1 // dwarf_reader.cc -- parse dwarf2/3 debug information
3 // Copyright (C) 2007-2014 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.
29 #include "elfcpp_swap.h"
33 #include "dwarf_reader.h"
34 #include "int_encoding.h"
35 #include "compressed_output.h"
39 // Class Sized_elf_reloc_mapper
41 // Initialize the relocation tracker for section RELOC_SHNDX.
43 template<int size, bool big_endian>
45 Sized_elf_reloc_mapper<size, big_endian>::do_initialize(
46 unsigned int reloc_shndx, unsigned int reloc_type)
48 this->reloc_type_ = reloc_type;
49 return this->track_relocs_.initialize(this->object_, reloc_shndx,
53 // Looks in the symtab to see what section a symbol is in.
55 template<int size, bool big_endian>
57 Sized_elf_reloc_mapper<size, big_endian>::symbol_section(
58 unsigned int symndx, Address* value, bool* is_ordinary)
60 const int symsize = elfcpp::Elf_sizes<size>::sym_size;
61 gold_assert(static_cast<off_t>((symndx + 1) * symsize) <= this->symtab_size_);
62 elfcpp::Sym<size, big_endian> elfsym(this->symtab_ + symndx * symsize);
63 *value = elfsym.get_st_value();
64 return this->object_->adjust_sym_shndx(symndx, elfsym.get_st_shndx(),
68 // Return the section index and offset within the section of
69 // the target of the relocation for RELOC_OFFSET.
71 template<int size, bool big_endian>
73 Sized_elf_reloc_mapper<size, big_endian>::do_get_reloc_target(
74 off_t reloc_offset, off_t* target_offset)
76 this->track_relocs_.advance(reloc_offset);
77 if (reloc_offset != this->track_relocs_.next_offset())
79 unsigned int symndx = this->track_relocs_.next_symndx();
80 typename elfcpp::Elf_types<size>::Elf_Addr value;
82 unsigned int target_shndx = this->symbol_section(symndx, &value,
86 if (this->reloc_type_ == elfcpp::SHT_RELA)
87 value += this->track_relocs_.next_addend();
88 *target_offset = value;
92 static inline Elf_reloc_mapper*
93 make_elf_reloc_mapper(Relobj* object, const unsigned char* symtab,
96 if (object->elfsize() == 32)
98 if (object->is_big_endian())
100 #ifdef HAVE_TARGET_32_BIG
101 return new Sized_elf_reloc_mapper<32, true>(object, symtab,
109 #ifdef HAVE_TARGET_32_LITTLE
110 return new Sized_elf_reloc_mapper<32, false>(object, symtab,
117 else if (object->elfsize() == 64)
119 if (object->is_big_endian())
121 #ifdef HAVE_TARGET_64_BIG
122 return new Sized_elf_reloc_mapper<64, true>(object, symtab,
130 #ifdef HAVE_TARGET_64_LITTLE
131 return new Sized_elf_reloc_mapper<64, false>(object, symtab,
142 // class Dwarf_abbrev_table
145 Dwarf_abbrev_table::clear_abbrev_codes()
147 for (unsigned int code = 0; code < this->low_abbrev_code_max_; ++code)
149 if (this->low_abbrev_codes_[code] != NULL)
151 delete this->low_abbrev_codes_[code];
152 this->low_abbrev_codes_[code] = NULL;
155 for (Abbrev_code_table::iterator it = this->high_abbrev_codes_.begin();
156 it != this->high_abbrev_codes_.end();
159 if (it->second != NULL)
162 this->high_abbrev_codes_.clear();
165 // Read the abbrev table from an object file.
168 Dwarf_abbrev_table::do_read_abbrevs(
170 unsigned int abbrev_shndx,
173 this->clear_abbrev_codes();
175 // If we don't have relocations, abbrev_shndx will be 0, and
176 // we'll have to hunt for the .debug_abbrev section.
177 if (abbrev_shndx == 0 && this->abbrev_shndx_ > 0)
178 abbrev_shndx = this->abbrev_shndx_;
179 else if (abbrev_shndx == 0)
181 for (unsigned int i = 1; i < object->shnum(); ++i)
183 std::string name = object->section_name(i);
184 if (name == ".debug_abbrev" || name == ".zdebug_abbrev")
187 // Correct the offset. For incremental update links, we have a
188 // relocated offset that is relative to the output section, but
189 // here we need an offset relative to the input section.
190 abbrev_offset -= object->output_section_offset(i);
194 if (abbrev_shndx == 0)
198 // Get the section contents and decompress if necessary.
199 if (abbrev_shndx != this->abbrev_shndx_)
201 if (this->owns_buffer_ && this->buffer_ != NULL)
203 delete[] this->buffer_;
204 this->owns_buffer_ = false;
207 section_size_type buffer_size;
209 object->decompressed_section_contents(abbrev_shndx,
211 &this->owns_buffer_);
212 this->buffer_end_ = this->buffer_ + buffer_size;
213 this->abbrev_shndx_ = abbrev_shndx;
216 this->buffer_pos_ = this->buffer_ + abbrev_offset;
220 // Lookup the abbrev code entry for CODE. This function is called
221 // only when the abbrev code is not in the direct lookup table.
222 // It may be in the hash table, it may not have been read yet,
223 // or it may not exist in the abbrev table.
225 const Dwarf_abbrev_table::Abbrev_code*
226 Dwarf_abbrev_table::do_get_abbrev(unsigned int code)
228 // See if the abbrev code is already in the hash table.
229 Abbrev_code_table::const_iterator it = this->high_abbrev_codes_.find(code);
230 if (it != this->high_abbrev_codes_.end())
233 // Read and store abbrev code definitions until we find the
234 // one we're looking for.
237 // Read the abbrev code. A zero here indicates the end of the
240 if (this->buffer_pos_ >= this->buffer_end_)
242 uint64_t nextcode = read_unsigned_LEB_128(this->buffer_pos_, &len);
245 this->buffer_pos_ = this->buffer_end_;
248 this->buffer_pos_ += len;
251 if (this->buffer_pos_ >= this->buffer_end_)
253 uint64_t tag = read_unsigned_LEB_128(this->buffer_pos_, &len);
254 this->buffer_pos_ += len;
256 // Read the has_children flag.
257 if (this->buffer_pos_ >= this->buffer_end_)
259 bool has_children = *this->buffer_pos_ == elfcpp::DW_CHILDREN_yes;
260 this->buffer_pos_ += 1;
262 // Read the list of (attribute, form) pairs.
263 Abbrev_code* entry = new Abbrev_code(tag, has_children);
266 // Read the attribute.
267 if (this->buffer_pos_ >= this->buffer_end_)
269 uint64_t attr = read_unsigned_LEB_128(this->buffer_pos_, &len);
270 this->buffer_pos_ += len;
273 if (this->buffer_pos_ >= this->buffer_end_)
275 uint64_t form = read_unsigned_LEB_128(this->buffer_pos_, &len);
276 this->buffer_pos_ += len;
278 // A (0,0) pair terminates the list.
279 if (attr == 0 && form == 0)
282 if (attr == elfcpp::DW_AT_sibling)
283 entry->has_sibling_attribute = true;
285 entry->add_attribute(attr, form);
288 this->store_abbrev(nextcode, entry);
289 if (nextcode == code)
296 // class Dwarf_ranges_table
298 // Read the ranges table from an object file.
301 Dwarf_ranges_table::read_ranges_table(
303 const unsigned char* symtab,
305 unsigned int ranges_shndx)
307 // If we've already read this abbrev table, return immediately.
308 if (this->ranges_shndx_ > 0
309 && this->ranges_shndx_ == ranges_shndx)
312 // If we don't have relocations, ranges_shndx will be 0, and
313 // we'll have to hunt for the .debug_ranges section.
314 if (ranges_shndx == 0 && this->ranges_shndx_ > 0)
315 ranges_shndx = this->ranges_shndx_;
316 else if (ranges_shndx == 0)
318 for (unsigned int i = 1; i < object->shnum(); ++i)
320 std::string name = object->section_name(i);
321 if (name == ".debug_ranges" || name == ".zdebug_ranges")
324 this->output_section_offset_ = object->output_section_offset(i);
328 if (ranges_shndx == 0)
332 // Get the section contents and decompress if necessary.
333 if (ranges_shndx != this->ranges_shndx_)
335 if (this->owns_ranges_buffer_ && this->ranges_buffer_ != NULL)
337 delete[] this->ranges_buffer_;
338 this->owns_ranges_buffer_ = false;
341 section_size_type buffer_size;
342 this->ranges_buffer_ =
343 object->decompressed_section_contents(ranges_shndx,
345 &this->owns_ranges_buffer_);
346 this->ranges_buffer_end_ = this->ranges_buffer_ + buffer_size;
347 this->ranges_shndx_ = ranges_shndx;
350 if (this->ranges_reloc_mapper_ != NULL)
352 delete this->ranges_reloc_mapper_;
353 this->ranges_reloc_mapper_ = NULL;
356 // For incremental objects, we have no relocations.
