]> git.ipfire.org Git - thirdparty/binutils-gdb.git/blob - gold/dwarf_reader.cc
2013-07-22 Sterling Augustine <saugustine@google.com>
[thirdparty/binutils-gdb.git] / gold / dwarf_reader.cc
1 // dwarf_reader.cc -- parse dwarf2/3 debug information
2
3 // Copyright 2007, 2008, 2009, 2010, 2011, 2012 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 <vector>
27
28 #include "elfcpp_swap.h"
29 #include "dwarf.h"
30 #include "object.h"
31 #include "reloc.h"
32 #include "dwarf_reader.h"
33 #include "int_encoding.h"
34 #include "compressed_output.h"
35
36 namespace gold {
37
38 // Class Sized_elf_reloc_mapper
39
40 // Initialize the relocation tracker for section RELOC_SHNDX.
41
42 template<int size, bool big_endian>
43 bool
44 Sized_elf_reloc_mapper<size, big_endian>::do_initialize(
45 unsigned int reloc_shndx, unsigned int reloc_type)
46 {
47 this->reloc_type_ = reloc_type;
48 return this->track_relocs_.initialize(this->object_, reloc_shndx,
49 reloc_type);
50 }
51
52 // Looks in the symtab to see what section a symbol is in.
53
54 template<int size, bool big_endian>
55 unsigned int
56 Sized_elf_reloc_mapper<size, big_endian>::symbol_section(
57 unsigned int symndx, Address* value, bool* is_ordinary)
58 {
59 const int symsize = elfcpp::Elf_sizes<size>::sym_size;
60 gold_assert(static_cast<off_t>((symndx + 1) * symsize) <= this->symtab_size_);
61 elfcpp::Sym<size, big_endian> elfsym(this->symtab_ + symndx * symsize);
62 *value = elfsym.get_st_value();
63 return this->object_->adjust_sym_shndx(symndx, elfsym.get_st_shndx(),
64 is_ordinary);
65 }
66
67 // Return the section index and offset within the section of
68 // the target of the relocation for RELOC_OFFSET.
69
70 template<int size, bool big_endian>
71 unsigned int
72 Sized_elf_reloc_mapper<size, big_endian>::do_get_reloc_target(
73 off_t reloc_offset, off_t* target_offset)
74 {
75 this->track_relocs_.advance(reloc_offset);
76 if (reloc_offset != this->track_relocs_.next_offset())
77 return 0;
78 unsigned int symndx = this->track_relocs_.next_symndx();
79 typename elfcpp::Elf_types<size>::Elf_Addr value;
80 bool is_ordinary;
81 unsigned int target_shndx = this->symbol_section(symndx, &value,
82 &is_ordinary);
83 if (!is_ordinary)
84 return 0;
85 if (this->reloc_type_ == elfcpp::SHT_RELA)
86 value += this->track_relocs_.next_addend();
87 *target_offset = value;
88 return target_shndx;
89 }
90
91 static inline Elf_reloc_mapper*
92 make_elf_reloc_mapper(Relobj* object, const unsigned char* symtab,
93 off_t symtab_size)
94 {
95 if (object->elfsize() == 32)
96 {
97 if (object->is_big_endian())
98 {
99 #ifdef HAVE_TARGET_32_BIG
100 return new Sized_elf_reloc_mapper<32, true>(object, symtab,
101 symtab_size);
102 #else
103 gold_unreachable();
104 #endif
105 }
106 else
107 {
108 #ifdef HAVE_TARGET_32_LITTLE
109 return new Sized_elf_reloc_mapper<32, false>(object, symtab,
110 symtab_size);
111 #else
112 gold_unreachable();
113 #endif
114 }
115 }
116 else if (object->elfsize() == 64)
117 {
118 if (object->is_big_endian())
119 {
120 #ifdef HAVE_TARGET_64_BIG
121 return new Sized_elf_reloc_mapper<64, true>(object, symtab,
122 symtab_size);
123 #else
124 gold_unreachable();
125 #endif
126 }
127 else
128 {
129 #ifdef HAVE_TARGET_64_LITTLE
130 return new Sized_elf_reloc_mapper<64, false>(object, symtab,
131 symtab_size);
132 #else
133 gold_unreachable();
134 #endif
135 }
136 }
137 else
138 gold_unreachable();
139 }
140
141 // class Dwarf_abbrev_table
142
143 void
144 Dwarf_abbrev_table::clear_abbrev_codes()
145 {
146 for (unsigned int code = 0; code < this->low_abbrev_code_max_; ++code)
147 {
148 if (this->low_abbrev_codes_[code] != NULL)
149 {
150 delete this->low_abbrev_codes_[code];
151 this->low_abbrev_codes_[code] = NULL;
152 }
153 }
154 for (Abbrev_code_table::iterator it = this->high_abbrev_codes_.begin();
155 it != this->high_abbrev_codes_.end();
156 ++it)
157 {
158 if (it->second != NULL)
159 delete it->second;
160 }
161 this->high_abbrev_codes_.clear();
162 }
163
164 // Read the abbrev table from an object file.
165
166 bool
167 Dwarf_abbrev_table::do_read_abbrevs(
168 Relobj* object,
169 unsigned int abbrev_shndx,
170 off_t abbrev_offset)
171 {
172 this->clear_abbrev_codes();
173
174 // If we don't have relocations, abbrev_shndx will be 0, and
175 // we'll have to hunt for the .debug_abbrev section.
176 if (abbrev_shndx == 0 && this->abbrev_shndx_ > 0)
177 abbrev_shndx = this->abbrev_shndx_;
178 else if (abbrev_shndx == 0)
179 {
180 for (unsigned int i = 1; i < object->shnum(); ++i)
181 {
182 std::string name = object->section_name(i);
183 if (name == ".debug_abbrev")
184 {
185 abbrev_shndx = i;
186 // Correct the offset. For incremental update links, we have a
187 // relocated offset that is relative to the output section, but
188 // here we need an offset relative to the input section.
189 abbrev_offset -= object->output_section_offset(i);
190 break;
191 }
192 }
193 if (abbrev_shndx == 0)
194 return false;
195 }
196
197 // Get the section contents and decompress if necessary.
198 if (abbrev_shndx != this->abbrev_shndx_)
199 {
200 if (this->owns_buffer_ && this->buffer_ != NULL)
201 {
202 delete[] this->buffer_;
203 this->owns_buffer_ = false;
204 }
205
206 section_size_type buffer_size;
207 this->buffer_ =
208 object->decompressed_section_contents(abbrev_shndx,
209 &buffer_size,
210 &this->owns_buffer_);
211 this->buffer_end_ = this->buffer_ + buffer_size;
212 this->abbrev_shndx_ = abbrev_shndx;
213 }
214
215 this->buffer_pos_ = this->buffer_ + abbrev_offset;
216 return true;
217 }
218
219 // Lookup the abbrev code entry for CODE. This function is called
220 // only when the abbrev code is not in the direct lookup table.
221 // It may be in the hash table, it may not have been read yet,
222 // or it may not exist in the abbrev table.
223
224 const Dwarf_abbrev_table::Abbrev_code*
225 Dwarf_abbrev_table::do_get_abbrev(unsigned int code)
226 {
227 // See if the abbrev code is already in the hash table.
228 Abbrev_code_table::const_iterator it = this->high_abbrev_codes_.find(code);
229 if (it != this->high_abbrev_codes_.end())
230 return it->second;
231
232 // Read and store abbrev code definitions until we find the
233 // one we're looking for.
234 for (;;)
235 {
236 // Read the abbrev code. A zero here indicates the end of the
237 // abbrev table.
238 size_t len;
239 if (this->buffer_pos_ >= this->buffer_end_)
240 return NULL;
241 uint64_t nextcode = read_unsigned_LEB_128(this->buffer_pos_, &len);
242 if (nextcode == 0)
243 {
244 this->buffer_pos_ = this->buffer_end_;
245 return NULL;
246 }
247 this->buffer_pos_ += len;
248
249 // Read the tag.
250 if (this->buffer_pos_ >= this->buffer_end_)
251 return NULL;
252 uint64_t tag = read_unsigned_LEB_128(this->buffer_pos_, &len);
253 this->buffer_pos_ += len;
254
255 // Read the has_children flag.
256 if (this->buffer_pos_ >= this->buffer_end_)
257 return NULL;
258 bool has_children = *this->buffer_pos_ == elfcpp::DW_CHILDREN_yes;
259 this->buffer_pos_ += 1;
260
261 // Read the list of (attribute, form) pairs.
262 Abbrev_code* entry = new Abbrev_code(tag, has_children);
263 for (;;)
264 {
265 // Read the attribute.
266 if (this->buffer_pos_ >= this->buffer_end_)
267 return NULL;
268 uint64_t attr = read_unsigned_LEB_128(this->buffer_pos_, &len);
269 this->buffer_pos_ += len;
270
271 // Read the form.
272 if (this->buffer_pos_ >= this->buffer_end_)
273 return NULL;
274 uint64_t form = read_unsigned_LEB_128(this->buffer_pos_, &len);
275 this->buffer_pos_ += len;
276
277 // A (0,0) pair terminates the list.
278 if (attr == 0 && form == 0)
279 break;
280
281 if (attr == elfcpp::DW_AT_sibling)
282 entry->has_sibling_attribute = true;
283
284 entry->add_attribute(attr, form);
285 }
286
287 this->store_abbrev(nextcode, entry);
288 if (nextcode == code)
289 return entry;
290 }
291
292 return NULL;
293 }
294
295 // class Dwarf_ranges_table
296
297 // Read the ranges table from an object file.
298
299 bool
300 Dwarf_ranges_table::read_ranges_table(
301 Relobj* object,
302 const unsigned char* symtab,
303 off_t symtab_size,
304 unsigned int ranges_shndx)
305 {
306 // If we've already read this abbrev table, return immediately.
307 if (this->ranges_shndx_ > 0
308 && this->ranges_shndx_ == ranges_shndx)
309 return true;
310
311 // If we don't have relocations, ranges_shndx will be 0, and
312 // we'll have to hunt for the .debug_ranges section.
313 if (ranges_shndx == 0 && this->ranges_shndx_ > 0)
314 ranges_shndx = this->ranges_shndx_;
315 else if (ranges_shndx == 0)
316 {
317 for (unsigned int i = 1; i < object->shnum(); ++i)
318 {
319 std::string name = object->section_name(i);
320 if (name == ".debug_ranges")
321 {
322 ranges_shndx = i;
323 this->output_section_offset_ = object->output_section_offset(i);
324 break;
325 }
326 }
327 if (ranges_shndx == 0)
328 return false;
329 }
330
331 // Get the section contents and decompress if necessary.
