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(
121 Sized_relobj
<size
, big_endian
>* object
)
122 : data_valid_(true), buffer_(NULL
), symtab_buffer_(NULL
),
123 directories_(1), files_(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
)
137 this->data_valid_
= false;
141 // Find the relocation section for ".debug_line".
142 bool got_relocs
= false;
143 for (unsigned int reloc_shndx
= 0;
144 reloc_shndx
< object
->shnum();
147 unsigned int reloc_sh_type
= object
->section_type(reloc_shndx
);
148 if ((reloc_sh_type
== elfcpp::SHT_REL
149 || reloc_sh_type
== elfcpp::SHT_RELA
)
150 && object
->section_info(reloc_shndx
) == debug_shndx
)
152 got_relocs
= this->track_relocs_
.initialize(object
, reloc_shndx
,
159 this->data_valid_
= false;
163 // Finally, we need the symtab section to interpret the relocs.
164 unsigned int symtab_shndx
;
165 for (symtab_shndx
= 0; symtab_shndx
< object
->shnum(); ++symtab_shndx
)
166 if (object
->section_type(symtab_shndx
) == elfcpp::SHT_SYMTAB
)
169 this->symtab_buffer_
= object
->section_contents(
170 symtab_shndx
, &symtab_size
, false);
171 this->symtab_buffer_end_
= this->symtab_buffer_
+ symtab_size
;
174 if (this->symtab_buffer_
== NULL
)
176 this->data_valid_
= false;
180 // Now that we have successfully read all the data, parse the debug
182 this->read_line_mappings();
185 // Read the DWARF header.
187 template<int size
, bool big_endian
>
189 Dwarf_line_info
<size
, big_endian
>::read_header_prolog(
190 const unsigned char* lineptr
)
192 uint32_t initial_length
= elfcpp::Swap
<32, big_endian
>::readval(lineptr
);
195 // In DWARF2/3, if the initial length is all 1 bits, then the offset
196 // size is 8 and we need to read the next 8 bytes for the real length.
197 if (initial_length
== 0xffffffff)
199 header_
.offset_size
= 8;
200 initial_length
= elfcpp::Swap
<64, big_endian
>::readval(lineptr
);
204 header_
.offset_size
= 4;
206 header_
.total_length
= initial_length
;
208 gold_assert(lineptr
+ header_
.total_length
<= buffer_end_
);
210 header_
.version
= elfcpp::Swap
<16, big_endian
>::readval(lineptr
);
213 if (header_
.offset_size
== 4)
214 header_
.prologue_length
= elfcpp::Swap
<32, big_endian
>::readval(lineptr
);
216 header_
.prologue_length
= elfcpp::Swap
<64, big_endian
>::readval(lineptr
);
217 lineptr
+= header_
.offset_size
;
219 header_
.min_insn_length
= *lineptr
;
222 header_
.default_is_stmt
= *lineptr
;
225 header_
.line_base
= *reinterpret_cast<const signed char*>(lineptr
);
228 header_
.line_range
= *lineptr
;
231 header_
.opcode_base
= *lineptr
;
234 header_
.std_opcode_lengths
.reserve(header_
.opcode_base
+ 1);
235 header_
.std_opcode_lengths
[0] = 0;
236 for (int i
= 1; i
< header_
.opcode_base
; i
++)
238 header_
.std_opcode_lengths
[i
] = *lineptr
;
245 // The header for a debug_line section is mildly complicated, because
246 // the line info is very tightly encoded.
248 template<int size
, bool big_endian
>
250 Dwarf_line_info
<size
, big_endian
>::read_header_tables(
251 const unsigned char* lineptr
)
253 // It is legal for the directory entry table to be empty.
259 const unsigned char* dirname
= lineptr
;
260 gold_assert(dirindex
== static_cast<int>(directories_
.size()));
261 directories_
.push_back(reinterpret_cast<const char*>(dirname
));
262 lineptr
+= directories_
.back().size() + 1;
268 // It is also legal for the file entry table to be empty.
275 const char* filename
= reinterpret_cast<const char*>(lineptr
);
276 lineptr
+= strlen(filename
) + 1;
278 uint64_t dirindex
= read_unsigned_LEB_128(lineptr
, &len
);
279 if (dirindex
>= directories_
.size())
283 read_unsigned_LEB_128(lineptr
, &len
); // mod_time
286 read_unsigned_LEB_128(lineptr
, &len
); // filelength
289 gold_assert(fileindex
== static_cast<int>(files_
.size()));
290 files_
.push_back(std::pair
<int, std::string
>(dirindex
, filename
));
299 // Process a single opcode in the .debug.line structure.
