* libtool.m4: Update to libtool 2.2.6.
[binutils.git] / gold / dwarf_reader.cc
blob3d0f65a4b5a1cfb4ad6469b3ba3116e63fba431c
1 // dwarf_reader.cc -- parse dwarf2/3 debug information
3 // Copyright 2007, 2008 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.
23 #include "gold.h"
25 #include <algorithm>
26 #include <vector>
28 #include "elfcpp_swap.h"
29 #include "dwarf.h"
30 #include "object.h"
31 #include "parameters.h"
32 #include "reloc.h"
33 #include "dwarf_reader.h"
35 namespace gold {
37 // Read an unsigned LEB128 number. Each byte contains 7 bits of
38 // information, plus one bit saying whether the number continues or
39 // not.
41 uint64_t
42 read_unsigned_LEB_128(const unsigned char* buffer, size_t* len)
44 uint64_t result = 0;
45 size_t num_read = 0;
46 unsigned int shift = 0;
47 unsigned char byte;
51 if (num_read >= 64 / 7)
53 gold_warning(_("Unusually large LEB128 decoded, "
54 "debug information may be corrupted"));
55 break;
57 byte = *buffer++;
58 num_read++;
59 result |= (static_cast<uint64_t>(byte & 0x7f)) << shift;
60 shift += 7;
62 while (byte & 0x80);
64 *len = num_read;
66 return result;
69 // Read a signed LEB128 number. These are like regular LEB128
70 // numbers, except the last byte may have a sign bit set.
72 int64_t
73 read_signed_LEB_128(const unsigned char* buffer, size_t* len)
75 int64_t result = 0;
76 int shift = 0;
77 size_t num_read = 0;
78 unsigned char byte;
82 if (num_read >= 64 / 7)
84 gold_warning(_("Unusually large LEB128 decoded, "
85 "debug information may be corrupted"));
86 break;
88 byte = *buffer++;
89 num_read++;
90 result |= (static_cast<uint64_t>(byte & 0x7f) << shift);
91 shift += 7;
93 while (byte & 0x80);
95 if ((shift < 8 * static_cast<int>(sizeof(result))) && (byte & 0x40))
96 result |= -((static_cast<int64_t>(1)) << shift);
97 *len = num_read;
98 return result;
101 // This is the format of a DWARF2/3 line state machine that we process
102 // opcodes using. There is no need for anything outside the lineinfo
103 // processor to know how this works.
105 struct LineStateMachine
107 int file_num;
108 uint64_t address;
109 int line_num;
110 int column_num;
111 unsigned int shndx; // the section address refers to
112 bool is_stmt; // stmt means statement.
113 bool basic_block;
114 bool end_sequence;
117 static void
118 ResetLineStateMachine(struct LineStateMachine* lsm, bool default_is_stmt)
120 lsm->file_num = 1;
121 lsm->address = 0;
122 lsm->line_num = 1;
123 lsm->column_num = 0;
124 lsm->shndx = -1U;
125 lsm->is_stmt = default_is_stmt;
126 lsm->basic_block = false;
127 lsm->end_sequence = false;
130 template<int size, bool big_endian>
131 Sized_dwarf_line_info<size, big_endian>::Sized_dwarf_line_info(Object* object,
132 off_t read_shndx)
133 : data_valid_(false), buffer_(NULL), symtab_buffer_(NULL),
134 directories_(), files_(), current_header_index_(-1)
136 unsigned int debug_shndx;
137 for (debug_shndx = 0; debug_shndx < object->shnum(); ++debug_shndx)
138 // FIXME: do this more efficiently: section_name() isn't super-fast
139 if (object->section_name(debug_shndx) == ".debug_line")
141 section_size_type buffer_size;
142 this->buffer_ = object->section_contents(debug_shndx, &buffer_size,
143 false);
144 this->buffer_end_ = this->buffer_ + buffer_size;
145 break;
147 if (this->buffer_ == NULL)
148 return;
150 // Find the relocation section for ".debug_line".
151 // We expect these for relobjs (.o's) but not dynobjs (.so's).