332 if (ranges_shndx != this->ranges_shndx_)
333 {
334 if (this->owns_ranges_buffer_ && this->ranges_buffer_ != NULL)
335 {
336 delete[] this->ranges_buffer_;
337 this->owns_ranges_buffer_ = false;
338 }
339
340 section_size_type buffer_size;
341 this->ranges_buffer_ =
342 object->decompressed_section_contents(ranges_shndx,
343 &buffer_size,
344 &this->owns_ranges_buffer_);
345 this->ranges_buffer_end_ = this->ranges_buffer_ + buffer_size;
346 this->ranges_shndx_ = ranges_shndx;
347 }
348
349 if (this->ranges_reloc_mapper_ != NULL)
350 {
351 delete this->ranges_reloc_mapper_;
352 this->ranges_reloc_mapper_ = NULL;
353 }
354
355 // For incremental objects, we have no relocations.
356 if (object->is_incremental())
357 return true;
358
359 // Find the relocation section for ".debug_ranges".
360 unsigned int reloc_shndx = 0;
361 unsigned int reloc_type = 0;
362 for (unsigned int i = 0; i < object->shnum(); ++i)
363 {
364 reloc_type = object->section_type(i);
365 if ((reloc_type == elfcpp::SHT_REL
366 || reloc_type == elfcpp::SHT_RELA)
367 && object->section_info(i) == ranges_shndx)
368 {
369 reloc_shndx = i;
370 break;
371 }
372 }
373
374 this->ranges_reloc_mapper_ = make_elf_reloc_mapper(object, symtab,
375 symtab_size);
376 this->ranges_reloc_mapper_->initialize(reloc_shndx, reloc_type);
377 this->reloc_type_ = reloc_type;
378
379 return true;
380 }
381
382 // Read a range list from section RANGES_SHNDX at offset RANGES_OFFSET.
383
384 Dwarf_range_list*
385 Dwarf_ranges_table::read_range_list(
386 Relobj* object,
387 const unsigned char* symtab,
388 off_t symtab_size,
389 unsigned int addr_size,
390 unsigned int ranges_shndx,
391 off_t offset)
392 {
393 Dwarf_range_list* ranges;
394
395 if (!this->read_ranges_table(object, symtab, symtab_size, ranges_shndx))
396 return NULL;
397
398 // Correct the offset. For incremental update links, we have a
399 // relocated offset that is relative to the output section, but
400 // here we need an offset relative to the input section.
401 offset -= this->output_section_offset_;
402
403 // Read the range list at OFFSET.
404 ranges = new Dwarf_range_list();
405 off_t base = 0;
406 for (;
407 this->ranges_buffer_ + offset < this->ranges_buffer_end_;
408 offset += 2 * addr_size)
409 {
410 off_t start;
411 off_t end;
412
413 // Read the raw contents of the section.
414 if (addr_size == 4)
415 {
416 start = this->dwinfo_->read_from_pointer<32>(this->ranges_buffer_
417 + offset);
418 end = this->dwinfo_->read_from_pointer<32>(this->ranges_buffer_
419 + offset + 4);
420 }
421 else
422 {
423 start = this->dwinfo_->read_from_pointer<64>(this->ranges_buffer_
424 + offset);
425 end = this->dwinfo_->read_from_pointer<64>(this->ranges_buffer_
426 + offset + 8);
427 }
428
429 // Check for relocations and adjust the values.
430 unsigned int shndx1 = 0;
431 unsigned int shndx2 = 0;
432 if (this->ranges_reloc_mapper_ != NULL)
433 {
434 shndx1 = this->lookup_reloc(offset, &start);
435 shndx2 = this->lookup_reloc(offset + addr_size, &end);
436 }
437
438 // End of list is marked by a pair of zeroes.
439 if (shndx1 == 0 && start == 0 && end == 0)
440 break;
441
442 // A "base address selection entry" is identified by
443 // 0xffffffff for the first value of the pair. The second
444 // value is used as a base for subsequent range list entries.
445 if (shndx1 == 0 && start == -1)
446 base = end;
447 else if (shndx1 == shndx2)
448 {
449 if (shndx1 == 0 || object->is_section_included(shndx1))
450 ranges->add(shndx1, base + start, base + end);
451 }
452 else
453 gold_warning(_("%s: DWARF info may be corrupt; offsets in a "
454 "range list entry are in different sections"),
455 object->name().c_str());
456 }
457
458 return ranges;
459 }
460
461 // Look for a relocation at offset OFF in the range table,
462 // and return the section index and offset of the target.
463
464 unsigned int
465 Dwarf_ranges_table::lookup_reloc(off_t off, off_t* target_off)
466 {
467 off_t value;
468 unsigned int shndx =
469 this->ranges_reloc_mapper_->get_reloc_target(off, &value);
470 if (shndx == 0)
471 return 0;
472 if (this->reloc_type_ == elfcpp::SHT_REL)
473 *target_off += value;
474 else
475 *target_off = value;
476 return shndx;
477 }
478
479 // class Dwarf_pubnames_table
480
481 // Read the pubnames section from the object file.
482
483 bool
484 Dwarf_pubnames_table::read_section(Relobj* object, const unsigned char* symtab,
485 off_t symtab_size)
486 {
487 section_size_type buffer_size;
488 unsigned int shndx = 0;
489
490 // Find the .debug_pubnames/pubtypes section.
491 const char* name = (this->is_pubtypes_
492 ? ".debug_pubtypes"
493 : ".debug_pubnames");
494 for (unsigned int i = 1; i < object->shnum(); ++i)
495 {
496 if (object->section_name(i) == name)
497 {
498 shndx = i;
499 this->output_section_offset_ = object->output_section_offset(i);
500 break;
501 }
502 }
503 if (shndx == 0)
504 return false;
505
506
507 this->buffer_ = object->decompressed_section_contents(shndx,
508 &buffer_size,
509 &this->owns_buffer_);
510 if (this->buffer_ == NULL)
511 return false;
512 this->buffer_end_ = this->buffer_ + buffer_size;
513
514 // For incremental objects, we have no relocations.
515 if (object->is_incremental())
516 return true;
517
518 // Find the relocation section
519 unsigned int reloc_shndx = 0;
520 unsigned int reloc_type = 0;
521 for (unsigned int i = 0; i < object->shnum(); ++i)
522 {
523 reloc_type = object->section_type(i);
524 if ((reloc_type == elfcpp::SHT_REL
525 || reloc_type == elfcpp::SHT_RELA)
526 && object->section_info(i) == shndx)
527 {
528 reloc_shndx = i;
529 break;
530 }
531 }
532
533 this->reloc_mapper_ = make_elf_reloc_mapper(object, symtab, symtab_size);
534 this->reloc_mapper_->initialize(reloc_shndx, reloc_type);
535 this->reloc_type_ = reloc_type;
536
537 return true;
538 }
539
540 // Read the header for the set at OFFSET.
541
542 bool
543 Dwarf_pubnames_table::read_header(off_t offset)
544 {
545 // Make sure we have actually read the section.
546 gold_assert(this->buffer_ != NULL);
547
548 // Correct the offset. For incremental update links, we have a
549 // relocated offset that is relative to the output section, but
550 // here we need an offset relative to the input section.
551 offset -= this->output_section_offset_;
552
553 if (offset < 0 || offset + 14 >= this->buffer_end_ - this->buffer_)
554 return false;
555
556 const unsigned char* pinfo = this->buffer_ + offset;
557
558 // Read the unit_length field.
559 uint64_t unit_length = this->dwinfo_->read_from_pointer<32>(pinfo);
560 pinfo += 4;
561 if (unit_length == 0xffffffff)
562 {
563 unit_length = this->dwinfo_->read_from_pointer<64>(pinfo);
564 this->unit_length_ = unit_length + 12;
565 pinfo += 8;
566 this->offset_size_ = 8;
567 }
568 else
569 {
570 this->unit_length_ = unit_length + 4;
571 this->offset_size_ = 4;
572 }
573
574 // Check the version.
575 unsigned int version = this->dwinfo_->read_from_pointer<16>(pinfo);
576 pinfo += 2;
577 if (version != 2)
578 return false;
579
580 this->reloc_mapper_->get_reloc_target(pinfo - this->buffer_,
581 &this->cu_offset_);
582
583 // Skip the debug_info_offset and debug_info_size fields.
584 pinfo += 2 * this->offset_size_;
585
586 if (pinfo >= this->buffer_end_)
587 return false;
588
589 this->pinfo_ = pinfo;
590 return true;
591 }
592
593 // Read the next name from the set.
594
595 const char*
596 Dwarf_pubnames_table::next_name()
597 {
598 const unsigned char* pinfo = this->pinfo_;
599
600 // Read the offset within the CU. If this is zero, we have reached
601 // the end of the list.
602 uint32_t offset;
603 if (this->offset_size_ == 4)
604 offset = this->dwinfo_->read_from_pointer<32>(&pinfo);
605 else
606 offset = this->dwinfo_->read_from_pointer<64>(&pinfo);
607 if (offset == 0)
608 return NULL;
609
610 // Return a pointer to the string at the current location,
611 // and advance the pointer to the next entry.
612 const char* ret = reinterpret_cast<const char*>(pinfo);
613 while (pinfo < this->buffer_end_ && *pinfo != '\0')
614 ++pinfo;
615 if (pinfo < this->buffer_end_)
616 ++pinfo;
617
618 this->pinfo_ = pinfo;
619 return ret;
620 }
621
622 // class Dwarf_die
623
624 Dwarf_die::Dwarf_die(
625 Dwarf_info_reader* dwinfo,
626 off_t die_offset,
627 Dwarf_die* parent)
628 : dwinfo_(dwinfo), parent_(parent), die_offset_(die_offset),
629 child_offset_(0), sibling_offset_(0), abbrev_code_(NULL), attributes_(),
630 attributes_read_(false), name_(NULL), name_off_(-1), linkage_name_(NULL),
631 linkage_name_off_(-1), string_shndx_(0), specification_(0),
632 abstract_origin_(0)
633 {
634 size_t len;
635 const unsigned char* pdie = dwinfo->buffer_at_offset(die_offset);
636 if (pdie == NULL)
637 return;
638 unsigned int code = read_unsigned_LEB_128(pdie, &len);
639 if (code == 0)
640 {
641 if (parent != NULL)
642 parent->set_sibling_offset(die_offset + len);
643 return;
644 }
645 this->attr_offset_ = len;
646
647 // Lookup the abbrev code in the abbrev table.
648 this->abbrev_code_ = dwinfo->get_abbrev(code);
649 }
650
651 // Read all the attributes of the DIE.
