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"
29 #include "dwarf_reader.h"
33 // Read an unsigned LEB128 number. Each byte contains 7 bits of
34 // information, plus one bit saying whether the number continues or
38 read_unsigned_LEB_128(const unsigned char* buffer
, size_t* len
)
42 unsigned int shift
= 0;
49 result
|= (static_cast<uint64_t>(byte
& 0x7f)) << shift
;
59 // Read a signed LEB128 number. These are like regular LEB128
60 // numbers, except the last byte may have a sign bit set.
63 read_signed_LEB_128(const unsigned char* buffer
, size_t* len
)
74 result
|= (static_cast<uint64_t>(byte
& 0x7f) << shift
);
79 if ((shift
< 8 * static_cast<int>(sizeof(result
))) && (byte
& 0x40))
80 result
|= -((static_cast<int64_t>(1)) << shift
);
85 } // End anonymous namespace.
90 // This is the format of a DWARF2/3 line state machine that we process
91 // opcodes using. There is no need for anything outside the lineinfo
92 // processor to know how this works.
94 struct LineStateMachine
100 unsigned int shndx
; // the section address refers to
101 bool is_stmt
; // stmt means statement.
107 ResetLineStateMachine(struct LineStateMachine
* lsm
, bool default_is_stmt
)
114 lsm
->is_stmt
= default_is_stmt
;
115 lsm
->basic_block
= false;
116 lsm
->end_sequence
= false;
119 template<int size
, bool big_endian
>
120 Dwarf_line_info
<size
, big_endian
>::Dwarf_line_info(Object
* object
)
121 : data_valid_(false), buffer_(NULL
), symtab_buffer_(NULL
),
122 directories_(1), files_(1)
124 unsigned int debug_shndx
;
125 for (debug_shndx
= 0; debug_shndx
< object
->shnum(); ++debug_shndx
)
126 if (object
->section_name(debug_shndx
) == ".debug_line")
129 this->buffer_
= object
->section_contents(
130 debug_shndx
, &buffer_size
, false);
131 this->buffer_end_
= this->buffer_
+ buffer_size
;
134 if (this->buffer_
== NULL
)
137 // Find the relocation section for ".debug_line".
138 bool got_relocs
= false;
139 for (unsigned int reloc_shndx
= 0;
140 reloc_shndx
< object
->shnum();
143 unsigned int reloc_sh_type
= object
->section_type(reloc_shndx
);
144 if ((reloc_sh_type
== elfcpp::SHT_REL
145 || reloc_sh_type
== elfcpp::SHT_RELA
)
146 && object
->section_info(reloc_shndx
) == debug_shndx
)
148 got_relocs
= this->track_relocs_
.initialize(object
, reloc_shndx
,
156 // Finally, we need the symtab section to interpret the relocs.
157 unsigned int symtab_shndx
;
158 for (symtab_shndx
= 0; symtab_shndx
< object
->shnum(); ++symtab_shndx
)
159 if (object
->section_type(symtab_shndx
) == elfcpp::SHT_SYMTAB
)
162 this->symtab_buffer_
= object
->section_contents(
163 symtab_shndx
, &symtab_size
, false);
164 this->symtab_buffer_end_
= this->symtab_buffer_
+ symtab_size
;
167 if (this->symtab_buffer_
== NULL
)
170 // Now that we have successfully read all the data, parse the debug
172 this->data_valid_
= true;
173 this->read_line_mappings();
176 // Read the DWARF header.
178 template<int size
, bool big_endian
>
180 Dwarf_line_info
<size
, big_endian
>::read_header_prolog(
181 const unsigned char* lineptr
)
183 uint32_t initial_length
= elfcpp::Swap
<32, big_endian
>::readval(lineptr
);
186 // In DWARF2/3, if the initial length is all 1 bits, then the offset
187 // size is 8 and we need to read the next 8 bytes for the real length.
