DW_LNE_end_sequence indicates that the address points to the byte
[deliverable/binutils-gdb.git] / gold / dwarf_reader.cc
1 // dwarf_reader.cc -- parse dwarf2/3 debug information
2
3 // Copyright 2007 Free Software Foundation, Inc.
4 // Written by Ian Lance Taylor <iant@google.com>.
5
6 // This file is part of gold.
7
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.
12
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.
17
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.
22
23 #include "gold.h"
24
25 #include "elfcpp_swap.h"
26 #include "dwarf.h"
27 #include "object.h"
28 #include "parameters.h"
29 #include "reloc.h"
30 #include "dwarf_reader.h"
31
32 namespace {
33
34 // Read an unsigned LEB128 number. Each byte contains 7 bits of
35 // information, plus one bit saying whether the number continues or
36 // not.
37
38 uint64_t
39 read_unsigned_LEB_128(const unsigned char* buffer, size_t* len)
40 {
41 uint64_t result = 0;
42 size_t num_read = 0;
43 unsigned int shift = 0;
44 unsigned char byte;
45
46 do
47 {
48 byte = *buffer++;
49 num_read++;
50 result |= (static_cast<uint64_t>(byte & 0x7f)) << shift;
51 shift += 7;
52 }
53 while (byte & 0x80);
54
55 *len = num_read;
56
57 return result;
58 }
59
60 // Read a signed LEB128 number. These are like regular LEB128
61 // numbers, except the last byte may have a sign bit set.
62
63 int64_t
64 read_signed_LEB_128(const unsigned char* buffer, size_t* len)
65 {
66 int64_t result = 0;
67 int shift = 0;
68 size_t num_read = 0;
69 unsigned char byte;
70
71 do
72 {
73 byte = *buffer++;
74 num_read++;
75 result |= (static_cast<uint64_t>(byte & 0x7f) << shift);
76 shift += 7;
77 }
78 while (byte & 0x80);
79
80 if ((shift < 8 * static_cast<int>(sizeof(result))) && (byte & 0x40))
81 result |= -((static_cast<int64_t>(1)) << shift);
82 *len = num_read;
83 return result;
84 }
85
86 } // End anonymous namespace.
87
88
89 namespace gold {
90
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.
94
95 struct LineStateMachine
96 {
97 int file_num;
98 uint64_t address;
99 int line_num;
100 int column_num;
101 unsigned int shndx; // the section address refers to
102 bool is_stmt; // stmt means statement.
103 bool basic_block;
104 bool end_sequence;
105 };
106
107 static void
108 ResetLineStateMachine(struct LineStateMachine* lsm, bool default_is_stmt)
109 {
110 lsm->file_num = 1;
111 lsm->address = 0;
112 lsm->line_num = 1;
113 lsm->column_num = 0;
114 lsm->shndx = -1U;
115 lsm->is_stmt = default_is_stmt;
116 lsm->basic_block = false;
117 lsm->end_sequence = false;
118 }
119
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)
124 {
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")
128 {
129 off_t buffer_size;
130 this->buffer_ = object->section_contents(
131 debug_shndx, &buffer_size, false);
132 this->buffer_end_ = this->buffer_ + buffer_size;
133 break;
134 }
135 if (this->buffer_ == NULL)
136 return;
137
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();
143 ++reloc_shndx)
144 {
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)
149 {
150 got_relocs = this->track_relocs_.initialize(object, reloc_shndx,
151 reloc_sh_type);
152 break;
153 }
154 }
155
156 // Finally, we need the symtab section to interpret the relocs.
157 if (got_relocs)
158 {
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)
162 {
163 this->symtab_buffer_ = object->section_contents(
164 symtab_shndx, &this->symtab_buffer_size_, false);
165 break;
166 }
167 if (this->symtab_buffer_ == NULL)
168 return;
169 }
170
171 // Now that we have successfully read all the data, parse the debug
172 // info.
173 this->data_valid_ = true;
174 this->read_line_mappings();
175 }
176
177 // Read the DWARF header.