652
653 bool
654 Dwarf_die::read_attributes()
655 {
656 if (this->attributes_read_)
657 return true;
658
659 gold_assert(this->abbrev_code_ != NULL);
660
661 const unsigned char* pdie =
662 this->dwinfo_->buffer_at_offset(this->die_offset_);
663 if (pdie == NULL)
664 return false;
665 const unsigned char* pattr = pdie + this->attr_offset_;
666
667 unsigned int nattr = this->abbrev_code_->attributes.size();
668 this->attributes_.reserve(nattr);
669 for (unsigned int i = 0; i < nattr; ++i)
670 {
671 size_t len;
672 unsigned int attr = this->abbrev_code_->attributes[i].attr;
673 unsigned int form = this->abbrev_code_->attributes[i].form;
674 if (form == elfcpp::DW_FORM_indirect)
675 {
676 form = read_unsigned_LEB_128(pattr, &len);
677 pattr += len;
678 }
679 off_t attr_off = this->die_offset_ + (pattr - pdie);
680 bool ref_form = false;
681 Attribute_value attr_value;
682 attr_value.attr = attr;
683 attr_value.form = form;
684 attr_value.aux.shndx = 0;
685 switch(form)
686 {
687 case elfcpp::DW_FORM_flag_present:
688 attr_value.val.intval = 1;
689 break;
690 case elfcpp::DW_FORM_strp:
691 {
692 off_t str_off;
693 if (this->dwinfo_->offset_size() == 4)
694 str_off = this->dwinfo_->read_from_pointer<32>(&pattr);
695 else
696 str_off = this->dwinfo_->read_from_pointer<64>(&pattr);
697 unsigned int shndx =
698 this->dwinfo_->lookup_reloc(attr_off, &str_off);
699 attr_value.aux.shndx = shndx;
700 attr_value.val.refval = str_off;
701 break;
702 }
703 case elfcpp::DW_FORM_sec_offset:
704 {
705 off_t sec_off;
706 if (this->dwinfo_->offset_size() == 4)
707 sec_off = this->dwinfo_->read_from_pointer<32>(&pattr);
708 else
709 sec_off = this->dwinfo_->read_from_pointer<64>(&pattr);
710 unsigned int shndx =
711 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
712 attr_value.aux.shndx = shndx;
713 attr_value.val.refval = sec_off;
714 ref_form = true;
715 break;
716 }
717 case elfcpp::DW_FORM_addr:
718 case elfcpp::DW_FORM_ref_addr:
719 {
720 off_t sec_off;
721 if (this->dwinfo_->address_size() == 4)
722 sec_off = this->dwinfo_->read_from_pointer<32>(&pattr);
723 else
724 sec_off = this->dwinfo_->read_from_pointer<64>(&pattr);
725 unsigned int shndx =
726 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
727 attr_value.aux.shndx = shndx;
728 attr_value.val.refval = sec_off;
729 ref_form = true;
730 break;
731 }
732 case elfcpp::DW_FORM_block1:
733 attr_value.aux.blocklen = *pattr++;
734 attr_value.val.blockval = pattr;
735 pattr += attr_value.aux.blocklen;
736 break;
737 case elfcpp::DW_FORM_block2:
738 attr_value.aux.blocklen =
739 this->dwinfo_->read_from_pointer<16>(&pattr);
740 attr_value.val.blockval = pattr;
741 pattr += attr_value.aux.blocklen;
742 break;
743 case elfcpp::DW_FORM_block4:
744 attr_value.aux.blocklen =
745 this->dwinfo_->read_from_pointer<32>(&pattr);
746 attr_value.val.blockval = pattr;
747 pattr += attr_value.aux.blocklen;
748 break;
749 case elfcpp::DW_FORM_block:
750 case elfcpp::DW_FORM_exprloc:
751 attr_value.aux.blocklen = read_unsigned_LEB_128(pattr, &len);
752 attr_value.val.blockval = pattr + len;
753 pattr += len + attr_value.aux.blocklen;
754 break;
755 case elfcpp::DW_FORM_data1:
756 case elfcpp::DW_FORM_flag:
757 attr_value.val.intval = *pattr++;
758 break;
759 case elfcpp::DW_FORM_ref1:
760 attr_value.val.refval = *pattr++;
761 ref_form = true;
762 break;
763 case elfcpp::DW_FORM_data2:
764 attr_value.val.intval =
765 this->dwinfo_->read_from_pointer<16>(&pattr);
766 break;
767 case elfcpp::DW_FORM_ref2:
768 attr_value.val.refval =
769 this->dwinfo_->read_from_pointer<16>(&pattr);
770 ref_form = true;
771 break;
772 case elfcpp::DW_FORM_data4:
773 {
774 off_t sec_off;
775 sec_off = this->dwinfo_->read_from_pointer<32>(&pattr);
776 unsigned int shndx =
777 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
778 attr_value.aux.shndx = shndx;
779 attr_value.val.intval = sec_off;
780 break;
781 }
782 case elfcpp::DW_FORM_ref4:
783 {
784 off_t sec_off;
785 sec_off = this->dwinfo_->read_from_pointer<32>(&pattr);
786 unsigned int shndx =
787 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
788 attr_value.aux.shndx = shndx;
789 attr_value.val.refval = sec_off;
790 ref_form = true;
791 break;
792 }
793 case elfcpp::DW_FORM_data8:
794 {
795 off_t sec_off;
796 sec_off = this->dwinfo_->read_from_pointer<64>(&pattr);
797 unsigned int shndx =
798 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
799 attr_value.aux.shndx = shndx;
800 attr_value.val.intval = sec_off;
801 break;
802 }
803 case elfcpp::DW_FORM_ref_sig8:
804 attr_value.val.uintval =
805 this->dwinfo_->read_from_pointer<64>(&pattr);
806 break;
807 case elfcpp::DW_FORM_ref8:
808 {
809 off_t sec_off;
810 sec_off = this->dwinfo_->read_from_pointer<64>(&pattr);
811 unsigned int shndx =
812 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
813 attr_value.aux.shndx = shndx;
814 attr_value.val.refval = sec_off;
815 ref_form = true;
816 break;
817 }
818 case elfcpp::DW_FORM_ref_udata:
819 attr_value.val.refval = read_unsigned_LEB_128(pattr, &len);
820 ref_form = true;
821 pattr += len;
822 break;
823 case elfcpp::DW_FORM_udata:
824 case elfcpp::DW_FORM_GNU_addr_index:
825 case elfcpp::DW_FORM_GNU_str_index:
826 attr_value.val.uintval = read_unsigned_LEB_128(pattr, &len);
827 pattr += len;
828 break;
829 case elfcpp::DW_FORM_sdata:
830 attr_value.val.intval = read_signed_LEB_128(pattr, &len);
831 pattr += len;
832 break;
833 case elfcpp::DW_FORM_string:
834 attr_value.val.stringval = reinterpret_cast<const char*>(pattr);
835 len = strlen(attr_value.val.stringval);
836 pattr += len + 1;
837 break;
838 default:
839 return false;
840 }
841
842 // Cache the most frequently-requested attributes.
843 switch (attr)
844 {
845 case elfcpp::DW_AT_name:
846 if (form == elfcpp::DW_FORM_string)
847 this->name_ = attr_value.val.stringval;
848 else if (form == elfcpp::DW_FORM_strp)
849 {
850 // All indirect strings should refer to the same
851 // string section, so we just save the last one seen.
852 this->string_shndx_ = attr_value.aux.shndx;
853 this->name_off_ = attr_value.val.refval;
854 }
855 break;
856 case elfcpp::DW_AT_linkage_name:
857 case elfcpp::DW_AT_MIPS_linkage_name:
858 if (form == elfcpp::DW_FORM_string)
859 this->linkage_name_ = attr_value.val.stringval;
860 else if (form == elfcpp::DW_FORM_strp)
861 {
862 // All indirect strings should refer to the same
863 // string section, so we just save the last one seen.
864 this->string_shndx_ = attr_value.aux.shndx;
865 this->linkage_name_off_ = attr_value.val.refval;
866 }
867 break;
868 case elfcpp::DW_AT_specification:
869 if (ref_form)
870 this->specification_ = attr_value.val.refval;
871 break;
872 case elfcpp::DW_AT_abstract_origin:
873 if (ref_form)
874 this->abstract_origin_ = attr_value.val.refval;
875 break;
876 case elfcpp::DW_AT_sibling:
877 if (ref_form && attr_value.aux.shndx == 0)
878 this->sibling_offset_ = attr_value.val.refval;
879 default:
880 break;
881 }
882
883 this->attributes_.push_back(attr_value);
884 }
885
886 // Now that we know where the next DIE begins, record the offset
887 // to avoid later recalculation.
888 if (this->has_children())
889 this->child_offset_ = this->die_offset_ + (pattr - pdie);
890 else
891 this->sibling_offset_ = this->die_offset_ + (pattr - pdie);
892
893 this->attributes_read_ = true;
894 return true;
895 }
896
897 // Skip all the attributes of the DIE and return the offset of the next DIE.