188 if (initial_length
== 0xffffffff)
190 header_
.offset_size
= 8;
191 initial_length
= elfcpp::Swap
<64, big_endian
>::readval(lineptr
);
195 header_
.offset_size
= 4;
197 header_
.total_length
= initial_length
;
199 gold_assert(lineptr
+ header_
.total_length
<= buffer_end_
);
201 header_
.version
= elfcpp::Swap
<16, big_endian
>::readval(lineptr
);
204 if (header_
.offset_size
== 4)
205 header_
.prologue_length
= elfcpp::Swap
<32, big_endian
>::readval(lineptr
);
207 header_
.prologue_length
= elfcpp::Swap
<64, big_endian
>::readval(lineptr
);
208 lineptr
+= header_
.offset_size
;
210 header_
.min_insn_length
= *lineptr
;
213 header_
.default_is_stmt
= *lineptr
;
216 header_
.line_base
= *reinterpret_cast<const signed char*>(lineptr
);
219 header_
.line_range
= *lineptr
;
222 header_
.opcode_base
= *lineptr
;
225 header_
.std_opcode_lengths
.reserve(header_
.opcode_base
+ 1);
226 header_
.std_opcode_lengths
[0] = 0;
227 for (int i
= 1; i
< header_
.opcode_base
; i
++)
229 header_
.std_opcode_lengths
[i
] = *lineptr
;
236 // The header for a debug_line section is mildly complicated, because
237 // the line info is very tightly encoded.
239 template<int size
, bool big_endian
>
241 Dwarf_line_info
<size
, big_endian
>::read_header_tables(
242 const unsigned char* lineptr
)
244 // It is legal for the directory entry table to be empty.
250 const unsigned char* dirname
= lineptr
;
251 gold_assert(dirindex
== static_cast<int>(directories_
.size()));
252 directories_
.push_back(reinterpret_cast<const char*>(dirname
));
253 lineptr
+= directories_
.back().size() + 1;
259 // It is also legal for the file entry table to be empty.
266 const char* filename
= reinterpret_cast<const char*>(lineptr
);
267 lineptr
+= strlen(filename
) + 1;
269 uint64_t dirindex
= read_unsigned_LEB_128(lineptr
, &len
);
270 if (dirindex
>= directories_
.size())
274 read_unsigned_LEB_128(lineptr
, &len
); // mod_time
277 read_unsigned_LEB_128(lineptr
, &len
); // filelength
280 gold_assert(fileindex
== static_cast<int>(files_
.size()));
281 files_
.push_back(std::pair
<int, std::string
>(dirindex
, filename
));
290 // Process a single opcode in the .debug.line structure.
292 // Templating on size and big_endian would yield more efficient (and
293 // simpler) code, but would bloat the binary. Speed isn't important
296 template<int size
, bool big_endian
>
298 Dwarf_line_info
<size
, big_endian
>::process_one_opcode(
299 const unsigned char* start
, struct LineStateMachine
* lsm
, size_t* len
)
303 unsigned char opcode
= *start
;
307 // If the opcode is great than the opcode_base, it is a special
308 // opcode. Most line programs consist mainly of special opcodes.
