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"
27 #include "dwarf_reader.h"
31 // Read an unsigned LEB128 number. Each byte contains 7 bits of
32 // information, plus one bit saying whether the number continues or
36 read_unsigned_LEB_128(const unsigned char* buffer
, size_t* len
)
40 unsigned int shift
= 0;
47 result
|= (static_cast<uint64_t>(byte
& 0x7f)) << shift
;
57 // Read a signed LEB128 number. These are like regular LEB128
58 // numbers, except the last byte may have a sign bit set.
61 read_signed_LEB_128(const unsigned char* buffer
, size_t* len
)
72 result
|= (static_cast<uint64_t>(byte
& 0x7f) << shift
);
77 if ((shift
< 8 * static_cast<int>(sizeof(result
))) && (byte
& 0x40))
78 result
|= -((static_cast<int64_t>(1)) << shift
);
83 } // End anonymous namespace.
88 // This is the format of a DWARF2/3 line state machine that we process
89 // opcodes using. There is no need for anything outside the lineinfo
90 // processor to know how this works.
92 struct LineStateMachine
98 unsigned int shndx
; // the section address refers to
99 bool is_stmt
; // stmt means statement.
105 ResetLineStateMachine(struct LineStateMachine
* lsm
, bool default_is_stmt
)
112 lsm
->is_stmt
= default_is_stmt
;
113 lsm
->basic_block
= false;
114 lsm
->end_sequence
= false;
117 // Read the DWARF header.
119 template<int size
, bool big_endian
>
121 Dwarf_line_info
<size
, big_endian
>::read_header_prolog(
122 const unsigned char* lineptr
)
124 uint32_t initial_length
= elfcpp::Swap
<32, big_endian
>::readval(lineptr
);
127 // In DWARF2/3, if the initial length is all 1 bits, then the offset
128 // size is 8 and we need to read the next 8 bytes for the real length.
129 if (initial_length
== 0xffffffff)
131 header_
.offset_size
= 8;
132 initial_length
= elfcpp::Swap
<64, big_endian
>::readval(lineptr
);
136 header_
.offset_size
= 4;
138 header_
.total_length
= initial_length
;
140 gold_assert(lineptr
+ header_
.total_length
<= buffer_end_
);
142 header_
.version
= elfcpp::Swap
<16, big_endian
>::readval(lineptr
);
145 if (header_
.offset_size
== 4)
146 header_
.prologue_length
= elfcpp::Swap
<32, big_endian
>::readval(lineptr
);
148 header_
.prologue_length
= elfcpp::Swap
<64, big_endian
>::readval(lineptr
);
149 lineptr
+= header_
.offset_size
;
151 header_
.min_insn_length
= *lineptr
;
154 header_
.default_is_stmt
= *lineptr
;
157 header_
.line_base
= *reinterpret_cast<const signed char*>(lineptr
);
160 header_
.line_range
= *lineptr
;
163 header_
.opcode_base
= *lineptr
;
166 header_
.std_opcode_lengths
.reserve(header_
.opcode_base
+ 1);
167 header_
.std_opcode_lengths
[0] = 0;
168 for (int i
= 1; i
< header_
.opcode_base
; i
++)
170 header_
.std_opcode_lengths
[i
] = *lineptr
;
177 // The header for a debug_line section is mildly complicated, because
178 // the line info is very tightly encoded.
180 template<int size
, bool big_endian
>
182 Dwarf_line_info
<size
, big_endian
>::read_header_tables(
183 const unsigned char* lineptr
)
185 // It is legal for the directory entry table to be empty.
191 const unsigned char* dirname
= lineptr
;
192 gold_assert(dirindex
== static_cast<int>(directories_
.size()));
193 directories_
.push_back(reinterpret_cast<const char*>(dirname
));
194 lineptr
+= directories_
.back().size() + 1;
200 // It is also legal for the file entry table to be empty.
