From Craig Silverstein: Clean up DWARF line reader code.
[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 lsm->end_sequence = true;
422 *len = oplen;
423 return true;
424
425 case elfcpp::DW_LNE_set_address:
426 {
427 lsm->address = elfcpp::Swap<size, big_endian>::readval(start);
428 typename Reloc_map::const_iterator it
429 = reloc_map_.find(start - this->buffer_);
430 if (it != reloc_map_.end())
431 {
432 // value + addend.
433 lsm->address += it->second.second;
434 lsm->shndx = it->second.first;
435 }
436 else
437 {
438 // If we're a normal .o file, with relocs, every
439 // set_address should have an associated relocation.
440 if (this->input_is_relobj())
441 this->data_valid_ = false;
442 }
443 break;
444 }
445 case elfcpp::DW_LNE_define_file:
446 {
447 const char* filename = reinterpret_cast<const char*>(start);
448 templen = strlen(filename) + 1;
449 start += templen;
450
451 uint64_t dirindex = read_unsigned_LEB_128(start, &templen);
452 oplen += templen;
453
454 if (dirindex >= this->directories_.back().size())
455 dirindex = 0;
456 int dirindexi = static_cast<int>(dirindex);
457
458 read_unsigned_LEB_128(start, &templen); // mod_time
459 oplen += templen;
460
461 read_unsigned_LEB_128(start, &templen); // filelength
462 oplen += templen;
463
464 this->files_.back().push_back(std::make_pair(dirindexi,
465 filename));
466 }
467 break;
468 }
469 }
470 break;
471
472 default:
473 {
474 // Ignore unknown opcode silently
475 for (int i = 0; i < header_.std_opcode_lengths[opcode]; i++)
476 {
477 size_t templen;
478 read_unsigned_LEB_128(start, &templen);
479 start += templen;
480 oplen += templen;
481 }
482 }
483 break;
484 }
485 *len = oplen;
486 return false;
487 }
488
489 // Read the debug information at LINEPTR and store it in the line
490 // number map.
491
492 template<int size, bool big_endian>
493 unsigned const char*
494 Sized_dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr)
495 {
496 struct LineStateMachine lsm;
497
498 // LENGTHSTART is the place the length field is based on. It is the
499 // point in the header after the initial length field.
500 const unsigned char* lengthstart = buffer_;
501
502 // In 64 bit dwarf, the initial length is 12 bytes, because of the
503 // 0xffffffff at the start.
504 if (header_.offset_size == 8)
505 lengthstart += 12;
506 else
507 lengthstart += 4;
508
509 while (lineptr < lengthstart + header_.total_length)
510 {
511 ResetLineStateMachine(&lsm, header_.default_is_stmt);
512 while (!lsm.end_sequence)
513 {
514 size_t oplength;
515 bool add_line = this->process_one_opcode(lineptr, &lsm, &oplength);
516 if (add_line)
517 {
518 Offset_to_lineno_entry entry
519 = { lsm.address, this->current_header_index_,
520 lsm.file_num, lsm.line_num };
521 line_number_map_[lsm.shndx].push_back(entry);
522 }
523 lineptr += oplength;
524 }
525 }
526
527 return lengthstart + header_.total_length;
528 }
529
530 // Looks in the symtab to see what section a symbol is in.
531
532 template<int size, bool big_endian>
533 unsigned int
534 Sized_dwarf_line_info<size, big_endian>::symbol_section(
535 unsigned int sym,
536 typename elfcpp::Elf_types<size>::Elf_Addr* value)
537 {
538 const int symsize = elfcpp::Elf_sizes<size>::sym_size;
539 gold_assert(sym * symsize < this->symtab_buffer_size_);
540 elfcpp::Sym<size, big_endian> elfsym(this->symtab_buffer_ + sym * symsize);
541 *value = elfsym.get_st_value();
542 return elfsym.get_st_shndx();
543 }
544
545 // Read the relocations into a Reloc_map.
546
547 template<int size, bool big_endian>
548 void
549 Sized_dwarf_line_info<size, big_endian>::read_relocs()
550 {
551 if (this->symtab_buffer_ == NULL)
552 return;
553
554 typename elfcpp::Elf_types<size>::Elf_Addr value;
555 off_t reloc_offset;
556 while ((reloc_offset = this->track_relocs_.next_offset()) != -1)
557 {
558 const unsigned int sym = this->track_relocs_.next_symndx();
559 const unsigned int shndx = this->symbol_section(sym, &value);
560 this->reloc_map_[reloc_offset] = std::make_pair(shndx, value);
561 this->track_relocs_.advance(reloc_offset + 1);
562 }
563 }
564
565 // Read the line number info.
566
567 template<int size, bool big_endian>
568 void
569 Sized_dwarf_line_info<size, big_endian>::read_line_mappings()
570 {
571 gold_assert(this->data_valid_ == true);
572
573 read_relocs();
574 while (this->buffer_ < this->buffer_end_)
575 {
576 const unsigned char* lineptr = this->buffer_;
577 lineptr = this->read_header_prolog(lineptr);
578 lineptr = this->read_header_tables(lineptr);
579 lineptr = this->read_lines(lineptr);
580 this->buffer_ = lineptr;
581 }
582
583 // Sort the lines numbers, so addr2line can use binary search.
