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 "dwarf_reader.h"
32 // Read an unsigned LEB128 number. Each byte contains 7 bits of
33 // information, plus one bit saying whether the number continues or
37 read_unsigned_LEB_128(const unsigned char* buffer, size_t* len)
41 unsigned int shift = 0;
48 result |= (static_cast<uint64_t>(byte & 0x7f)) << shift;
58 // Read a signed LEB128 number. These are like regular LEB128
59 // numbers, except the last byte may have a sign bit set.
62 read_signed_LEB_128(const unsigned char* buffer, size_t* len)
73 result |= (static_cast<uint64_t>(byte & 0x7f) << shift);
78 if ((shift < 8 * static_cast<int>(sizeof(result))) && (byte & 0x40))
79 result |= -((static_cast<int64_t>(1)) << shift);
84 } // End anonymous namespace.
89 // This is the format of a DWARF2/3 line state machine that we process
90 // opcodes using. There is no need for anything outside the lineinfo
91 // processor to know how this works.
93 struct LineStateMachine
99 unsigned int shndx; // the section address refers to
100 bool is_stmt; // stmt means statement.
106 ResetLineStateMachine(struct LineStateMachine* lsm, bool default_is_stmt)
113 lsm->is_stmt = default_is_stmt;
114 lsm->basic_block = false;
115 lsm->end_sequence = false;
118 // Read the DWARF header.
120 template<int size, bool big_endian>
122 Dwarf_line_info<size, big_endian>::read_header_prolog(
123 const unsigned char* lineptr)
125 uint32_t initial_length = elfcpp::Swap<32, big_endian>::readval(lineptr);
128 // In DWARF2/3, if the initial length is all 1 bits, then the offset
129 // size is 8 and we need to read the next 8 bytes for the real length.
130 if (initial_length == 0xffffffff)
132 header_.offset_size = 8;
133 initial_length = elfcpp::Swap<64, big_endian>::readval(lineptr);
137 header_.offset_size = 4;
139 header_.total_length = initial_length;
141 gold_assert(lineptr + header_.total_length <= buffer_end_);
143 header_.version = elfcpp::Swap<16, big_endian>::readval(lineptr);
146 if (header_.offset_size == 4)
147 header_.prologue_length = elfcpp::Swap<32, big_endian>::readval(lineptr);
149 header_.prologue_length = elfcpp::Swap<64, big_endian>::readval(lineptr);
150 lineptr += header_.offset_size;
152 header_.min_insn_length = *lineptr;
155 header_.default_is_stmt = *lineptr;
158 header_.line_base = *reinterpret_cast<const signed char*>(lineptr);
161 header_.line_range = *lineptr;
164 header_.opcode_base = *lineptr;
167 header_.std_opcode_lengths.reserve(header_.opcode_base + 1);
168 header_.std_opcode_lengths[0] = 0;
169 for (int i = 1; i < header_.opcode_base; i++)
171 header_.std_opcode_lengths[i] = *lineptr;
178 // The header for a debug_line section is mildly complicated, because
179 // the line info is very tightly encoded.
181 template<int size, bool big_endian>
183 Dwarf_line_info<size, big_endian>::read_header_tables(
184 const unsigned char* lineptr)
186 // It is legal for the directory entry table to be empty.
192 const unsigned char* dirname = lineptr;
193 gold_assert(dirindex == static_cast<int>(directories_.size()));
194 directories_.push_back(reinterpret_cast<const char*>(dirname));
195 lineptr += directories_.back().size() + 1;
201 // It is also legal for the file entry table to be empty.
208 const char* filename = reinterpret_cast<const char*>(lineptr);
209 lineptr += strlen(filename) + 1;
211 uint64_t dirindex = read_unsigned_LEB_128(lineptr, &len);
212 if (dirindex >= directories_.size())
216 read_unsigned_LEB_128(lineptr, &len); // mod_time
219 read_unsigned_LEB_128(lineptr, &len); // filelength
222 gold_assert(fileindex == static_cast<int>(files_.size()));
223 files_.push_back(std::pair<int, std::string>(dirindex, filename));
232 // Process a single opcode in the .debug.line structure.