207 const char* filename
= reinterpret_cast<const char*>(lineptr
);
208 lineptr
+= strlen(filename
) + 1;
210 uint64_t dirindex
= read_unsigned_LEB_128(lineptr
, &len
);
211 if (dirindex
>= directories_
.size())
215 read_unsigned_LEB_128(lineptr
, &len
); // mod_time
218 read_unsigned_LEB_128(lineptr
, &len
); // filelength
221 gold_assert(fileindex
== static_cast<int>(files_
.size()));
222 files_
.push_back(std::pair
<int, std::string
>(dirindex
, filename
));
231 // Process a single opcode in the .debug.line structure.
233 // Templating on size and big_endian would yield more efficient (and
234 // simpler) code, but would bloat the binary. Speed isn't important
237 template<int size
, bool big_endian
>
239 Dwarf_line_info
<size
, big_endian
>::process_one_opcode(
240 const unsigned char* start
, struct LineStateMachine
* lsm
, size_t* len
)
244 unsigned char opcode
= *start
;
248 // If the opcode is great than the opcode_base, it is a special
249 // opcode. Most line programs consist mainly of special opcodes.
250 if (opcode
>= header_
.opcode_base
)
252 opcode
-= header_
.opcode_base
;
253 const int advance_address
= ((opcode
/ header_
.line_range
)
254 * header_
.min_insn_length
);
255 lsm
->address
+= advance_address
;
257 const int advance_line
= ((opcode
% header_
.line_range
)
258 + header_
.line_base
);
259 lsm
->line_num
+= advance_line
;
260 lsm
->basic_block
= true;
265 // Otherwise, we have the regular opcodes
268 case elfcpp::DW_LNS_copy
:
269 lsm
->basic_block
= false;
273 case elfcpp::DW_LNS_advance_pc
:
275 const uint64_t advance_address
276 = read_unsigned_LEB_128(start
, &templen
);
278 lsm
->address
+= header_
.min_insn_length
* advance_address
;
282 case elfcpp::DW_LNS_advance_line
:
284 const uint64_t advance_line
= read_signed_LEB_128(start
, &templen
);
286 lsm
->line_num
+= advance_line
;
290 case elfcpp::DW_LNS_set_file
:
292 const uint64_t fileno
= read_unsigned_LEB_128(start
, &templen
);
294 lsm
->file_num
= fileno
;
298 case elfcpp::DW_LNS_set_column
:
300 const uint64_t colno
= read_unsigned_LEB_128(start
, &templen
);
302 lsm
->column_num
= colno
;
306 case elfcpp::DW_LNS_negate_stmt
:
307 lsm
->is_stmt
= !lsm
->is_stmt
;
310 case elfcpp::DW_LNS_set_basic_block
:
311 lsm
->basic_block
= true;
314 case elfcpp::DW_LNS_fixed_advance_pc
:
317 advance_address
= elfcpp::Swap
<16, big_endian
>::readval(start
);
319 lsm
->address
+= advance_address
;
323 case elfcpp::DW_LNS_const_add_pc
:
325 const int advance_address
= (header_
.min_insn_length
326 * ((255 - header_
.opcode_base
)
327 / header_
.line_range
));
328 lsm
->address
+= advance_address
;
332 case elfcpp::DW_LNS_extended_op
:
334 const uint64_t extended_op_len
335 = read_unsigned_LEB_128(start
, &templen
);
337 oplen
+= templen
+ extended_op_len
;
339 const unsigned char extended_op
= *start
;
344 case elfcpp::DW_LNE_end_sequence
:
345 lsm
->end_sequence
= true;
349 case elfcpp::DW_LNE_set_address
:
350 // FIXME: modify the address based on the reloc
351 lsm
->address
= elfcpp::Swap
<size
, big_endian
>::readval(start
);
352 // FIXME: set lsm->shndx from the reloc
356 case elfcpp::DW_LNE_define_file
:
358 const char* filename
= reinterpret_cast<const char*>(start
);
359 templen
= strlen(filename
) + 1;
362 uint64_t dirindex
= read_unsigned_LEB_128(start
, &templen
);
363 if (dirindex
>= directories_
.size())
367 read_unsigned_LEB_128(start
, &templen
); // mod_time
370 read_unsigned_LEB_128(start
, &templen
); // filelength
373 files_
.push_back(std::pair
<int, std::string
>(dirindex
,
383 // Ignore unknown opcode silently
384 for (int i
= 0; i
< header_
.std_opcode_lengths
[opcode
]; i
++)
387 read_unsigned_LEB_128(start
, &templen
);
398 // Read the debug information at LINEPTR and store it in the line
401 template<int size
, bool big_endian
>
403 Dwarf_line_info
<size
, big_endian
>::read_lines(unsigned const char* lineptr
)
405 struct LineStateMachine lsm
;
407 // LENGTHSTART is the place the length field is based on. It is the
408 // point in the header after the initial length field.
