1 // dwarf_reader.cc -- parse dwarf2/3 debug information
3 // Copyright 2007, 2008, 2009, 2010, 2011, 2012 Free Software Foundation, Inc.
4 // Written by Ian Lance Taylor <iant@google.com>.
6 // This file is part of gold.
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.
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.
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.
28 #include "elfcpp_swap.h"
32 #include "dwarf_reader.h"
33 #include "int_encoding.h"
34 #include "compressed_output.h"
38 // Class Sized_elf_reloc_mapper
40 // Initialize the relocation tracker for section RELOC_SHNDX.
42 template<int size, bool big_endian>
44 Sized_elf_reloc_mapper<size, big_endian>::do_initialize(
45 unsigned int reloc_shndx, unsigned int reloc_type)
47 this->reloc_type_ = reloc_type;
48 return this->track_relocs_.initialize(this->object_, reloc_shndx,
52 // Looks in the symtab to see what section a symbol is in.
54 template<int size, bool big_endian>
56 Sized_elf_reloc_mapper<size, big_endian>::symbol_section(
57 unsigned int symndx, Address* value, bool* is_ordinary)
59 const int symsize = elfcpp::Elf_sizes<size>::sym_size;
60 gold_assert((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(),
67 // Return the section index and offset within the section of
68 // the target of the relocation for RELOC_OFFSET.
70 template<int size, bool big_endian>
72 Sized_elf_reloc_mapper<size, big_endian>::do_get_reloc_target(
73 off_t reloc_offset, off_t* target_offset)
75 this->track_relocs_.advance(reloc_offset);
76 if (reloc_offset != this->track_relocs_.next_offset())
78 unsigned int symndx = this->track_relocs_.next_symndx();
79 typename elfcpp::Elf_types<size>::Elf_Addr value;
81 unsigned int target_shndx = this->symbol_section(symndx, &value,
85 if (this->reloc_type_ == elfcpp::SHT_RELA)
86 value += this->track_relocs_.next_addend();
87 *target_offset = value;
91 static inline Elf_reloc_mapper*
92 make_elf_reloc_mapper(Relobj* object, const unsigned char* symtab,
95 if (object->elfsize() == 32)
97 if (object->is_big_endian())
99 #ifdef HAVE_TARGET_32_BIG
100 return new Sized_elf_reloc_mapper<32, true>(object, symtab,
108 #ifdef HAVE_TARGET_32_LITTLE
109 return new Sized_elf_reloc_mapper<32, false>(object, symtab,
116 else if (object->elfsize() == 64)
118 if (object->is_big_endian())
120 #ifdef HAVE_TARGET_64_BIG
121 return new Sized_elf_reloc_mapper<64, true>(object, symtab,
129 #ifdef HAVE_TARGET_64_LITTLE
130 return new Sized_elf_reloc_mapper<64, false>(object, symtab,
141 // class Dwarf_abbrev_table
144 Dwarf_abbrev_table::clear_abbrev_codes()
146 for (unsigned int code = 0; code < this->low_abbrev_code_max_; ++code)
148 if (this->low_abbrev_codes_[code] != NULL)
150 delete this->low_abbrev_codes_[code];
151 this->low_abbrev_codes_[code] = NULL;
154 for (Abbrev_code_table::iterator it = this->high_abbrev_codes_.begin();
155 it != this->high_abbrev_codes_.end();
158 if (it->second != NULL)
161 this->high_abbrev_codes_.clear();
164 // Read the abbrev table from an object file.
167 Dwarf_abbrev_table::do_read_abbrevs(
169 unsigned int abbrev_shndx,
172 this->clear_abbrev_codes();
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)
180 for (unsigned int i = 1; i < object->shnum(); ++i)
182 std::string name = object->section_name(i);
183 if (name == ".debug_abbrev")
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);
193 if (abbrev_shndx == 0)
197 // Get the section contents and decompress if necessary.
198 if (abbrev_shndx != this->abbrev_shndx_)
200 if (this->owns_buffer_ && this->buffer_ != NULL)
202 delete[] this->buffer_;
203 this->owns_buffer_ = false;
206 section_size_type buffer_size;
208 object->decompressed_section_contents(abbrev_shndx,
210 &this->owns_buffer_);
211 this->buffer_end_ = this->buffer_ + buffer_size;
212 this->abbrev_shndx_ = abbrev_shndx;
215 this->buffer_pos_ = this->buffer_ + abbrev_offset;
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.
224 const Dwarf_abbrev_table::Abbrev_code*
225 Dwarf_abbrev_table::do_get_abbrev(unsigned int code)
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())
232 // Read and store abbrev code definitions until we find the
233 // one we're looking for.
236 // Read the abbrev code. A zero here indicates the end of the
239 if (this->buffer_pos_ >= this->buffer_end_)
241 uint64_t nextcode = read_unsigned_LEB_128(this->buffer_pos_, &len);
244 this->buffer_pos_ = this->buffer_end_;
247 this->buffer_pos_ += len;
250 if (this->buffer_pos_ >= this->buffer_end_)
252 uint64_t tag = read_unsigned_LEB_128(this->buffer_pos_, &len);
253 this->buffer_pos_ += len;
255 // Read the has_children flag.
256 if (this->buffer_pos_ >= this->buffer_end_)
258 bool has_children = *this->buffer_pos_ == elfcpp::DW_CHILDREN_yes;
259 this->buffer_pos_ += 1;
261 // Read the list of (attribute, form) pairs.
262 Abbrev_code* entry = new Abbrev_code(tag, has_children);
265 // Read the attribute.
266 if (this->buffer_pos_ >= this->buffer_end_)
268 uint64_t attr = read_unsigned_LEB_128(this->buffer_pos_, &len);
269 this->buffer_pos_ += len;
272 if (this->buffer_pos_ >= this->buffer_end_)
274 uint64_t form = read_unsigned_LEB_128(this->buffer_pos_, &len);
275 this->buffer_pos_ += len;
277 // A (0,0) pair terminates the list.
278 if (attr == 0 && form == 0)
281 if (attr == elfcpp::DW_AT_sibling)
282 entry->has_sibling_attribute = true;
284 entry->add_attribute(attr, form);
287 this->store_abbrev(nextcode, entry);
288 if (nextcode == code)
295 // class Dwarf_ranges_table
297 // Read the ranges table from an object file.
300 Dwarf_ranges_table::read_ranges_table(
302 const unsigned char* symtab,
304 unsigned int ranges_shndx)
306 // If we've already read this abbrev table, return immediately.
307 if (this->ranges_shndx_ > 0
308 && this->ranges_shndx_ == ranges_shndx)
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)
317 for (unsigned int i = 1; i < object->shnum(); ++i)
319 std::string name = object->section_name(i);
320 if (name == ".debug_ranges")
323 this->output_section_offset_ = object->output_section_offset(i);
327 if (ranges_shndx == 0)
331 // Get the section contents and decompress if necessary.
332 if (ranges_shndx != this->ranges_shndx_)
334 if (this->owns_ranges_buffer_ && this->ranges_buffer_ != NULL)
336 delete[] this->ranges_buffer_;
337 this->owns_ranges_buffer_ = false;
340 section_size_type buffer_size;
341 this->ranges_buffer_ =
342 object->decompressed_section_contents(ranges_shndx,
344 &this->owns_ranges_buffer_);
345 this->ranges_buffer_end_ = this->ranges_buffer_ + buffer_size;
346 this->ranges_shndx_ = ranges_shndx;
349 if (this->ranges_reloc_mapper_ != NULL)
351 delete this->ranges_reloc_mapper_;
352 this->ranges_reloc_mapper_ = NULL;
355 // For incremental objects, we have no relocations.
356 if (object->is_incremental())
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)
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)
374 this->ranges_reloc_mapper_ = make_elf_reloc_mapper(object, symtab,
376 this->ranges_reloc_mapper_->initialize(reloc_shndx, reloc_type);
381 // Read a range list from section RANGES_SHNDX at offset RANGES_OFFSET.