898
899 off_t
900 Dwarf_die::skip_attributes()
901 {
902 gold_assert(this->abbrev_code_ != NULL);
903
904 const unsigned char* pdie =
905 this->dwinfo_->buffer_at_offset(this->die_offset_);
906 if (pdie == NULL)
907 return 0;
908 const unsigned char* pattr = pdie + this->attr_offset_;
909
910 for (unsigned int i = 0; i < this->abbrev_code_->attributes.size(); ++i)
911 {
912 size_t len;
913 unsigned int form = this->abbrev_code_->attributes[i].form;
914 if (form == elfcpp::DW_FORM_indirect)
915 {
916 form = read_unsigned_LEB_128(pattr, &len);
917 pattr += len;
918 }
919 switch(form)
920 {
921 case elfcpp::DW_FORM_flag_present:
922 break;
923 case elfcpp::DW_FORM_strp:
924 case elfcpp::DW_FORM_sec_offset:
925 pattr += this->dwinfo_->offset_size();
926 break;
927 case elfcpp::DW_FORM_addr:
928 case elfcpp::DW_FORM_ref_addr:
929 pattr += this->dwinfo_->address_size();
930 break;
931 case elfcpp::DW_FORM_block1:
932 pattr += 1 + *pattr;
933 break;
934 case elfcpp::DW_FORM_block2:
935 {
936 uint16_t block_size;
937 block_size = this->dwinfo_->read_from_pointer<16>(&pattr);
938 pattr += block_size;
939 break;
940 }
941 case elfcpp::DW_FORM_block4:
942 {
943 uint32_t block_size;
944 block_size = this->dwinfo_->read_from_pointer<32>(&pattr);
945 pattr += block_size;
946 break;
947 }
948 case elfcpp::DW_FORM_block:
949 case elfcpp::DW_FORM_exprloc:
950 {
951 uint64_t block_size;
952 block_size = read_unsigned_LEB_128(pattr, &len);
953 pattr += len + block_size;
954 break;
955 }
956 case elfcpp::DW_FORM_data1:
957 case elfcpp::DW_FORM_ref1:
958 case elfcpp::DW_FORM_flag:
959 pattr += 1;
960 break;
961 case elfcpp::DW_FORM_data2:
962 case elfcpp::DW_FORM_ref2:
963 pattr += 2;
964 break;
965 case elfcpp::DW_FORM_data4:
966 case elfcpp::DW_FORM_ref4:
967 pattr += 4;
968 break;
969 case elfcpp::DW_FORM_data8:
970 case elfcpp::DW_FORM_ref8:
971 case elfcpp::DW_FORM_ref_sig8:
972 pattr += 8;
973 break;
974 case elfcpp::DW_FORM_ref_udata:
975 case elfcpp::DW_FORM_udata:
976 case elfcpp::DW_FORM_GNU_addr_index:
977 case elfcpp::DW_FORM_GNU_str_index:
978 read_unsigned_LEB_128(pattr, &len);
979 pattr += len;
980 break;
981 case elfcpp::DW_FORM_sdata:
982 read_signed_LEB_128(pattr, &len);
983 pattr += len;
984 break;
985 case elfcpp::DW_FORM_string:
986 len = strlen(reinterpret_cast<const char*>(pattr));
987 pattr += len + 1;
988 break;
989 default:
990 return 0;
991 }
992 }
993
994 return this->die_offset_ + (pattr - pdie);
995 }
996
997 // Get the name of the DIE and cache it.
998
999 void
1000 Dwarf_die::set_name()
1001 {
1002 if (this->name_ != NULL || !this->read_attributes())
1003 return;
1004 if (this->name_off_ != -1)
1005 this->name_ = this->dwinfo_->get_string(this->name_off_,
1006 this->string_shndx_);
1007 }
1008
1009 // Get the linkage name of the DIE and cache it.
1010
1011 void
1012 Dwarf_die::set_linkage_name()
1013 {
1014 if (this->linkage_name_ != NULL || !this->read_attributes())
1015 return;
1016 if (this->linkage_name_off_ != -1)
1017 this->linkage_name_ = this->dwinfo_->get_string(this->linkage_name_off_,
1018 this->string_shndx_);
1019 }
1020
1021 // Return the value of attribute ATTR.
1022
1023 const Dwarf_die::Attribute_value*
1024 Dwarf_die::attribute(unsigned int attr)
1025 {
1026 if (!this->read_attributes())
1027 return NULL;
1028 for (unsigned int i = 0; i < this->attributes_.size(); ++i)
1029 {
1030 if (this->attributes_[i].attr == attr)
1031 return &this->attributes_[i];
1032 }
1033 return NULL;
1034 }
1035
1036 const char*
1037 Dwarf_die::string_attribute(unsigned int attr)
1038 {
1039 const Attribute_value* attr_val = this->attribute(attr);
1040 if (attr_val == NULL)
1041 return NULL;
1042 switch (attr_val->form)
1043 {
1044 case elfcpp::DW_FORM_string:
1045 return attr_val->val.stringval;
1046 case elfcpp::DW_FORM_strp:
1047 return this->dwinfo_->get_string(attr_val->val.refval,
1048 attr_val->aux.shndx);
1049 default:
1050 return NULL;
1051 }
1052 }
1053
1054 int64_t
1055 Dwarf_die::int_attribute(unsigned int attr)
1056 {
1057 const Attribute_value* attr_val = this->attribute(attr);
1058 if (attr_val == NULL)
1059 return 0;
1060 switch (attr_val->form)
1061 {
1062 case elfcpp::DW_FORM_flag_present:
1063 case elfcpp::DW_FORM_data1:
1064 case elfcpp::DW_FORM_flag:
1065 case elfcpp::DW_FORM_data2:
1066 case elfcpp::DW_FORM_data4:
1067 case elfcpp::DW_FORM_data8:
1068 case elfcpp::DW_FORM_sdata:
1069 return attr_val->val.intval;
1070 default:
1071 return 0;
1072 }
1073 }
1074
1075 uint64_t
1076 Dwarf_die::uint_attribute(unsigned int attr)
1077 {
1078 const Attribute_value* attr_val = this->attribute(attr);
1079 if (attr_val == NULL)
1080 return 0;
1081 switch (attr_val->form)
1082 {
1083 case elfcpp::DW_FORM_flag_present:
1084 case elfcpp::DW_FORM_data1:
1085 case elfcpp::DW_FORM_flag:
1086 case elfcpp::DW_FORM_data4:
1087 case elfcpp::DW_FORM_data8:
1088 case elfcpp::DW_FORM_ref_sig8:
1089 case elfcpp::DW_FORM_udata:
1090 return attr_val->val.uintval;
1091 default:
1092 return 0;
1093 }
1094 }
1095
1096 off_t
1097 Dwarf_die::ref_attribute(unsigned int attr, unsigned int* shndx)
1098 {
1099 const Attribute_value* attr_val = this->attribute(attr);
1100 if (attr_val == NULL)
1101 return -1;
1102 switch (attr_val->form)
1103 {
1104 case elfcpp::DW_FORM_sec_offset:
1105 case elfcpp::DW_FORM_addr:
1106 case elfcpp::DW_FORM_ref_addr:
1107 case elfcpp::DW_FORM_ref1:
1108 case elfcpp::DW_FORM_ref2:
1109 case elfcpp::DW_FORM_ref4:
1110 case elfcpp::DW_FORM_ref8:
1111 case elfcpp::DW_FORM_ref_udata:
1112 *shndx = attr_val->aux.shndx;
1113 return attr_val->val.refval;
1114 case elfcpp::DW_FORM_ref_sig8:
1115 *shndx = attr_val->aux.shndx;
1116 return attr_val->val.uintval;
1117 case elfcpp::DW_FORM_data4:
1118 case elfcpp::DW_FORM_data8:
1119 *shndx = attr_val->aux.shndx;
1120 return attr_val->val.intval;
1121 default:
1122 return -1;
1123 }
1124 }
1125
1126 off_t
1127 Dwarf_die::address_attribute(unsigned int attr, unsigned int* shndx)
1128 {
1129 const Attribute_value* attr_val = this->attribute(attr);
1130 if (attr_val == NULL || attr_val->form != elfcpp::DW_FORM_addr)
1131 return -1;
1132
1133 *shndx = attr_val->aux.shndx;
1134 return attr_val->val.refval;
1135 }
1136
1137 // Return the offset of this DIE's first child.
1138
1139 off_t
1140 Dwarf_die::child_offset()
1141 {
1142 gold_assert(this->abbrev_code_ != NULL);
1143 if (!this->has_children())
1144 return 0;
1145 if (this->child_offset_ == 0)
1146 this->child_offset_ = this->skip_attributes();
1147 return this->child_offset_;
1148 }
1149
1150 // Return the offset of this DIE's next sibling.
1151
1152 off_t
1153 Dwarf_die::sibling_offset()
1154 {
1155 gold_assert(this->abbrev_code_ != NULL);
1156
1157 if (this->sibling_offset_ != 0)
1158 return this->sibling_offset_;
1159
1160 if (!this->has_children())
1161 {
1162 this->sibling_offset_ = this->skip_attributes();
1163 return this->sibling_offset_;
1164 }
1165
1166 if (this->has_sibling_attribute())
1167 {
1168 if (!this->read_attributes())
1169 return 0;
1170 if (this->sibling_offset_ != 0)
1171 return this->sibling_offset_;
1172 }
1173
1174 // Skip over the children.
1175 off_t child_offset = this->child_offset();
1176 while (child_offset > 0)
1177 {
1178 Dwarf_die die(this->dwinfo_, child_offset, this);
1179 // The Dwarf_die ctor will set this DIE's sibling offset
1180 // when it reads a zero abbrev code.
1181 if (die.tag() == 0)
1182 break;
1183 child_offset = die.sibling_offset();
1184 }
1185
1186 // This should be set by now. If not, there was a problem reading
1187 // the DWARF info, and we return 0.
1188 return this->sibling_offset_;
1189 }
1190
1191 // class Dwarf_info_reader
1192
1193 // Begin parsing the debug info. This calls visit_compilation_unit()
1194 // or visit_type_unit() for each compilation or type unit found in the
1195 // section, and visit_die() for each top-level DIE.
1196
1197 void
1198 Dwarf_info_reader::parse()
1199 {
1200 if (this->object_->is_big_endian())
1201 {
1202 #if defined(HAVE_TARGET_32_BIG) || defined(HAVE_TARGET_64_BIG)
1203 this->do_parse<true>();
1204 #else
1205 gold_unreachable();
1206 #endif
1207 }
1208 else
1209 {
1210 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_64_LITTLE)
1211 this->do_parse<false>();
1212 #else
1213 gold_unreachable();
1214 #endif
1215 }
1216 }
1217
1218 template<bool big_endian>
1219 void
1220 Dwarf_info_reader::do_parse()
1221 {
1222 // Get the section contents and decompress if necessary.
1223 section_size_type buffer_size;
1224 bool buffer_is_new;
1225 this->buffer_ = this->object_->decompressed_section_contents(this->shndx_,
1226 &buffer_size,
1227 &buffer_is_new);
1228 if (this->buffer_ == NULL || buffer_size == 0)
1229 return;
1230 this->buffer_end_ = this->buffer_ + buffer_size;
1231
1232 // The offset of this input section in the output section.
1233 off_t section_offset = this->object_->output_section_offset(this->shndx_);
1234
1235 // Start tracking relocations for this section.
1236 this->reloc_mapper_ = make_elf_reloc_mapper(this->object_, this->symtab_,
1237 this->symtab_size_);
1238 this->reloc_mapper_->initialize(this->reloc_shndx_, this->reloc_type_);
1239
1240 // Loop over compilation units (or type units).
1241 unsigned int abbrev_shndx = this->abbrev_shndx_;
1242 off_t abbrev_offset = 0;
1243 const unsigned char* pinfo = this->buffer_;
1244 while (pinfo < this->buffer_end_)
1245 {
1246 // Read the compilation (or type) unit header.
1247 const unsigned char* cu_start = pinfo;
1248 this->cu_offset_ = cu_start - this->buffer_;
1249 this->cu_length_ = this->buffer_end_ - cu_start;
1250
1251 // Read unit_length (4 or 12 bytes).