357 if (object->is_incremental())
360 // Find the relocation section for ".debug_ranges".
361 unsigned int reloc_shndx = 0;
362 unsigned int reloc_type = 0;
363 for (unsigned int i = 0; i < object->shnum(); ++i)
365 reloc_type = object->section_type(i);
366 if ((reloc_type == elfcpp::SHT_REL
367 || reloc_type == elfcpp::SHT_RELA)
368 && object->section_info(i) == ranges_shndx)
375 this->ranges_reloc_mapper_ = make_elf_reloc_mapper(object, symtab,
377 this->ranges_reloc_mapper_->initialize(reloc_shndx, reloc_type);
378 this->reloc_type_ = reloc_type;
383 // Read a range list from section RANGES_SHNDX at offset RANGES_OFFSET.
386 Dwarf_ranges_table::read_range_list(
388 const unsigned char* symtab,
390 unsigned int addr_size,
391 unsigned int ranges_shndx,
394 Dwarf_range_list* ranges;
396 if (!this->read_ranges_table(object, symtab, symtab_size, ranges_shndx))
399 // Correct the offset. For incremental update links, we have a
400 // relocated offset that is relative to the output section, but
401 // here we need an offset relative to the input section.
402 offset -= this->output_section_offset_;
404 // Read the range list at OFFSET.
405 ranges = new Dwarf_range_list();
408 this->ranges_buffer_ + offset < this->ranges_buffer_end_;
409 offset += 2 * addr_size)
414 // Read the raw contents of the section.
417 start = this->dwinfo_->read_from_pointer<32>(this->ranges_buffer_
419 end = this->dwinfo_->read_from_pointer<32>(this->ranges_buffer_
424 start = this->dwinfo_->read_from_pointer<64>(this->ranges_buffer_
426 end = this->dwinfo_->read_from_pointer<64>(this->ranges_buffer_
430 // Check for relocations and adjust the values.
431 unsigned int shndx1 = 0;
432 unsigned int shndx2 = 0;
433 if (this->ranges_reloc_mapper_ != NULL)
435 shndx1 = this->lookup_reloc(offset, &start);
436 shndx2 = this->lookup_reloc(offset + addr_size, &end);
439 // End of list is marked by a pair of zeroes.
440 if (shndx1 == 0 && start == 0 && end == 0)
443 // A "base address selection entry" is identified by
444 // 0xffffffff for the first value of the pair. The second
445 // value is used as a base for subsequent range list entries.
446 if (shndx1 == 0 && start == -1)
448 else if (shndx1 == shndx2)
450 if (shndx1 == 0 || object->is_section_included(shndx1))
451 ranges->add(shndx1, base + start, base + end);
454 gold_warning(_("%s: DWARF info may be corrupt; offsets in a "
455 "range list entry are in different sections"),
456 object->name().c_str());
462 // Look for a relocation at offset OFF in the range table,
463 // and return the section index and offset of the target.
466 Dwarf_ranges_table::lookup_reloc(off_t off, off_t* target_off)
470 this->ranges_reloc_mapper_->get_reloc_target(off, &value);
473 if (this->reloc_type_ == elfcpp::SHT_REL)
474 *target_off += value;
480 // class Dwarf_pubnames_table
482 // Read the pubnames section from the object file.
485 Dwarf_pubnames_table::read_section(Relobj* object, const unsigned char* symtab,
488 section_size_type buffer_size;
489 unsigned int shndx = 0;
490 const char* name = this->is_pubtypes_ ? "pubtypes" : "pubnames";
491 const char* gnu_name = (this->is_pubtypes_
495 for (unsigned int i = 1; i < object->shnum(); ++i)
497 std::string section_name = object->section_name(i);
498 const char* section_name_suffix = section_name.c_str();
499 if (is_prefix_of(".debug_", section_name_suffix))
500 section_name_suffix += 7;
501 else if (is_prefix_of(".zdebug_", section_name_suffix))
502 section_name_suffix += 8;
505 if (strcmp(section_name_suffix, name) == 0)
508 this->output_section_offset_ = object->output_section_offset(i);
511 else if (strcmp(section_name_suffix, gnu_name) == 0)
514 this->output_section_offset_ = object->output_section_offset(i);
515 this->is_gnu_style_ = true;
522 this->buffer_ = object->decompressed_section_contents(shndx,
524 &this->owns_buffer_);
525 if (this->buffer_ == NULL)
527 this->buffer_end_ = this->buffer_ + buffer_size;
529 // For incremental objects, we have no relocations.
530 if (object->is_incremental())
533 // Find the relocation section
534 unsigned int reloc_shndx = 0;
535 unsigned int reloc_type = 0;
536 for (unsigned int i = 0; i < object->shnum(); ++i)
538 reloc_type = object->section_type(i);
539 if ((reloc_type == elfcpp::SHT_REL
540 || reloc_type == elfcpp::SHT_RELA)
541 && object->section_info(i) == shndx)
548 this->reloc_mapper_ = make_elf_reloc_mapper(object, symtab, symtab_size);
549 this->reloc_mapper_->initialize(reloc_shndx, reloc_type);
550 this->reloc_type_ = reloc_type;
555 // Read the header for the set at OFFSET.
558 Dwarf_pubnames_table::read_header(off_t offset)
560 // Make sure we have actually read the section.
561 gold_assert(this->buffer_ != NULL);
563 // Correct the offset. For incremental update links, we have a
564 // relocated offset that is relative to the output section, but
565 // here we need an offset relative to the input section.
566 offset -= this->output_section_offset_;
568 if (offset < 0 || offset + 14 >= this->buffer_end_ - this->buffer_)
571 const unsigned char* pinfo = this->buffer_ + offset;
573 // Read the unit_length field.
574 uint64_t unit_length = this->dwinfo_->read_from_pointer<32>(pinfo);
576 if (unit_length == 0xffffffff)
578 unit_length = this->dwinfo_->read_from_pointer<64>(pinfo);
579 this->unit_length_ = unit_length + 12;
581 this->offset_size_ = 8;
585 this->unit_length_ = unit_length + 4;
586 this->offset_size_ = 4;
588 this->end_of_table_ = pinfo + unit_length;
590 // Check the version.
591 unsigned int version = this->dwinfo_->read_from_pointer<16>(pinfo);
596 this->reloc_mapper_->get_reloc_target(pinfo - this->buffer_,
599 // Skip the debug_info_offset and debug_info_size fields.
600 pinfo += 2 * this->offset_size_;
602 if (pinfo >= this->buffer_end_)
605 this->pinfo_ = pinfo;
609 // Read the next name from the set.
612 Dwarf_pubnames_table::next_name(uint8_t* flag_byte)
614 const unsigned char* pinfo = this->pinfo_;
616 // Check for end of list. The table should be terminated by an
617 // entry containing nothing but a DIE offset of 0.
618 if (pinfo + this->offset_size_ >= this->end_of_table_)
621 // Skip the offset within the CU. If this is zero, but we're not
622 // at the end of the table, then we have a real pubnames entry
623 // whose DIE offset is 0 (likely to be a GCC bug). Since we
624 // don't actually use the DIE offset in building .gdb_index,
626 pinfo += this->offset_size_;
628 if (this->is_gnu_style_)
629 *flag_byte = *pinfo++;
633 // Return a pointer to the string at the current location,
634 // and advance the pointer to the next entry.
635 const char* ret = reinterpret_cast<const char*>(pinfo);
636 while (pinfo < this->buffer_end_ && *pinfo != '\0')
638 if (pinfo < this->buffer_end_)
641 this->pinfo_ = pinfo;
647 Dwarf_die::Dwarf_die(
648 Dwarf_info_reader* dwinfo,
651 : dwinfo_(dwinfo), parent_(parent), die_offset_(die_offset),
652 child_offset_(0), sibling_offset_(0), abbrev_code_(NULL), attributes_(),
653 attributes_read_(false), name_(NULL), name_off_(-1), linkage_name_(NULL),
654 linkage_name_off_(-1), string_shndx_(0), specification_(0),
658 const unsigned char* pdie = dwinfo->buffer_at_offset(die_offset);
661 unsigned int code = read_unsigned_LEB_128(pdie, &len);
665 parent->set_sibling_offset(die_offset + len);
668 this->attr_offset_ = len;
670 // Lookup the abbrev code in the abbrev table.
671 this->abbrev_code_ = dwinfo->get_abbrev(code);
674 // Read all the attributes of the DIE.