234 // Templating on size and big_endian would yield more efficient (and
235 // simpler) code, but would bloat the binary. Speed isn't important
238 template<int size, bool big_endian>
240 Dwarf_line_info<size, big_endian>::process_one_opcode(
241 const unsigned char* start, struct LineStateMachine* lsm, size_t* len)
245 unsigned char opcode = *start;
249 // If the opcode is great than the opcode_base, it is a special
250 // opcode. Most line programs consist mainly of special opcodes.
251 if (opcode >= header_.opcode_base)
253 opcode -= header_.opcode_base;
254 const int advance_address = ((opcode / header_.line_range)
255 * header_.min_insn_length);
256 lsm->address += advance_address;
258 const int advance_line = ((opcode % header_.line_range)
259 + header_.line_base);
260 lsm->line_num += advance_line;
261 lsm->basic_block = true;
266 // Otherwise, we have the regular opcodes
269 case elfcpp::DW_LNS_copy:
270 lsm->basic_block = false;
274 case elfcpp::DW_LNS_advance_pc:
276 const uint64_t advance_address
277 = read_unsigned_LEB_128(start, &templen);
279 lsm->address += header_.min_insn_length * advance_address;
283 case elfcpp::DW_LNS_advance_line:
285 const uint64_t advance_line = read_signed_LEB_128(start, &templen);
287 lsm->line_num += advance_line;
291 case elfcpp::DW_LNS_set_file:
293 const uint64_t fileno = read_unsigned_LEB_128(start, &templen);
295 lsm->file_num = fileno;
299 case elfcpp::DW_LNS_set_column:
301 const uint64_t colno = read_unsigned_LEB_128(start, &templen);
303 lsm->column_num = colno;
307 case elfcpp::DW_LNS_negate_stmt:
308 lsm->is_stmt = !lsm->is_stmt;
311 case elfcpp::DW_LNS_set_basic_block:
312 lsm->basic_block = true;
315 case elfcpp::DW_LNS_fixed_advance_pc:
318 advance_address = elfcpp::Swap<16, big_endian>::readval(start);
320 lsm->address += advance_address;
324 case elfcpp::DW_LNS_const_add_pc:
326 const int advance_address = (header_.min_insn_length
327 * ((255 - header_.opcode_base)
328 / header_.line_range));
329 lsm->address += advance_address;
333 case elfcpp::DW_LNS_extended_op:
335 const uint64_t extended_op_len
336 = read_unsigned_LEB_128(start, &templen);
338 oplen += templen + extended_op_len;
340 const unsigned char extended_op = *start;
345 case elfcpp::DW_LNE_end_sequence:
346 lsm->end_sequence = true;
350 case elfcpp::DW_LNE_set_address:
352 typename Reloc_map::const_iterator it
353 = reloc_map_.find(start - this->buffer_);
354 if (it != reloc_map_.end())
358 (elfcpp::Swap<size, big_endian>::readval(start)
359 + it->second.second);
360 lsm->shndx = it->second.first;
364 // Every set_address should have an associated
366 this->data_valid_ = false;
370 case elfcpp::DW_LNE_define_file:
372 const char* filename = reinterpret_cast<const char*>(start);
373 templen = strlen(filename) + 1;
376 uint64_t dirindex = read_unsigned_LEB_128(start, &templen);
377 if (dirindex >= directories_.size())
381 read_unsigned_LEB_128(start, &templen); // mod_time
384 read_unsigned_LEB_128(start, &templen); // filelength
387 files_.push_back(std::pair<int, std::string>(dirindex,
397 // Ignore unknown opcode silently
398 for (int i = 0; i < header_.std_opcode_lengths[opcode]; i++)
401 read_unsigned_LEB_128(start, &templen);
412 // Read the debug information at LINEPTR and store it in the line
415 template<int size, bool big_endian>
417 Dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr)
419 struct LineStateMachine lsm;
421 // LENGTHSTART is the place the length field is based on. It is the
422 // point in the header after the initial length field.
423 const unsigned char* lengthstart = buffer_;
425 // In 64 bit dwarf, the initial length is 12 bytes, because of the
426 // 0xffffffff at the start.