384 Dwarf_ranges_table::read_range_list(
386 const unsigned char* symtab,
388 unsigned int addr_size,
389 unsigned int ranges_shndx,
392 Dwarf_range_list* ranges;
394 if (!this->read_ranges_table(object, symtab, symtab_size, ranges_shndx))
397 // Correct the offset. For incremental update links, we have a
398 // relocated offset that is relative to the output section, but
399 // here we need an offset relative to the input section.
400 offset -= this->output_section_offset_;
402 // Read the range list at OFFSET.
403 ranges = new Dwarf_range_list();
406 this->ranges_buffer_ + offset < this->ranges_buffer_end_;
407 offset += 2 * addr_size)
412 // Read the raw contents of the section.
415 start = read_from_pointer<32>(this->ranges_buffer_ + offset);
416 end = read_from_pointer<32>(this->ranges_buffer_ + offset + 4);
420 start = read_from_pointer<64>(this->ranges_buffer_ + offset);
421 end = read_from_pointer<64>(this->ranges_buffer_ + offset + 8);
424 // Check for relocations and adjust the values.
425 unsigned int shndx1 = 0;
426 unsigned int shndx2 = 0;
427 if (this->ranges_reloc_mapper_ != NULL)
430 this->ranges_reloc_mapper_->get_reloc_target(offset, &start);
432 this->ranges_reloc_mapper_->get_reloc_target(offset + addr_size,
436 // End of list is marked by a pair of zeroes.
437 if (shndx1 == 0 && start == 0 && end == 0)
440 // A "base address selection entry" is identified by
441 // 0xffffffff for the first value of the pair. The second
442 // value is used as a base for subsequent range list entries.
443 if (shndx1 == 0 && start == -1)
445 else if (shndx1 == shndx2)
447 if (shndx1 == 0 || object->is_section_included(shndx1))
448 ranges->add(shndx1, base + start, base + end);
451 gold_warning(_("%s: DWARF info may be corrupt; offsets in a "
452 "range list entry are in different sections"),
453 object->name().c_str());
459 // class Dwarf_pubnames_table
461 // Read the pubnames section SHNDX from the object file.
464 Dwarf_pubnames_table::read_section(Relobj* object, unsigned int shndx)
466 section_size_type buffer_size;
468 // If we don't have relocations, shndx will be 0, and
469 // we'll have to hunt for the .debug_pubnames/pubtypes section.
472 const char* name = (this->is_pubtypes_
474 : ".debug_pubnames");
475 for (unsigned int i = 1; i < object->shnum(); ++i)
477 if (object->section_name(i) == name)
480 this->output_section_offset_ = object->output_section_offset(i);
488 this->buffer_ = object->decompressed_section_contents(shndx,
490 &this->owns_buffer_);
491 if (this->buffer_ == NULL)
493 this->buffer_end_ = this->buffer_ + buffer_size;
497 // Read the header for the set at OFFSET.
500 Dwarf_pubnames_table::read_header(off_t offset)
502 // Correct the offset. For incremental update links, we have a
503 // relocated offset that is relative to the output section, but
504 // here we need an offset relative to the input section.
505 offset -= this->output_section_offset_;
507 if (offset < 0 || offset + 14 >= this->buffer_end_ - this->buffer_)
510 const unsigned char* pinfo = this->buffer_ + offset;
512 // Read the unit_length field.
513 uint32_t unit_length = read_from_pointer<32>(pinfo);
515 if (unit_length == 0xffffffff)
517 unit_length = read_from_pointer<64>(pinfo);
519 this->offset_size_ = 8;
522 this->offset_size_ = 4;
524 // Check the version.
525 unsigned int version = read_from_pointer<16>(pinfo);
530 // Skip the debug_info_offset and debug_info_size fields.
531 pinfo += 2 * this->offset_size_;
533 if (pinfo >= this->buffer_end_)
536 this->pinfo_ = pinfo;
540 // Read the next name from the set.
543 Dwarf_pubnames_table::next_name()
545 const unsigned char* pinfo = this->pinfo_;
547 // Read the offset within the CU. If this is zero, we have reached
548 // the end of the list.
550 if (this->offset_size_ == 4)
551 offset = read_from_pointer<32>(&pinfo);
553 offset = read_from_pointer<64>(&pinfo);
557 // Return a pointer to the string at the current location,
558 // and advance the pointer to the next entry.
559 const char* ret = reinterpret_cast<const char*>(pinfo);
560 while (pinfo < this->buffer_end_ && *pinfo != '\0')
562 if (pinfo < this->buffer_end_)
565 this->pinfo_ = pinfo;
571 Dwarf_die::Dwarf_die(
572 Dwarf_info_reader* dwinfo,
575 : dwinfo_(dwinfo), parent_(parent), die_offset_(die_offset),
576 child_offset_(0), sibling_offset_(0), abbrev_code_(NULL), attributes_(),
577 attributes_read_(false), name_(NULL), name_off_(-1), linkage_name_(NULL),
578 linkage_name_off_(-1), string_shndx_(0), specification_(0),
582 const unsigned char* pdie = dwinfo->buffer_at_offset(die_offset);
585 unsigned int code = read_unsigned_LEB_128(pdie, &len);
589 parent->set_sibling_offset(die_offset + len);
592 this->attr_offset_ = len;
594 // Lookup the abbrev code in the abbrev table.
595 this->abbrev_code_ = dwinfo->get_abbrev(code);
598 // Read all the attributes of the DIE.
601 Dwarf_die::read_attributes()
603 if (this->attributes_read_)
606 gold_assert(this->abbrev_code_ != NULL);
608 const unsigned char* pdie =
609 this->dwinfo_->buffer_at_offset(this->die_offset_);
612 const unsigned char* pattr = pdie + this->attr_offset_;
614 unsigned int nattr = this->abbrev_code_->attributes.size();
615 this->attributes_.reserve(nattr);
616 for (unsigned int i = 0; i < nattr; ++i)
619 unsigned int attr = this->abbrev_code_->attributes[i].attr;
620 unsigned int form = this->abbrev_code_->attributes[i].form;
621 if (form == elfcpp::DW_FORM_indirect)
623 form = read_unsigned_LEB_128(pattr, &len);
626 off_t attr_off = this->die_offset_ + (pattr - pdie);
627 bool ref_form = false;
628 Attribute_value attr_value;
629 attr_value.attr = attr;
630 attr_value.form = form;
631 attr_value.aux.shndx = 0;
634 case elfcpp::DW_FORM_flag_present:
635 attr_value.val.intval = 1;
637 case elfcpp::DW_FORM_strp:
640 if (this->dwinfo_->offset_size() == 4)
641 str_off = read_from_pointer<32>(&pattr);
643 str_off = read_from_pointer<64>(&pattr);
645 this->dwinfo_->lookup_reloc(attr_off, &str_off);
646 attr_value.aux.shndx = shndx;
647 attr_value.val.refval = str_off;
650 case elfcpp::DW_FORM_sec_offset:
653 if (this->dwinfo_->offset_size() == 4)
654 sec_off = read_from_pointer<32>(&pattr);
656 sec_off = read_from_pointer<64>(&pattr);
658 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
659 attr_value.aux.shndx = shndx;
660 attr_value.val.refval = sec_off;
664 case elfcpp::DW_FORM_addr:
665 case elfcpp::DW_FORM_ref_addr:
668 if (this->dwinfo_->address_size() == 4)
669 sec_off = read_from_pointer<32>(&pattr);
671 sec_off = read_from_pointer<64>(&pattr);