178
179 template<int size, bool big_endian>
180 const unsigned char*
181 Sized_dwarf_line_info<size, big_endian>::read_header_prolog(
182 const unsigned char* lineptr)
183 {
184 uint32_t initial_length = elfcpp::Swap<32, big_endian>::readval(lineptr);
185 lineptr += 4;
186
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)
190 {
191 header_.offset_size = 8;
192 initial_length = elfcpp::Swap<64, big_endian>::readval(lineptr);
193 lineptr += 8;
194 }
195 else
196 header_.offset_size = 4;
197
198 header_.total_length = initial_length;
199
200 gold_assert(lineptr + header_.total_length <= buffer_end_);
201
202 header_.version = elfcpp::Swap<16, big_endian>::readval(lineptr);
203 lineptr += 2;
204
205 if (header_.offset_size == 4)
206 header_.prologue_length = elfcpp::Swap<32, big_endian>::readval(lineptr);
207 else
208 header_.prologue_length = elfcpp::Swap<64, big_endian>::readval(lineptr);
209 lineptr += header_.offset_size;
210
211 header_.min_insn_length = *lineptr;
212 lineptr += 1;
213
214 header_.default_is_stmt = *lineptr;
215 lineptr += 1;
216
217 header_.line_base = *reinterpret_cast<const signed char*>(lineptr);
218 lineptr += 1;
219
220 header_.line_range = *lineptr;
221 lineptr += 1;
222
223 header_.opcode_base = *lineptr;
224 lineptr += 1;
225
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++)
229 {
230 header_.std_opcode_lengths[i] = *lineptr;
231 lineptr += 1;
232 }
233
234 return lineptr;
235 }
236
237 // The header for a debug_line section is mildly complicated, because
238 // the line info is very tightly encoded.
239
240 template<int size, bool big_endian>
241 const unsigned char*
242 Sized_dwarf_line_info<size, big_endian>::read_header_tables(
243 const unsigned char* lineptr)
244 {
245 ++this->current_header_index_;
246
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));
256
257 // It is legal for the directory entry table to be empty.
258 if (*lineptr)
259 {
260 int dirindex = 1;
261 while (*lineptr)
262 {
263 const char* dirname = reinterpret_cast<const char*>(lineptr);
264 gold_assert(dirindex
265 == static_cast<int>(this->directories_.back().size()));
266 this->directories_.back().push_back(dirname);
267 lineptr += this->directories_.back().back().size() + 1;
268 dirindex++;
269 }
270 }
271 lineptr++;
272
273 // It is also legal for the file entry table to be empty.
274 if (*lineptr)
275 {
276 int fileindex = 1;
277 size_t len;
278 while (*lineptr)
279 {
280 const char* filename = reinterpret_cast<const char*>(lineptr);
281 lineptr += strlen(filename) + 1;
282
283 uint64_t dirindex = read_unsigned_LEB_128(lineptr, &len);
284 lineptr += len;
285
286 if (dirindex >= this->directories_.back().size())
287 dirindex = 0;
288 int dirindexi = static_cast<int>(dirindex);
289
290 read_unsigned_LEB_128(lineptr, &len); // mod_time
291 lineptr += len;
292
293 read_unsigned_LEB_128(lineptr, &len); // filelength
294 lineptr += len;
295
296 gold_assert(fileindex
297 == static_cast<int>(this->files_.back().size()));
298 this->files_.back().push_back(std::make_pair(dirindexi, filename));
299 fileindex++;
300 }
301 }
302 lineptr++;
303
304 return lineptr;
305 }
306
307 // Process a single opcode in the .debug.line structure.
308
309 // Templating on size and big_endian would yield more efficient (and
310 // simpler) code, but would bloat the binary. Speed isn't important
311 // here.