1252 if (!this->check_buffer(pinfo + 4))
1253 break;
1254 uint32_t unit_length =
1255 elfcpp::Swap_unaligned<32, big_endian>::readval(pinfo);
1256 pinfo += 4;
1257 if (unit_length == 0xffffffff)
1258 {
1259 if (!this->check_buffer(pinfo + 8))
1260 break;
1261 unit_length = elfcpp::Swap_unaligned<64, big_endian>::readval(pinfo);
1262 pinfo += 8;
1263 this->offset_size_ = 8;
1264 }
1265 else
1266 this->offset_size_ = 4;
1267 if (!this->check_buffer(pinfo + unit_length))
1268 break;
1269 const unsigned char* cu_end = pinfo + unit_length;
1270 this->cu_length_ = cu_end - cu_start;
1271 if (!this->check_buffer(pinfo + 2 + this->offset_size_ + 1))
1272 break;
1273
1274 // Read version (2 bytes).
1275 this->cu_version_ =
1276 elfcpp::Swap_unaligned<16, big_endian>::readval(pinfo);
1277 pinfo += 2;
1278
1279 // Read debug_abbrev_offset (4 or 8 bytes).
1280 if (this->offset_size_ == 4)
1281 abbrev_offset = elfcpp::Swap_unaligned<32, big_endian>::readval(pinfo);
1282 else
1283 abbrev_offset = elfcpp::Swap_unaligned<64, big_endian>::readval(pinfo);
1284 if (this->reloc_shndx_ > 0)
1285 {
1286 off_t reloc_offset = pinfo - this->buffer_;
1287 off_t value;
1288 abbrev_shndx =
1289 this->reloc_mapper_->get_reloc_target(reloc_offset, &value);
1290 if (abbrev_shndx == 0)
1291 return;
1292 if (this->reloc_type_ == elfcpp::SHT_REL)
1293 abbrev_offset += value;
1294 else
1295 abbrev_offset = value;
1296 }
1297 pinfo += this->offset_size_;
1298
1299 // Read address_size (1 byte).
1300 this->address_size_ = *pinfo++;
1301
1302 // For type units, read the two extra fields.
1303 uint64_t signature = 0;
1304 off_t type_offset = 0;
1305 if (this->is_type_unit_)
1306 {
1307 if (!this->check_buffer(pinfo + 8 + this->offset_size_))
1308 break;
1309
1310 // Read type_signature (8 bytes).
1311 signature = elfcpp::Swap_unaligned<64, big_endian>::readval(pinfo);
1312 pinfo += 8;
1313
1314 // Read type_offset (4 or 8 bytes).
1315 if (this->offset_size_ == 4)
1316 type_offset =
1317 elfcpp::Swap_unaligned<32, big_endian>::readval(pinfo);
1318 else
1319 type_offset =
1320 elfcpp::Swap_unaligned<64, big_endian>::readval(pinfo);
1321 pinfo += this->offset_size_;
1322 }
1323
1324 // Read the .debug_abbrev table.
1325 this->abbrev_table_.read_abbrevs(this->object_, abbrev_shndx,
1326 abbrev_offset);
1327
1328 // Visit the root DIE.
1329 Dwarf_die root_die(this,
1330 pinfo - (this->buffer_ + this->cu_offset_),
1331 NULL);
1332 if (root_die.tag() != 0)
1333 {
1334 // Visit the CU or TU.
1335 if (this->is_type_unit_)
1336 this->visit_type_unit(section_offset + this->cu_offset_,
1337 cu_end - cu_start, type_offset, signature,
1338 &root_die);
1339 else
1340 this->visit_compilation_unit(section_offset + this->cu_offset_,
1341 cu_end - cu_start, &root_die);
1342 }
1343
1344 // Advance to the next CU.
1345 pinfo = cu_end;
1346 }
1347
1348 if (buffer_is_new)
1349 {
1350 delete[] this->buffer_;
1351 this->buffer_ = NULL;
1352 }
1353 }
1354
1355 // Read the DWARF string table.
1356
1357 bool
1358 Dwarf_info_reader::do_read_string_table(unsigned int string_shndx)
1359 {
1360 Relobj* object = this->object_;
1361
1362 // If we don't have relocations, string_shndx will be 0, and
1363 // we'll have to hunt for the .debug_str section.
1364 if (string_shndx == 0)
1365 {
1366 for (unsigned int i = 1; i < this->object_->shnum(); ++i)
1367 {
1368 std::string name = object->section_name(i);
1369 if (name == ".debug_str")
1370 {
1371 string_shndx = i;
1372 this->string_output_section_offset_ =
1373 object->output_section_offset(i);
1374 break;
1375 }
1376 }
1377 if (string_shndx == 0)
1378 return false;
1379 }
1380
1381 if (this->owns_string_buffer_ && this->string_buffer_ != NULL)
1382 {
1383 delete[] this->string_buffer_;
1384 this->owns_string_buffer_ = false;
1385 }
1386
1387 // Get the secton contents and decompress if necessary.
1388 section_size_type buffer_size;
1389 const unsigned char* buffer =
1390 object->decompressed_section_contents(string_shndx,
1391 &buffer_size,
1392 &this->owns_string_buffer_);
1393 this->string_buffer_ = reinterpret_cast<const char*>(buffer);
1394 this->string_buffer_end_ = this->string_buffer_ + buffer_size;
1395 this->string_shndx_ = string_shndx;
1396 return true;
1397 }
1398
1399 // Read a possibly unaligned integer of SIZE.
1400 template <int valsize>
1401 inline typename elfcpp::Valtype_base<valsize>::Valtype
1402 Dwarf_info_reader::read_from_pointer(const unsigned char* source)
1403 {
1404 typename elfcpp::Valtype_base<valsize>::Valtype return_value;
1405 if (this->object_->is_big_endian())
1406 return_value = elfcpp::Swap_unaligned<valsize, true>::readval(source);
1407 else
1408 return_value = elfcpp::Swap_unaligned<valsize, false>::readval(source);
1409 return return_value;
1410 }
1411
1412 // Read a possibly unaligned integer of SIZE. Update SOURCE after read.
1413 template <int valsize>
1414 inline typename elfcpp::Valtype_base<valsize>::Valtype
1415 Dwarf_info_reader::read_from_pointer(const unsigned char** source)
1416 {
1417 typename elfcpp::Valtype_base<valsize>::Valtype return_value;
1418 if (this->object_->is_big_endian())
1419 return_value = elfcpp::Swap_unaligned<valsize, true>::readval(*source);
1420 else
1421 return_value = elfcpp::Swap_unaligned<valsize, false>::readval(*source);
1422 *source += valsize / 8;
1423 return return_value;
1424 }
1425
1426 // Look for a relocation at offset ATTR_OFF in the dwarf info,
1427 // and return the section index and offset of the target.
1428
1429 unsigned int
1430 Dwarf_info_reader::lookup_reloc(off_t attr_off, off_t* target_off)
1431 {
1432 off_t value;
1433 attr_off += this->cu_offset_;
1434 unsigned int shndx = this->reloc_mapper_->get_reloc_target(attr_off, &value);
1435 if (shndx == 0)
1436 return 0;
1437 if (this->reloc_type_ == elfcpp::SHT_REL)
1438 *target_off += value;
1439 else
1440 *target_off = value;
1441 return shndx;
1442 }
1443
1444 // Return a string from the DWARF string table.
1445
1446 const char*
1447 Dwarf_info_reader::get_string(off_t str_off, unsigned int string_shndx)
1448 {
1449 if (!this->read_string_table(string_shndx))
1450 return NULL;
1451
1452 // Correct the offset. For incremental update links, we have a
1453 // relocated offset that is relative to the output section, but
1454 // here we need an offset relative to the input section.
1455 str_off -= this->string_output_section_offset_;
1456
1457 const char* p = this->string_buffer_ + str_off;
1458
1459 if (p < this->string_buffer_ || p >= this->string_buffer_end_)
1460 return NULL;
1461
1462 return p;
1463 }
1464
1465 // The following are default, do-nothing, implementations of the
1466 // hook methods normally provided by a derived class. We provide
1467 // default implementations rather than no implementation so that
1468 // a derived class needs to implement only the hooks that it needs
1469 // to use.
1470
1471 // Process a compilation unit and parse its child DIE.
1472
1473 void
1474 Dwarf_info_reader::visit_compilation_unit(off_t, off_t, Dwarf_die*)
1475 {
1476 }
1477
1478 // Process a type unit and parse its child DIE.
1479
1480 void
1481 Dwarf_info_reader::visit_type_unit(off_t, off_t, off_t, uint64_t, Dwarf_die*)
1482 {
1483 }
1484
1485 // Print a warning about a corrupt debug section.
1486
1487 void
1488 Dwarf_info_reader::warn_corrupt_debug_section() const
1489 {
1490 gold_warning(_("%s: corrupt debug info in %s"),
1491 this->object_->name().c_str(),
1492 this->object_->section_name(this->shndx_).c_str());
1493 }
1494
1495 // class Sized_dwarf_line_info
1496
1497 struct LineStateMachine
1498 {
1499 int file_num;
1500 uint64_t address;
1501 int line_num;
1502 int column_num;
1503 unsigned int shndx; // the section address refers to
1504 bool is_stmt; // stmt means statement.
1505 bool basic_block;
1506 bool end_sequence;
1507 };
1508
1509 static void
1510 ResetLineStateMachine(struct LineStateMachine* lsm, bool default_is_stmt)
1511 {
1512 lsm->file_num = 1;
1513 lsm->address = 0;
1514 lsm->line_num = 1;
1515 lsm->column_num = 0;
1516 lsm->shndx = -1U;
1517 lsm->is_stmt = default_is_stmt;
1518 lsm->basic_block = false;
1519 lsm->end_sequence = false;
1520 }
1521
1522 template<int size, bool big_endian>
1523 Sized_dwarf_line_info<size, big_endian>::Sized_dwarf_line_info(
1524 Object* object,
1525 unsigned int read_shndx)
1526 : data_valid_(false), buffer_(NULL), buffer_start_(NULL),
1527 reloc_mapper_(NULL), symtab_buffer_(NULL), directories_(), files_(),
1528 current_header_index_(-1)
1529 {
1530 unsigned int debug_shndx;
1531
1532 for (debug_shndx = 1; debug_shndx < object->shnum(); ++debug_shndx)
1533 {
1534 // FIXME: do this more efficiently: section_name() isn't super-fast
1535 std::string name = object->section_name(debug_shndx);
1536 if (name == ".debug_line" || name == ".zdebug_line")
1537 {
1538 section_size_type buffer_size;
1539 bool is_new = false;
1540 this->buffer_ = object->decompressed_section_contents(debug_shndx,
1541 &buffer_size,
1542 &is_new);
1543 if (is_new)
1544 this->buffer_start_ = this->buffer_;
1545 this->buffer_end_ = this->buffer_ + buffer_size;
1546 break;
1547 }
1548 }
1549 if (this->buffer_ == NULL)
1550 return;
1551
1552 // Find the relocation section for ".debug_line".