673 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
674 attr_value.aux.shndx = shndx;
675 attr_value.val.refval = sec_off;
679 case elfcpp::DW_FORM_block1:
680 attr_value.aux.blocklen = *pattr++;
681 attr_value.val.blockval = pattr;
682 pattr += attr_value.aux.blocklen;
684 case elfcpp::DW_FORM_block2:
685 attr_value.aux.blocklen = read_from_pointer<16>(&pattr);
686 attr_value.val.blockval = pattr;
687 pattr += attr_value.aux.blocklen;
689 case elfcpp::DW_FORM_block4:
690 attr_value.aux.blocklen = read_from_pointer<32>(&pattr);
691 attr_value.val.blockval = pattr;
692 pattr += attr_value.aux.blocklen;
694 case elfcpp::DW_FORM_block:
695 case elfcpp::DW_FORM_exprloc:
696 attr_value.aux.blocklen = read_unsigned_LEB_128(pattr, &len);
697 attr_value.val.blockval = pattr + len;
698 pattr += len + attr_value.aux.blocklen;
700 case elfcpp::DW_FORM_data1:
701 case elfcpp::DW_FORM_flag:
702 attr_value.val.intval = *pattr++;
704 case elfcpp::DW_FORM_ref1:
705 attr_value.val.refval = *pattr++;
708 case elfcpp::DW_FORM_data2:
709 attr_value.val.intval = read_from_pointer<16>(&pattr);
711 case elfcpp::DW_FORM_ref2:
712 attr_value.val.refval = read_from_pointer<16>(&pattr);
715 case elfcpp::DW_FORM_data4:
718 sec_off = read_from_pointer<32>(&pattr);
720 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
721 attr_value.aux.shndx = shndx;
722 attr_value.val.intval = sec_off;
725 case elfcpp::DW_FORM_ref4:
728 sec_off = read_from_pointer<32>(&pattr);
730 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
731 attr_value.aux.shndx = shndx;
732 attr_value.val.refval = sec_off;
736 case elfcpp::DW_FORM_data8:
739 sec_off = read_from_pointer<64>(&pattr);
741 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
742 attr_value.aux.shndx = shndx;
743 attr_value.val.intval = sec_off;
746 case elfcpp::DW_FORM_ref_sig8:
747 attr_value.val.uintval = read_from_pointer<64>(&pattr);
749 case elfcpp::DW_FORM_ref8:
752 sec_off = read_from_pointer<64>(&pattr);
754 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
755 attr_value.aux.shndx = shndx;
756 attr_value.val.refval = sec_off;
760 case elfcpp::DW_FORM_ref_udata:
761 attr_value.val.refval = read_unsigned_LEB_128(pattr, &len);
765 case elfcpp::DW_FORM_udata:
766 case elfcpp::DW_FORM_GNU_addr_index:
767 case elfcpp::DW_FORM_GNU_str_index:
768 attr_value.val.uintval = read_unsigned_LEB_128(pattr, &len);
771 case elfcpp::DW_FORM_sdata:
772 attr_value.val.intval = read_signed_LEB_128(pattr, &len);
775 case elfcpp::DW_FORM_string:
776 attr_value.val.stringval = reinterpret_cast<const char*>(pattr);
777 len = strlen(attr_value.val.stringval);
784 // Cache the most frequently-requested attributes.
787 case elfcpp::DW_AT_name:
788 if (form == elfcpp::DW_FORM_string)
789 this->name_ = attr_value.val.stringval;
790 else if (form == elfcpp::DW_FORM_strp)
792 // All indirect strings should refer to the same
793 // string section, so we just save the last one seen.
794 this->string_shndx_ = attr_value.aux.shndx;
795 this->name_off_ = attr_value.val.refval;
798 case elfcpp::DW_AT_linkage_name:
799 case elfcpp::DW_AT_MIPS_linkage_name:
800 if (form == elfcpp::DW_FORM_string)
801 this->linkage_name_ = attr_value.val.stringval;
802 else if (form == elfcpp::DW_FORM_strp)
804 // All indirect strings should refer to the same
805 // string section, so we just save the last one seen.
806 this->string_shndx_ = attr_value.aux.shndx;
807 this->linkage_name_off_ = attr_value.val.refval;
810 case elfcpp::DW_AT_specification:
812 this->specification_ = attr_value.val.refval;
814 case elfcpp::DW_AT_abstract_origin:
816 this->abstract_origin_ = attr_value.val.refval;
818 case elfcpp::DW_AT_sibling:
819 if (ref_form && attr_value.aux.shndx == 0)
820 this->sibling_offset_ = attr_value.val.refval;
825 this->attributes_.push_back(attr_value);
828 // Now that we know where the next DIE begins, record the offset
829 // to avoid later recalculation.
830 if (this->has_children())
831 this->child_offset_ = this->die_offset_ + (pattr - pdie);
833 this->sibling_offset_ = this->die_offset_ + (pattr - pdie);
835 this->attributes_read_ = true;
839 // Skip all the attributes of the DIE and return the offset of the next DIE.
842 Dwarf_die::skip_attributes()
844 gold_assert(this->abbrev_code_ != NULL);
846 const unsigned char* pdie =
847 this->dwinfo_->buffer_at_offset(this->die_offset_);
850 const unsigned char* pattr = pdie + this->attr_offset_;
852 for (unsigned int i = 0; i < this->abbrev_code_->attributes.size(); ++i)
855 unsigned int form = this->abbrev_code_->attributes[i].form;
856 if (form == elfcpp::DW_FORM_indirect)
858 form = read_unsigned_LEB_128(pattr, &len);
863 case elfcpp::DW_FORM_flag_present:
865 case elfcpp::DW_FORM_strp:
866 case elfcpp::DW_FORM_sec_offset:
867 pattr += this->dwinfo_->offset_size();
869 case elfcpp::DW_FORM_addr:
870 case elfcpp::DW_FORM_ref_addr:
871 pattr += this->dwinfo_->address_size();
873 case elfcpp::DW_FORM_block1:
876 case elfcpp::DW_FORM_block2:
879 block_size = read_from_pointer<16>(&pattr);
883 case elfcpp::DW_FORM_block4:
886 block_size = read_from_pointer<32>(&pattr);
890 case elfcpp::DW_FORM_block:
891 case elfcpp::DW_FORM_exprloc:
894 block_size = read_unsigned_LEB_128(pattr, &len);
895 pattr += len + block_size;
898 case elfcpp::DW_FORM_data1:
899 case elfcpp::DW_FORM_ref1:
900 case elfcpp::DW_FORM_flag:
903 case elfcpp::DW_FORM_data2:
904 case elfcpp::DW_FORM_ref2:
907 case elfcpp::DW_FORM_data4:
908 case elfcpp::DW_FORM_ref4:
911 case elfcpp::DW_FORM_data8:
912 case elfcpp::DW_FORM_ref8:
913 case elfcpp::DW_FORM_ref_sig8:
916 case elfcpp::DW_FORM_ref_udata:
917 case elfcpp::DW_FORM_udata:
918 case elfcpp::DW_FORM_GNU_addr_index:
919 case elfcpp::DW_FORM_GNU_str_index:
920 read_unsigned_LEB_128(pattr, &len);
923 case elfcpp::DW_FORM_sdata:
924 read_signed_LEB_128(pattr, &len);
927 case elfcpp::DW_FORM_string:
928 len = strlen(reinterpret_cast<const char*>(pattr));
936 return this->die_offset_ + (pattr - pdie);
939 // Get the name of the DIE and cache it.
942 Dwarf_die::set_name()
944 if (this->name_ != NULL || !this->read_attributes())
946 if (this->name_off_ != -1)
947 this->name_ = this->dwinfo_->get_string(this->name_off_,
948 this->string_shndx_);
951 // Get the linkage name of the DIE and cache it.
954 Dwarf_die::set_linkage_name()
956 if (this->linkage_name_ != NULL || !this->read_attributes())
958 if (this->linkage_name_off_ != -1)
959 this->linkage_name_ = this->dwinfo_->get_string(this->linkage_name_off_,
960 this->string_shndx_);
963 // Return the value of attribute ATTR.
