1 // dwarf_reader.cc -- parse dwarf2/3 debug information
3 // Copyright 2007 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.
25 #include "elfcpp_swap.h"
28 #include "parameters.h"
30 #include "dwarf_reader.h"
34 // Read an unsigned LEB128 number. Each byte contains 7 bits of
35 // information, plus one bit saying whether the number continues or
39 read_unsigned_LEB_128(const unsigned char* buffer, size_t* len)
43 unsigned int shift = 0;
50 result |= (static_cast<uint64_t>(byte & 0x7f)) << shift;
60 // Read a signed LEB128 number. These are like regular LEB128
61 // numbers, except the last byte may have a sign bit set.
64 read_signed_LEB_128(const unsigned char* buffer, size_t* len)
75 result |= (static_cast<uint64_t>(byte & 0x7f) << shift);
80 if ((shift < 8 * static_cast<int>(sizeof(result))) && (byte & 0x40))
81 result |= -((static_cast<int64_t>(1)) << shift);
86 } // End anonymous namespace.
91 // This is the format of a DWARF2/3 line state machine that we process
92 // opcodes using. There is no need for anything outside the lineinfo
93 // processor to know how this works.
95 struct LineStateMachine
101 unsigned int shndx; // the section address refers to
102 bool is_stmt; // stmt means statement.
108 ResetLineStateMachine(struct LineStateMachine* lsm, bool default_is_stmt)
115 lsm->is_stmt = default_is_stmt;
116 lsm->basic_block = false;
117 lsm->end_sequence = false;
120 template<int size, bool big_endian>
121 Sized_dwarf_line_info<size, big_endian>::Sized_dwarf_line_info(Object* object)
122 : data_valid_(false), buffer_(NULL), symtab_buffer_(NULL),
123 directories_(), files_(), current_header_index_(-1)
125 unsigned int debug_shndx;
126 for (debug_shndx = 0; debug_shndx < object->shnum(); ++debug_shndx)
127 if (object->section_name(debug_shndx) == ".debug_line")
130 this->buffer_ = object->section_contents(
131 debug_shndx, &buffer_size, false);
132 this->buffer_end_ = this->buffer_ + buffer_size;
135 if (this->buffer_ == NULL)
138 // Find the relocation section for ".debug_line".
139 // We expect these for relobjs (.o's) but not dynobjs (.so's).
140 bool got_relocs = false;
141 for (unsigned int reloc_shndx = 0;
142 reloc_shndx < object->shnum();
145 unsigned int reloc_sh_type = object->section_type(reloc_shndx);
146 if ((reloc_sh_type == elfcpp::SHT_REL
147 || reloc_sh_type == elfcpp::SHT_RELA)
148 && object->section_info(reloc_shndx) == debug_shndx)
150 got_relocs = this->track_relocs_.initialize(object, reloc_shndx,
156 // Finally, we need the symtab section to interpret the relocs.
159 unsigned int symtab_shndx;
160 for (symtab_shndx = 0; symtab_shndx < object->shnum(); ++symtab_shndx)
161 if (object->section_type(symtab_shndx) == elfcpp::SHT_SYMTAB)
163 this->symtab_buffer_ = object->section_contents(
164 symtab_shndx, &this->symtab_buffer_size_, false);
167 if (this->symtab_buffer_ == NULL)
171 // Now that we have successfully read all the data, parse the debug
173 this->data_valid_ = true;
174 this->read_line_mappings();
177 // Read the DWARF header.
179 template<int size, bool big_endian>
181 Sized_dwarf_line_info<size, big_endian>::read_header_prolog(
182 const unsigned char* lineptr)
184 uint32_t initial_length = elfcpp::Swap<32, big_endian>::readval(lineptr);
187 // In DWARF2/3, if the initial length is all 1 bits, then the offset
188 // size is 8 and we need to read the next 8 bytes for the real length.