152 bool got_relocs = false;
153 for (unsigned int reloc_shndx = 0;
154 reloc_shndx < object->shnum();
155 ++reloc_shndx)
157 unsigned int reloc_sh_type = object->section_type(reloc_shndx);
158 if ((reloc_sh_type == elfcpp::SHT_REL
159 || reloc_sh_type == elfcpp::SHT_RELA)
160 && object->section_info(reloc_shndx) == debug_shndx)
162 got_relocs = this->track_relocs_.initialize(object, reloc_shndx,
163 reloc_sh_type);
164 break;
168 // Finally, we need the symtab section to interpret the relocs.
169 if (got_relocs)
171 unsigned int symtab_shndx;
172 for (symtab_shndx = 0; symtab_shndx < object->shnum(); ++symtab_shndx)
173 if (object->section_type(symtab_shndx) == elfcpp::SHT_SYMTAB)
175 this->symtab_buffer_ = object->section_contents(
176 symtab_shndx, &this->symtab_buffer_size_, false);
177 break;
179 if (this->symtab_buffer_ == NULL)
180 return;
183 // Now that we have successfully read all the data, parse the debug
184 // info.
185 this->data_valid_ = true;
186 this->read_line_mappings(object, read_shndx);
189 // Read the DWARF header.
191 template<int size, bool big_endian>
192 const unsigned char*
193 Sized_dwarf_line_info<size, big_endian>::read_header_prolog(
194 const unsigned char* lineptr)
196 uint32_t initial_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr);
197 lineptr += 4;
199 // In DWARF2/3, if the initial length is all 1 bits, then the offset
200 // size is 8 and we need to read the next 8 bytes for the real length.
201 if (initial_length == 0xffffffff)
203 header_.offset_size = 8;
204 initial_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr);
205 lineptr += 8;
207 else
208 header_.offset_size = 4;
210 header_.total_length = initial_length;
212 gold_assert(lineptr + header_.total_length <= buffer_end_);
214 header_.version = elfcpp::Swap_unaligned<16, big_endian>::readval(lineptr);
215 lineptr += 2;
217 if (header_.offset_size == 4)
218 header_.prologue_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr);
219 else
220 header_.prologue_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr);
221 lineptr += header_.offset_size;
223 header_.min_insn_length = *lineptr;
224 lineptr += 1;
226 header_.default_is_stmt = *lineptr;
227 lineptr += 1;
229 header_.line_base = *reinterpret_cast<const signed char*>(lineptr);
230 lineptr += 1;
232 header_.line_range = *lineptr;
233 lineptr += 1;
235 header_.opcode_base = *lineptr;
236 lineptr += 1;
238 header_.std_opcode_lengths.reserve(header_.opcode_base + 1);
239 header_.std_opcode_lengths[0] = 0;
240 for (int i = 1; i < header_.opcode_base; i++)
242 header_.std_opcode_lengths[i] = *lineptr;
243 lineptr += 1;
246 return lineptr;
249 // The header for a debug_line section is mildly complicated, because
250 // the line info is very tightly encoded.
252 template<int size, bool big_endian>
253 const unsigned char*
254 Sized_dwarf_line_info<size, big_endian>::read_header_tables(
255 const unsigned char* lineptr)
257 ++this->current_header_index_;
259 // Create a new directories_ entry and a new files_ entry for our new
260 // header. We initialize each with a single empty element, because
261 // dwarf indexes directory and filenames starting at 1.
262 gold_assert(static_cast<int>(this->directories_.size())
263 == this->current_header_index_);
264 gold_assert(static_cast<int>(this->files_.size())
265 == this->current_header_index_);
266 this->directories_.push_back(std::vector<std::string>(1));
267 this->files_.push_back(std::vector<std::pair<int, std::string> >(1));
269 // It is legal for the directory entry table to be empty.
270 if (*lineptr)
272 int dirindex = 1;
273 while (*lineptr)
275 const char* dirname = reinterpret_cast<const char*>(lineptr);
276 gold_assert(dirindex
277 == static_cast<int>(this->directories_.back().size()));
278 this->directories_.back().push_back(dirname);
279 lineptr += this->directories_.back().back().size() + 1;
280 dirindex++;
283 lineptr++;
285 // It is also legal for the file entry table to be empty.