965 const Dwarf_die::Attribute_value*
966 Dwarf_die::attribute(unsigned int attr)
968 if (!this->read_attributes())
970 for (unsigned int i = 0; i < this->attributes_.size(); ++i)
972 if (this->attributes_[i].attr == attr)
973 return &this->attributes_[i];
979 Dwarf_die::string_attribute(unsigned int attr)
981 const Attribute_value* attr_val = this->attribute(attr);
982 if (attr_val == NULL)
984 switch (attr_val->form)
986 case elfcpp::DW_FORM_string:
987 return attr_val->val.stringval;
988 case elfcpp::DW_FORM_strp:
989 return this->dwinfo_->get_string(attr_val->val.refval,
990 attr_val->aux.shndx);
997 Dwarf_die::int_attribute(unsigned int attr)
999 const Attribute_value* attr_val = this->attribute(attr);
1000 if (attr_val == NULL)
1002 switch (attr_val->form)
1004 case elfcpp::DW_FORM_flag_present:
1005 case elfcpp::DW_FORM_data1:
1006 case elfcpp::DW_FORM_flag:
1007 case elfcpp::DW_FORM_data2:
1008 case elfcpp::DW_FORM_data4:
1009 case elfcpp::DW_FORM_data8:
1010 case elfcpp::DW_FORM_sdata:
1011 return attr_val->val.intval;
1018 Dwarf_die::uint_attribute(unsigned int attr)
1020 const Attribute_value* attr_val = this->attribute(attr);
1021 if (attr_val == NULL)
1023 switch (attr_val->form)
1025 case elfcpp::DW_FORM_flag_present:
1026 case elfcpp::DW_FORM_data1:
1027 case elfcpp::DW_FORM_flag:
1028 case elfcpp::DW_FORM_data4:
1029 case elfcpp::DW_FORM_data8:
1030 case elfcpp::DW_FORM_ref_sig8:
1031 case elfcpp::DW_FORM_udata:
1032 return attr_val->val.uintval;
1039 Dwarf_die::ref_attribute(unsigned int attr, unsigned int* shndx)
1041 const Attribute_value* attr_val = this->attribute(attr);
1042 if (attr_val == NULL)
1044 switch (attr_val->form)
1046 case elfcpp::DW_FORM_sec_offset:
1047 case elfcpp::DW_FORM_addr:
1048 case elfcpp::DW_FORM_ref_addr:
1049 case elfcpp::DW_FORM_ref1:
1050 case elfcpp::DW_FORM_ref2:
1051 case elfcpp::DW_FORM_ref4:
1052 case elfcpp::DW_FORM_ref8:
1053 case elfcpp::DW_FORM_ref_udata:
1054 *shndx = attr_val->aux.shndx;
1055 return attr_val->val.refval;
1056 case elfcpp::DW_FORM_ref_sig8:
1057 *shndx = attr_val->aux.shndx;
1058 return attr_val->val.uintval;
1059 case elfcpp::DW_FORM_data4:
1060 case elfcpp::DW_FORM_data8:
1061 *shndx = attr_val->aux.shndx;
1062 return attr_val->val.intval;
1069 Dwarf_die::address_attribute(unsigned int attr, unsigned int* shndx)
1071 const Attribute_value* attr_val = this->attribute(attr);
1072 if (attr_val == NULL || attr_val->form != elfcpp::DW_FORM_addr)
1075 *shndx = attr_val->aux.shndx;
1076 return attr_val->val.refval;
1079 // Return the offset of this DIE's first child.
1082 Dwarf_die::child_offset()
1084 gold_assert(this->abbrev_code_ != NULL);
1085 if (!this->has_children())
1087 if (this->child_offset_ == 0)
1088 this->child_offset_ = this->skip_attributes();
1089 return this->child_offset_;
1092 // Return the offset of this DIE's next sibling.
1095 Dwarf_die::sibling_offset()
1097 gold_assert(this->abbrev_code_ != NULL);
1099 if (this->sibling_offset_ != 0)
1100 return this->sibling_offset_;
1102 if (!this->has_children())
1104 this->sibling_offset_ = this->skip_attributes();
1105 return this->sibling_offset_;
1108 if (this->has_sibling_attribute())
1110 if (!this->read_attributes())
1112 if (this->sibling_offset_ != 0)
1113 return this->sibling_offset_;
1116 // Skip over the children.
1117 off_t child_offset = this->child_offset();
1118 while (child_offset > 0)
1120 Dwarf_die die(this->dwinfo_, child_offset, this);
1121 // The Dwarf_die ctor will set this DIE's sibling offset
1122 // when it reads a zero abbrev code.
1125 child_offset = die.sibling_offset();
1128 // This should be set by now. If not, there was a problem reading
1129 // the DWARF info, and we return 0.
1130 return this->sibling_offset_;
1133 // class Dwarf_info_reader
1135 // Check that the pointer P is within the current compilation unit.
1138 Dwarf_info_reader::check_buffer(const unsigned char* p) const
1140 if (p > this->buffer_ + this->cu_offset_ + this->cu_length_)
1142 gold_warning(_("%s: corrupt debug info in %s"),
1143 this->object_->name().c_str(),
1144 this->object_->section_name(this->shndx_).c_str());
1150 // Begin parsing the debug info. This calls visit_compilation_unit()
1151 // or visit_type_unit() for each compilation or type unit found in the
1152 // section, and visit_die() for each top-level DIE.
1155 Dwarf_info_reader::parse()
1157 if (this->object_->is_big_endian())
1159 #if defined(HAVE_TARGET_32_BIG) || defined(HAVE_TARGET_64_BIG)
1160 this->do_parse<true>();
1167 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_64_LITTLE)
1168 this->do_parse<false>();
1175 template<bool big_endian>
1177 Dwarf_info_reader::do_parse()
1179 // Get the section contents and decompress if necessary.
1180 section_size_type buffer_size;
1182 this->buffer_ = this->object_->decompressed_section_contents(this->shndx_,
1185 if (this->buffer_ == NULL || buffer_size == 0)
1187 this->buffer_end_ = this->buffer_ + buffer_size;
1189 // The offset of this input section in the output section.
1190 off_t section_offset = this->object_->output_section_offset(this->shndx_);
1192 // Start tracking relocations for this section.
1193 this->reloc_mapper_ = make_elf_reloc_mapper(this->object_, this->symtab_,
1194 this->symtab_size_);
1195 this->reloc_mapper_->initialize(this->reloc_shndx_, this->reloc_type_);
1197 // Loop over compilation units (or type units).
1198 unsigned int abbrev_shndx = 0;
1199 off_t abbrev_offset = 0;
1200 const unsigned char* pinfo = this->buffer_;
1201 while (pinfo < this->buffer_end_)
1203 // Read the compilation (or type) unit header.
1204 const unsigned char* cu_start = pinfo;
1205 this->cu_offset_ = cu_start - this->buffer_;
1206 this->cu_length_ = this->buffer_end_ - cu_start;
1208 // Read unit_length (4 or 12 bytes).
1209 if (!this->check_buffer(pinfo + 4))
1211 uint32_t unit_length =
1212 elfcpp::Swap_unaligned<32, big_endian>::readval(pinfo);
1214 if (unit_length == 0xffffffff)
1216 if (!this->check_buffer(pinfo + 8))
1218 unit_length = elfcpp::Swap_unaligned<64, big_endian>::readval(pinfo);
1220 this->offset_size_ = 8;
1223 this->offset_size_ = 4;
1224 if (!this->check_buffer(pinfo + unit_length))
1226 const unsigned char* cu_end = pinfo + unit_length;
1227 this->cu_length_ = cu_end - cu_start;
1228 if (!this->check_buffer(pinfo + 2 + this->offset_size_ + 1))
1231 // Read version (2 bytes).
1233 elfcpp::Swap_unaligned<16, big_endian>::readval(pinfo);
1236 // Read debug_abbrev_offset (4 or 8 bytes).
1237 if (this->offset_size_ == 4)
1238 abbrev_offset = elfcpp::Swap_unaligned<32, big_endian>::readval(pinfo);
1240 abbrev_offset = elfcpp::Swap_unaligned<64, big_endian>::readval(pinfo);
1241 if (this->reloc_shndx_ > 0)
1243 off_t reloc_offset = pinfo - this->buffer_;
1246 this->reloc_mapper_->get_reloc_target(reloc_offset, &value);
1247 if (abbrev_shndx == 0)
1249 if (this->reloc_type_ == elfcpp::SHT_REL)
1250 abbrev_offset += value;
1252 abbrev_offset = value;
1254 pinfo += this->offset_size_;
1256 // Read address_size (1 byte).