309 if (opcode
>= header_
.opcode_base
)
311 opcode
-= header_
.opcode_base
;
312 const int advance_address
= ((opcode
/ header_
.line_range
)
313 * header_
.min_insn_length
);
314 lsm
->address
+= advance_address
;
316 const int advance_line
= ((opcode
% header_
.line_range
)
317 + header_
.line_base
);
318 lsm
->line_num
+= advance_line
;
319 lsm
->basic_block
= true;
324 // Otherwise, we have the regular opcodes
327 case elfcpp::DW_LNS_copy
:
328 lsm
->basic_block
= false;
332 case elfcpp::DW_LNS_advance_pc
:
334 const uint64_t advance_address
335 = read_unsigned_LEB_128(start
, &templen
);
337 lsm
->address
+= header_
.min_insn_length
* advance_address
;
341 case elfcpp::DW_LNS_advance_line
:
343 const uint64_t advance_line
= read_signed_LEB_128(start
, &templen
);
345 lsm
->line_num
+= advance_line
;
349 case elfcpp::DW_LNS_set_file
:
351 const uint64_t fileno
= read_unsigned_LEB_128(start
, &templen
);
353 lsm
->file_num
= fileno
;
357 case elfcpp::DW_LNS_set_column
:
359 const uint64_t colno
= read_unsigned_LEB_128(start
, &templen
);
361 lsm
->column_num
= colno
;
365 case elfcpp::DW_LNS_negate_stmt
:
366 lsm
->is_stmt
= !lsm
->is_stmt
;
369 case elfcpp::DW_LNS_set_basic_block
:
370 lsm
->basic_block
= true;
373 case elfcpp::DW_LNS_fixed_advance_pc
:
376 advance_address
= elfcpp::Swap
<16, big_endian
>::readval(start
);
378 lsm
->address
+= advance_address
;
382 case elfcpp::DW_LNS_const_add_pc
:
384 const int advance_address
= (header_
.min_insn_length
385 * ((255 - header_
.opcode_base
)
386 / header_
.line_range
));
387 lsm
->address
+= advance_address
;
391 case elfcpp::DW_LNS_extended_op
:
393 const uint64_t extended_op_len
394 = read_unsigned_LEB_128(start
, &templen
);
396 oplen
+= templen
+ extended_op_len
;
398 const unsigned char extended_op
= *start
;
403 case elfcpp::DW_LNE_end_sequence
:
404 lsm
->end_sequence
= true;
408 case elfcpp::DW_LNE_set_address
:
410 typename
Reloc_map::const_iterator it
411 = reloc_map_
.find(start
- this->buffer_
);
412 if (it
!= reloc_map_
.end())
416 (elfcpp::Swap
<size
, big_endian
>::readval(start
)
417 + it
->second
.second
);
418 lsm
->shndx
= it
->second
.first
;
422 // Every set_address should have an associated
424 this->data_valid_
= false;
428 case elfcpp::DW_LNE_define_file
:
430 const char* filename
= reinterpret_cast<const char*>(start
);
431 templen
= strlen(filename
) + 1;
434 uint64_t dirindex
= read_unsigned_LEB_128(start
, &templen
);
435 if (dirindex
>= directories_
.size())
439 read_unsigned_LEB_128(start
, &templen
); // mod_time
442 read_unsigned_LEB_128(start
, &templen
); // filelength
445 files_
.push_back(std::pair
<int, std::string
>(dirindex
,
455 // Ignore unknown opcode silently
456 for (int i
= 0; i
< header_
.std_opcode_lengths
[opcode
]; i
++)
459 read_unsigned_LEB_128(start
, &templen
);
470 // Read the debug information at LINEPTR and store it in the line
473 template<int size
, bool big_endian
>
475 Dwarf_line_info
<size
, big_endian
>::read_lines(unsigned const char* lineptr
)
477 struct LineStateMachine lsm
;
479 // LENGTHSTART is the place the length field is based on. It is the
480 // point in the header after the initial length field.
481 const unsigned char* lengthstart
= buffer_
;
483 // In 64 bit dwarf, the initial length is 12 bytes, because of the
484 // 0xffffffff at the start.
485 if (header_
.offset_size
== 8)
490 while (lineptr
< lengthstart
+ header_
.total_length
)
492 ResetLineStateMachine(&lsm
, header_
.default_is_stmt
);
493 while (!lsm
.end_sequence
)
496 bool add_line
= this->process_one_opcode(lineptr
, &lsm
, &oplength
);
499 Offset_to_lineno_entry entry
500 = { lsm
.address
, lsm
.file_num
, lsm
.line_num
};
501 line_number_map_
[lsm
.shndx
].push_back(entry
);
507 return lengthstart
+ header_
.total_length
;
510 // Looks in the symtab to see what section a symbol is in.