286 if (*lineptr)
288 int fileindex = 1;
289 size_t len;
290 while (*lineptr)
292 const char* filename = reinterpret_cast<const char*>(lineptr);
293 lineptr += strlen(filename) + 1;
295 uint64_t dirindex = read_unsigned_LEB_128(lineptr, &len);
296 lineptr += len;
298 if (dirindex >= this->directories_.back().size())
299 dirindex = 0;
300 int dirindexi = static_cast<int>(dirindex);
302 read_unsigned_LEB_128(lineptr, &len); // mod_time
303 lineptr += len;
305 read_unsigned_LEB_128(lineptr, &len); // filelength
306 lineptr += len;
308 gold_assert(fileindex
309 == static_cast<int>(this->files_.back().size()));
310 this->files_.back().push_back(std::make_pair(dirindexi, filename));
311 fileindex++;
314 lineptr++;
316 return lineptr;
319 // Process a single opcode in the .debug.line structure.
321 // Templating on size and big_endian would yield more efficient (and
322 // simpler) code, but would bloat the binary. Speed isn't important
323 // here.
325 template<int size, bool big_endian>
326 bool
327 Sized_dwarf_line_info<size, big_endian>::process_one_opcode(
328 const unsigned char* start, struct LineStateMachine* lsm, size_t* len)
330 size_t oplen = 0;
331 size_t templen;
332 unsigned char opcode = *start;
333 oplen++;
334 start++;
336 // If the opcode is great than the opcode_base, it is a special
337 // opcode. Most line programs consist mainly of special opcodes.
338 if (opcode >= header_.opcode_base)
340 opcode -= header_.opcode_base;
341 const int advance_address = ((opcode / header_.line_range)
342 * header_.min_insn_length);
343 lsm->address += advance_address;
345 const int advance_line = ((opcode % header_.line_range)
346 + header_.line_base);
347 lsm->line_num += advance_line;
348 lsm->basic_block = true;
349 *len = oplen;
350 return true;
353 // Otherwise, we have the regular opcodes
354 switch (opcode)
356 case elfcpp::DW_LNS_copy:
357 lsm->basic_block = false;
358 *len = oplen;
359 return true;
361 case elfcpp::DW_LNS_advance_pc:
363 const uint64_t advance_address
364 = read_unsigned_LEB_128(start, &templen);
365 oplen += templen;
366 lsm->address += header_.min_insn_length * advance_address;
368 break;
370 case elfcpp::DW_LNS_advance_line:
372 const uint64_t advance_line = read_signed_LEB_128(start, &templen);
373 oplen += templen;
374 lsm->line_num += advance_line;
376 break;
378 case elfcpp::DW_LNS_set_file:
380 const uint64_t fileno = read_unsigned_LEB_128(start, &templen);
381 oplen += templen;
382 lsm->file_num = fileno;
384 break;
386 case elfcpp::DW_LNS_set_column:
388 const uint64_t colno = read_unsigned_LEB_128(start, &templen);
389 oplen += templen;
390 lsm->column_num = colno;
392 break;
394 case elfcpp::DW_LNS_negate_stmt:
395 lsm->is_stmt = !lsm->is_stmt;
396 break;
398 case elfcpp::DW_LNS_set_basic_block:
399 lsm->basic_block = true;
400 break;
402 case elfcpp::DW_LNS_fixed_advance_pc:
404 int advance_address;
405 advance_address = elfcpp::Swap_unaligned<16, big_endian>::readval(start);
406 oplen += 2;
407 lsm->address += advance_address;
409 break;
411 case elfcpp::DW_LNS_const_add_pc:
413 const int advance_address = (header_.min_insn_length
414 * ((255 - header_.opcode_base)
415 / header_.line_range));
416 lsm->address += advance_address;
418 break;
420 case elfcpp::DW_LNS_extended_op:
422 const uint64_t extended_op_len
423 = read_unsigned_LEB_128(start, &templen);
424 start += templen;
425 oplen += templen + extended_op_len;
427 const unsigned char extended_op = *start;
428 start++;
430 switch (extended_op)
432 case elfcpp::DW_LNE_end_sequence:
433 // This means that the current byte is the one immediately
434 // after a set of instructions. Record the current line
435 // for up to one less than the current address.
436 lsm->line_num = -1;
437 lsm->end_sequence = true;
438 *len = oplen;
439 return true;
441 case elfcpp::DW_LNE_set_address:
443 lsm->address = elfcpp::Swap_unaligned<size, big_endian>::readval(start);
444 typename Reloc_map::const_iterator it
445 = reloc_map_.find(start - this->buffer_);
446 if (it != reloc_map_.end())
448 // value + addend.