1257 this->address_size_ = *pinfo++;
1259 // For type units, read the two extra fields.
1260 uint64_t signature = 0;
1261 off_t type_offset = 0;
1262 if (this->is_type_unit_)
1264 if (!this->check_buffer(pinfo + 8 + this->offset_size_))
1267 // Read type_signature (8 bytes).
1268 signature = elfcpp::Swap_unaligned<64, big_endian>::readval(pinfo);
1271 // Read type_offset (4 or 8 bytes).
1272 if (this->offset_size_ == 4)
1274 elfcpp::Swap_unaligned<32, big_endian>::readval(pinfo);
1277 elfcpp::Swap_unaligned<64, big_endian>::readval(pinfo);
1278 pinfo += this->offset_size_;
1281 // Read the .debug_abbrev table.
1282 this->abbrev_table_.read_abbrevs(this->object_, abbrev_shndx,
1285 // Visit the root DIE.
1286 Dwarf_die root_die(this,
1287 pinfo - (this->buffer_ + this->cu_offset_),
1289 if (root_die.tag() != 0)
1291 // Visit the CU or TU.
1292 if (this->is_type_unit_)
1293 this->visit_type_unit(section_offset + this->cu_offset_,
1294 type_offset, signature, &root_die);
1296 this->visit_compilation_unit(section_offset + this->cu_offset_,
1297 cu_end - cu_start, &root_die);
1300 // Advance to the next CU.
1306 delete[] this->buffer_;
1307 this->buffer_ = NULL;
1311 // Read the DWARF string table.
1314 Dwarf_info_reader::do_read_string_table(unsigned int string_shndx)
1316 Relobj* object = this->object_;
1318 // If we don't have relocations, string_shndx will be 0, and
1319 // we'll have to hunt for the .debug_str section.
1320 if (string_shndx == 0)
1322 for (unsigned int i = 1; i < this->object_->shnum(); ++i)
1324 std::string name = object->section_name(i);
1325 if (name == ".debug_str")
1328 this->string_output_section_offset_ =
1329 object->output_section_offset(i);
1333 if (string_shndx == 0)
1337 if (this->owns_string_buffer_ && this->string_buffer_ != NULL)
1339 delete[] this->string_buffer_;
1340 this->owns_string_buffer_ = false;
1343 // Get the secton contents and decompress if necessary.
1344 section_size_type buffer_size;
1345 const unsigned char* buffer =
1346 object->decompressed_section_contents(string_shndx,
1348 &this->owns_string_buffer_);
1349 this->string_buffer_ = reinterpret_cast<const char*>(buffer);
1350 this->string_buffer_end_ = this->string_buffer_ + buffer_size;
1351 this->string_shndx_ = string_shndx;
1355 // Look for a relocation at offset ATTR_OFF in the dwarf info,
1356 // and return the section index and offset of the target.
1359 Dwarf_info_reader::lookup_reloc(off_t attr_off, off_t* target_off)
1362 attr_off += this->cu_offset_;
1363 unsigned int shndx = this->reloc_mapper_->get_reloc_target(attr_off, &value);
1366 if (this->reloc_type_ == elfcpp::SHT_REL)
1367 *target_off += value;
1369 *target_off = value;
1373 // Return a string from the DWARF string table.
1376 Dwarf_info_reader::get_string(off_t str_off, unsigned int string_shndx)
1378 if (!this->read_string_table(string_shndx))
1381 // Correct the offset. For incremental update links, we have a
1382 // relocated offset that is relative to the output section, but
1383 // here we need an offset relative to the input section.
1384 str_off -= this->string_output_section_offset_;
1386 const char* p = this->string_buffer_ + str_off;
1388 if (p < this->string_buffer_ || p >= this->string_buffer_end_)
1394 // The following are default, do-nothing, implementations of the
1395 // hook methods normally provided by a derived class. We provide
1396 // default implementations rather than no implementation so that
1397 // a derived class needs to implement only the hooks that it needs
1400 // Process a compilation unit and parse its child DIE.
1403 Dwarf_info_reader::visit_compilation_unit(off_t, off_t, Dwarf_die*)
1407 // Process a type unit and parse its child DIE.
1410 Dwarf_info_reader::visit_type_unit(off_t, off_t, uint64_t, Dwarf_die*)
1414 // class Sized_dwarf_line_info
1416 struct LineStateMachine
1422 unsigned int shndx; // the section address refers to
1423 bool is_stmt; // stmt means statement.
1429 ResetLineStateMachine(struct LineStateMachine* lsm, bool default_is_stmt)
1434 lsm->column_num = 0;
1436 lsm->is_stmt = default_is_stmt;
1437 lsm->basic_block = false;
1438 lsm->end_sequence = false;
1441 template<int size, bool big_endian>
1442 Sized_dwarf_line_info<size, big_endian>::Sized_dwarf_line_info(
1444 unsigned int read_shndx)
1445 : data_valid_(false), buffer_(NULL), buffer_start_(NULL),
1446 reloc_mapper_(NULL), symtab_buffer_(NULL), directories_(), files_(),
1447 current_header_index_(-1)
1449 unsigned int debug_shndx;
1451 for (debug_shndx = 1; debug_shndx < object->shnum(); ++debug_shndx)
1453 // FIXME: do this more efficiently: section_name() isn't super-fast
1454 std::string name = object->section_name(debug_shndx);
1455 if (name == ".debug_line" || name == ".zdebug_line")
1457 section_size_type buffer_size;
1458 bool is_new = false;
1459 this->buffer_ = object->decompressed_section_contents(debug_shndx,
1463 this->buffer_start_ = this->buffer_;
1464 this->buffer_end_ = this->buffer_ + buffer_size;
1468 if (this->buffer_ == NULL)
1471 // Find the relocation section for ".debug_line".
1472 // We expect these for relobjs (.o's) but not dynobjs (.so's).
1473 unsigned int reloc_shndx = 0;
1474 for (unsigned int i = 0; i < object->shnum(); ++i)
1476 unsigned int reloc_sh_type = object->section_type(i);
1477 if ((reloc_sh_type == elfcpp::SHT_REL
1478 || reloc_sh_type == elfcpp::SHT_RELA)
1479 && object->section_info(i) == debug_shndx)
1482 this->track_relocs_type_ = reloc_sh_type;
1487 // Finally, we need the symtab section to interpret the relocs.
1488 if (reloc_shndx != 0)
1490 unsigned int symtab_shndx;
1491 for (symtab_shndx = 0; symtab_shndx < object->shnum(); ++symtab_shndx)
1492 if (object->section_type(symtab_shndx) == elfcpp::SHT_SYMTAB)
1494 this->symtab_buffer_ = object->section_contents(
1495 symtab_shndx, &this->symtab_buffer_size_, false);
1498 if (this->symtab_buffer_ == NULL)
1502 this->reloc_mapper_ =
1503 new Sized_elf_reloc_mapper<size, big_endian>(object,
1504 this->symtab_buffer_,
1505 this->symtab_buffer_size_);
1506 if (!this->reloc_mapper_->initialize(reloc_shndx, this->track_relocs_type_))
1509 // Now that we have successfully read all the data, parse the debug
1511 this->data_valid_ = true;
1512 this->read_line_mappings(read_shndx);
1515 // Read the DWARF header.
1517 template<int size, bool big_endian>
1518 const unsigned char*
1519 Sized_dwarf_line_info<size, big_endian>::read_header_prolog(
1520 const unsigned char* lineptr)
1522 uint32_t initial_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr);
1525 // In DWARF2/3, if the initial length is all 1 bits, then the offset
1526 // size is 8 and we need to read the next 8 bytes for the real length.