312
313 template<int size, bool big_endian>
314 bool
315 Sized_dwarf_line_info<size, big_endian>::process_one_opcode(
316 const unsigned char* start, struct LineStateMachine* lsm, size_t* len)
317 {
318 size_t oplen = 0;
319 size_t templen;
320 unsigned char opcode = *start;
321 oplen++;
322 start++;
323
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)
327 {
328 opcode -= header_.opcode_base;
329 const int advance_address = ((opcode / header_.line_range)
330 * header_.min_insn_length);
331 lsm->address += advance_address;
332
333 const int advance_line = ((opcode % header_.line_range)
334 + header_.line_base);
335 lsm->line_num += advance_line;
336 lsm->basic_block = true;
337 *len = oplen;
338 return true;
339 }
340
341 // Otherwise, we have the regular opcodes
342 switch (opcode)
343 {
344 case elfcpp::DW_LNS_copy:
345 lsm->basic_block = false;
346 *len = oplen;
347 return true;
348
349 case elfcpp::DW_LNS_advance_pc:
350 {
351 const uint64_t advance_address
352 = read_unsigned_LEB_128(start, &templen);
353 oplen += templen;
354 lsm->address += header_.min_insn_length * advance_address;
355 }
356 break;
357
358 case elfcpp::DW_LNS_advance_line:
359 {
360 const uint64_t advance_line = read_signed_LEB_128(start, &templen);
361 oplen += templen;
362 lsm->line_num += advance_line;
363 }
364 break;
365
366 case elfcpp::DW_LNS_set_file:
367 {
368 const uint64_t fileno = read_unsigned_LEB_128(start, &templen);
369 oplen += templen;
370 lsm->file_num = fileno;
371 }
372 break;
373
374 case elfcpp::DW_LNS_set_column:
375 {
376 const uint64_t colno = read_unsigned_LEB_128(start, &templen);
377 oplen += templen;
378 lsm->column_num = colno;
379 }
380 break;
381
382 case elfcpp::DW_LNS_negate_stmt:
383 lsm->is_stmt = !lsm->is_stmt;
384 break;
385
386 case elfcpp::DW_LNS_set_basic_block:
387 lsm->basic_block = true;
388 break;
389
390 case elfcpp::DW_LNS_fixed_advance_pc:
391 {
392 int advance_address;
393 advance_address = elfcpp::Swap<16, big_endian>::readval(start);
394 oplen += 2;
395 lsm->address += advance_address;
396 }
397 break;
398
399 case elfcpp::DW_LNS_const_add_pc:
400 {
401 const int advance_address = (header_.min_insn_length
402 * ((255 - header_.opcode_base)
403 / header_.line_range));
404 lsm->address += advance_address;
405 }
406 break;
407
408 case elfcpp::DW_LNS_extended_op:
409 {
410 const uint64_t extended_op_len
411 = read_unsigned_LEB_128(start, &templen);
412 start += templen;
413 oplen += templen + extended_op_len;
414
415 const unsigned char extended_op = *start;
416 start++;
417
418 switch (extended_op)
419 {
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.
424 lsm->address -= 1;
425 lsm->end_sequence = true;
426 *len = oplen;
427 return true;
428
429 case elfcpp::DW_LNE_set_address:
430 {
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())
435 {
436 // value + addend.
437 lsm->address += it->second.second;
438 lsm->shndx = it->second.first;
439 }
440 else
441 {
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;
446 }
447 break;
448 }
449 case elfcpp::DW_LNE_define_file:
450 {
451 const char* filename = reinterpret_cast<const char*>(start);
452 templen = strlen(filename) + 1;
453 start += templen;
454
455 uint64_t dirindex = read_unsigned_LEB_128(start, &templen);
456 oplen += templen;
457
458 if (dirindex >= this->directories_.back().size())
459 dirindex = 0;
460 int dirindexi = static_cast<int>(dirindex);
461
462 read_unsigned_LEB_128(start, &templen); // mod_time
463 oplen += templen;
464
465 read_unsigned_LEB_128(start, &templen); // filelength
466 oplen += templen;
467
468 this->files_.back().push_back(std::make_pair(dirindexi,
469 filename));
470 }
471 break;
472 }
473 }
474 break;
475
476 default:
477 {
478 // Ignore unknown opcode silently
479 for (int i = 0; i < header_.std_opcode_lengths[opcode]; i++)
480 {
481 size_t templen;
482 read_unsigned_LEB_128(start, &templen);
483 start += templen;
484 oplen += templen;
485 }
486 }
487 break;
488 }
489 *len = oplen;
490 return false;
491 }
492
493 // Read the debug information at LINEPTR and store it in the line
494 // number map.
495
496 template<int size, bool big_endian>
497 unsigned const char*
498 Sized_dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr)
499 {
500 struct LineStateMachine lsm;
501
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_;
505
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)
509 lengthstart += 12;
510 else
511 lengthstart += 4;
512
513 while (lineptr < lengthstart + header_.total_length)
514 {
515 ResetLineStateMachine(&lsm, header_.default_is_stmt);
516 while (!lsm.end_sequence)
517 {
518 size_t oplength;
519 bool add_line = this->process_one_opcode(lineptr, &lsm, &oplength);
520 if (add_line)
521 {
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);
526 }
527 lineptr += oplength;
528 }
529 }
530
531 return lengthstart + header_.total_length;
532 }
533
534 // Looks in the symtab to see what section a symbol is in.