584 for (typename Lineno_map::iterator it = line_number_map_.begin();
585 it != line_number_map_.end();
586 ++it)
587 // Each vector needs to be sorted by offset.
588 std::sort(it->second.begin(), it->second.end());
589 }
590
591 // Some processing depends on whether the input is a .o file or not.
592 // For instance, .o files have relocs, and have .debug_lines
593 // information on a per section basis. .so files, on the other hand,
594 // lack relocs, and offsets are unique, so we can ignore the section
595 // information.
596
597 template<int size, bool big_endian>
598 bool
599 Sized_dwarf_line_info<size, big_endian>::input_is_relobj()
600 {
601 // Only .o files have relocs and the symtab buffer that goes with them.
602 return this->symtab_buffer_ != NULL;
603 }
604
605
606 // Return a string for a file name and line number.
607
608 template<int size, bool big_endian>
609 std::string
610 Sized_dwarf_line_info<size, big_endian>::do_addr2line(unsigned int shndx,
611 off_t offset)
612 {
613 if (this->data_valid_ == false)
614 return "";
615
616 const Offset_to_lineno_entry lookup_key = { offset, 0, 0, 0 };
617 const std::vector<Offset_to_lineno_entry>* offsets;
618 // If we do not have reloc information, then our input is a .so or
619 // some similar data structure where all the information is held in
620 // the offset. In that case, we ignore the input shndx.
621 if (this->input_is_relobj())
622 offsets = &this->line_number_map_[shndx];
623 else
624 offsets = &this->line_number_map_[-1U];
625 if (offsets->empty())
626 return "";
627
628 typename std::vector<Offset_to_lineno_entry>::const_iterator it
629 = std::lower_bound(offsets->begin(), offsets->end(), lookup_key);
630
631 // If we found an exact match, great, otherwise find the last entry
632 // before the passed-in offset.
633 if (it == offsets->end() || it->offset > offset)
634 {
635 if (it == offsets->begin())
636 return "";
637 --it;
638 gold_assert(it->offset < offset);
639 }
640
641 // Convert the file_num + line_num into a string.
642 std::string ret;
643
644 gold_assert(it->header_num < static_cast<int>(this->files_.size()));
645 gold_assert(it->file_num
646 < static_cast<int>(this->files_[it->header_num].size()));
647 const std::pair<int, std::string>& filename_pair
648 = this->files_[it->header_num][it->file_num];
649 const std::string& filename = filename_pair.second;
650
651 gold_assert(it->header_num < static_cast<int>(this->directories_.size()));
652 gold_assert(filename_pair.first
653 < static_cast<int>(this->directories_[it->header_num].size()));
654 const std::string& dirname
655 = this->directories_[it->header_num][filename_pair.first];
656
657 if (!dirname.empty())
658 {
659 ret += dirname;
660 ret += "/";
661 }
662 ret += filename;
663 if (ret.empty())
664 ret = "(unknown)";
665
666 char buffer[64]; // enough to hold a line number
667 snprintf(buffer, sizeof(buffer), "%d", it->line_num);
668 ret += ":";
669 ret += buffer;
670
671 return ret;
672 }
673
674 // Dwarf_line_info routines.
675
676 std::string
677 Dwarf_line_info::one_addr2line(Object* object,
678 unsigned int shndx, off_t offset)
679 {
680 if (parameters->get_size() == 32 && !parameters->is_big_endian())
681 #ifdef HAVE_TARGET_32_LITTLE
682 return Sized_dwarf_line_info<32, false>(object).addr2line(shndx, offset);
683 #else
684 gold_unreachable();
685 #endif
686 else if (parameters->get_size() == 32 && parameters->is_big_endian())
687 #ifdef HAVE_TARGET_32_BIG
688 return Sized_dwarf_line_info<32, true>(object).addr2line(shndx, offset);
689 #else
690 gold_unreachable();
691 #endif
692 else if (parameters->get_size() == 64 && !parameters->is_big_endian())
693 #ifdef HAVE_TARGET_64_LITTLE
694 return Sized_dwarf_line_info<64, false>(object).addr2line(shndx, offset);
695 #else
696 gold_unreachable();
697 #endif
698 else if (parameters->get_size() == 64 && parameters->is_big_endian())
699 #ifdef HAVE_TARGET_64_BIT
700 return Sized_dwarf_line_info<64, true>(object).addr2line(shndx, offset);
701 #else
702 gold_unreachable();
703 #endif
704 else
705 gold_unreachable();
706 }
707
708 #ifdef HAVE_TARGET_32_LITTLE
709 template
710 class Sized_dwarf_line_info<32, false>;
711 #endif
712
713 #ifdef HAVE_TARGET_32_BIG
714 template
715 class Sized_dwarf_line_info<32, true>;
716 #endif
717
718 #ifdef HAVE_TARGET_64_LITTLE
719 template
720 class Sized_dwarf_line_info<64, false>;
721 #endif
722
723 #ifdef HAVE_TARGET_64_BIG
724 template
725 class Sized_dwarf_line_info<64, true>;
726 #endif
727
728 } // End namespace gold.
This page took 0.04471 seconds and 5 git commands to generate.