1527 if (initial_length == 0xffffffff)
1529 header_.offset_size = 8;
1530 initial_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr);
1534 header_.offset_size = 4;
1536 header_.total_length = initial_length;
1538 gold_assert(lineptr + header_.total_length <= buffer_end_);
1540 header_.version = elfcpp::Swap_unaligned<16, big_endian>::readval(lineptr);
1543 if (header_.offset_size == 4)
1544 header_.prologue_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr);
1546 header_.prologue_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr);
1547 lineptr += header_.offset_size;
1549 header_.min_insn_length = *lineptr;
1552 header_.default_is_stmt = *lineptr;
1555 header_.line_base = *reinterpret_cast<const signed char*>(lineptr);
1558 header_.line_range = *lineptr;
1561 header_.opcode_base = *lineptr;
1564 header_.std_opcode_lengths.resize(header_.opcode_base + 1);
1565 header_.std_opcode_lengths[0] = 0;
1566 for (int i = 1; i < header_.opcode_base; i++)
1568 header_.std_opcode_lengths[i] = *lineptr;
1575 // The header for a debug_line section is mildly complicated, because
1576 // the line info is very tightly encoded.
1578 template<int size, bool big_endian>
1579 const unsigned char*
1580 Sized_dwarf_line_info<size, big_endian>::read_header_tables(
1581 const unsigned char* lineptr)
1583 ++this->current_header_index_;
1585 // Create a new directories_ entry and a new files_ entry for our new
1586 // header. We initialize each with a single empty element, because
1587 // dwarf indexes directory and filenames starting at 1.
1588 gold_assert(static_cast<int>(this->directories_.size())
1589 == this->current_header_index_);
1590 gold_assert(static_cast<int>(this->files_.size())
1591 == this->current_header_index_);
1592 this->directories_.push_back(std::vector<std::string>(1));
1593 this->files_.push_back(std::vector<std::pair<int, std::string> >(1));
1595 // It is legal for the directory entry table to be empty.
1601 const char* dirname = reinterpret_cast<const char*>(lineptr);
1602 gold_assert(dirindex
1603 == static_cast<int>(this->directories_.back().size()));
1604 this->directories_.back().push_back(dirname);
1605 lineptr += this->directories_.back().back().size() + 1;
1611 // It is also legal for the file entry table to be empty.
1618 const char* filename = reinterpret_cast<const char*>(lineptr);
1619 lineptr += strlen(filename) + 1;
1621 uint64_t dirindex = read_unsigned_LEB_128(lineptr, &len);
1624 if (dirindex >= this->directories_.back().size())
1626 int dirindexi = static_cast<int>(dirindex);
1628 read_unsigned_LEB_128(lineptr, &len); // mod_time
1631 read_unsigned_LEB_128(lineptr, &len); // filelength
1634 gold_assert(fileindex
1635 == static_cast<int>(this->files_.back().size()));
1636 this->files_.back().push_back(std::make_pair(dirindexi, filename));
1645 // Process a single opcode in the .debug.line structure.
1647 template<int size, bool big_endian>
1649 Sized_dwarf_line_info<size, big_endian>::process_one_opcode(
1650 const unsigned char* start, struct LineStateMachine* lsm, size_t* len)
1654 unsigned char opcode = *start;
1658 // If the opcode is great than the opcode_base, it is a special
1659 // opcode. Most line programs consist mainly of special opcodes.
1660 if (opcode >= header_.opcode_base)
1662 opcode -= header_.opcode_base;
1663 const int advance_address = ((opcode / header_.line_range)
1664 * header_.min_insn_length);
1665 lsm->address += advance_address;
1667 const int advance_line = ((opcode % header_.line_range)
1668 + header_.line_base);
1669 lsm->line_num += advance_line;
1670 lsm->basic_block = true;
1675 // Otherwise, we have the regular opcodes
1678 case elfcpp::DW_LNS_copy:
1679 lsm->basic_block = false;
1683 case elfcpp::DW_LNS_advance_pc:
1685 const uint64_t advance_address
1686 = read_unsigned_LEB_128(start, &templen);
1688 lsm->address += header_.min_insn_length * advance_address;
1692 case elfcpp::DW_LNS_advance_line:
1694 const uint64_t advance_line = read_signed_LEB_128(start, &templen);
1696 lsm->line_num += advance_line;
1700 case elfcpp::DW_LNS_set_file:
1702 const uint64_t fileno = read_unsigned_LEB_128(start, &templen);
1704 lsm->file_num = fileno;
1708 case elfcpp::DW_LNS_set_column:
1710 const uint64_t colno = read_unsigned_LEB_128(start, &templen);
1712 lsm->column_num = colno;
1716 case elfcpp::DW_LNS_negate_stmt:
1717 lsm->is_stmt = !lsm->is_stmt;
1720 case elfcpp::DW_LNS_set_basic_block:
1721 lsm->basic_block = true;
1724 case elfcpp::DW_LNS_fixed_advance_pc:
1726 int advance_address;
1727 advance_address = elfcpp::Swap_unaligned<16, big_endian>::readval(start);
1729 lsm->address += advance_address;
1733 case elfcpp::DW_LNS_const_add_pc:
1735 const int advance_address = (header_.min_insn_length
1736 * ((255 - header_.opcode_base)
1737 / header_.line_range));
1738 lsm->address += advance_address;
1742 case elfcpp::DW_LNS_extended_op:
1744 const uint64_t extended_op_len
1745 = read_unsigned_LEB_128(start, &templen);
1747 oplen += templen + extended_op_len;
1749 const unsigned char extended_op = *start;
1752 switch (extended_op)
1754 case elfcpp::DW_LNE_end_sequence:
1755 // This means that the current byte is the one immediately
1756 // after a set of instructions. Record the current line
1757 // for up to one less than the current address.
1759 lsm->end_sequence = true;
1763 case elfcpp::DW_LNE_set_address:
1766 elfcpp::Swap_unaligned<size, big_endian>::readval(start);
1767 typename Reloc_map::const_iterator it
1768 = this->reloc_map_.find(start - this->buffer_);
1769 if (it != reloc_map_.end())
1771 // If this is a SHT_RELA section, then ignore the
1772 // section contents. This assumes that this is a
1773 // straight reloc which just uses the reloc addend.
1774 // The reloc addend has already been included in the
1776 if (this->track_relocs_type_ == elfcpp::SHT_RELA)
1778 // Add in the symbol value.
1779 lsm->address += it->second.second;
1780 lsm->shndx = it->second.first;
1784 // If we're a normal .o file, with relocs, every
1785 // set_address should have an associated relocation.
1786 if (this->input_is_relobj())
1787 this->data_valid_ = false;
1791 case elfcpp::DW_LNE_define_file:
1793 const char* filename = reinterpret_cast<const char*>(start);
1794 templen = strlen(filename) + 1;
1797 uint64_t dirindex = read_unsigned_LEB_128(start, &templen);
1799 if (dirindex >= this->directories_.back().size())
1801 int dirindexi = static_cast<int>(dirindex);
1803 // This opcode takes two additional ULEB128 parameters
1804 // (mod_time and filelength), but we don't use those
1805 // values. Because OPLEN already tells us how far to
1806 // skip to the next opcode, we don't need to read
1809 this->files_.back().push_back(std::make_pair(dirindexi,
1819 // Ignore unknown opcode silently
1820 for (int i = 0; i < header_.std_opcode_lengths[opcode]; i++)
1823 read_unsigned_LEB_128(start, &templen);
1834 // Read the debug information at LINEPTR and store it in the line
1837 template<int size, bool big_endian>
1838 unsigned const char*
1839 Sized_dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr,
1842 struct LineStateMachine lsm;
1844 // LENGTHSTART is the place the length field is based on. It is the
1845 // point in the header after the initial length field.
1846 const unsigned char* lengthstart = buffer_;
1848 // In 64 bit dwarf, the initial length is 12 bytes, because of the
1849 // 0xffffffff at the start.