677 Dwarf_die::read_attributes()
679 if (this->attributes_read_)
682 gold_assert(this->abbrev_code_ != NULL);
684 const unsigned char* pdie =
685 this->dwinfo_->buffer_at_offset(this->die_offset_);
688 const unsigned char* pattr = pdie + this->attr_offset_;
690 unsigned int nattr = this->abbrev_code_->attributes.size();
691 this->attributes_.reserve(nattr);
692 for (unsigned int i = 0; i < nattr; ++i)
695 unsigned int attr = this->abbrev_code_->attributes[i].attr;
696 unsigned int form = this->abbrev_code_->attributes[i].form;
697 if (form == elfcpp::DW_FORM_indirect)
699 form = read_unsigned_LEB_128(pattr, &len);
702 off_t attr_off = this->die_offset_ + (pattr - pdie);
703 bool ref_form = false;
704 Attribute_value attr_value;
705 attr_value.attr = attr;
706 attr_value.form = form;
707 attr_value.aux.shndx = 0;
710 case elfcpp::DW_FORM_flag_present:
711 attr_value.val.intval = 1;
713 case elfcpp::DW_FORM_strp:
716 if (this->dwinfo_->offset_size() == 4)
717 str_off = this->dwinfo_->read_from_pointer<32>(&pattr);
719 str_off = this->dwinfo_->read_from_pointer<64>(&pattr);
721 this->dwinfo_->lookup_reloc(attr_off, &str_off);
722 attr_value.aux.shndx = shndx;
723 attr_value.val.refval = str_off;
726 case elfcpp::DW_FORM_sec_offset:
729 if (this->dwinfo_->offset_size() == 4)
730 sec_off = this->dwinfo_->read_from_pointer<32>(&pattr);
732 sec_off = this->dwinfo_->read_from_pointer<64>(&pattr);
734 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
735 attr_value.aux.shndx = shndx;
736 attr_value.val.refval = sec_off;
740 case elfcpp::DW_FORM_addr:
741 case elfcpp::DW_FORM_ref_addr:
744 if (this->dwinfo_->address_size() == 4)
745 sec_off = this->dwinfo_->read_from_pointer<32>(&pattr);
747 sec_off = this->dwinfo_->read_from_pointer<64>(&pattr);
749 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
750 attr_value.aux.shndx = shndx;
751 attr_value.val.refval = sec_off;
755 case elfcpp::DW_FORM_block1:
756 attr_value.aux.blocklen = *pattr++;
757 attr_value.val.blockval = pattr;
758 pattr += attr_value.aux.blocklen;
760 case elfcpp::DW_FORM_block2:
761 attr_value.aux.blocklen =
762 this->dwinfo_->read_from_pointer<16>(&pattr);
763 attr_value.val.blockval = pattr;
764 pattr += attr_value.aux.blocklen;
766 case elfcpp::DW_FORM_block4:
767 attr_value.aux.blocklen =
768 this->dwinfo_->read_from_pointer<32>(&pattr);
769 attr_value.val.blockval = pattr;
770 pattr += attr_value.aux.blocklen;
772 case elfcpp::DW_FORM_block:
773 case elfcpp::DW_FORM_exprloc:
774 attr_value.aux.blocklen = read_unsigned_LEB_128(pattr, &len);
775 attr_value.val.blockval = pattr + len;
776 pattr += len + attr_value.aux.blocklen;
778 case elfcpp::DW_FORM_data1:
779 case elfcpp::DW_FORM_flag:
780 attr_value.val.intval = *pattr++;
782 case elfcpp::DW_FORM_ref1:
783 attr_value.val.refval = *pattr++;
786 case elfcpp::DW_FORM_data2:
787 attr_value.val.intval =
788 this->dwinfo_->read_from_pointer<16>(&pattr);
790 case elfcpp::DW_FORM_ref2:
791 attr_value.val.refval =
792 this->dwinfo_->read_from_pointer<16>(&pattr);
795 case elfcpp::DW_FORM_data4:
798 sec_off = this->dwinfo_->read_from_pointer<32>(&pattr);
800 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
801 attr_value.aux.shndx = shndx;
802 attr_value.val.intval = sec_off;
805 case elfcpp::DW_FORM_ref4:
808 sec_off = this->dwinfo_->read_from_pointer<32>(&pattr);
810 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
811 attr_value.aux.shndx = shndx;
812 attr_value.val.refval = sec_off;
816 case elfcpp::DW_FORM_data8:
819 sec_off = this->dwinfo_->read_from_pointer<64>(&pattr);
821 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
822 attr_value.aux.shndx = shndx;
823 attr_value.val.intval = sec_off;
826 case elfcpp::DW_FORM_ref_sig8:
827 attr_value.val.uintval =
828 this->dwinfo_->read_from_pointer<64>(&pattr);
830 case elfcpp::DW_FORM_ref8:
833 sec_off = this->dwinfo_->read_from_pointer<64>(&pattr);
835 this->dwinfo_->lookup_reloc(attr_off, &sec_off);
836 attr_value.aux.shndx = shndx;
837 attr_value.val.refval = sec_off;
841 case elfcpp::DW_FORM_ref_udata:
842 attr_value.val.refval = read_unsigned_LEB_128(pattr, &len);
846 case elfcpp::DW_FORM_udata:
847 case elfcpp::DW_FORM_GNU_addr_index:
848 case elfcpp::DW_FORM_GNU_str_index:
849 attr_value.val.uintval = read_unsigned_LEB_128(pattr, &len);
852 case elfcpp::DW_FORM_sdata:
853 attr_value.val.intval = read_signed_LEB_128(pattr, &len);
856 case elfcpp::DW_FORM_string:
857 attr_value.val.stringval = reinterpret_cast<const char*>(pattr);
858 len = strlen(attr_value.val.stringval);
865 // Cache the most frequently-requested attributes.
868 case elfcpp::DW_AT_name:
869 if (form == elfcpp::DW_FORM_string)
870 this->name_ = attr_value.val.stringval;
871 else if (form == elfcpp::DW_FORM_strp)
873 // All indirect strings should refer to the same
874 // string section, so we just save the last one seen.
875 this->string_shndx_ = attr_value.aux.shndx;
876 this->name_off_ = attr_value.val.refval;
879 case elfcpp::DW_AT_linkage_name:
880 case elfcpp::DW_AT_MIPS_linkage_name:
881 if (form == elfcpp::DW_FORM_string)
882 this->linkage_name_ = attr_value.val.stringval;
883 else if (form == elfcpp::DW_FORM_strp)
885 // All indirect strings should refer to the same
886 // string section, so we just save the last one seen.
887 this->string_shndx_ = attr_value.aux.shndx;
888 this->linkage_name_off_ = attr_value.val.refval;
891 case elfcpp::DW_AT_specification:
893 this->specification_ = attr_value.val.refval;
895 case elfcpp::DW_AT_abstract_origin:
897 this->abstract_origin_ = attr_value.val.refval;
899 case elfcpp::DW_AT_sibling:
900 if (ref_form && attr_value.aux.shndx == 0)
901 this->sibling_offset_ = attr_value.val.refval;
906 this->attributes_.push_back(attr_value);
909 // Now that we know where the next DIE begins, record the offset
910 // to avoid later recalculation.
911 if (this->has_children())
912 this->child_offset_ = this->die_offset_ + (pattr - pdie);
914 this->sibling_offset_ = this->die_offset_ + (pattr - pdie);
916 this->attributes_read_ = true;
920 // Skip all the attributes of the DIE and return the offset of the next DIE.