449 lsm->address += it->second.second;
450 lsm->shndx = it->second.first;
452 else
454 // If we're a normal .o file, with relocs, every
455 // set_address should have an associated relocation.
456 if (this->input_is_relobj())
457 this->data_valid_ = false;
459 break;
461 case elfcpp::DW_LNE_define_file:
463 const char* filename = reinterpret_cast<const char*>(start);
464 templen = strlen(filename) + 1;
465 start += templen;
467 uint64_t dirindex = read_unsigned_LEB_128(start, &templen);
468 oplen += templen;
470 if (dirindex >= this->directories_.back().size())
471 dirindex = 0;
472 int dirindexi = static_cast<int>(dirindex);
474 read_unsigned_LEB_128(start, &templen); // mod_time
475 oplen += templen;
477 read_unsigned_LEB_128(start, &templen); // filelength
478 oplen += templen;
480 this->files_.back().push_back(std::make_pair(dirindexi,
481 filename));
483 break;
486 break;
488 default:
490 // Ignore unknown opcode silently
491 for (int i = 0; i < header_.std_opcode_lengths[opcode]; i++)
493 size_t templen;
494 read_unsigned_LEB_128(start, &templen);
495 start += templen;
496 oplen += templen;
499 break;
501 *len = oplen;
502 return false;
505 // Read the debug information at LINEPTR and store it in the line
506 // number map.
508 template<int size, bool big_endian>
509 unsigned const char*
510 Sized_dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr,
511 off_t shndx)
513 struct LineStateMachine lsm;
515 // LENGTHSTART is the place the length field is based on. It is the
516 // point in the header after the initial length field.
517 const unsigned char* lengthstart = buffer_;
519 // In 64 bit dwarf, the initial length is 12 bytes, because of the
520 // 0xffffffff at the start.
521 if (header_.offset_size == 8)
522 lengthstart += 12;
523 else
524 lengthstart += 4;
526 while (lineptr < lengthstart + header_.total_length)
528 ResetLineStateMachine(&lsm, header_.default_is_stmt);
529 while (!lsm.end_sequence)
531 size_t oplength;
532 bool add_line = this->process_one_opcode(lineptr, &lsm, &oplength);
533 if (add_line
534 && (shndx == -1U || lsm.shndx == -1U || shndx == lsm.shndx))
536 Offset_to_lineno_entry entry
537 = { lsm.address, this->current_header_index_,
538 lsm.file_num, lsm.line_num };
539 line_number_map_[lsm.shndx].push_back(entry);
541 lineptr += oplength;
545 return lengthstart + header_.total_length;
548 // Looks in the symtab to see what section a symbol is in.
550 template<int size, bool big_endian>
551 unsigned int
552 Sized_dwarf_line_info<size, big_endian>::symbol_section(
553 Object* object,
554 unsigned int sym,
555 typename elfcpp::Elf_types<size>::Elf_Addr* value,
556 bool* is_ordinary)
558 const int symsize = elfcpp::Elf_sizes<size>::sym_size;
559 gold_assert(sym * symsize < this->symtab_buffer_size_);
560 elfcpp::Sym<size, big_endian> elfsym(this->symtab_buffer_ + sym * symsize);
561 *value = elfsym.get_st_value();
562 return object->adjust_sym_shndx(sym, elfsym.get_st_shndx(), is_ordinary);
565 // Read the relocations into a Reloc_map.
567 template<int size, bool big_endian>
568 void
569 Sized_dwarf_line_info<size, big_endian>::read_relocs(Object* object)
571 if (this->symtab_buffer_ == NULL)
572 return;
574 typename elfcpp::Elf_types<size>::Elf_Addr value;
575 off_t reloc_offset;
576 while ((reloc_offset = this->track_relocs_.next_offset()) != -1)
578 const unsigned int sym = this->track_relocs_.next_symndx();
580 bool is_ordinary;
581 const unsigned int shndx = this->symbol_section(object, sym, &value,
582 &is_ordinary);
584 // There is no reason to record non-ordinary section indexes, or
585 // SHN_UNDEF, because they will never match the real section.
586 if (is_ordinary && shndx != elfcpp::SHN_UNDEF)
587 this->reloc_map_[reloc_offset] = std::make_pair(shndx, value);
589 this->track_relocs_.advance(reloc_offset + 1);
593 // Read the line number info.