1850 if (header_.offset_size == 8)
1855 while (lineptr < lengthstart + header_.total_length)
1857 ResetLineStateMachine(&lsm, header_.default_is_stmt);
1858 while (!lsm.end_sequence)
1861 bool add_line = this->process_one_opcode(lineptr, &lsm, &oplength);
1863 && (shndx == -1U || lsm.shndx == -1U || shndx == lsm.shndx))
1865 Offset_to_lineno_entry entry
1866 = { static_cast<off_t>(lsm.address),
1867 this->current_header_index_,
1868 static_cast<unsigned int>(lsm.file_num),
1869 true, lsm.line_num };
1870 std::vector<Offset_to_lineno_entry>&
1871 map(this->line_number_map_[lsm.shndx]);
1872 // If we see two consecutive entries with the same
1873 // offset and a real line number, then mark the first
1874 // one as non-canonical.
1876 && (map.back().offset == static_cast<off_t>(lsm.address))
1877 && lsm.line_num != -1
1878 && map.back().line_num != -1)
1879 map.back().last_line_for_offset = false;
1880 map.push_back(entry);
1882 lineptr += oplength;
1886 return lengthstart + header_.total_length;
1889 // Read the relocations into a Reloc_map.
1891 template<int size, bool big_endian>
1893 Sized_dwarf_line_info<size, big_endian>::read_relocs()
1895 if (this->symtab_buffer_ == NULL)
1900 while ((reloc_offset = this->reloc_mapper_->next_offset()) != -1)
1902 const unsigned int shndx =
1903 this->reloc_mapper_->get_reloc_target(reloc_offset, &value);
1905 // There is no reason to record non-ordinary section indexes, or
1906 // SHN_UNDEF, because they will never match the real section.
1908 this->reloc_map_[reloc_offset] = std::make_pair(shndx, value);
1910 this->reloc_mapper_->advance(reloc_offset + 1);
1914 // Read the line number info.
1916 template<int size, bool big_endian>
1918 Sized_dwarf_line_info<size, big_endian>::read_line_mappings(unsigned int shndx)
1920 gold_assert(this->data_valid_ == true);
1922 this->read_relocs();
1923 while (this->buffer_ < this->buffer_end_)
1925 const unsigned char* lineptr = this->buffer_;
1926 lineptr = this->read_header_prolog(lineptr);
1927 lineptr = this->read_header_tables(lineptr);
1928 lineptr = this->read_lines(lineptr, shndx);
1929 this->buffer_ = lineptr;
1932 // Sort the lines numbers, so addr2line can use binary search.
1933 for (typename Lineno_map::iterator it = line_number_map_.begin();
1934 it != line_number_map_.end();
1936 // Each vector needs to be sorted by offset.
1937 std::sort(it->second.begin(), it->second.end());
1940 // Some processing depends on whether the input is a .o file or not.
1941 // For instance, .o files have relocs, and have .debug_lines
1942 // information on a per section basis. .so files, on the other hand,
1943 // lack relocs, and offsets are unique, so we can ignore the section
1946 template<int size, bool big_endian>
1948 Sized_dwarf_line_info<size, big_endian>::input_is_relobj()
1950 // Only .o files have relocs and the symtab buffer that goes with them.
1951 return this->symtab_buffer_ != NULL;
1954 // Given an Offset_to_lineno_entry vector, and an offset, figure out
1955 // if the offset points into a function according to the vector (see
1956 // comments below for the algorithm). If it does, return an iterator
1957 // into the vector that points to the line-number that contains that
1958 // offset. If not, it returns vector::end().
1960 static std::vector<Offset_to_lineno_entry>::const_iterator
1961 offset_to_iterator(const std::vector<Offset_to_lineno_entry>* offsets,
1964 const Offset_to_lineno_entry lookup_key = { offset, 0, 0, true, 0 };
1966 // lower_bound() returns the smallest offset which is >= lookup_key.
1967 // If no offset in offsets is >= lookup_key, returns end().
1968 std::vector<Offset_to_lineno_entry>::const_iterator it
1969 = std::lower_bound(offsets->begin(), offsets->end(), lookup_key);
1971 // This code is easiest to understand with a concrete example.
1972 // Here's a possible offsets array:
1973 // {{offset = 3211, header_num = 0, file_num = 1, last, line_num = 16}, // 0
1974 // {offset = 3224, header_num = 0, file_num = 1, last, line_num = 20}, // 1
1975 // {offset = 3226, header_num = 0, file_num = 1, last, line_num = 22}, // 2
1976 // {offset = 3231, header_num = 0, file_num = 1, last, line_num = 25}, // 3
1977 // {offset = 3232, header_num = 0, file_num = 1, last, line_num = -1}, // 4
1978 // {offset = 3232, header_num = 0, file_num = 1, last, line_num = 65}, // 5
1979 // {offset = 3235, header_num = 0, file_num = 1, last, line_num = 66}, // 6
1980 // {offset = 3236, header_num = 0, file_num = 1, last, line_num = -1}, // 7
1981 // {offset = 5764, header_num = 0, file_num = 1, last, line_num = 48}, // 8
1982 // {offset = 5764, header_num = 0, file_num = 1,!last, line_num = 47}, // 9
1983 // {offset = 5765, header_num = 0, file_num = 1, last, line_num = 49}, // 10
1984 // {offset = 5767, header_num = 0, file_num = 1, last, line_num = 50}, // 11
1985 // {offset = 5768, header_num = 0, file_num = 1, last, line_num = 51}, // 12
1986 // {offset = 5773, header_num = 0, file_num = 1, last, line_num = -1}, // 13
1987 // {offset = 5787, header_num = 1, file_num = 1, last, line_num = 19}, // 14
1988 // {offset = 5790, header_num = 1, file_num = 1, last, line_num = 20}, // 15
1989 // {offset = 5793, header_num = 1, file_num = 1, last, line_num = 67}, // 16
1990 // {offset = 5793, header_num = 1, file_num = 1, last, line_num = -1}, // 17
1991 // {offset = 5793, header_num = 1, file_num = 1,!last, line_num = 66}, // 18
1992 // {offset = 5795, header_num = 1, file_num = 1, last, line_num = 68}, // 19
1993 // {offset = 5798, header_num = 1, file_num = 1, last, line_num = -1}, // 20
1994 // The entries with line_num == -1 mark the end of a function: the
1995 // associated offset is one past the last instruction in the
1996 // function. This can correspond to the beginning of the next
1997 // function (as is true for offset 3232); alternately, there can be
1998 // a gap between the end of one function and the start of the next
1999 // (as is true for some others, most obviously from 3236->5764).
2001 // Case 1: lookup_key has offset == 10. lower_bound returns
2002 // offsets[0]. Since it's not an exact match and we're
2003 // at the beginning of offsets, we return end() (invalid).
2004 // Case 2: lookup_key has offset 10000. lower_bound returns
2005 // offset[21] (end()). We return end() (invalid).
2006 // Case 3: lookup_key has offset == 3211. lower_bound matches
2007 // offsets[0] exactly, and that's the entry we return.
2008 // Case 4: lookup_key has offset == 3232. lower_bound returns
2009 // offsets[4]. That's an exact match, but indicates
2010 // end-of-function. We check if offsets[5] is also an
2011 // exact match but not end-of-function. It is, so we
2012 // return offsets[5].
2013 // Case 5: lookup_key has offset == 3214. lower_bound returns
2014 // offsets[1]. Since it's not an exact match, we back
2015 // up to the offset that's < lookup_key, offsets[0].
2016 // We note offsets[0] is a valid entry (not end-of-function),
2017 // so that's the entry we return.
2018 // Case 6: lookup_key has offset == 4000. lower_bound returns
2019 // offsets[8]. Since it's not an exact match, we back
2020 // up to offsets[7]. Since offsets[7] indicates
2021 // end-of-function, we know lookup_key is between
2022 // functions, so we return end() (not a valid offset).
2023 // Case 7: lookup_key has offset == 5794. lower_bound returns
2024 // offsets[19]. Since it's not an exact match, we back
2025 // up to offsets[16]. Note we back up to the *first*
2026 // entry with offset 5793, not just offsets[19-1].
2027 // We note offsets[16] is a valid entry, so we return it.