301 // Templating on size and big_endian would yield more efficient (and
302 // simpler) code, but would bloat the binary. Speed isn't important
305 template<int size
, bool big_endian
>
307 Dwarf_line_info
<size
, big_endian
>::process_one_opcode(
308 const unsigned char* start
, struct LineStateMachine
* lsm
, size_t* len
)
312 unsigned char opcode
= *start
;
316 // If the opcode is great than the opcode_base, it is a special
317 // opcode. Most line programs consist mainly of special opcodes.
318 if (opcode
>= header_
.opcode_base
)
320 opcode
-= header_
.opcode_base
;
321 const int advance_address
= ((opcode
/ header_
.line_range
)
322 * header_
.min_insn_length
);
323 lsm
->address
+= advance_address
;
325 const int advance_line
= ((opcode
% header_
.line_range
)
326 + header_
.line_base
);
327 lsm
->line_num
+= advance_line
;
328 lsm
->basic_block
= true;
333 // Otherwise, we have the regular opcodes
336 case elfcpp::DW_LNS_copy
:
337 lsm
->basic_block
= false;
341 case elfcpp::DW_LNS_advance_pc
:
343 const uint64_t advance_address
344 = read_unsigned_LEB_128(start
, &templen
);
346 lsm
->address
+= header_
.min_insn_length
* advance_address
;
350 case elfcpp::DW_LNS_advance_line
:
352 const uint64_t advance_line
= read_signed_LEB_128(start
, &templen
);
354 lsm
->line_num
+= advance_line
;
358 case elfcpp::DW_LNS_set_file
:
360 const uint64_t fileno
= read_unsigned_LEB_128(start
, &templen
);
362 lsm
->file_num
= fileno
;
366 case elfcpp::DW_LNS_set_column
:
368 const uint64_t colno
= read_unsigned_LEB_128(start
, &templen
);
370 lsm
->column_num
= colno
;
374 case elfcpp::DW_LNS_negate_stmt
:
375 lsm
->is_stmt
= !lsm
->is_stmt
;
378 case elfcpp::DW_LNS_set_basic_block
:
379 lsm
->basic_block
= true;
382 case elfcpp::DW_LNS_fixed_advance_pc
:
385 advance_address
= elfcpp::Swap
<16, big_endian
>::readval(start
);
387 lsm
->address
+= advance_address
;
391 case elfcpp::DW_LNS_const_add_pc
:
393 const int advance_address
= (header_
.min_insn_length
394 * ((255 - header_
.opcode_base
)
395 / header_
.line_range
));
396 lsm
->address
+= advance_address
;
400 case elfcpp::DW_LNS_extended_op
:
402 const uint64_t extended_op_len
403 = read_unsigned_LEB_128(start
, &templen
);
405 oplen
+= templen
+ extended_op_len
;
407 const unsigned char extended_op
= *start
;
412 case elfcpp::DW_LNE_end_sequence
:
413 lsm
->end_sequence
= true;
417 case elfcpp::DW_LNE_set_address
:
419 typename
Reloc_map::const_iterator it
420 = reloc_map_
.find(start
- this->buffer_
);
421 if (it
!= reloc_map_
.end())
425 (elfcpp::Swap
<size
, big_endian
>::readval(start
)
426 + it
->second
.second
);
427 lsm
->shndx
= it
->second
.first
;
431 // Every set_address should have an associated
433 this->data_valid_
= false;
437 case elfcpp::DW_LNE_define_file
:
439 const char* filename
= reinterpret_cast<const char*>(start
);
440 templen
= strlen(filename
) + 1;
443 uint64_t dirindex
= read_unsigned_LEB_128(start
, &templen
);
444 if (dirindex
>= directories_
.size())
448 read_unsigned_LEB_128(start
, &templen
); // mod_time
451 read_unsigned_LEB_128(start
, &templen
); // filelength
454 files_
.push_back(std::pair
<int, std::string
>(dirindex
,
464 // Ignore unknown opcode silently
465 for (int i
= 0; i
< header_
.std_opcode_lengths
[opcode
]; i
++)
468 read_unsigned_LEB_128(start
, &templen
);
479 // Read the debug information at LINEPTR and store it in the line
482 template<int size
, bool big_endian
>
484 Dwarf_line_info
<size
, big_endian
>::read_lines(unsigned const char* lineptr
)
486 struct LineStateMachine lsm
;
488 // LENGTHSTART is the place the length field is based on. It is the
489 // point in the header after the initial length field.
490 const unsigned char* lengthstart
= buffer_
;
492 // In 64 bit dwarf, the initial length is 12 bytes, because of the
493 // 0xffffffff at the start.
