
2046 | // information on a per section basis. .so files, on the other hand, | |
2047 | // lack relocs, and offsets are unique, so we can ignore the section | |
2048 | // information. | |
2049 | ||
2050 | template<int size, bool big_endian> | |
2051 | bool | |
a55ce7fe | 2052 | Sized_dwarf_line_info<size, big_endian>::input_is_relobj() |
af674d1d ILT |
2053 | { |
2054 | // Only .o files have relocs and the symtab buffer that goes with them. | |
2055 | return this->symtab_buffer_ != NULL; | |
2056 | } | |
2057 | ||
79e052ea ILT |
2058 | // Given an Offset_to_lineno_entry vector, and an offset, figure out |
2059 | // if the offset points into a function according to the vector (see | |
2060 | // comments below for the algorithm). If it does, return an iterator | |
2061 | // into the vector that points to the line-number that contains that | |
2062 | // offset. If not, it returns vector::end(). | |
2063 | ||
2064 | static std::vector<Offset_to_lineno_entry>::const_iterator | |
2065 | offset_to_iterator(const std::vector<Offset_to_lineno_entry>* offsets, | |
2066 | off_t offset) | |
2067 | { | |
71ff8986 | 2068 | const Offset_to_lineno_entry lookup_key = { offset, 0, 0, true, 0 }; |
79e052ea ILT |
2069 | |
2070 | // lower_bound() returns the smallest offset which is >= lookup_key. | |
2071 | // If no offset in offsets is >= lookup_key, returns end(). | |
2072 | std::vector<Offset_to_lineno_entry>::const_iterator it | |
2073 | = std::lower_bound(offsets->begin(), offsets->end(), lookup_key); | |
2074 | ||
2075 | // This code is easiest to understand with a concrete example. | |
2076 | // Here's a possible offsets array: | |
71ff8986 ILT |
2077 | // {{offset = 3211, header_num = 0, file_num = 1, last, line_num = 16}, // 0 |
2078 | // {offset = 3224, header_num = 0, file_num = 1, last, line_num = 20}, // 1 | |
2079 | // {offset = 3226, header_num = 0, file_num = 1, last, line_num = 22}, // 2 | |
2080 | // {offset = 3231, header_num = 0, file_num = 1, last, line_num = 25}, // 3 | |
2081 | // {offset = 3232, header_num = 0, file_num = 1, last, line_num = -1}, // 4 | |
2082 | // {offset = 3232, header_num = 0, file_num = 1, last, line_num = 65}, // 5 | |
2083 | // {offset = 3235, header_num = 0, file_num = 1, last, line_num = 66}, // 6 | |
2084 | // {offset = 3236, header_num = 0, file_num = 1, last, line_num = -1}, // 7 | |
2085 | // {offset = 5764, header_num = 0, file_num = 1, last, line_num = 48}, // 8 | |
2086 | // {offset = 5764, header_num = 0, file_num = 1,!last, line_num = 47}, // 9 | |
2087 | // {offset = 5765, header_num = 0, file_num = 1, last, line_num = 49}, // 10 | |
2088 | // {offset = 5767, header_num = 0, file_num = 1, last, line_num = 50}, // 11 | |
2089 | // {offset = 5768, header_num = 0, file_num = 1, last, line_num = 51}, // 12 | |
2090 | // {offset = 5773, header_num = 0, file_num = 1, last, line_num = -1}, // 13 | |
2091 | // {offset = 5787, header_num = 1, file_num = 1, last, line_num = 19}, // 14 | |
2092 | // {offset = 5790, header_num = 1, file_num = 1, last, line_num = 20}, // 15 | |
2093 | // {offset = 5793, header_num = 1, file_num = 1, last, line_num = 67}, // 16 | |
2094 | // {offset = 5793, header_num = 1, file_num = 1, last, line_num = -1}, // 17 | |
2095 | // {offset = 5793, header_num = 1, file_num = 1,!last, line_num = 66}, // 18 | |