1553 // We expect these for relobjs (.o's) but not dynobjs (.so's).
1554 unsigned int reloc_shndx = 0;
1555 for (unsigned int i = 0; i < object->shnum(); ++i)
1556 {
1557 unsigned int reloc_sh_type = object->section_type(i);
1558 if ((reloc_sh_type == elfcpp::SHT_REL
1559 || reloc_sh_type == elfcpp::SHT_RELA)
1560 && object->section_info(i) == debug_shndx)
1561 {
1562 reloc_shndx = i;
1563 this->track_relocs_type_ = reloc_sh_type;
1564 break;
1565 }
1566 }
1567
1568 // Finally, we need the symtab section to interpret the relocs.
1569 if (reloc_shndx != 0)
1570 {
1571 unsigned int symtab_shndx;
1572 for (symtab_shndx = 0; symtab_shndx < object->shnum(); ++symtab_shndx)
1573 if (object->section_type(symtab_shndx) == elfcpp::SHT_SYMTAB)
1574 {
1575 this->symtab_buffer_ = object->section_contents(
1576 symtab_shndx, &this->symtab_buffer_size_, false);
1577 break;
1578 }
1579 if (this->symtab_buffer_ == NULL)
1580 return;
1581 }
1582
1583 this->reloc_mapper_ =
1584 new Sized_elf_reloc_mapper<size, big_endian>(object,
1585 this->symtab_buffer_,
1586 this->symtab_buffer_size_);
1587 if (!this->reloc_mapper_->initialize(reloc_shndx, this->track_relocs_type_))
1588 return;
1589
1590 // Now that we have successfully read all the data, parse the debug
1591 // info.
1592 this->data_valid_ = true;
1593 this->read_line_mappings(read_shndx);
1594 }
1595
1596 // Read the DWARF header.
1597
1598 template<int size, bool big_endian>
1599 const unsigned char*
1600 Sized_dwarf_line_info<size, big_endian>::read_header_prolog(
1601 const unsigned char* lineptr)
1602 {
1603 uint32_t initial_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr);
1604 lineptr += 4;
1605
1606 // In DWARF2/3, if the initial length is all 1 bits, then the offset
1607 // size is 8 and we need to read the next 8 bytes for the real length.
1608 if (initial_length == 0xffffffff)
1609 {
1610 header_.offset_size = 8;
1611 initial_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr);
1612 lineptr += 8;
1613 }
1614 else
1615 header_.offset_size = 4;
1616
1617 header_.total_length = initial_length;
1618
1619 gold_assert(lineptr + header_.total_length <= buffer_end_);
1620
1621 header_.version = elfcpp::Swap_unaligned<16, big_endian>::readval(lineptr);
1622 lineptr += 2;
1623
1624 if (header_.offset_size == 4)
1625 header_.prologue_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr);
1626 else
1627 header_.prologue_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr);
1628 lineptr += header_.offset_size;
1629
1630 header_.min_insn_length = *lineptr;
1631 lineptr += 1;
1632
1633 header_.default_is_stmt = *lineptr;
1634 lineptr += 1;
1635
1636 header_.line_base = *reinterpret_cast<const signed char*>(lineptr);
1637 lineptr += 1;
1638
1639 header_.line_range = *lineptr;
1640 lineptr += 1;
1641
1642 header_.opcode_base = *lineptr;
1643 lineptr += 1;
1644
1645 header_.std_opcode_lengths.resize(header_.opcode_base + 1);
1646 header_.std_opcode_lengths[0] = 0;
1647 for (int i = 1; i < header_.opcode_base; i++)
1648 {
1649 header_.std_opcode_lengths[i] = *lineptr;
1650 lineptr += 1;
1651 }
1652
1653 return lineptr;
1654 }
1655
1656 // The header for a debug_line section is mildly complicated, because
1657 // the line info is very tightly encoded.
1658
1659 template<int size, bool big_endian>
1660 const unsigned char*
1661 Sized_dwarf_line_info<size, big_endian>::read_header_tables(
1662 const unsigned char* lineptr)
1663 {
1664 ++this->current_header_index_;
1665
1666 // Create a new directories_ entry and a new files_ entry for our new
1667 // header. We initialize each with a single empty element, because
1668 // dwarf indexes directory and filenames starting at 1.
1669 gold_assert(static_cast<int>(this->directories_.size())
1670 == this->current_header_index_);
1671 gold_assert(static_cast<int>(this->files_.size())
1672 == this->current_header_index_);
1673 this->directories_.push_back(std::vector<std::string>(1));
1674 this->files_.push_back(std::vector<std::pair<int, std::string> >(1));
1675
1676 // It is legal for the directory entry table to be empty.
1677 if (*lineptr)
1678 {
1679 int dirindex = 1;
1680 while (*lineptr)
1681 {
1682 const char* dirname = reinterpret_cast<const char*>(lineptr);
1683 gold_assert(dirindex
1684 == static_cast<int>(this->directories_.back().size()));
1685 this->directories_.back().push_back(dirname);
1686 lineptr += this->directories_.back().back().size() + 1;
1687 dirindex++;
1688 }
1689 }
1690 lineptr++;
1691
1692 // It is also legal for the file entry table to be empty.
1693 if (*lineptr)
1694 {
1695 int fileindex = 1;
1696 size_t len;
1697 while (*lineptr)
1698 {
1699 const char* filename = reinterpret_cast<const char*>(lineptr);
1700 lineptr += strlen(filename) + 1;
1701
1702 uint64_t dirindex = read_unsigned_LEB_128(lineptr, &len);
1703 lineptr += len;
1704
1705 if (dirindex >= this->directories_.back().size())
1706 dirindex = 0;
1707 int dirindexi = static_cast<int>(dirindex);
1708
1709 read_unsigned_LEB_128(lineptr, &len); // mod_time
1710 lineptr += len;
1711
1712 read_unsigned_LEB_128(lineptr, &len); // filelength
1713 lineptr += len;
1714
1715 gold_assert(fileindex
1716 == static_cast<int>(this->files_.back().size()));
1717 this->files_.back().push_back(std::make_pair(dirindexi, filename));
1718 fileindex++;
1719 }
1720 }
1721 lineptr++;
1722
1723 return lineptr;
1724 }
1725
1726 // Process a single opcode in the .debug.line structure.
1727
1728 template<int size, bool big_endian>
1729 bool
1730 Sized_dwarf_line_info<size, big_endian>::process_one_opcode(
1731 const unsigned char* start, struct LineStateMachine* lsm, size_t* len)
1732 {
1733 size_t oplen = 0;
1734 size_t templen;
1735 unsigned char opcode = *start;
1736 oplen++;
1737 start++;
1738
1739 // If the opcode is great than the opcode_base, it is a special
1740 // opcode. Most line programs consist mainly of special opcodes.
1741 if (opcode >= header_.opcode_base)
1742 {
1743 opcode -= header_.opcode_base;
1744 const int advance_address = ((opcode / header_.line_range)
1745 * header_.min_insn_length);
1746 lsm->address += advance_address;
1747
1748 const int advance_line = ((opcode % header_.line_range)
1749 + header_.line_base);
1750 lsm->line_num += advance_line;
1751 lsm->basic_block = true;
1752 *len = oplen;
1753 return true;
1754 }
1755
1756 // Otherwise, we have the regular opcodes
1757 switch (opcode)
1758 {
1759 case elfcpp::DW_LNS_copy:
1760 lsm->basic_block = false;
1761 *len = oplen;
1762 return true;
1763
1764 case elfcpp::DW_LNS_advance_pc:
1765 {
1766 const uint64_t advance_address
1767 = read_unsigned_LEB_128(start, &templen);
1768 oplen += templen;
1769 lsm->address += header_.min_insn_length * advance_address;
1770 }
1771 break;
1772
1773 case elfcpp::DW_LNS_advance_line:
1774 {
1775 const uint64_t advance_line = read_signed_LEB_128(start, &templen);
1776 oplen += templen;
1777 lsm->line_num += advance_line;
1778 }
1779 break;
1780
1781 case elfcpp::DW_LNS_set_file:
1782 {
1783 const uint64_t fileno = read_unsigned_LEB_128(start, &templen);
1784 oplen += templen;
1785 lsm->file_num = fileno;
1786 }
1787 break;
1788
1789 case elfcpp::DW_LNS_set_column:
1790 {
1791 const uint64_t colno = read_unsigned_LEB_128(start, &templen);
1792 oplen += templen;
1793 lsm->column_num = colno;
1794 }
1795 break;
1796
1797 case elfcpp::DW_LNS_negate_stmt:
1798 lsm->is_stmt = !lsm->is_stmt;
1799 break;
1800
1801 case elfcpp::DW_LNS_set_basic_block:
1802 lsm->basic_block = true;
1803 break;
1804
1805 case elfcpp::DW_LNS_fixed_advance_pc:
1806 {
1807 int advance_address;
1808 advance_address = elfcpp::Swap_unaligned<16, big_endian>::readval(start);
1809 oplen += 2;
1810 lsm->address += advance_address;
1811 }
1812 break;
1813
1814 case elfcpp::DW_LNS_const_add_pc:
1815 {
1816 const int advance_address = (header_.min_insn_length
1817 * ((255 - header_.opcode_base)
1818 / header_.line_range));
1819 lsm->address += advance_address;
1820 }
1821 break;
1822
1823 case elfcpp::DW_LNS_extended_op:
1824 {
1825 const uint64_t extended_op_len
1826 = read_unsigned_LEB_128(start, &templen);
1827 start += templen;
1828 oplen += templen + extended_op_len;
1829
1830 const unsigned char extended_op = *start;
1831 start++;
1832
1833 switch (extended_op)
1834 {
1835 case elfcpp::DW_LNE_end_sequence:
1836 // This means that the current byte is the one immediately
1837 // after a set of instructions. Record the current line
1838 // for up to one less than the current address.
1839 lsm->line_num = -1;
1840 lsm->end_sequence = true;
1841 *len = oplen;
1842 return true;
1843
1844 case elfcpp::DW_LNE_set_address:
1845 {
1846 lsm->address =
1847 elfcpp::Swap_unaligned<size, big_endian>::readval(start);
1848 typename Reloc_map::const_iterator it
1849 = this->reloc_map_.find(start - this->buffer_);
1850 if (it != reloc_map_.end())
1851 {
1852 // If this is a SHT_RELA section, then ignore the
1853 // section contents. This assumes that this is a
1854 // straight reloc which just uses the reloc addend.
