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