2096 | // {offset = 5795, header_num = 1, file_num = 1, last, line_num = 68}, // 19 | |
2097 | // {offset = 5798, header_num = 1, file_num = 1, last, line_num = -1}, // 20 | |
79e052ea ILT |
2098 | // The entries with line_num == -1 mark the end of a function: the |
2099 | // associated offset is one past the last instruction in the | |
2100 | // function. This can correspond to the beginning of the next | |
2101 | // function (as is true for offset 3232); alternately, there can be | |
2102 | // a gap between the end of one function and the start of the next | |
ef04e392 | 2103 | // (as is true for some others, most obviously from 3236->5764). |
79e052ea ILT |
2104 | // |
2105 | // Case 1: lookup_key has offset == 10. lower_bound returns | |
2106 | // offsets[0]. Since it's not an exact match and we're | |
ef04e392 | 2107 | // at the beginning of offsets, we return end() (invalid). |
79e052ea | 2108 | // Case 2: lookup_key has offset 10000. lower_bound returns |
71ff8986 | 2109 | // offset[21] (end()). We return end() (invalid). |
79e052ea ILT |
2110 | // Case 3: lookup_key has offset == 3211. lower_bound matches |
2111 | // offsets[0] exactly, and that's the entry we return. | |
2112 | // Case 4: lookup_key has offset == 3232. lower_bound returns | |
2113 | // offsets[4]. That's an exact match, but indicates | |
2114 | // end-of-function. We check if offsets[5] is also an | |
2115 | // exact match but not end-of-function. It is, so we | |
2116 | // return offsets[5]. | |
2117 | // Case 5: lookup_key has offset == 3214. lower_bound returns | |
2118 | // offsets[1]. Since it's not an exact match, we back | |
2119 | // up to the offset that's < lookup_key, offsets[0]. | |
2120 | // We note offsets[0] is a valid entry (not end-of-function), | |
2121 | // so that's the entry we return. | |
2122 | // Case 6: lookup_key has offset == 4000. lower_bound returns | |
2123 | // offsets[8]. Since it's not an exact match, we back | |
2124 | // up to offsets[7]. Since offsets[7] indicates | |
2125 | // end-of-function, we know lookup_key is between | |
ef04e392 | 2126 | // functions, so we return end() (not a valid offset). |
79e052ea | 2127 | // Case 7: lookup_key has offset == 5794. lower_bound returns |
71ff8986 ILT |
2128 | // offsets[19]. Since it's not an exact match, we back |
2129 | // up to offsets[16]. Note we back up to the *first* | |
2130 | // entry with offset 5793, not just offsets[19-1]. | |
2131 | // We note offsets[16] is a valid entry, so we return it. | |
2132 | // If offsets[16] had had line_num == -1, we would have | |
2133 | // checked offsets[17]. The reason for this is that | |
2134 | // 16 and 17 can be in an arbitrary order, since we sort | |
2135 | // only by offset and last_line_for_offset. (Note it | |
2136 | // doesn't help to use line_number as a tertiary sort key, | |
2137 | // since sometimes we want the -1 to be first and sometimes | |
2138 | // we want it to be last.) | |
79e052ea ILT |
2139 | |
2140 | // This deals with cases (1) and (2). | |
2141 | if ((it == offsets->begin() && offset < it->offset) | |
2142 | || it == offsets->end()) | |
2143 | return offsets->end(); | |
2144 | ||
2145 | // This deals with cases (3) and (4). | |
2146 | if (offset == it->offset) | |
2147 | { | |
2148 | while (it != offsets->end() | |
2149 | && it->offset == offset | |
2150 | && it->line_num == -1) | |
2151 | ++it; | |