427 if (header_.offset_size == 8)
432 while (lineptr < lengthstart + header_.total_length)
434 ResetLineStateMachine(&lsm, header_.default_is_stmt);
435 while (!lsm.end_sequence)
438 bool add_line = this->process_one_opcode(lineptr, &lsm, &oplength);
441 Offset_to_lineno_entry entry
442 = { lsm.address, lsm.file_num, lsm.line_num };
443 line_number_map_[lsm.shndx].push_back(entry);
449 return lengthstart + header_.total_length;
452 // Looks in the symtab to see what section a symbol is in.
454 template<int size, bool big_endian>
456 Dwarf_line_info<size, big_endian>::symbol_section(
458 typename elfcpp::Elf_types<size>::Elf_Addr* value)
460 const int symsize = elfcpp::Elf_sizes<size>::sym_size;
461 gold_assert(this->symtab_buffer_ + sym * symsize < this->symtab_buffer_end_);
462 elfcpp::Sym<size, big_endian> elfsym(this->symtab_buffer_ + sym * symsize);
463 *value = elfsym.get_st_value();
464 return elfsym.get_st_shndx();
467 // Read the relocations into a Reloc_map.
469 template<int size, bool big_endian>
471 Dwarf_line_info<size, big_endian>::read_relocs()
473 if (this->symtab_buffer_ == NULL)
476 typename elfcpp::Elf_types<size>::Elf_Addr value;
478 while ((reloc_offset = this->track_relocs_->next_offset()) != -1)
480 const unsigned int sym = this->track_relocs_->next_symndx();
481 const unsigned int shndx = this->symbol_section(sym, &value);
482 this->reloc_map_[reloc_offset] = std::make_pair(shndx, value);
483 this->track_relocs_->advance(reloc_offset + 1);
487 // Read the line number info.
489 template<int size, bool big_endian>
491 Dwarf_line_info<size, big_endian>::read_line_mappings()
494 while (this->buffer_ < this->buffer_end_)
496 const unsigned char* lineptr = this->buffer_;
497 lineptr = this->read_header_prolog(lineptr);
498 lineptr = this->read_header_tables(lineptr);
499 lineptr = this->read_lines(lineptr);
500 this->buffer_ = lineptr;
503 // Sort the lines numbers, so addr2line can use binary search.
504 for (typename Lineno_map::iterator it = line_number_map_.begin();
505 it != line_number_map_.end();
507 // Each vector needs to be sorted by offset.
508 std::sort(it->second.begin(), it->second.end());
511 // Return a string for a file name and line number.
513 template<int size, bool big_endian>
515 Dwarf_line_info<size, big_endian>::addr2line(unsigned int shndx, off_t offset)
517 if (this->data_valid_ == false)
520 const Offset_to_lineno_entry lookup_key = { offset, 0, 0 };
521 std::vector<Offset_to_lineno_entry>& offsets = this->line_number_map_[shndx];
525 typename std::vector<Offset_to_lineno_entry>::const_iterator it
526 = std::lower_bound(offsets.begin(), offsets.end(), lookup_key);
528 // If we found an exact match, great, otherwise find the last entry
529 // before the passed-in offset.
530 if (it->offset > offset)
532 if (it == offsets.begin())
535 gold_assert(it->offset < offset);
538 // Convert the file_num + line_num into a string.
540 gold_assert(it->file_num < static_cast<int>(files_.size()));
541 const std::pair<int, std::string>& filename_pair = files_[it->file_num];
542 gold_assert(filename_pair.first < static_cast<int>(directories_.size()));
543 const std::string& dirname = directories_[filename_pair.first];
544 const std::string& filename = filename_pair.second;
545 if (!dirname.empty())
554 char buffer[64]; // enough to hold a line number
555 snprintf(buffer, sizeof(buffer), "%d", it->line_num);
562 #ifdef HAVE_TARGET_32_LITTLE
564 class Dwarf_line_info<32, false>;
567 #ifdef HAVE_TARGET_32_BIG
569 class Dwarf_line_info<32, true>;
572 #ifdef HAVE_TARGET_64_LITTLE
574 class Dwarf_line_info<64, false>;
577 #ifdef HAVE_TARGET_64_BIG
579 class Dwarf_line_info<64, true>;
582 } // End namespace gold.