512 template<int size
, bool big_endian
>
514 Dwarf_line_info
<size
, big_endian
>::symbol_section(
516 typename
elfcpp::Elf_types
<size
>::Elf_Addr
* value
)
518 const int symsize
= elfcpp::Elf_sizes
<size
>::sym_size
;
519 gold_assert(this->symtab_buffer_
+ sym
* symsize
< this->symtab_buffer_end_
);
520 elfcpp::Sym
<size
, big_endian
> elfsym(this->symtab_buffer_
+ sym
* symsize
);
521 *value
= elfsym
.get_st_value();
522 return elfsym
.get_st_shndx();
525 // Read the relocations into a Reloc_map.
527 template<int size
, bool big_endian
>
529 Dwarf_line_info
<size
, big_endian
>::read_relocs()
531 if (this->symtab_buffer_
== NULL
)
534 typename
elfcpp::Elf_types
<size
>::Elf_Addr value
;
536 while ((reloc_offset
= this->track_relocs_
.next_offset()) != -1)
538 const unsigned int sym
= this->track_relocs_
.next_symndx();
539 const unsigned int shndx
= this->symbol_section(sym
, &value
);
540 this->reloc_map_
[reloc_offset
] = std::make_pair(shndx
, value
);
541 this->track_relocs_
.advance(reloc_offset
+ 1);
545 // Read the line number info.
547 template<int size
, bool big_endian
>
549 Dwarf_line_info
<size
, big_endian
>::read_line_mappings()
551 gold_assert(this->data_valid_
== true);
554 while (this->buffer_
< this->buffer_end_
)
556 const unsigned char* lineptr
= this->buffer_
;
557 lineptr
= this->read_header_prolog(lineptr
);
558 lineptr
= this->read_header_tables(lineptr
);
559 lineptr
= this->read_lines(lineptr
);
560 this->buffer_
= lineptr
;
563 // Sort the lines numbers, so addr2line can use binary search.
564 for (typename
Lineno_map::iterator it
= line_number_map_
.begin();
565 it
!= line_number_map_
.end();
567 // Each vector needs to be sorted by offset.
568 std::sort(it
->second
.begin(), it
->second
.end());
571 // Return a string for a file name and line number.
573 template<int size
, bool big_endian
>
575 Dwarf_line_info
<size
, big_endian
>::addr2line(unsigned int shndx
, off_t offset
)
577 if (this->data_valid_
== false)
580 const Offset_to_lineno_entry lookup_key
= { offset
, 0, 0 };
581 std::vector
<Offset_to_lineno_entry
>& offsets
= this->line_number_map_
[shndx
];
585 typename
std::vector
<Offset_to_lineno_entry
>::const_iterator it
586 = std::lower_bound(offsets
.begin(), offsets
.end(), lookup_key
);
588 // If we found an exact match, great, otherwise find the last entry
589 // before the passed-in offset.
590 if (it
->offset
> offset
)
592 if (it
== offsets
.begin())
595 gold_assert(it
->offset
< offset
);
598 // Convert the file_num + line_num into a string.
600 gold_assert(it
->file_num
< static_cast<int>(files_
.size()));
601 const std::pair
<int, std::string
>& filename_pair
= files_
[it
->file_num
];
602 gold_assert(filename_pair
.first
< static_cast<int>(directories_
.size()));
603 const std::string
& dirname
= directories_
[filename_pair
.first
];
604 const std::string
& filename
= filename_pair
.second
;
605 if (!dirname
.empty())
614 char buffer
[64]; // enough to hold a line number
615 snprintf(buffer
, sizeof(buffer
), "%d", it
->line_num
);
622 #ifdef HAVE_TARGET_32_LITTLE
624 class Dwarf_line_info
<32, false>;
627 #ifdef HAVE_TARGET_32_BIG
629 class Dwarf_line_info
<32, true>;
632 #ifdef HAVE_TARGET_64_LITTLE
634 class Dwarf_line_info
<64, false>;
637 #ifdef HAVE_TARGET_64_BIG
639 class Dwarf_line_info
<64, true>;
642 } // End namespace gold.