409 const unsigned char* lengthstart
= buffer_
;
411 // In 64 bit dwarf, the initial length is 12 bytes, because of the
412 // 0xffffffff at the start.
413 if (header_
.offset_size
== 8)
418 while (lineptr
< lengthstart
+ header_
.total_length
)
420 ResetLineStateMachine(&lsm
, header_
.default_is_stmt
);
421 while (!lsm
.end_sequence
)
424 bool add_line
= this->process_one_opcode(lineptr
, &lsm
, &oplength
);
427 Offset_to_lineno_entry entry
428 = { lsm
.address
, lsm
.file_num
, lsm
.line_num
};
429 line_number_map_
[lsm
.shndx
].push_back(entry
);
435 return lengthstart
+ header_
.total_length
;
438 template<int size
, bool big_endian
>
440 Dwarf_line_info
<size
, big_endian
>::read_line_mappings()
442 while (buffer_
< buffer_end_
)
444 const unsigned char* lineptr
= buffer_
;
445 lineptr
= this->read_header_prolog(lineptr
);
446 lineptr
= this->read_header_tables(lineptr
);
447 lineptr
= this->read_lines(lineptr
);
451 // Sort the lines numbers, so addr2line can use binary search.
452 for (typename
Lineno_map::iterator it
= line_number_map_
.begin();
453 it
!= line_number_map_
.end();
455 // Each vector needs to be sorted by offset.
456 sort(it
->second
.begin(), it
->second
.end());
459 // Return a string for a file name and line number.
461 template<int size
, bool big_endian
>
463 Dwarf_line_info
<size
, big_endian
>::addr2line(unsigned int shndx
, off_t offset
)
465 const Offset_to_lineno_entry lookup_key
= { offset
, 0, 0 };
466 std::vector
<Offset_to_lineno_entry
>& offsets
= line_number_map_
[shndx
];
467 typename
std::vector
<Offset_to_lineno_entry
>::const_iterator it
468 = std::lower_bound(offsets
.begin(), offsets
.end(), lookup_key
);
470 // If we found an exact match, great, otherwise find the last entry
471 // before the passed-in offset.
472 if (it
->offset
> offset
)
474 if (it
== offsets
.begin())
477 gold_assert(it
->offset
< offset
);
480 // Convert the file_num + line_num into a string.
482 gold_assert(it
->file_num
< static_cast<int>(files_
.size()));
483 const std::pair
<int, std::string
>& filename_pair
= files_
[it
->file_num
];
484 gold_assert(filename_pair
.first
< static_cast<int>(directories_
.size()));
485 const std::string
& dirname
= directories_
[filename_pair
.first
];
486 const std::string
& filename
= filename_pair
.second
;
487 if (!dirname
.empty())
496 char buffer
[64]; // enough to hold a line number
497 snprintf(buffer
, sizeof(buffer
), "%d", it
->line_num
);
504 #ifdef HAVE_TARGET_32_LITTLE
506 class Dwarf_line_info
<32, false>;
509 #ifdef HAVE_TARGET_32_BIG
511 class Dwarf_line_info
<32, true>;
514 #ifdef HAVE_TARGET_64_LITTLE
516 class Dwarf_line_info
<64, false>;
519 #ifdef HAVE_TARGET_64_BIG
521 class Dwarf_line_info
<64, true>;
524 } // End namespace gold.