923 Dwarf_die::skip_attributes()
925 gold_assert(this->abbrev_code_ != NULL);
927 const unsigned char* pdie =
928 this->dwinfo_->buffer_at_offset(this->die_offset_);
931 const unsigned char* pattr = pdie + this->attr_offset_;
933 for (unsigned int i = 0; i < this->abbrev_code_->attributes.size(); ++i)
936 unsigned int form = this->abbrev_code_->attributes[i].form;
937 if (form == elfcpp::DW_FORM_indirect)
939 form = read_unsigned_LEB_128(pattr, &len);
944 case elfcpp::DW_FORM_flag_present:
946 case elfcpp::DW_FORM_strp:
947 case elfcpp::DW_FORM_sec_offset:
948 pattr += this->dwinfo_->offset_size();
950 case elfcpp::DW_FORM_addr:
951 case elfcpp::DW_FORM_ref_addr:
952 pattr += this->dwinfo_->address_size();
954 case elfcpp::DW_FORM_block1:
957 case elfcpp::DW_FORM_block2:
960 block_size = this->dwinfo_->read_from_pointer<16>(&pattr);
964 case elfcpp::DW_FORM_block4:
967 block_size = this->dwinfo_->read_from_pointer<32>(&pattr);
971 case elfcpp::DW_FORM_block:
972 case elfcpp::DW_FORM_exprloc:
975 block_size = read_unsigned_LEB_128(pattr, &len);
976 pattr += len + block_size;
979 case elfcpp::DW_FORM_data1:
980 case elfcpp::DW_FORM_ref1:
981 case elfcpp::DW_FORM_flag:
984 case elfcpp::DW_FORM_data2:
985 case elfcpp::DW_FORM_ref2:
988 case elfcpp::DW_FORM_data4:
989 case elfcpp::DW_FORM_ref4:
992 case elfcpp::DW_FORM_data8:
993 case elfcpp::DW_FORM_ref8:
994 case elfcpp::DW_FORM_ref_sig8:
997 case elfcpp::DW_FORM_ref_udata:
998 case elfcpp::DW_FORM_udata:
999 case elfcpp::DW_FORM_GNU_addr_index:
1000 case elfcpp::DW_FORM_GNU_str_index:
1001 read_unsigned_LEB_128(pattr, &len);
1004 case elfcpp::DW_FORM_sdata:
1005 read_signed_LEB_128(pattr, &len);
1008 case elfcpp::DW_FORM_string:
1009 len = strlen(reinterpret_cast<const char*>(pattr));
1017 return this->die_offset_ + (pattr - pdie);
1020 // Get the name of the DIE and cache it.
1023 Dwarf_die::set_name()
1025 if (this->name_ != NULL || !this->read_attributes())
1027 if (this->name_off_ != -1)
1028 this->name_ = this->dwinfo_->get_string(this->name_off_,
1029 this->string_shndx_);
1032 // Get the linkage name of the DIE and cache it.
1035 Dwarf_die::set_linkage_name()
1037 if (this->linkage_name_ != NULL || !this->read_attributes())
1039 if (this->linkage_name_off_ != -1)
1040 this->linkage_name_ = this->dwinfo_->get_string(this->linkage_name_off_,
1041 this->string_shndx_);
1044 // Return the value of attribute ATTR.
1046 const Dwarf_die::Attribute_value*
1047 Dwarf_die::attribute(unsigned int attr)
1049 if (!this->read_attributes())
1051 for (unsigned int i = 0; i < this->attributes_.size(); ++i)
1053 if (this->attributes_[i].attr == attr)
1054 return &this->attributes_[i];
1060 Dwarf_die::string_attribute(unsigned int attr)
1062 const Attribute_value* attr_val = this->attribute(attr);
1063 if (attr_val == NULL)
1065 switch (attr_val->form)
1067 case elfcpp::DW_FORM_string:
1068 return attr_val->val.stringval;
1069 case elfcpp::DW_FORM_strp:
1070 return this->dwinfo_->get_string(attr_val->val.refval,
1071 attr_val->aux.shndx);
1078 Dwarf_die::int_attribute(unsigned int attr)
1080 const Attribute_value* attr_val = this->attribute(attr);
1081 if (attr_val == NULL)
1083 switch (attr_val->form)
1085 case elfcpp::DW_FORM_flag_present:
1086 case elfcpp::DW_FORM_data1:
1087 case elfcpp::DW_FORM_flag:
1088 case elfcpp::DW_FORM_data2:
1089 case elfcpp::DW_FORM_data4:
1090 case elfcpp::DW_FORM_data8:
1091 case elfcpp::DW_FORM_sdata:
1092 return attr_val->val.intval;
1099 Dwarf_die::uint_attribute(unsigned int attr)
1101 const Attribute_value* attr_val = this->attribute(attr);
1102 if (attr_val == NULL)
1104 switch (attr_val->form)
1106 case elfcpp::DW_FORM_flag_present:
1107 case elfcpp::DW_FORM_data1:
1108 case elfcpp::DW_FORM_flag:
1109 case elfcpp::DW_FORM_data4:
1110 case elfcpp::DW_FORM_data8:
1111 case elfcpp::DW_FORM_ref_sig8:
1112 case elfcpp::DW_FORM_udata:
1113 return attr_val->val.uintval;
1120 Dwarf_die::ref_attribute(unsigned int attr, unsigned int* shndx)
1122 const Attribute_value* attr_val = this->attribute(attr);
1123 if (attr_val == NULL)
1125 switch (attr_val->form)
1127 case elfcpp::DW_FORM_sec_offset:
1128 case elfcpp::DW_FORM_addr:
1129 case elfcpp::DW_FORM_ref_addr:
1130 case elfcpp::DW_FORM_ref1:
1131 case elfcpp::DW_FORM_ref2:
1132 case elfcpp::DW_FORM_ref4:
1133 case elfcpp::DW_FORM_ref8:
1134 case elfcpp::DW_FORM_ref_udata:
1135 *shndx = attr_val->aux.shndx;
1136 return attr_val->val.refval;
1137 case elfcpp::DW_FORM_ref_sig8:
1138 *shndx = attr_val->aux.shndx;
1139 return attr_val->val.uintval;
1140 case elfcpp::DW_FORM_data4:
1141 case elfcpp::DW_FORM_data8:
1142 *shndx = attr_val->aux.shndx;
1143 return attr_val->val.intval;
1150 Dwarf_die::address_attribute(unsigned int attr, unsigned int* shndx)
1152 const Attribute_value* attr_val = this->attribute(attr);
1153 if (attr_val == NULL || attr_val->form != elfcpp::DW_FORM_addr)
1156 *shndx = attr_val->aux.shndx;
1157 return attr_val->val.refval;
1160 // Return the offset of this DIE's first child.
1163 Dwarf_die::child_offset()
1165 gold_assert(this->abbrev_code_ != NULL);
1166 if (!this->has_children())
1168 if (this->child_offset_ == 0)
1169 this->child_offset_ = this->skip_attributes();
1170 return this->child_offset_;
1173 // Return the offset of this DIE's next sibling.
1176 Dwarf_die::sibling_offset()
1178 gold_assert(this->abbrev_code_ != NULL);
1180 if (this->sibling_offset_ != 0)
1181 return this->sibling_offset_;
1183 if (!this->has_children())
1185 this->sibling_offset_ = this->skip_attributes();
1186 return this->sibling_offset_;
1189 if (this->has_sibling_attribute())
1191 if (!this->read_attributes())
1193 if (this->sibling_offset_ != 0)
1194 return this->sibling_offset_;
1197 // Skip over the children.
1198 off_t child_offset = this->child_offset();
1199 while (child_offset > 0)
1201 Dwarf_die die(this->dwinfo_, child_offset, this);
1202 // The Dwarf_die ctor will set this DIE's sibling offset
1203 // when it reads a zero abbrev code.
1206 child_offset = die.sibling_offset();
1209 // This should be set by now. If not, there was a problem reading
1210 // the DWARF info, and we return 0.
1211 return this->sibling_offset_;
1214 // class Dwarf_info_reader
1216 // Begin parsing the debug info. This calls visit_compilation_unit()
1217 // or visit_type_unit() for each compilation or type unit found in the
1218 // section, and visit_die() for each top-level DIE.
1221 Dwarf_info_reader::parse()
1223 if (this->object_->is_big_endian())
1225 #if defined(HAVE_TARGET_32_BIG) || defined(HAVE_TARGET_64_BIG)
1226 this->do_parse<true>();
1233 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_64_LITTLE)
1234 this->do_parse<false>();
1241 template<bool big_endian>
1243 Dwarf_info_reader::do_parse()
1245 // Get the section contents and decompress if necessary.
1246 section_size_type buffer_size;
1248 this->buffer_ = this->object_->decompressed_section_contents(this->shndx_,
1251 if (this->buffer_ == NULL || buffer_size == 0)
1253 this->buffer_end_ = this->buffer_ + buffer_size;
1255 // The offset of this input section in the output section.
1256 off_t section_offset = this->object_->output_section_offset(this->shndx_);
1258 // Start tracking relocations for this section.
1259 this->reloc_mapper_ = make_elf_reloc_mapper(this->object_, this->symtab_,
1260 this->symtab_size_);
1261 this->reloc_mapper_->initialize(this->reloc_shndx_, this->reloc_type_);
1263 // Loop over compilation units (or type units).
1264 unsigned int abbrev_shndx = this->abbrev_shndx_;
1265 off_t abbrev_offset = 0;
1266 const unsigned char* pinfo = this->buffer_;
1267 while (pinfo < this->buffer_end_)
1269 // Read the compilation (or type) unit header.
1270 const unsigned char* cu_start = pinfo;
1271 this->cu_offset_ = cu_start - this->buffer_;
1272 this->cu_length_ = this->buffer_end_ - cu_start;
1274 // Read unit_length (4 or 12 bytes).