2152 | if (it == offsets->end() || it->offset != offset) | |
2153 | return offsets->end(); | |
2154 | else | |
2155 | return it; | |
2156 | } | |
2157 | ||
2158 | // This handles the first part of case (7) -- we back up to the | |
2159 | // *first* entry that has the offset that's behind us. | |
2160 | gold_assert(it != offsets->begin()); | |
2161 | std::vector<Offset_to_lineno_entry>::const_iterator range_end = it; | |
2162 | --it; | |
2163 | const off_t range_value = it->offset; | |
2164 | while (it != offsets->begin() && (it-1)->offset == range_value) | |
2165 | --it; | |
2166 | ||
2167 | // This handles cases (5), (6), and (7): if any entry in the | |
2168 | // equal_range [it, range_end) has a line_num != -1, it's a valid | |
71ff8986 ILT |
2169 | // match. If not, we're not in a function. The line number we saw |
2170 | // last for an offset will be sorted first, so it'll get returned if | |
2171 | // it's present. | |
79e052ea ILT |
2172 | for (; it != range_end; ++it) |
2173 | if (it->line_num != -1) | |
2174 | return it; | |
2175 | return offsets->end(); | |
2176 | } | |
af674d1d | 2177 | |
71ff8986 ILT |
2178 | // Returns the canonical filename:lineno for the address passed in. |
2179 | // If other_lines is not NULL, appends the non-canonical lines | |
2180 | // assigned to the same address. | |
5c2c6c95 | 2181 | |
e43872e9 | 2182 | template<int size, bool big_endian> |
5c2c6c95 | 2183 | std::string |
71ff8986 ILT |
2184 | Sized_dwarf_line_info<size, big_endian>::do_addr2line( |
2185 | unsigned int shndx, | |
2186 | off_t offset, | |
2187 | std::vector<std::string>* other_lines) | |
5c2c6c95 | 2188 | { |
4c50553d ILT |
2189 | if (this->data_valid_ == false) |
2190 | return ""; | |
2191 | ||
af674d1d ILT |
2192 | const std::vector<Offset_to_lineno_entry>* offsets; |
2193 | // If we do not have reloc information, then our input is a .so or | |
2194 | // some similar data structure where all the information is held in | |
2195 | // the offset. In that case, we ignore the input shndx. | |
2196 | if (this->input_is_relobj()) | |
2197 | offsets = &this->line_number_map_[shndx]; | |
2198 | else | |
2199 | offsets = &this->line_number_map_[-1U]; | |
2200 | if (offsets->empty()) | |
4c50553d ILT |
2201 | return ""; |
2202 | ||
e43872e9 | 2203 | typename std::vector<Offset_to_lineno_entry>::const_iterator it |
79e052ea ILT |
2204 | = offset_to_iterator(offsets, offset); |
2205 | if (it == offsets->end()) | |
2206 | return ""; | |
5c2c6c95 | 2207 | |
71ff8986 ILT |
2208 | std::string result = this->format_file_lineno(*it); |
2209 | if (other_lines != NULL) | |
2210 | for (++it; it != offsets->end() && it->offset == offset; ++it) | |
2211 | { | |
2212 | if (it->line_num == -1) | |
2213 | continue; // The end of a previous function. | |
2214 | other_lines->push_back(this->format_file_lineno(*it)); | |
2215 | } | |
2216 | return result; | |
2217 | } | |
2218 | ||
2219 | // Convert the file_num + line_num into a string. | |
2220 | ||
2221 | template<int size, bool big_endian> | |
2222 | std::string | |
2223 | Sized_dwarf_line_info<size, big_endian>::format_file_lineno( | |
2224 | const Offset_to_lineno_entry& loc) const | |
2225 | { | |
5c2c6c95 | 2226 | std::string ret; |
af674d1d | 2227 | |
71ff8986 ILT |
2228 | gold_assert(loc.header_num < static_cast<int>(this->files_.size())); |