189 if (initial_length == 0xffffffff)
191 header_.offset_size = 8;
192 initial_length = elfcpp::Swap<64, big_endian>::readval(lineptr);
196 header_.offset_size = 4;
198 header_.total_length = initial_length;
200 gold_assert(lineptr + header_.total_length <= buffer_end_);
202 header_.version = elfcpp::Swap<16, big_endian>::readval(lineptr);
205 if (header_.offset_size == 4)
206 header_.prologue_length = elfcpp::Swap<32, big_endian>::readval(lineptr);
208 header_.prologue_length = elfcpp::Swap<64, big_endian>::readval(lineptr);
209 lineptr += header_.offset_size;
211 header_.min_insn_length = *lineptr;
214 header_.default_is_stmt = *lineptr;
217 header_.line_base = *reinterpret_cast<const signed char*>(lineptr);
220 header_.line_range = *lineptr;
223 header_.opcode_base = *lineptr;
226 header_.std_opcode_lengths.reserve(header_.opcode_base + 1);
227 header_.std_opcode_lengths[0] = 0;
228 for (int i = 1; i < header_.opcode_base; i++)
230 header_.std_opcode_lengths[i] = *lineptr;
237 // The header for a debug_line section is mildly complicated, because
238 // the line info is very tightly encoded.
240 template<int size, bool big_endian>
242 Sized_dwarf_line_info<size, big_endian>::read_header_tables(
243 const unsigned char* lineptr)
245 ++this->current_header_index_;
247 // Create a new directories_ entry and a new files_ entry for our new
248 // header. We initialize each with a single empty element, because
249 // dwarf indexes directory and filenames starting at 1.
250 gold_assert(static_cast<int>(this->directories_.size())
251 == this->current_header_index_);
252 gold_assert(static_cast<int>(this->files_.size())
253 == this->current_header_index_);
254 this->directories_.push_back(std::vector<std::string>(1));
255 this->files_.push_back(std::vector<std::pair<int, std::string> >(1));
257 // It is legal for the directory entry table to be empty.
263 const char* dirname = reinterpret_cast<const char*>(lineptr);
265 == static_cast<int>(this->directories_.back().size()));
266 this->directories_.back().push_back(dirname);
267 lineptr += this->directories_.back().back().size() + 1;
273 // It is also legal for the file entry table to be empty.
280 const char* filename = reinterpret_cast<const char*>(lineptr);
281 lineptr += strlen(filename) + 1;
283 uint64_t dirindex = read_unsigned_LEB_128(lineptr, &len);
286 if (dirindex >= this->directories_.back().size())
288 int dirindexi = static_cast<int>(dirindex);
290 read_unsigned_LEB_128(lineptr, &len); // mod_time
293 read_unsigned_LEB_128(lineptr, &len); // filelength
296 gold_assert(fileindex
297 == static_cast<int>(this->files_.back().size()));
298 this->files_.back().push_back(std::make_pair(dirindexi, filename));
307 // Process a single opcode in the .debug.line structure.
309 // Templating on size and big_endian would yield more efficient (and
310 // simpler) code, but would bloat the binary. Speed isn't important
313 template<int size, bool big_endian>
315 Sized_dwarf_line_info<size, big_endian>::process_one_opcode(
316 const unsigned char* start, struct LineStateMachine* lsm, size_t* len)
320 unsigned char opcode = *start;
324 // If the opcode is great than the opcode_base, it is a special
325 // opcode. Most line programs consist mainly of special opcodes.
326 if (opcode >= header_.opcode_base)
328 opcode -= header_.opcode_base;
329 const int advance_address = ((opcode / header_.line_range)
330 * header_.min_insn_length);
331 lsm->address += advance_address;
333 const int advance_line = ((opcode % header_.line_range)
334 + header_.line_base);
335 lsm->line_num += advance_line;
336 lsm->basic_block = true;
341 // Otherwise, we have the regular opcodes
344 case elfcpp::DW_LNS_copy:
345 lsm->basic_block = false;
349 case elfcpp::DW_LNS_advance_pc:
351 const uint64_t advance_address
352 = read_unsigned_LEB_128(start, &templen);
354 lsm->address += header_.min_insn_length * advance_address;
358 case elfcpp::DW_LNS_advance_line:
360 const uint64_t advance_line = read_signed_LEB_128(start, &templen);
362 lsm->line_num += advance_line;
366 case elfcpp::DW_LNS_set_file:
368 const uint64_t fileno = read_unsigned_LEB_128(start, &templen);
370 lsm->file_num = fileno;
374 case elfcpp::DW_LNS_set_column:
376 const uint64_t colno = read_unsigned_LEB_128(start, &templen);
378 lsm->column_num = colno;
382 case elfcpp::DW_LNS_negate_stmt:
383 lsm->is_stmt = !lsm->is_stmt;
386 case elfcpp::DW_LNS_set_basic_block:
387 lsm->basic_block = true;
390 case elfcpp::DW_LNS_fixed_advance_pc:
393 advance_address = elfcpp::Swap<16, big_endian>::readval(start);
395 lsm->address += advance_address;
399 case elfcpp::DW_LNS_const_add_pc:
401 const int advance_address = (header_.min_insn_length
402 * ((255 - header_.opcode_base)
403 / header_.line_range));
404 lsm->address += advance_address;
408 case elfcpp::DW_LNS_extended_op:
410 const uint64_t extended_op_len
411 = read_unsigned_LEB_128(start, &templen);
413 oplen += templen + extended_op_len;
415 const unsigned char extended_op = *start;
420 case elfcpp::DW_LNE_end_sequence:
421 // This means that the current byte is the one immediately
422 // after a set of instructions. Record the current line
423 // for up to one less than the current address.