1275 if (!this->check_buffer(pinfo + 4))
1277 uint32_t unit_length =
1278 elfcpp::Swap_unaligned<32, big_endian>::readval(pinfo);
1280 if (unit_length == 0xffffffff)
1282 if (!this->check_buffer(pinfo + 8))
1284 unit_length = elfcpp::Swap_unaligned<64, big_endian>::readval(pinfo);
1286 this->offset_size_ = 8;
1289 this->offset_size_ = 4;
1290 if (!this->check_buffer(pinfo + unit_length))
1292 const unsigned char* cu_end = pinfo + unit_length;
1293 this->cu_length_ = cu_end - cu_start;
1294 if (!this->check_buffer(pinfo + 2 + this->offset_size_ + 1))
1297 // Read version (2 bytes).
1299 elfcpp::Swap_unaligned<16, big_endian>::readval(pinfo);
1302 // Read debug_abbrev_offset (4 or 8 bytes).
1303 if (this->offset_size_ == 4)
1304 abbrev_offset = elfcpp::Swap_unaligned<32, big_endian>::readval(pinfo);
1306 abbrev_offset = elfcpp::Swap_unaligned<64, big_endian>::readval(pinfo);
1307 if (this->reloc_shndx_ > 0)
1309 off_t reloc_offset = pinfo - this->buffer_;
1312 this->reloc_mapper_->get_reloc_target(reloc_offset, &value);
1313 if (abbrev_shndx == 0)
1315 if (this->reloc_type_ == elfcpp::SHT_REL)
1316 abbrev_offset += value;
1318 abbrev_offset = value;
1320 pinfo += this->offset_size_;
1322 // Read address_size (1 byte).
1323 this->address_size_ = *pinfo++;
1325 // For type units, read the two extra fields.
1326 uint64_t signature = 0;
1327 off_t type_offset = 0;
1328 if (this->is_type_unit_)
1330 if (!this->check_buffer(pinfo + 8 + this->offset_size_))
1333 // Read type_signature (8 bytes).
1334 signature = elfcpp::Swap_unaligned<64, big_endian>::readval(pinfo);
1337 // Read type_offset (4 or 8 bytes).
1338 if (this->offset_size_ == 4)
1340 elfcpp::Swap_unaligned<32, big_endian>::readval(pinfo);
1343 elfcpp::Swap_unaligned<64, big_endian>::readval(pinfo);
1344 pinfo += this->offset_size_;
1347 // Read the .debug_abbrev table.
1348 this->abbrev_table_.read_abbrevs(this->object_, abbrev_shndx,
1351 // Visit the root DIE.
1352 Dwarf_die root_die(this,
1353 pinfo - (this->buffer_ + this->cu_offset_),
1355 if (root_die.tag() != 0)
1357 // Visit the CU or TU.
1358 if (this->is_type_unit_)
1359 this->visit_type_unit(section_offset + this->cu_offset_,
1360 cu_end - cu_start, type_offset, signature,
1363 this->visit_compilation_unit(section_offset + this->cu_offset_,
1364 cu_end - cu_start, &root_die);
1367 // Advance to the next CU.
1373 delete[] this->buffer_;
1374 this->buffer_ = NULL;
1378 // Read the DWARF string table.
1381 Dwarf_info_reader::do_read_string_table(unsigned int string_shndx)
1383 Relobj* object = this->object_;
1385 // If we don't have relocations, string_shndx will be 0, and
1386 // we'll have to hunt for the .debug_str section.
1387 if (string_shndx == 0)
1389 for (unsigned int i = 1; i < this->object_->shnum(); ++i)
1391 std::string name = object->section_name(i);
1392 if (name == ".debug_str" || name == ".zdebug_str")
1395 this->string_output_section_offset_ =
1396 object->output_section_offset(i);
1400 if (string_shndx == 0)
1404 if (this->owns_string_buffer_ && this->string_buffer_ != NULL)
1406 delete[] this->string_buffer_;
1407 this->owns_string_buffer_ = false;
1410 // Get the secton contents and decompress if necessary.
1411 section_size_type buffer_size;
1412 const unsigned char* buffer =
1413 object->decompressed_section_contents(string_shndx,
1415 &this->owns_string_buffer_);
1416 this->string_buffer_ = reinterpret_cast<const char*>(buffer);
1417 this->string_buffer_end_ = this->string_buffer_ + buffer_size;
1418 this->string_shndx_ = string_shndx;
1422 // Read a possibly unaligned integer of SIZE.
1423 template <int valsize>
1424 inline typename elfcpp::Valtype_base<valsize>::Valtype
1425 Dwarf_info_reader::read_from_pointer(const unsigned char* source)
1427 typename elfcpp::Valtype_base<valsize>::Valtype return_value;
1428 if (this->object_->is_big_endian())
1429 return_value = elfcpp::Swap_unaligned<valsize, true>::readval(source);
1431 return_value = elfcpp::Swap_unaligned<valsize, false>::readval(source);
1432 return return_value;
1435 // Read a possibly unaligned integer of SIZE. Update SOURCE after read.
1436 template <int valsize>
1437 inline typename elfcpp::Valtype_base<valsize>::Valtype
1438 Dwarf_info_reader::read_from_pointer(const unsigned char** source)
1440 typename elfcpp::Valtype_base<valsize>::Valtype return_value;
1441 if (this->object_->is_big_endian())
1442 return_value = elfcpp::Swap_unaligned<valsize, true>::readval(*source);
1444 return_value = elfcpp::Swap_unaligned<valsize, false>::readval(*source);
1445 *source += valsize / 8;
1446 return return_value;
1449 // Look for a relocation at offset ATTR_OFF in the dwarf info,
1450 // and return the section index and offset of the target.
1453 Dwarf_info_reader::lookup_reloc(off_t attr_off, off_t* target_off)
1456 attr_off += this->cu_offset_;
1457 unsigned int shndx = this->reloc_mapper_->get_reloc_target(attr_off, &value);
1460 if (this->reloc_type_ == elfcpp::SHT_REL)
1461 *target_off += value;
1463 *target_off = value;
1467 // Return a string from the DWARF string table.
1470 Dwarf_info_reader::get_string(off_t str_off, unsigned int string_shndx)
1472 if (!this->read_string_table(string_shndx))
1475 // Correct the offset. For incremental update links, we have a
1476 // relocated offset that is relative to the output section, but
1477 // here we need an offset relative to the input section.
1478 str_off -= this->string_output_section_offset_;
1480 const char* p = this->string_buffer_ + str_off;
1482 if (p < this->string_buffer_ || p >= this->string_buffer_end_)
1488 // The following are default, do-nothing, implementations of the
1489 // hook methods normally provided by a derived class. We provide
1490 // default implementations rather than no implementation so that
1491 // a derived class needs to implement only the hooks that it needs
1494 // Process a compilation unit and parse its child DIE.
1497 Dwarf_info_reader::visit_compilation_unit(off_t, off_t, Dwarf_die*)
1501 // Process a type unit and parse its child DIE.
1504 Dwarf_info_reader::visit_type_unit(off_t, off_t, off_t, uint64_t, Dwarf_die*)
1508 // Print a warning about a corrupt debug section.
1511 Dwarf_info_reader::warn_corrupt_debug_section() const
1513 gold_warning(_("%s: corrupt debug info in %s"),
1514 this->object_->name().c_str(),
1515 this->object_->section_name(this->shndx_).c_str());
1518 // class Sized_dwarf_line_info
1520 struct LineStateMachine
1526 unsigned int shndx; // the section address refers to
1527 bool is_stmt; // stmt means statement.
1533 ResetLineStateMachine(struct LineStateMachine* lsm, bool default_is_stmt)
1538 lsm->column_num = 0;
1540 lsm->is_stmt = default_is_stmt;
1541 lsm->basic_block = false;
1542 lsm->end_sequence = false;
1545 template<int size, bool big_endian>
1546 Sized_dwarf_line_info<size, big_endian>::Sized_dwarf_line_info(
1548 unsigned int read_shndx)
1549 : data_valid_(false), buffer_(NULL), buffer_start_(NULL),
1550 reloc_mapper_(NULL), symtab_buffer_(NULL), directories_(), files_(),
1551 current_header_index_(-1)
1553 unsigned int debug_shndx;
1555 for (debug_shndx = 1; debug_shndx < object->shnum(); ++debug_shndx)
1557 // FIXME: do this more efficiently: section_name() isn't super-fast
1558 std::string name = object->section_name(debug_shndx);
1559 if (name == ".debug_line" || name == ".zdebug_line")
1561 section_size_type buffer_size;
1562 bool is_new = false;
1563 this->buffer_ = object->decompressed_section_contents(debug_shndx,
1567 this->buffer_start_ = this->buffer_;
1568 this->buffer_end_ = this->buffer_ + buffer_size;
1572 if (this->buffer_ == NULL)
1575 // Find the relocation section for ".debug_line".