2229 | gold_assert(loc.file_num | |
c1027032 | 2230 | < static_cast<unsigned int>(this->files_[loc.header_num].size())); |
af674d1d | 2231 | const std::pair<int, std::string>& filename_pair |
71ff8986 | 2232 | = this->files_[loc.header_num][loc.file_num]; |
5c2c6c95 | 2233 | const std::string& filename = filename_pair.second; |
af674d1d | 2234 | |
71ff8986 | 2235 | gold_assert(loc.header_num < static_cast<int>(this->directories_.size())); |
af674d1d | 2236 | gold_assert(filename_pair.first |
71ff8986 | 2237 | < static_cast<int>(this->directories_[loc.header_num].size())); |
af674d1d | 2238 | const std::string& dirname |
71ff8986 | 2239 | = this->directories_[loc.header_num][filename_pair.first]; |
af674d1d | 2240 | |
5c2c6c95 ILT |
2241 | if (!dirname.empty()) |
2242 | { | |
2243 | ret += dirname; | |
2244 | ret += "/"; | |
2245 | } | |
2246 | ret += filename; | |
2247 | if (ret.empty()) | |
2248 | ret = "(unknown)"; | |
2249 | ||
2250 | char buffer[64]; // enough to hold a line number | |
71ff8986 | 2251 | snprintf(buffer, sizeof(buffer), "%d", loc.line_num); |
5c2c6c95 ILT |
2252 | ret += ":"; |
2253 | ret += buffer; | |
2254 | ||
2255 | return ret; | |
2256 | } | |
2257 | ||
a55ce7fe ILT |
2258 | // Dwarf_line_info routines. |
2259 | ||
e4e5049b CS |
2260 | static unsigned int next_generation_count = 0; |
2261 | ||
2262 | struct Addr2line_cache_entry | |
2263 | { | |
2264 | Object* object; | |
2265 | unsigned int shndx; | |
2266 | Dwarf_line_info* dwarf_line_info; | |
2267 | unsigned int generation_count; | |
2268 | unsigned int access_count; | |
2269 | ||
2270 | Addr2line_cache_entry(Object* o, unsigned int s, Dwarf_line_info* d) | |
2271 | : object(o), shndx(s), dwarf_line_info(d), | |
2272 | generation_count(next_generation_count), access_count(0) | |
2273 | { | |
2274 | if (next_generation_count < (1U << 31)) | |
2275 | ++next_generation_count; | |
2276 | } | |
2277 | }; | |
2278 | // We expect this cache to be small, so don't bother with a hashtable | |
2279 | // or priority queue or anything: just use a simple vector. | |
2280 | static std::vector<Addr2line_cache_entry> addr2line_cache; | |
2281 | ||
a55ce7fe ILT |
2282 | std::string |
2283 | Dwarf_line_info::one_addr2line(Object* object, | |
e4e5049b | 2284 | unsigned int shndx, off_t offset, |
71ff8986 ILT |
2285 | size_t cache_size, |
2286 | std::vector<std::string>* other_lines) | |
a55ce7fe | 2287 | { |
e4e5049b CS |
2288 | Dwarf_line_info* lineinfo = NULL; |
2289 | std::vector<Addr2line_cache_entry>::iterator it; | |
2290 | ||
2291 | // First, check the cache. If we hit, update the counts. | |
2292 | for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it) | |
8851ecca | 2293 | { |
e4e5049b CS |
2294 | if (it->object == object && it->shndx == shndx) |
2295 | { | |
2296 | lineinfo = it->dwarf_line_info; | |
2297 | it->generation_count = next_generation_count; | |
2298 | // We cap generation_count at 2^31 -1 to avoid overflow. | |
2299 | if (next_generation_count < (1U << 31)) | |
2300 | ++next_generation_count; | |
2301 | // We cap access_count at 31 so 2^access_count doesn't overflow | |
2302 | if (it->access_count < 31) | |
2303 | ++it->access_count; | |
2304 | break; | |
2305 | } | |
2306 | } | |
2307 | ||
2308 | // If we don't hit the cache, create a new object and insert into the | |
2309 | // cache. | |
2310 | if (lineinfo == NULL) | |