425 lsm->end_sequence = true;
429 case elfcpp::DW_LNE_set_address:
431 lsm->address = elfcpp::Swap<size, big_endian>::readval(start);
432 typename Reloc_map::const_iterator it
433 = reloc_map_.find(start - this->buffer_);
434 if (it != reloc_map_.end())
437 lsm->address += it->second.second;
438 lsm->shndx = it->second.first;
442 // If we're a normal .o file, with relocs, every
443 // set_address should have an associated relocation.
444 if (this->input_is_relobj())
445 this->data_valid_ = false;
449 case elfcpp::DW_LNE_define_file:
451 const char* filename = reinterpret_cast<const char*>(start);
452 templen = strlen(filename) + 1;
455 uint64_t dirindex = read_unsigned_LEB_128(start, &templen);
458 if (dirindex >= this->directories_.back().size())
460 int dirindexi = static_cast<int>(dirindex);
462 read_unsigned_LEB_128(start, &templen); // mod_time
465 read_unsigned_LEB_128(start, &templen); // filelength
468 this->files_.back().push_back(std::make_pair(dirindexi,
478 // Ignore unknown opcode silently
479 for (int i = 0; i < header_.std_opcode_lengths[opcode]; i++)
482 read_unsigned_LEB_128(start, &templen);
493 // Read the debug information at LINEPTR and store it in the line
496 template<int size, bool big_endian>
498 Sized_dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr)
500 struct LineStateMachine lsm;
502 // LENGTHSTART is the place the length field is based on. It is the
503 // point in the header after the initial length field.
504 const unsigned char* lengthstart = buffer_;
506 // In 64 bit dwarf, the initial length is 12 bytes, because of the
507 // 0xffffffff at the start.
508 if (header_.offset_size == 8)
513 while (lineptr < lengthstart + header_.total_length)
515 ResetLineStateMachine(&lsm, header_.default_is_stmt);
516 while (!lsm.end_sequence)
519 bool add_line = this->process_one_opcode(lineptr, &lsm, &oplength);
522 Offset_to_lineno_entry entry
523 = { lsm.address, this->current_header_index_,
524 lsm.file_num, lsm.line_num };
525 line_number_map_[lsm.shndx].push_back(entry);
531 return lengthstart + header_.total_length;
534 // Looks in the symtab to see what section a symbol is in.
536 template<int size, bool big_endian>
538 Sized_dwarf_line_info<size, big_endian>::symbol_section(
540 typename elfcpp::Elf_types<size>::Elf_Addr* value)
542 const int symsize = elfcpp::Elf_sizes<size>::sym_size;
543 gold_assert(sym * symsize < this->symtab_buffer_size_);
544 elfcpp::Sym<size, big_endian> elfsym(this->symtab_buffer_ + sym * symsize);
545 *value = elfsym.get_st_value();
546 return elfsym.get_st_shndx();
549 // Read the relocations into a Reloc_map.
551 template<int size, bool big_endian>
553 Sized_dwarf_line_info<size, big_endian>::read_relocs()
555 if (this->symtab_buffer_ == NULL)
558 typename elfcpp::Elf_types<size>::Elf_Addr value;
560 while ((reloc_offset = this->track_relocs_.next_offset()) != -1)
562 const unsigned int sym = this->track_relocs_.next_symndx();
563 const unsigned int shndx = this->symbol_section(sym, &value);
564 this->reloc_map_[reloc_offset] = std::make_pair(shndx, value);
565 this->track_relocs_.advance(reloc_offset + 1);
569 // Read the line number info.