1576 // We expect these for relobjs (.o's) but not dynobjs (.so's).
1577 unsigned int reloc_shndx = 0;
1578 for (unsigned int i = 0; i < object->shnum(); ++i)
1580 unsigned int reloc_sh_type = object->section_type(i);
1581 if ((reloc_sh_type == elfcpp::SHT_REL
1582 || reloc_sh_type == elfcpp::SHT_RELA)
1583 && object->section_info(i) == debug_shndx)
1586 this->track_relocs_type_ = reloc_sh_type;
1591 // Finally, we need the symtab section to interpret the relocs.
1592 if (reloc_shndx != 0)
1594 unsigned int symtab_shndx;
1595 for (symtab_shndx = 0; symtab_shndx < object->shnum(); ++symtab_shndx)
1596 if (object->section_type(symtab_shndx) == elfcpp::SHT_SYMTAB)
1598 this->symtab_buffer_ = object->section_contents(
1599 symtab_shndx, &this->symtab_buffer_size_, false);
1602 if (this->symtab_buffer_ == NULL)
1606 this->reloc_mapper_ =
1607 new Sized_elf_reloc_mapper<size, big_endian>(object,
1608 this->symtab_buffer_,
1609 this->symtab_buffer_size_);
1610 if (!this->reloc_mapper_->initialize(reloc_shndx, this->track_relocs_type_))
1613 // Now that we have successfully read all the data, parse the debug
1615 this->data_valid_ = true;
1616 this->read_line_mappings(read_shndx);
1619 // Read the DWARF header.
1621 template<int size, bool big_endian>
1622 const unsigned char*
1623 Sized_dwarf_line_info<size, big_endian>::read_header_prolog(
1624 const unsigned char* lineptr)
1626 uint32_t initial_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr);
1629 // In DWARF2/3, if the initial length is all 1 bits, then the offset
1630 // size is 8 and we need to read the next 8 bytes for the real length.
1631 if (initial_length == 0xffffffff)
1633 header_.offset_size = 8;
1634 initial_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr);
1638 header_.offset_size = 4;
1640 header_.total_length = initial_length;
1642 gold_assert(lineptr + header_.total_length <= buffer_end_);
1644 header_.version = elfcpp::Swap_unaligned<16, big_endian>::readval(lineptr);
1647 if (header_.offset_size == 4)
1648 header_.prologue_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr);
1650 header_.prologue_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr);
1651 lineptr += header_.offset_size;
1653 header_.min_insn_length = *lineptr;
1656 header_.default_is_stmt = *lineptr;
1659 header_.line_base = *reinterpret_cast<const signed char*>(lineptr);
1662 header_.line_range = *lineptr;
1665 header_.opcode_base = *lineptr;
1668 header_.std_opcode_lengths.resize(header_.opcode_base + 1);
1669 header_.std_opcode_lengths[0] = 0;
1670 for (int i = 1; i < header_.opcode_base; i++)
1672 header_.std_opcode_lengths[i] = *lineptr;
1679 // The header for a debug_line section is mildly complicated, because
1680 // the line info is very tightly encoded.
1682 template<int size, bool big_endian>
1683 const unsigned char*
1684 Sized_dwarf_line_info<size, big_endian>::read_header_tables(
1685 const unsigned char* lineptr)
1687 ++this->current_header_index_;
1689 // Create a new directories_ entry and a new files_ entry for our new
1690 // header. We initialize each with a single empty element, because
1691 // dwarf indexes directory and filenames starting at 1.
1692 gold_assert(static_cast<int>(this->directories_.size())
1693 == this->current_header_index_);
1694 gold_assert(static_cast<int>(this->files_.size())
1695 == this->current_header_index_);
1696 this->directories_.push_back(std::vector<std::string>(1));
1697 this->files_.push_back(std::vector<std::pair<int, std::string> >(1));
1699 // It is legal for the directory entry table to be empty.
1705 const char* dirname = reinterpret_cast<const char*>(lineptr);
1706 gold_assert(dirindex
1707 == static_cast<int>(this->directories_.back().size()));
1708 this->directories_.back().push_back(dirname);
1709 lineptr += this->directories_.back().back().size() + 1;
1715 // It is also legal for the file entry table to be empty.
1722 const char* filename = reinterpret_cast<const char*>(lineptr);
1723 lineptr += strlen(filename) + 1;
1725 uint64_t dirindex = read_unsigned_LEB_128(lineptr, &len);
1728 if (dirindex >= this->directories_.back().size())
1730 int dirindexi = static_cast<int>(dirindex);
1732 read_unsigned_LEB_128(lineptr, &len); // mod_time
1735 read_unsigned_LEB_128(lineptr, &len); // filelength
1738 gold_assert(fileindex
1739 == static_cast<int>(this->files_.back().size()));
1740 this->files_.back().push_back(std::make_pair(dirindexi, filename));
1749 // Process a single opcode in the .debug.line structure.
1751 template<int size, bool big_endian>
1753 Sized_dwarf_line_info<size, big_endian>::process_one_opcode(
1754 const unsigned char* start, struct LineStateMachine* lsm, size_t* len)
1758 unsigned char opcode = *start;
1762 // If the opcode is great than the opcode_base, it is a special
1763 // opcode. Most line programs consist mainly of special opcodes.
1764 if (opcode >= header_.opcode_base)
1766 opcode -= header_.opcode_base;
1767 const int advance_address = ((opcode / header_.line_range)
1768 * header_.min_insn_length);
1769 lsm->address += advance_address;
1771 const int advance_line = ((opcode % header_.line_range)
1772 + header_.line_base);
1773 lsm->line_num += advance_line;
1774 lsm->basic_block = true;
1779 // Otherwise, we have the regular opcodes
1782 case elfcpp::DW_LNS_copy:
1783 lsm->basic_block = false;
1787 case elfcpp::DW_LNS_advance_pc:
1789 const uint64_t advance_address
1790 = read_unsigned_LEB_128(start, &templen);
1792 lsm->address += header_.min_insn_length * advance_address;
1796 case elfcpp::DW_LNS_advance_line:
1798 const uint64_t advance_line = read_signed_LEB_128(start, &templen);
1800 lsm->line_num += advance_line;
1804 case elfcpp::DW_LNS_set_file:
1806 const uint64_t fileno = read_unsigned_LEB_128(start, &templen);
1808 lsm->file_num = fileno;
1812 case elfcpp::DW_LNS_set_column:
1814 const uint64_t colno = read_unsigned_LEB_128(start, &templen);
1816 lsm->column_num = colno;
1820 case elfcpp::DW_LNS_negate_stmt:
1821 lsm->is_stmt = !lsm->is_stmt;
1824 case elfcpp::DW_LNS_set_basic_block:
1825 lsm->basic_block = true;
1828 case elfcpp::DW_LNS_fixed_advance_pc:
1830 int advance_address;
1831 advance_address = elfcpp::Swap_unaligned<16, big_endian>::readval(start);
1833 lsm->address += advance_address;
1837 case elfcpp::DW_LNS_const_add_pc:
1839 const int advance_address = (header_.min_insn_length
1840 * ((255 - header_.opcode_base)
1841 / header_.line_range));
1842 lsm->address += advance_address;
1846 case elfcpp::DW_LNS_extended_op:
1848 const uint64_t extended_op_len
1849 = read_unsigned_LEB_128(start, &templen);
1851 oplen += templen + extended_op_len;
1853 const unsigned char extended_op = *start;
1856 switch (extended_op)
1858 case elfcpp::DW_LNE_end_sequence:
1859 // This means that the current byte is the one immediately
1860 // after a set of instructions. Record the current line
1861 // for up to one less than the current address.
1863 lsm->end_sequence = true;
1867 case elfcpp::DW_LNE_set_address:
1870 elfcpp::Swap_unaligned<size, big_endian>::readval(start);
1871 typename Reloc_map::const_iterator it
1872 = this->reloc_map_.find(start - this->buffer_);
1873 if (it != reloc_map_.end())
1875 // If this is a SHT_RELA section, then ignore the
1876 // section contents. This assumes that this is a
1877 // straight reloc which just uses the reloc addend.
1878 // The reloc addend has already been included in the
1880 if (this->track_relocs_type_ == elfcpp::SHT_RELA)
1882 // Add in the symbol value.
1883 lsm->address += it->second.second;
1884 lsm->shndx = it->second.first;
1888 // If we're a normal .o file, with relocs, every
1889 // set_address should have an associated relocation.