1855 // The reloc addend has already been included in the
1856 // symbol value.
1857 if (this->track_relocs_type_ == elfcpp::SHT_RELA)
1858 lsm->address = 0;
1859 // Add in the symbol value.
1860 lsm->address += it->second.second;
1861 lsm->shndx = it->second.first;
1862 }
1863 else
1864 {
1865 // If we're a normal .o file, with relocs, every
1866 // set_address should have an associated relocation.
1867 if (this->input_is_relobj())
1868 this->data_valid_ = false;
1869 }
1870 break;
1871 }
1872 case elfcpp::DW_LNE_define_file:
1873 {
1874 const char* filename = reinterpret_cast<const char*>(start);
1875 templen = strlen(filename) + 1;
1876 start += templen;
1877
1878 uint64_t dirindex = read_unsigned_LEB_128(start, &templen);
1879
1880 if (dirindex >= this->directories_.back().size())
1881 dirindex = 0;
1882 int dirindexi = static_cast<int>(dirindex);
1883
1884 // This opcode takes two additional ULEB128 parameters
1885 // (mod_time and filelength), but we don't use those
1886 // values. Because OPLEN already tells us how far to
1887 // skip to the next opcode, we don't need to read
1888 // them at all.
1889
1890 this->files_.back().push_back(std::make_pair(dirindexi,
1891 filename));
1892 }
1893 break;
1894 }
1895 }
1896 break;
1897
1898 default:
1899 {
1900 // Ignore unknown opcode silently
1901 for (int i = 0; i < header_.std_opcode_lengths[opcode]; i++)
1902 {
1903 size_t templen;
1904 read_unsigned_LEB_128(start, &templen);
1905 start += templen;
1906 oplen += templen;
1907 }
1908 }
1909 break;
1910 }
1911 *len = oplen;
1912 return false;
1913 }
1914
1915 // Read the debug information at LINEPTR and store it in the line
1916 // number map.
1917
1918 template<int size, bool big_endian>
1919 unsigned const char*
1920 Sized_dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr,
1921 unsigned int shndx)
1922 {
1923 struct LineStateMachine lsm;
1924
1925 // LENGTHSTART is the place the length field is based on. It is the
1926 // point in the header after the initial length field.
1927 const unsigned char* lengthstart = buffer_;
1928
1929 // In 64 bit dwarf, the initial length is 12 bytes, because of the
1930 // 0xffffffff at the start.
1931 if (header_.offset_size == 8)
1932 lengthstart += 12;
1933 else
1934 lengthstart += 4;
1935
1936 while (lineptr < lengthstart + header_.total_length)
1937 {
1938 ResetLineStateMachine(&lsm, header_.default_is_stmt);
1939 while (!lsm.end_sequence)
1940 {
1941 size_t oplength;
1942 bool add_line = this->process_one_opcode(lineptr, &lsm, &oplength);
1943 if (add_line
1944 && (shndx == -1U || lsm.shndx == -1U || shndx == lsm.shndx))
1945 {
1946 Offset_to_lineno_entry entry
1947 = { static_cast<off_t>(lsm.address),
1948 this->current_header_index_,
1949 static_cast<unsigned int>(lsm.file_num),
1950 true, lsm.line_num };
1951 std::vector<Offset_to_lineno_entry>&
1952 map(this->line_number_map_[lsm.shndx]);
1953 // If we see two consecutive entries with the same
1954 // offset and a real line number, then mark the first
1955 // one as non-canonical.
1956 if (!map.empty()
1957 && (map.back().offset == static_cast<off_t>(lsm.address))
1958 && lsm.line_num != -1
1959 && map.back().line_num != -1)
1960 map.back().last_line_for_offset = false;
1961 map.push_back(entry);
1962 }
1963 lineptr += oplength;
1964 }
1965 }
1966
1967 return lengthstart + header_.total_length;
1968 }
1969
1970 // Read the relocations into a Reloc_map.
1971
1972 template<int size, bool big_endian>
1973 void
1974 Sized_dwarf_line_info<size, big_endian>::read_relocs()
1975 {
1976 if (this->symtab_buffer_ == NULL)
1977 return;
1978
1979 off_t value;
1980 off_t reloc_offset;
1981 while ((reloc_offset = this->reloc_mapper_->next_offset()) != -1)
1982 {
1983 const unsigned int shndx =
1984 this->reloc_mapper_->get_reloc_target(reloc_offset, &value);
1985
1986 // There is no reason to record non-ordinary section indexes, or
1987 // SHN_UNDEF, because they will never match the real section.
1988 if (shndx != 0)
1989 this->reloc_map_[reloc_offset] = std::make_pair(shndx, value);
1990
1991 this->reloc_mapper_->advance(reloc_offset + 1);
1992 }
1993 }
1994
1995 // Read the line number info.
1996
1997 template<int size, bool big_endian>
1998 void
1999 Sized_dwarf_line_info<size, big_endian>::read_line_mappings(unsigned int shndx)
2000 {
2001 gold_assert(this->data_valid_ == true);
2002
2003 this->read_relocs();
2004 while (this->buffer_ < this->buffer_end_)
2005 {
2006 const unsigned char* lineptr = this->buffer_;
2007 lineptr = this->read_header_prolog(lineptr);
2008 lineptr = this->read_header_tables(lineptr);
2009 lineptr = this->read_lines(lineptr, shndx);
2010 this->buffer_ = lineptr;
2011 }
2012
2013 // Sort the lines numbers, so addr2line can use binary search.
2014 for (typename Lineno_map::iterator it = line_number_map_.begin();
2015 it != line_number_map_.end();
2016 ++it)
2017 // Each vector needs to be sorted by offset.
2018 std::sort(it->second.begin(), it->second.end());
2019 }
2020
2021 // Some processing depends on whether the input is a .o file or not.
2022 // For instance, .o files have relocs, and have .debug_lines
2023 // information on a per section basis. .so files, on the other hand,
2024 // lack relocs, and offsets are unique, so we can ignore the section
2025 // information.
2026
2027 template<int size, bool big_endian>
2028 bool
2029 Sized_dwarf_line_info<size, big_endian>::input_is_relobj()
2030 {
2031 // Only .o files have relocs and the symtab buffer that goes with them.
2032 return this->symtab_buffer_ != NULL;
2033 }
2034
2035 // Given an Offset_to_lineno_entry vector, and an offset, figure out
2036 // if the offset points into a function according to the vector (see
2037 // comments below for the algorithm). If it does, return an iterator
2038 // into the vector that points to the line-number that contains that
2039 // offset. If not, it returns vector::end().
2040
2041 static std::vector<Offset_to_lineno_entry>::const_iterator
2042 offset_to_iterator(const std::vector<Offset_to_lineno_entry>* offsets,
2043 off_t offset)
2044 {
2045 const Offset_to_lineno_entry lookup_key = { offset, 0, 0, true, 0 };
2046
2047 // lower_bound() returns the smallest offset which is >= lookup_key.
2048 // If no offset in offsets is >= lookup_key, returns end().
2049 std::vector<Offset_to_lineno_entry>::const_iterator it
2050 = std::lower_bound(offsets->begin(), offsets->end(), lookup_key);
2051
2052 // This code is easiest to understand with a concrete example.
2053 // Here's a possible offsets array:
2054 // {{offset = 3211, header_num = 0, file_num = 1, last, line_num = 16}, // 0
2055 // {offset = 3224, header_num = 0, file_num = 1, last, line_num = 20}, // 1
2056 // {offset = 3226, header_num = 0, file_num = 1, last, line_num = 22}, // 2
2057 // {offset = 3231, header_num = 0, file_num = 1, last, line_num = 25}, // 3
2058 // {offset = 3232, header_num = 0, file_num = 1, last, line_num = -1}, // 4
2059 // {offset = 3232, header_num = 0, file_num = 1, last, line_num = 65}, // 5
2060 // {offset = 3235, header_num = 0, file_num = 1, last, line_num = 66}, // 6
2061 // {offset = 3236, header_num = 0, file_num = 1, last, line_num = -1}, // 7
2062 // {offset = 5764, header_num = 0, file_num = 1, last, line_num = 48}, // 8
2063 // {offset = 5764, header_num = 0, file_num = 1,!last, line_num = 47}, // 9
2064 // {offset = 5765, header_num = 0, file_num = 1, last, line_num = 49}, // 10
2065 // {offset = 5767, header_num = 0, file_num = 1, last, line_num = 50}, // 11
2066 // {offset = 5768, header_num = 0, file_num = 1, last, line_num = 51}, // 12
2067 // {offset = 5773, header_num = 0, file_num = 1, last, line_num = -1}, // 13
2068 // {offset = 5787, header_num = 1, file_num = 1, last, line_num = 19}, // 14
2069 // {offset = 5790, header_num = 1, file_num = 1, last, line_num = 20}, // 15
2070 // {offset = 5793, header_num = 1, file_num = 1, last, line_num = 67}, // 16
2071 // {offset = 5793, header_num = 1, file_num = 1, last, line_num = -1}, // 17
2072 // {offset = 5793, header_num = 1, file_num = 1,!last, line_num = 66}, // 18
2073 // {offset = 5795, header_num = 1, file_num = 1, last, line_num = 68}, // 19
2074 // {offset = 5798, header_num = 1, file_num = 1, last, line_num = -1}, // 20
2075 // The entries with line_num == -1 mark the end of a function: the
2076 // associated offset is one past the last instruction in the
2077 // function. This can correspond to the beginning of the next
2078 // function (as is true for offset 3232); alternately, there can be
2079 // a gap between the end of one function and the start of the next
2080 // (as is true for some others, most obviously from 3236->5764).
2081 //
2082 // Case 1: lookup_key has offset == 10. lower_bound returns
2083 // offsets[0]. Since it's not an exact match and we're
2084 // at the beginning of offsets, we return end() (invalid).
2085 // Case 2: lookup_key has offset 10000. lower_bound returns
2086 // offset[21] (end()). We return end() (invalid).
2087 // Case 3: lookup_key has offset == 3211. lower_bound matches
2088 // offsets[0] exactly, and that's the entry we return.
2089 // Case 4: lookup_key has offset == 3232. lower_bound returns
2090 // offsets[4]. That's an exact match, but indicates
2091 // end-of-function. We check if offsets[5] is also an
2092 // exact match but not end-of-function. It is, so we
2093 // return offsets[5].