535
536 template<int size, bool big_endian>
537 unsigned int
538 Sized_dwarf_line_info<size, big_endian>::symbol_section(
539 unsigned int sym,
540 typename elfcpp::Elf_types<size>::Elf_Addr* value)
541 {
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();
547 }
548
549 // Read the relocations into a Reloc_map.
550
551 template<int size, bool big_endian>
552 void
553 Sized_dwarf_line_info<size, big_endian>::read_relocs()
554 {
555 if (this->symtab_buffer_ == NULL)
556 return;
557
558 typename elfcpp::Elf_types<size>::Elf_Addr value;
559 off_t reloc_offset;
560 while ((reloc_offset = this->track_relocs_.next_offset()) != -1)
561 {
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);
566 }
567 }
568
569 // Read the line number info.
570
571 template<int size, bool big_endian>
572 void
573 Sized_dwarf_line_info<size, big_endian>::read_line_mappings()
574 {
575 gold_assert(this->data_valid_ == true);
576
577 read_relocs();
578 while (this->buffer_ < this->buffer_end_)
579 {
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;
585 }
586
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();
590 ++it)
591 // Each vector needs to be sorted by offset.
592 std::sort(it->second.begin(), it->second.end());
593 }
594
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
599 // information.
600
601 template<int size, bool big_endian>
602 bool
603 Sized_dwarf_line_info<size, big_endian>::input_is_relobj()
604 {
605 // Only .o files have relocs and the symtab buffer that goes with them.
606 return this->symtab_buffer_ != NULL;
607 }
608
609
610 // Return a string for a file name and line number.
611
612 template<int size, bool big_endian>
613 std::string
614 Sized_dwarf_line_info<size, big_endian>::do_addr2line(unsigned int shndx,
615 off_t offset)
616 {
617 if (this->data_valid_ == false)
618 return "";
619
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];
627 else
628 offsets = &this->line_number_map_[-1U];
629 if (offsets->empty())
630 return "";
631
632 typename std::vector<Offset_to_lineno_entry>::const_iterator it
633 = std::lower_bound(offsets->begin(), offsets->end(), lookup_key);
634
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)
638 {
639 if (it == offsets->begin())
640 return "";
641 --it;
642 gold_assert(it->offset < offset);
643 }
644
645 // Convert the file_num + line_num into a string.
646 std::string ret;
647
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;
654
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];
660
661 if (!dirname.empty())
662 {
663 ret += dirname;
664 ret += "/";
665 }
666 ret += filename;
667 if (ret.empty())
668 ret = "(unknown)";
669
670 char buffer[64]; // enough to hold a line number
671 snprintf(buffer, sizeof(buffer), "%d", it->line_num);
672 ret += ":";
673 ret += buffer;
674
675 return ret;
676 }
677
678 // Dwarf_line_info routines.
679
680 std::string
681 Dwarf_line_info::one_addr2line(Object* object,
682 unsigned int shndx, off_t offset)
683 {
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);
687 #else
688 gold_unreachable();
689 #endif
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);
693 #else
694 gold_unreachable();
695 #endif
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);
699 #else
700 gold_unreachable();
701 #endif
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);
705 #else
706 gold_unreachable();
707 #endif
708 else
709 gold_unreachable();
710 }
711
712 #ifdef HAVE_TARGET_32_LITTLE
713 template
714 class Sized_dwarf_line_info<32, false>;
715 #endif
716
717 #ifdef HAVE_TARGET_32_BIG
718 template
719 class Sized_dwarf_line_info<32, true>;
720 #endif
721
722 #ifdef HAVE_TARGET_64_LITTLE
723 template
724 class Sized_dwarf_line_info<64, false>;
725 #endif
726
727 #ifdef HAVE_TARGET_64_BIG
728 template
729 class Sized_dwarf_line_info<64, true>;
730 #endif
731
732 } // End namespace gold.
This page took 0.045486 seconds and 5 git commands to generate.