1890 if (this->input_is_relobj())
1891 this->data_valid_ = false;
1895 case elfcpp::DW_LNE_define_file:
1897 const char* filename = reinterpret_cast<const char*>(start);
1898 templen = strlen(filename) + 1;
1901 uint64_t dirindex = read_unsigned_LEB_128(start, &templen);
1903 if (dirindex >= this->directories_.back().size())
1905 int dirindexi = static_cast<int>(dirindex);
1907 // This opcode takes two additional ULEB128 parameters
1908 // (mod_time and filelength), but we don't use those
1909 // values. Because OPLEN already tells us how far to
1910 // skip to the next opcode, we don't need to read
1913 this->files_.back().push_back(std::make_pair(dirindexi,
1923 // Ignore unknown opcode silently
1924 for (int i = 0; i < header_.std_opcode_lengths[opcode]; i++)
1927 read_unsigned_LEB_128(start, &templen);
1938 // Read the debug information at LINEPTR and store it in the line
1941 template<int size, bool big_endian>
1942 unsigned const char*
1943 Sized_dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr,
1946 struct LineStateMachine lsm;
1948 // LENGTHSTART is the place the length field is based on. It is the
1949 // point in the header after the initial length field.
1950 const unsigned char* lengthstart = buffer_;
1952 // In 64 bit dwarf, the initial length is 12 bytes, because of the
1953 // 0xffffffff at the start.
1954 if (header_.offset_size == 8)
1959 while (lineptr < lengthstart + header_.total_length)
1961 ResetLineStateMachine(&lsm, header_.default_is_stmt);
1962 while (!lsm.end_sequence)
1965 bool add_line = this->process_one_opcode(lineptr, &lsm, &oplength);
1967 && (shndx == -1U || lsm.shndx == -1U || shndx == lsm.shndx))
1969 Offset_to_lineno_entry entry
1970 = { static_cast<off_t>(lsm.address),
1971 this->current_header_index_,
1972 static_cast<unsigned int>(lsm.file_num),
1973 true, lsm.line_num };
1974 std::vector<Offset_to_lineno_entry>&
1975 map(this->line_number_map_[lsm.shndx]);
1976 // If we see two consecutive entries with the same
1977 // offset and a real line number, then mark the first
1978 // one as non-canonical.
1980 && (map.back().offset == static_cast<off_t>(lsm.address))
1981 && lsm.line_num != -1
1982 && map.back().line_num != -1)
1983 map.back().last_line_for_offset = false;
1984 map.push_back(entry);
1986 lineptr += oplength;
1990 return lengthstart + header_.total_length;
1993 // Read the relocations into a Reloc_map.
1995 template<int size, bool big_endian>
1997 Sized_dwarf_line_info<size, big_endian>::read_relocs()
1999 if (this->symtab_buffer_ == NULL)
2004 while ((reloc_offset = this->reloc_mapper_->next_offset()) != -1)
2006 const unsigned int shndx =
2007 this->reloc_mapper_->get_reloc_target(reloc_offset, &value);
2009 // There is no reason to record non-ordinary section indexes, or
2010 // SHN_UNDEF, because they will never match the real section.
2012 this->reloc_map_[reloc_offset] = std::make_pair(shndx, value);
2014 this->reloc_mapper_->advance(reloc_offset + 1);
2018 // Read the line number info.
2020 template<int size, bool big_endian>
2022 Sized_dwarf_line_info<size, big_endian>::read_line_mappings(unsigned int shndx)
2024 gold_assert(this->data_valid_ == true);
2026 this->read_relocs();
2027 while (this->buffer_ < this->buffer_end_)
2029 const unsigned char* lineptr = this->buffer_;
2030 lineptr = this->read_header_prolog(lineptr);
2031 lineptr = this->read_header_tables(lineptr);
2032 lineptr = this->read_lines(lineptr, shndx);
2033 this->buffer_ = lineptr;
2036 // Sort the lines numbers, so addr2line can use binary search.
2037 for (typename Lineno_map::iterator it = line_number_map_.begin();
2038 it != line_number_map_.end();
2040 // Each vector needs to be sorted by offset.
2041 std::sort(it->second.begin(), it->second.end());
2044 // Some processing depends on whether the input is a .o file or not.
2045 // For instance, .o files have relocs, and have .debug_lines
2046 // information on a per section basis. .so files, on the other hand,
2047 // lack relocs, and offsets are unique, so we can ignore the section
2050 template<int size, bool big_endian>
2052 Sized_dwarf_line_info<size, big_endian>::input_is_relobj()
2054 // Only .o files have relocs and the symtab buffer that goes with them.
2055 return this->symtab_buffer_ != NULL;
2058 // Given an Offset_to_lineno_entry vector, and an offset, figure out
2059 // if the offset points into a function according to the vector (see
2060 // comments below for the algorithm). If it does, return an iterator
2061 // into the vector that points to the line-number that contains that
2062 // offset. If not, it returns vector::end().
2064 static std::vector<Offset_to_lineno_entry>::const_iterator
2065 offset_to_iterator(const std::vector<Offset_to_lineno_entry>* offsets,
2068 const Offset_to_lineno_entry lookup_key = { offset, 0, 0, true, 0 };
2070 // lower_bound() returns the smallest offset which is >= lookup_key.
2071 // If no offset in offsets is >= lookup_key, returns end().
2072 std::vector<Offset_to_lineno_entry>::const_iterator it
2073 = std::lower_bound(offsets->begin(), offsets->end(), lookup_key);
2075 // This code is easiest to understand with a concrete example.
2076 // Here's a possible offsets array:
2077 // {{offset = 3211, header_num = 0, file_num = 1, last, line_num = 16}, // 0
2078 // {offset = 3224, header_num = 0, file_num = 1, last, line_num = 20}, // 1
2079 // {offset = 3226, header_num = 0, file_num = 1, last, line_num = 22}, // 2
2080 // {offset = 3231, header_num = 0, file_num = 1, last, line_num = 25}, // 3
2081 // {offset = 3232, header_num = 0, file_num = 1, last, line_num = -1}, // 4
2082 // {offset = 3232, header_num = 0, file_num = 1, last, line_num = 65}, // 5
2083 // {offset = 3235, header_num = 0, file_num = 1, last, line_num = 66}, // 6
2084 // {offset = 3236, header_num = 0, file_num = 1, last, line_num = -1}, // 7
2085 // {offset = 5764, header_num = 0, file_num = 1, last, line_num = 48}, // 8
2086 // {offset = 5764, header_num = 0, file_num = 1,!last, line_num = 47}, // 9
2087 // {offset = 5765, header_num = 0, file_num = 1, last, line_num = 49}, // 10
2088 // {offset = 5767, header_num = 0, file_num = 1, last, line_num = 50}, // 11
2089 // {offset = 5768, header_num = 0, file_num = 1, last, line_num = 51}, // 12
2090 // {offset = 5773, header_num = 0, file_num = 1, last, line_num = -1}, // 13
2091 // {offset = 5787, header_num = 1, file_num = 1, last, line_num = 19}, // 14
2092 // {offset = 5790, header_num = 1, file_num = 1, last, line_num = 20}, // 15
2093 // {offset = 5793, header_num = 1, file_num = 1, last, line_num = 67}, // 16
2094 // {offset = 5793, header_num = 1, file_num = 1, last, line_num = -1}, // 17
2095 // {offset = 5793, header_num = 1, file_num = 1,!last, line_num = 66}, // 18
2096 // {offset = 5795, header_num = 1, file_num = 1, last, line_num = 68}, // 19
2097 // {offset = 5798, header_num = 1, file_num = 1, last, line_num = -1}, // 20
2098 // The entries with line_num == -1 mark the end of a function: the
2099 // associated offset is one past the last instruction in the
2100 // function. This can correspond to the beginning of the next
2101 // function (as is true for offset 3232); alternately, there can be
2102 // a gap between the end of one function and the start of the next
2103 // (as is true for some others, most obviously from 3236->5764).
2105 // Case 1: lookup_key has offset == 10. lower_bound returns
2106 // offsets[0]. Since it's not an exact match and we're
2107 // at the beginning of offsets, we return end() (invalid).
2108 // Case 2: lookup_key has offset 10000. lower_bound returns
2109 // offset[21] (end()). We return end() (invalid).
2110 // Case 3: lookup_key has offset == 3211. lower_bound matches
2111 // offsets[0] exactly, and that's the entry we return.
2112 // Case 4: lookup_key has offset == 3232. lower_bound returns
2113 // offsets[4]. That's an exact match, but indicates
2114 // end-of-function. We check if offsets[5] is also an
2115 // exact match but not end-of-function. It is, so we
2116 // return offsets[5].
2117 // Case 5: lookup_key has offset == 3214. lower_bound returns
2118 // offsets[1]. Since it's not an exact match, we back
2119 // up to the offset that's < lookup_key, offsets[0].
2120 // We note offsets[0] is a valid entry (not end-of-function),
2121 // so that's the entry we return.