2028 // If offsets[16] had had line_num == -1, we would have
2029 // checked offsets[17]. The reason for this is that
2030 // 16 and 17 can be in an arbitrary order, since we sort
2031 // only by offset and last_line_for_offset. (Note it
2032 // doesn't help to use line_number as a tertiary sort key,
2033 // since sometimes we want the -1 to be first and sometimes
2034 // we want it to be last.)
2036 // This deals with cases (1) and (2).
2037 if ((it == offsets->begin() && offset < it->offset)
2038 || it == offsets->end())
2039 return offsets->end();
2041 // This deals with cases (3) and (4).
2042 if (offset == it->offset)
2044 while (it != offsets->end()
2045 && it->offset == offset
2046 && it->line_num == -1)
2048 if (it == offsets->end() || it->offset != offset)
2049 return offsets->end();
2054 // This handles the first part of case (7) -- we back up to the
2055 // *first* entry that has the offset that's behind us.
2056 gold_assert(it != offsets->begin());
2057 std::vector<Offset_to_lineno_entry>::const_iterator range_end = it;
2059 const off_t range_value = it->offset;
2060 while (it != offsets->begin() && (it-1)->offset == range_value)
2063 // This handles cases (5), (6), and (7): if any entry in the
2064 // equal_range [it, range_end) has a line_num != -1, it's a valid
2065 // match. If not, we're not in a function. The line number we saw
2066 // last for an offset will be sorted first, so it'll get returned if
2068 for (; it != range_end; ++it)
2069 if (it->line_num != -1)
2071 return offsets->end();
2074 // Returns the canonical filename:lineno for the address passed in.
2075 // If other_lines is not NULL, appends the non-canonical lines
2076 // assigned to the same address.
2078 template<int size, bool big_endian>
2080 Sized_dwarf_line_info<size, big_endian>::do_addr2line(
2083 std::vector<std::string>* other_lines)
2085 if (this->data_valid_ == false)
2088 const std::vector<Offset_to_lineno_entry>* offsets;
2089 // If we do not have reloc information, then our input is a .so or
2090 // some similar data structure where all the information is held in
2091 // the offset. In that case, we ignore the input shndx.
2092 if (this->input_is_relobj())
2093 offsets = &this->line_number_map_[shndx];
2095 offsets = &this->line_number_map_[-1U];
2096 if (offsets->empty())
2099 typename std::vector<Offset_to_lineno_entry>::const_iterator it
2100 = offset_to_iterator(offsets, offset);
2101 if (it == offsets->end())
2104 std::string result = this->format_file_lineno(*it);
2105 if (other_lines != NULL)
2106 for (++it; it != offsets->end() && it->offset == offset; ++it)
2108 if (it->line_num == -1)
2109 continue; // The end of a previous function.
2110 other_lines->push_back(this->format_file_lineno(*it));
2115 // Convert the file_num + line_num into a string.
2117 template<int size, bool big_endian>
2119 Sized_dwarf_line_info<size, big_endian>::format_file_lineno(
2120 const Offset_to_lineno_entry& loc) const
2124 gold_assert(loc.header_num < static_cast<int>(this->files_.size()));
2125 gold_assert(loc.file_num
2126 < static_cast<unsigned int>(this->files_[loc.header_num].size()));
2127 const std::pair<int, std::string>& filename_pair
2128 = this->files_[loc.header_num][loc.file_num];
2129 const std::string& filename = filename_pair.second;
2131 gold_assert(loc.header_num < static_cast<int>(this->directories_.size()));
2132 gold_assert(filename_pair.first
2133 < static_cast<int>(this->directories_[loc.header_num].size()));
2134 const std::string& dirname
2135 = this->directories_[loc.header_num][filename_pair.first];
2137 if (!dirname.empty())
2146 char buffer[64]; // enough to hold a line number
2147 snprintf(buffer, sizeof(buffer), "%d", loc.line_num);
2154 // Dwarf_line_info routines.
2156 static unsigned int next_generation_count = 0;
2158 struct Addr2line_cache_entry
2162 Dwarf_line_info* dwarf_line_info;
2163 unsigned int generation_count;
2164 unsigned int access_count;
2166 Addr2line_cache_entry(Object* o, unsigned int s, Dwarf_line_info* d)
2167 : object(o), shndx(s), dwarf_line_info(d),
2168 generation_count(next_generation_count), access_count(0)
2170 if (next_generation_count < (1U << 31))
2171 ++next_generation_count;
2174 // We expect this cache to be small, so don't bother with a hashtable
2175 // or priority queue or anything: just use a simple vector.
2176 static std::vector<Addr2line_cache_entry> addr2line_cache;
2179 Dwarf_line_info::one_addr2line(Object* object,
2180 unsigned int shndx, off_t offset,
2182 std::vector<std::string>* other_lines)
2184 Dwarf_line_info* lineinfo = NULL;
2185 std::vector<Addr2line_cache_entry>::iterator it;
2187 // First, check the cache. If we hit, update the counts.
2188 for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it)
2190 if (it->object == object && it->shndx == shndx)
2192 lineinfo = it->dwarf_line_info;
2193 it->generation_count = next_generation_count;
2194 // We cap generation_count at 2^31 -1 to avoid overflow.
2195 if (next_generation_count < (1U << 31))
2196 ++next_generation_count;
2197 // We cap access_count at 31 so 2^access_count doesn't overflow
2198 if (it->access_count < 31)
2204 // If we don't hit the cache, create a new object and insert into the
2206 if (lineinfo == NULL)
2208 switch (parameters->size_and_endianness())
2210 #ifdef HAVE_TARGET_32_LITTLE
2211 case Parameters::TARGET_32_LITTLE:
2212 lineinfo = new Sized_dwarf_line_info<32, false>(object, shndx); break;
2214 #ifdef HAVE_TARGET_32_BIG
2215 case Parameters::TARGET_32_BIG:
2216 lineinfo = new Sized_dwarf_line_info<32, true>(object, shndx); break;
2218 #ifdef HAVE_TARGET_64_LITTLE
2219 case Parameters::TARGET_64_LITTLE:
2220 lineinfo = new Sized_dwarf_line_info<64, false>(object, shndx); break;
2222 #ifdef HAVE_TARGET_64_BIG
2223 case Parameters::TARGET_64_BIG:
2224 lineinfo = new Sized_dwarf_line_info<64, true>(object, shndx); break;
2229 addr2line_cache.push_back(Addr2line_cache_entry(object, shndx, lineinfo));
2232 // Now that we have our object, figure out the answer
2233 std::string retval = lineinfo->addr2line(shndx, offset, other_lines);
2235 // Finally, if our cache has grown too big, delete old objects. We
2236 // assume the common (probably only) case is deleting only one object.
2237 // We use a pretty simple scheme to evict: function of LRU and MFU.
2238 while (addr2line_cache.size() > cache_size)
2240 unsigned int lowest_score = ~0U;
2241 std::vector<Addr2line_cache_entry>::iterator lowest
2242 = addr2line_cache.end();
2243 for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it)
2245 const unsigned int score = (it->generation_count
2246 + (1U << it->access_count));
2247 if (score < lowest_score)
2249 lowest_score = score;
2253 if (lowest != addr2line_cache.end())
2255 delete lowest->dwarf_line_info;
2256 addr2line_cache.erase(lowest);
2264 Dwarf_line_info::clear_addr2line_cache()
2266 for (std::vector<Addr2line_cache_entry>::iterator it = addr2line_cache.begin();
2267 it != addr2line_cache.end();
2269 delete it->dwarf_line_info;
2270 addr2line_cache.clear();
2273 #ifdef HAVE_TARGET_32_LITTLE
2275 class Sized_dwarf_line_info<32, false>;
2278 #ifdef HAVE_TARGET_32_BIG
2280 class Sized_dwarf_line_info<32, true>;
2283 #ifdef HAVE_TARGET_64_LITTLE
2285 class Sized_dwarf_line_info<64, false>;
2288 #ifdef HAVE_TARGET_64_BIG
2290 class Sized_dwarf_line_info<64, true>;
2293 } // End namespace gold.