2311 | { | |
2312 | switch (parameters->size_and_endianness()) | |
2313 | { | |
a55ce7fe | 2314 | #ifdef HAVE_TARGET_32_LITTLE |
e4e5049b CS |
2315 | case Parameters::TARGET_32_LITTLE: |
2316 | lineinfo = new Sized_dwarf_line_info<32, false>(object, shndx); break; | |
a55ce7fe | 2317 | #endif |
a55ce7fe | 2318 | #ifdef HAVE_TARGET_32_BIG |
e4e5049b CS |
2319 | case Parameters::TARGET_32_BIG: |
2320 | lineinfo = new Sized_dwarf_line_info<32, true>(object, shndx); break; | |
a55ce7fe | 2321 | #endif |
a55ce7fe | 2322 | #ifdef HAVE_TARGET_64_LITTLE |
e4e5049b CS |
2323 | case Parameters::TARGET_64_LITTLE: |
2324 | lineinfo = new Sized_dwarf_line_info<64, false>(object, shndx); break; | |
a55ce7fe | 2325 | #endif |
8851ecca | 2326 | #ifdef HAVE_TARGET_64_BIG |
e4e5049b CS |
2327 | case Parameters::TARGET_64_BIG: |
2328 | lineinfo = new Sized_dwarf_line_info<64, true>(object, shndx); break; | |
a55ce7fe | 2329 | #endif |
e4e5049b CS |
2330 | default: |
2331 | gold_unreachable(); | |
2332 | } | |
2333 | addr2line_cache.push_back(Addr2line_cache_entry(object, shndx, lineinfo)); | |
2334 | } | |
2335 | ||
2336 | // Now that we have our object, figure out the answer | |
71ff8986 | 2337 | std::string retval = lineinfo->addr2line(shndx, offset, other_lines); |
e4e5049b CS |
2338 | |
2339 | // Finally, if our cache has grown too big, delete old objects. We | |
2340 | // assume the common (probably only) case is deleting only one object. | |
2341 | // We use a pretty simple scheme to evict: function of LRU and MFU. | |
2342 | while (addr2line_cache.size() > cache_size) | |
2343 | { | |
2344 | unsigned int lowest_score = ~0U; | |
2345 | std::vector<Addr2line_cache_entry>::iterator lowest | |
2346 | = addr2line_cache.end(); | |
2347 | for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it) | |
2348 | { | |
2349 | const unsigned int score = (it->generation_count | |
2350 | + (1U << it->access_count)); | |
2351 | if (score < lowest_score) | |
2352 | { | |
2353 | lowest_score = score; | |
2354 | lowest = it; | |
2355 | } | |
2356 | } | |
2357 | if (lowest != addr2line_cache.end()) | |
2358 | { | |
2359 | delete lowest->dwarf_line_info; | |
2360 | addr2line_cache.erase(lowest); | |
2361 | } | |
8851ecca | 2362 | } |
e4e5049b CS |
2363 | |
2364 | return retval; | |
2365 | } | |
2366 | ||
2367 | void | |
2368 | Dwarf_line_info::clear_addr2line_cache() | |
2369 | { | |
2370 | for (std::vector<Addr2line_cache_entry>::iterator it = addr2line_cache.begin(); | |
2371 | it != addr2line_cache.end(); | |
2372 | ++it) | |
2373 | delete it->dwarf_line_info; | |
2374 | addr2line_cache.clear(); | |
a55ce7fe ILT |
2375 | } |
2376 | ||
5c2c6c95 ILT |
2377 | #ifdef HAVE_TARGET_32_LITTLE |
2378 | template | |
a55ce7fe | 2379 | class Sized_dwarf_line_info<32, false>; |
5c2c6c95 ILT |
2380 | #endif |
2381 | ||
2382 | #ifdef HAVE_TARGET_32_BIG | |
2383 | template | |
a55ce7fe | 2384 | class Sized_dwarf_line_info<32, true>; |
5c2c6c95 ILT |
2385 | #endif |
2386 | ||
2387 | #ifdef HAVE_TARGET_64_LITTLE | |
2388 | template | |
a55ce7fe | 2389 | class Sized_dwarf_line_info<64, false>; |
5c2c6c95 ILT |
2390 | #endif |
2391 | ||
2392 | #ifdef HAVE_TARGET_64_BIG | |
2393 | template | |
a55ce7fe | 2394 | class Sized_dwarf_line_info<64, true>; |
5c2c6c95 ILT |
2395 | #endif |
2396 | ||
2397 | } // End namespace gold. |