494 if (header_
.offset_size
== 8)
499 while (lineptr
< lengthstart
+ header_
.total_length
)
501 ResetLineStateMachine(&lsm
, header_
.default_is_stmt
);
502 while (!lsm
.end_sequence
)
505 bool add_line
= this->process_one_opcode(lineptr
, &lsm
, &oplength
);
508 Offset_to_lineno_entry entry
509 = { lsm
.address
, lsm
.file_num
, lsm
.line_num
};
510 line_number_map_
[lsm
.shndx
].push_back(entry
);
516 return lengthstart
+ header_
.total_length
;
519 // Looks in the symtab to see what section a symbol is in.
521 template<int size
, bool big_endian
>
523 Dwarf_line_info
<size
, big_endian
>::symbol_section(
525 typename
elfcpp::Elf_types
<size
>::Elf_Addr
* value
)
527 const int symsize
= elfcpp::Elf_sizes
<size
>::sym_size
;
528 gold_assert(this->symtab_buffer_
+ sym
* symsize
< this->symtab_buffer_end_
);
529 elfcpp::Sym
<size
, big_endian
> elfsym(this->symtab_buffer_
+ sym
* symsize
);
530 *value
= elfsym
.get_st_value();
531 return elfsym
.get_st_shndx();
534 // Read the relocations into a Reloc_map.
536 template<int size
, bool big_endian
>
538 Dwarf_line_info
<size
, big_endian
>::read_relocs()
540 if (this->symtab_buffer_
== NULL
)
543 typename
elfcpp::Elf_types
<size
>::Elf_Addr value
;
545 while ((reloc_offset
= this->track_relocs_
.next_offset()) != -1)
547 const unsigned int sym
= this->track_relocs_
.next_symndx();
548 const unsigned int shndx
= this->symbol_section(sym
, &value
);
549 this->reloc_map_
[reloc_offset
] = std::make_pair(shndx
, value
);
550 this->track_relocs_
.advance(reloc_offset
+ 1);
554 // Read the line number info.
556 template<int size
, bool big_endian
>
558 Dwarf_line_info
<size
, big_endian
>::read_line_mappings()
560 if (this->data_valid_
== false)
564 while (this->buffer_
< this->buffer_end_
)
566 const unsigned char* lineptr
= this->buffer_
;
567 lineptr
= this->read_header_prolog(lineptr
);
568 lineptr
= this->read_header_tables(lineptr
);
569 lineptr
= this->read_lines(lineptr
);
570 this->buffer_
= lineptr
;
573 // Sort the lines numbers, so addr2line can use binary search.
574 for (typename
Lineno_map::iterator it
= line_number_map_
.begin();
575 it
!= line_number_map_
.end();
577 // Each vector needs to be sorted by offset.
578 std::sort(it
->second
.begin(), it
->second
.end());
581 // Return a string for a file name and line number.
583 template<int size
, bool big_endian
>
585 Dwarf_line_info
<size
, big_endian
>::addr2line(unsigned int shndx
, off_t offset
)
587 if (this->data_valid_
== false)
590 const Offset_to_lineno_entry lookup_key
= { offset
, 0, 0 };
591 std::vector
<Offset_to_lineno_entry
>& offsets
= this->line_number_map_
[shndx
];
595 typename
std::vector
<Offset_to_lineno_entry
>::const_iterator it
596 = std::lower_bound(offsets
.begin(), offsets
.end(), lookup_key
);
598 // If we found an exact match, great, otherwise find the last entry
599 // before the passed-in offset.
600 if (it
->offset
> offset
)
602 if (it
== offsets
.begin())
605 gold_assert(it
->offset
< offset
);
608 // Convert the file_num + line_num into a string.
610 gold_assert(it
->file_num
< static_cast<int>(files_
.size()));
611 const std::pair
<int, std::string
>& filename_pair
= files_
[it
->file_num
];
612 gold_assert(filename_pair
.first
< static_cast<int>(directories_
.size()));
613 const std::string
& dirname
= directories_
[filename_pair
.first
];
614 const std::string
& filename
= filename_pair
.second
;
615 if (!dirname
.empty())
624 char buffer
[64]; // enough to hold a line number
625 snprintf(buffer
, sizeof(buffer
), "%d", it
->line_num
);
632 #ifdef HAVE_TARGET_32_LITTLE
634 class Dwarf_line_info
<32, false>;
637 #ifdef HAVE_TARGET_32_BIG
639 class Dwarf_line_info
<32, true>;
642 #ifdef HAVE_TARGET_64_LITTLE
644 class Dwarf_line_info
<64, false>;
647 #ifdef HAVE_TARGET_64_BIG
649 class Dwarf_line_info
<64, true>;
652 } // End namespace gold.