2094 // Case 5: lookup_key has offset == 3214. lower_bound returns
2095 // offsets[1]. Since it's not an exact match, we back
2096 // up to the offset that's < lookup_key, offsets[0].
2097 // We note offsets[0] is a valid entry (not end-of-function),
2098 // so that's the entry we return.
2099 // Case 6: lookup_key has offset == 4000. lower_bound returns
2100 // offsets[8]. Since it's not an exact match, we back
2101 // up to offsets[7]. Since offsets[7] indicates
2102 // end-of-function, we know lookup_key is between
2103 // functions, so we return end() (not a valid offset).
2104 // Case 7: lookup_key has offset == 5794. lower_bound returns
2105 // offsets[19]. Since it's not an exact match, we back
2106 // up to offsets[16]. Note we back up to the *first*
2107 // entry with offset 5793, not just offsets[19-1].
2108 // We note offsets[16] is a valid entry, so we return it.
2109 // If offsets[16] had had line_num == -1, we would have
2110 // checked offsets[17]. The reason for this is that
2111 // 16 and 17 can be in an arbitrary order, since we sort
2112 // only by offset and last_line_for_offset. (Note it
2113 // doesn't help to use line_number as a tertiary sort key,
2114 // since sometimes we want the -1 to be first and sometimes
2115 // we want it to be last.)
2116
2117 // This deals with cases (1) and (2).
2118 if ((it == offsets->begin() && offset < it->offset)
2119 || it == offsets->end())
2120 return offsets->end();
2121
2122 // This deals with cases (3) and (4).
2123 if (offset == it->offset)
2124 {
2125 while (it != offsets->end()
2126 && it->offset == offset
2127 && it->line_num == -1)
2128 ++it;
2129 if (it == offsets->end() || it->offset != offset)
2130 return offsets->end();
2131 else
2132 return it;
2133 }
2134
2135 // This handles the first part of case (7) -- we back up to the
2136 // *first* entry that has the offset that's behind us.
2137 gold_assert(it != offsets->begin());
2138 std::vector<Offset_to_lineno_entry>::const_iterator range_end = it;
2139 --it;
2140 const off_t range_value = it->offset;
2141 while (it != offsets->begin() && (it-1)->offset == range_value)
2142 --it;
2143
2144 // This handles cases (5), (6), and (7): if any entry in the
2145 // equal_range [it, range_end) has a line_num != -1, it's a valid
2146 // match. If not, we're not in a function. The line number we saw
2147 // last for an offset will be sorted first, so it'll get returned if
2148 // it's present.
2149 for (; it != range_end; ++it)
2150 if (it->line_num != -1)
2151 return it;
2152 return offsets->end();
2153 }
2154
2155 // Returns the canonical filename:lineno for the address passed in.
2156 // If other_lines is not NULL, appends the non-canonical lines
2157 // assigned to the same address.
2158
2159 template<int size, bool big_endian>
2160 std::string
2161 Sized_dwarf_line_info<size, big_endian>::do_addr2line(
2162 unsigned int shndx,
2163 off_t offset,
2164 std::vector<std::string>* other_lines)
2165 {
2166 if (this->data_valid_ == false)
2167 return "";
2168
2169 const std::vector<Offset_to_lineno_entry>* offsets;
2170 // If we do not have reloc information, then our input is a .so or
2171 // some similar data structure where all the information is held in
2172 // the offset. In that case, we ignore the input shndx.
2173 if (this->input_is_relobj())
2174 offsets = &this->line_number_map_[shndx];
2175 else
2176 offsets = &this->line_number_map_[-1U];
2177 if (offsets->empty())
2178 return "";
2179
2180 typename std::vector<Offset_to_lineno_entry>::const_iterator it
2181 = offset_to_iterator(offsets, offset);
2182 if (it == offsets->end())
2183 return "";
2184
2185 std::string result = this->format_file_lineno(*it);
2186 if (other_lines != NULL)
2187 for (++it; it != offsets->end() && it->offset == offset; ++it)
2188 {
2189 if (it->line_num == -1)
2190 continue; // The end of a previous function.
2191 other_lines->push_back(this->format_file_lineno(*it));
2192 }
2193 return result;
2194 }
2195
2196 // Convert the file_num + line_num into a string.
2197
2198 template<int size, bool big_endian>
2199 std::string
2200 Sized_dwarf_line_info<size, big_endian>::format_file_lineno(
2201 const Offset_to_lineno_entry& loc) const
2202 {
2203 std::string ret;
2204
2205 gold_assert(loc.header_num < static_cast<int>(this->files_.size()));
2206 gold_assert(loc.file_num
2207 < static_cast<unsigned int>(this->files_[loc.header_num].size()));
2208 const std::pair<int, std::string>& filename_pair
2209 = this->files_[loc.header_num][loc.file_num];
2210 const std::string& filename = filename_pair.second;
2211
2212 gold_assert(loc.header_num < static_cast<int>(this->directories_.size()));
2213 gold_assert(filename_pair.first
2214 < static_cast<int>(this->directories_[loc.header_num].size()));
2215 const std::string& dirname
2216 = this->directories_[loc.header_num][filename_pair.first];
2217
2218 if (!dirname.empty())
2219 {
2220 ret += dirname;
2221 ret += "/";
2222 }
2223 ret += filename;
2224 if (ret.empty())
2225 ret = "(unknown)";
2226
2227 char buffer[64]; // enough to hold a line number
2228 snprintf(buffer, sizeof(buffer), "%d", loc.line_num);
2229 ret += ":";
2230 ret += buffer;
2231
2232 return ret;
2233 }
2234
2235 // Dwarf_line_info routines.
2236
2237 static unsigned int next_generation_count = 0;
2238
2239 struct Addr2line_cache_entry
2240 {
2241 Object* object;
2242 unsigned int shndx;
2243 Dwarf_line_info* dwarf_line_info;
2244 unsigned int generation_count;
2245 unsigned int access_count;
2246
2247 Addr2line_cache_entry(Object* o, unsigned int s, Dwarf_line_info* d)
2248 : object(o), shndx(s), dwarf_line_info(d),
2249 generation_count(next_generation_count), access_count(0)
2250 {
2251 if (next_generation_count < (1U << 31))
2252 ++next_generation_count;
2253 }
2254 };
2255 // We expect this cache to be small, so don't bother with a hashtable
2256 // or priority queue or anything: just use a simple vector.
2257 static std::vector<Addr2line_cache_entry> addr2line_cache;
2258
2259 std::string
2260 Dwarf_line_info::one_addr2line(Object* object,
2261 unsigned int shndx, off_t offset,
2262 size_t cache_size,
2263 std::vector<std::string>* other_lines)
2264 {
2265 Dwarf_line_info* lineinfo = NULL;
2266 std::vector<Addr2line_cache_entry>::iterator it;
2267
2268 // First, check the cache. If we hit, update the counts.
2269 for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it)
2270 {
2271 if (it->object == object && it->shndx == shndx)
2272 {
2273 lineinfo = it->dwarf_line_info;
2274 it->generation_count = next_generation_count;
2275 // We cap generation_count at 2^31 -1 to avoid overflow.
2276 if (next_generation_count < (1U << 31))
2277 ++next_generation_count;
2278 // We cap access_count at 31 so 2^access_count doesn't overflow
2279 if (it->access_count < 31)
2280 ++it->access_count;
2281 break;
2282 }
2283 }
2284
2285 // If we don't hit the cache, create a new object and insert into the
2286 // cache.
2287 if (lineinfo == NULL)
2288 {
2289 switch (parameters->size_and_endianness())
2290 {
2291 #ifdef HAVE_TARGET_32_LITTLE
2292 case Parameters::TARGET_32_LITTLE:
2293 lineinfo = new Sized_dwarf_line_info<32, false>(object, shndx); break;
2294 #endif
2295 #ifdef HAVE_TARGET_32_BIG
2296 case Parameters::TARGET_32_BIG:
2297 lineinfo = new Sized_dwarf_line_info<32, true>(object, shndx); break;
2298 #endif
2299 #ifdef HAVE_TARGET_64_LITTLE
2300 case Parameters::TARGET_64_LITTLE:
2301 lineinfo = new Sized_dwarf_line_info<64, false>(object, shndx); break;
2302 #endif
2303 #ifdef HAVE_TARGET_64_BIG
2304 case Parameters::TARGET_64_BIG:
2305 lineinfo = new Sized_dwarf_line_info<64, true>(object, shndx); break;
2306 #endif
2307 default:
2308 gold_unreachable();
2309 }
2310 addr2line_cache.push_back(Addr2line_cache_entry(object, shndx, lineinfo));
2311 }
2312
2313 // Now that we have our object, figure out the answer
2314 std::string retval = lineinfo->addr2line(shndx, offset, other_lines);
2315
2316 // Finally, if our cache has grown too big, delete old objects. We
2317 // assume the common (probably only) case is deleting only one object.
2318 // We use a pretty simple scheme to evict: function of LRU and MFU.
2319 while (addr2line_cache.size() > cache_size)
2320 {
2321 unsigned int lowest_score = ~0U;
2322 std::vector<Addr2line_cache_entry>::iterator lowest
2323 = addr2line_cache.end();
2324 for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it)
2325 {
2326 const unsigned int score = (it->generation_count
2327 + (1U << it->access_count));
2328 if (score < lowest_score)
2329 {
2330 lowest_score = score;
2331 lowest = it;
2332 }
2333 }
2334 if (lowest != addr2line_cache.end())
2335 {
2336 delete lowest->dwarf_line_info;
2337 addr2line_cache.erase(lowest);
2338 }
2339 }
2340
2341 return retval;
2342 }
2343
2344 void
2345 Dwarf_line_info::clear_addr2line_cache()
2346 {
2347 for (std::vector<Addr2line_cache_entry>::iterator it = addr2line_cache.begin();
2348 it != addr2line_cache.end();
2349 ++it)
2350 delete it->dwarf_line_info;
2351 addr2line_cache.clear();
2352 }
2353
2354 #ifdef HAVE_TARGET_32_LITTLE
2355 template
2356 class Sized_dwarf_line_info<32, false>;
2357 #endif
2358
2359 #ifdef HAVE_TARGET_32_BIG
2360 template
2361 class Sized_dwarf_line_info<32, true>;
2362 #endif
2363
2364 #ifdef HAVE_TARGET_64_LITTLE
2365 template
2366 class Sized_dwarf_line_info<64, false>;
2367 #endif
2368
2369 #ifdef HAVE_TARGET_64_BIG
2370 template
2371 class Sized_dwarf_line_info<64, true>;
2372 #endif
2373
2374 } // End namespace gold.