571 template<int size, bool big_endian>
573 Sized_dwarf_line_info<size, big_endian>::read_line_mappings()
575 gold_assert(this->data_valid_ == true);
578 while (this->buffer_ < this->buffer_end_)
580 const unsigned char* lineptr = this->buffer_;
581 lineptr = this->read_header_prolog(lineptr);
582 lineptr = this->read_header_tables(lineptr);
583 lineptr = this->read_lines(lineptr);
584 this->buffer_ = lineptr;
587 // Sort the lines numbers, so addr2line can use binary search.
588 for (typename Lineno_map::iterator it = line_number_map_.begin();
589 it != line_number_map_.end();
591 // Each vector needs to be sorted by offset.
592 std::sort(it->second.begin(), it->second.end());
595 // Some processing depends on whether the input is a .o file or not.
596 // For instance, .o files have relocs, and have .debug_lines
597 // information on a per section basis. .so files, on the other hand,
598 // lack relocs, and offsets are unique, so we can ignore the section
601 template<int size, bool big_endian>
603 Sized_dwarf_line_info<size, big_endian>::input_is_relobj()
605 // Only .o files have relocs and the symtab buffer that goes with them.
606 return this->symtab_buffer_ != NULL;
610 // Return a string for a file name and line number.
612 template<int size, bool big_endian>
614 Sized_dwarf_line_info<size, big_endian>::do_addr2line(unsigned int shndx,
617 if (this->data_valid_ == false)
620 const Offset_to_lineno_entry lookup_key = { offset, 0, 0, 0 };
621 const std::vector<Offset_to_lineno_entry>* offsets;
622 // If we do not have reloc information, then our input is a .so or
623 // some similar data structure where all the information is held in
624 // the offset. In that case, we ignore the input shndx.
625 if (this->input_is_relobj())
626 offsets = &this->line_number_map_[shndx];
628 offsets = &this->line_number_map_[-1U];
629 if (offsets->empty())
632 typename std::vector<Offset_to_lineno_entry>::const_iterator it
633 = std::lower_bound(offsets->begin(), offsets->end(), lookup_key);
635 // If we found an exact match, great, otherwise find the last entry
636 // before the passed-in offset.
637 if (it == offsets->end() || it->offset > offset)
639 if (it == offsets->begin())
642 gold_assert(it->offset < offset);
645 // Convert the file_num + line_num into a string.
648 gold_assert(it->header_num < static_cast<int>(this->files_.size()));
649 gold_assert(it->file_num
650 < static_cast<int>(this->files_[it->header_num].size()));
651 const std::pair<int, std::string>& filename_pair
652 = this->files_[it->header_num][it->file_num];
653 const std::string& filename = filename_pair.second;
655 gold_assert(it->header_num < static_cast<int>(this->directories_.size()));
656 gold_assert(filename_pair.first
657 < static_cast<int>(this->directories_[it->header_num].size()));
658 const std::string& dirname
659 = this->directories_[it->header_num][filename_pair.first];
661 if (!dirname.empty())
670 char buffer[64]; // enough to hold a line number
671 snprintf(buffer, sizeof(buffer), "%d", it->line_num);
678 // Dwarf_line_info routines.
681 Dwarf_line_info::one_addr2line(Object* object,
682 unsigned int shndx, off_t offset)
684 if (parameters->get_size() == 32 && !parameters->is_big_endian())
685 #ifdef HAVE_TARGET_32_LITTLE
686 return Sized_dwarf_line_info<32, false>(object).addr2line(shndx, offset);
690 else if (parameters->get_size() == 32 && parameters->is_big_endian())
691 #ifdef HAVE_TARGET_32_BIG
692 return Sized_dwarf_line_info<32, true>(object).addr2line(shndx, offset);
696 else if (parameters->get_size() == 64 && !parameters->is_big_endian())
697 #ifdef HAVE_TARGET_64_LITTLE
698 return Sized_dwarf_line_info<64, false>(object).addr2line(shndx, offset);
702 else if (parameters->get_size() == 64 && parameters->is_big_endian())
703 #ifdef HAVE_TARGET_64_BIT
704 return Sized_dwarf_line_info<64, true>(object).addr2line(shndx, offset);
712 #ifdef HAVE_TARGET_32_LITTLE
714 class Sized_dwarf_line_info<32, false>;
717 #ifdef HAVE_TARGET_32_BIG
719 class Sized_dwarf_line_info<32, true>;
722 #ifdef HAVE_TARGET_64_LITTLE
724 class Sized_dwarf_line_info<64, false>;
727 #ifdef HAVE_TARGET_64_BIG
729 class Sized_dwarf_line_info<64, true>;
732 } // End namespace gold.