2122 // Case 6: lookup_key has offset == 4000. lower_bound returns
2123 // offsets[8]. Since it's not an exact match, we back
2124 // up to offsets[7]. Since offsets[7] indicates
2125 // end-of-function, we know lookup_key is between
2126 // functions, so we return end() (not a valid offset).
2127 // Case 7: lookup_key has offset == 5794. lower_bound returns
2128 // offsets[19]. Since it's not an exact match, we back
2129 // up to offsets[16]. Note we back up to the *first*
2130 // entry with offset 5793, not just offsets[19-1].
2131 // We note offsets[16] is a valid entry, so we return it.
2132 // If offsets[16] had had line_num == -1, we would have
2133 // checked offsets[17]. The reason for this is that
2134 // 16 and 17 can be in an arbitrary order, since we sort
2135 // only by offset and last_line_for_offset. (Note it
2136 // doesn't help to use line_number as a tertiary sort key,
2137 // since sometimes we want the -1 to be first and sometimes
2138 // we want it to be last.)
2140 // This deals with cases (1) and (2).
2141 if ((it == offsets->begin() && offset < it->offset)
2142 || it == offsets->end())
2143 return offsets->end();
2145 // This deals with cases (3) and (4).
2146 if (offset == it->offset)
2148 while (it != offsets->end()
2149 && it->offset == offset
2150 && it->line_num == -1)
2152 if (it == offsets->end() || it->offset != offset)
2153 return offsets->end();
2158 // This handles the first part of case (7) -- we back up to the
2159 // *first* entry that has the offset that's behind us.
2160 gold_assert(it != offsets->begin());
2161 std::vector<Offset_to_lineno_entry>::const_iterator range_end = it;
2163 const off_t range_value = it->offset;
2164 while (it != offsets->begin() && (it-1)->offset == range_value)
2167 // This handles cases (5), (6), and (7): if any entry in the
2168 // equal_range [it, range_end) has a line_num != -1, it's a valid
2169 // match. If not, we're not in a function. The line number we saw
2170 // last for an offset will be sorted first, so it'll get returned if
2172 for (; it != range_end; ++it)
2173 if (it->line_num != -1)
2175 return offsets->end();
2178 // Returns the canonical filename:lineno for the address passed in.
2179 // If other_lines is not NULL, appends the non-canonical lines
2180 // assigned to the same address.
2182 template<int size, bool big_endian>
2184 Sized_dwarf_line_info<size, big_endian>::do_addr2line(
2187 std::vector<std::string>* other_lines)
2189 if (this->data_valid_ == false)
2192 const std::vector<Offset_to_lineno_entry>* offsets;
2193 // If we do not have reloc information, then our input is a .so or
2194 // some similar data structure where all the information is held in
2195 // the offset. In that case, we ignore the input shndx.
2196 if (this->input_is_relobj())
2197 offsets = &this->line_number_map_[shndx];
2199 offsets = &this->line_number_map_[-1U];
2200 if (offsets->empty())
2203 typename std::vector<Offset_to_lineno_entry>::const_iterator it
2204 = offset_to_iterator(offsets, offset);
2205 if (it == offsets->end())
2208 std::string result = this->format_file_lineno(*it);
2209 if (other_lines != NULL)
2210 for (++it; it != offsets->end() && it->offset == offset; ++it)
2212 if (it->line_num == -1)
2213 continue; // The end of a previous function.
2214 other_lines->push_back(this->format_file_lineno(*it));
2219 // Convert the file_num + line_num into a string.
2221 template<int size, bool big_endian>
2223 Sized_dwarf_line_info<size, big_endian>::format_file_lineno(
2224 const Offset_to_lineno_entry& loc) const
2228 gold_assert(loc.header_num < static_cast<int>(this->files_.size()));
2229 gold_assert(loc.file_num
2230 < static_cast<unsigned int>(this->files_[loc.header_num].size()));
2231 const std::pair<int, std::string>& filename_pair
2232 = this->files_[loc.header_num][loc.file_num];
2233 const std::string& filename = filename_pair.second;
2235 gold_assert(loc.header_num < static_cast<int>(this->directories_.size()));
2236 gold_assert(filename_pair.first
2237 < static_cast<int>(this->directories_[loc.header_num].size()));
2238 const std::string& dirname
2239 = this->directories_[loc.header_num][filename_pair.first];
2241 if (!dirname.empty())
2250 char buffer[64]; // enough to hold a line number
2251 snprintf(buffer, sizeof(buffer), "%d", loc.line_num);
2258 // Dwarf_line_info routines.
2260 static unsigned int next_generation_count = 0;
2262 struct Addr2line_cache_entry
2266 Dwarf_line_info* dwarf_line_info;
2267 unsigned int generation_count;
2268 unsigned int access_count;
2270 Addr2line_cache_entry(Object* o, unsigned int s, Dwarf_line_info* d)
2271 : object(o), shndx(s), dwarf_line_info(d),
2272 generation_count(next_generation_count), access_count(0)
2274 if (next_generation_count < (1U << 31))
2275 ++next_generation_count;
2278 // We expect this cache to be small, so don't bother with a hashtable
2279 // or priority queue or anything: just use a simple vector.
2280 static std::vector<Addr2line_cache_entry> addr2line_cache;
2283 Dwarf_line_info::one_addr2line(Object* object,
2284 unsigned int shndx, off_t offset,
2286 std::vector<std::string>* other_lines)
2288 Dwarf_line_info* lineinfo = NULL;
2289 std::vector<Addr2line_cache_entry>::iterator it;
2291 // First, check the cache. If we hit, update the counts.
2292 for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it)
2294 if (it->object == object && it->shndx == shndx)
2296 lineinfo = it->dwarf_line_info;
2297 it->generation_count = next_generation_count;
2298 // We cap generation_count at 2^31 -1 to avoid overflow.
2299 if (next_generation_count < (1U << 31))
2300 ++next_generation_count;
2301 // We cap access_count at 31 so 2^access_count doesn't overflow
2302 if (it->access_count < 31)
2308 // If we don't hit the cache, create a new object and insert into the
2310 if (lineinfo == NULL)
2312 switch (parameters->size_and_endianness())
2314 #ifdef HAVE_TARGET_32_LITTLE
2315 case Parameters::TARGET_32_LITTLE:
2316 lineinfo = new Sized_dwarf_line_info<32, false>(object, shndx); break;
2318 #ifdef HAVE_TARGET_32_BIG
2319 case Parameters::TARGET_32_BIG:
2320 lineinfo = new Sized_dwarf_line_info<32, true>(object, shndx); break;
2322 #ifdef HAVE_TARGET_64_LITTLE
2323 case Parameters::TARGET_64_LITTLE:
2324 lineinfo = new Sized_dwarf_line_info<64, false>(object, shndx); break;
2326 #ifdef HAVE_TARGET_64_BIG
2327 case Parameters::TARGET_64_BIG:
2328 lineinfo = new Sized_dwarf_line_info<64, true>(object, shndx); break;
2333 addr2line_cache.push_back(Addr2line_cache_entry(object, shndx, lineinfo));
2336 // Now that we have our object, figure out the answer
2337 std::string retval = lineinfo->addr2line(shndx, offset, other_lines);
2339 // Finally, if our cache has grown too big, delete old objects. We
2340 // assume the common (probably only) case is deleting only one object.
2341 // We use a pretty simple scheme to evict: function of LRU and MFU.
2342 while (addr2line_cache.size() > cache_size)
2344 unsigned int lowest_score = ~0U;
2345 std::vector<Addr2line_cache_entry>::iterator lowest
2346 = addr2line_cache.end();
2347 for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it)
2349 const unsigned int score = (it->generation_count
2350 + (1U << it->access_count));
2351 if (score < lowest_score)
2353 lowest_score = score;
2357 if (lowest != addr2line_cache.end())
2359 delete lowest->dwarf_line_info;
2360 addr2line_cache.erase(lowest);
2368 Dwarf_line_info::clear_addr2line_cache()
2370 for (std::vector<Addr2line_cache_entry>::iterator it = addr2line_cache.begin();
2371 it != addr2line_cache.end();
2373 delete it->dwarf_line_info;
2374 addr2line_cache.clear();
2377 #ifdef HAVE_TARGET_32_LITTLE
2379 class Sized_dwarf_line_info<32, false>;
2382 #ifdef HAVE_TARGET_32_BIG
2384 class Sized_dwarf_line_info<32, true>;
2387 #ifdef HAVE_TARGET_64_LITTLE
2389 class Sized_dwarf_line_info<64, false>;
2392 #ifdef HAVE_TARGET_64_BIG
2394 class Sized_dwarf_line_info<64, true>;
2397 } // End namespace gold.