595 template<int size, bool big_endian>
596 void
597 Sized_dwarf_line_info<size, big_endian>::read_line_mappings(Object* object,
598 off_t shndx)
600 gold_assert(this->data_valid_ == true);
602 this->read_relocs(object);
603 while (this->buffer_ < this->buffer_end_)
605 const unsigned char* lineptr = this->buffer_;
606 lineptr = this->read_header_prolog(lineptr);
607 lineptr = this->read_header_tables(lineptr);
608 lineptr = this->read_lines(lineptr, shndx);
609 this->buffer_ = lineptr;
612 // Sort the lines numbers, so addr2line can use binary search.
613 for (typename Lineno_map::iterator it = line_number_map_.begin();
614 it != line_number_map_.end();
615 ++it)
616 // Each vector needs to be sorted by offset.
617 std::sort(it->second.begin(), it->second.end());
620 // Some processing depends on whether the input is a .o file or not.
621 // For instance, .o files have relocs, and have .debug_lines
622 // information on a per section basis. .so files, on the other hand,
623 // lack relocs, and offsets are unique, so we can ignore the section
624 // information.
626 template<int size, bool big_endian>
627 bool
628 Sized_dwarf_line_info<size, big_endian>::input_is_relobj()
630 // Only .o files have relocs and the symtab buffer that goes with them.
631 return this->symtab_buffer_ != NULL;
634 // Given an Offset_to_lineno_entry vector, and an offset, figure out
635 // if the offset points into a function according to the vector (see
636 // comments below for the algorithm). If it does, return an iterator
637 // into the vector that points to the line-number that contains that
638 // offset. If not, it returns vector::end().
640 static std::vector<Offset_to_lineno_entry>::const_iterator
641 offset_to_iterator(const std::vector<Offset_to_lineno_entry>* offsets,
642 off_t offset)
644 const Offset_to_lineno_entry lookup_key = { offset, 0, 0, 0 };
646 // lower_bound() returns the smallest offset which is >= lookup_key.
647 // If no offset in offsets is >= lookup_key, returns end().
648 std::vector<Offset_to_lineno_entry>::const_iterator it
649 = std::lower_bound(offsets->begin(), offsets->end(), lookup_key);
651 // This code is easiest to understand with a concrete example.
652 // Here's a possible offsets array:
653 // {{offset = 3211, header_num = 0, file_num = 1, line_num = 16}, // 0
654 // {offset = 3224, header_num = 0, file_num = 1, line_num = 20}, // 1
655 // {offset = 3226, header_num = 0, file_num = 1, line_num = 22}, // 2
656 // {offset = 3231, header_num = 0, file_num = 1, line_num = 25}, // 3
657 // {offset = 3232, header_num = 0, file_num = 1, line_num = -1}, // 4
658 // {offset = 3232, header_num = 0, file_num = 1, line_num = 65}, // 5
659 // {offset = 3235, header_num = 0, file_num = 1, line_num = 66}, // 6
660 // {offset = 3236, header_num = 0, file_num = 1, line_num = -1}, // 7
661 // {offset = 5764, header_num = 0, file_num = 1, line_num = 47}, // 8
662 // {offset = 5765, header_num = 0, file_num = 1, line_num = 48}, // 9
663 // {offset = 5767, header_num = 0, file_num = 1, line_num = 49}, // 10
664 // {offset = 5768, header_num = 0, file_num = 1, line_num = 50}, // 11
665 // {offset = 5773, header_num = 0, file_num = 1, line_num = -1}, // 12
666 // {offset = 5787, header_num = 1, file_num = 1, line_num = 19}, // 13
667 // {offset = 5790, header_num = 1, file_num = 1, line_num = 20}, // 14
668 // {offset = 5793, header_num = 1, file_num = 1, line_num = 67}, // 15
669 // {offset = 5793, header_num = 1, file_num = 1, line_num = -1}, // 16
670 // {offset = 5795, header_num = 1, file_num = 1, line_num = 68}, // 17
671 // {offset = 5798, header_num = 1, file_num = 1, line_num = -1}, // 18
672 // The entries with line_num == -1 mark the end of a function: the
673 // associated offset is one past the last instruction in the
674 // function. This can correspond to the beginning of the next
675 // function (as is true for offset 3232); alternately, there can be
676 // a gap between the end of one function and the start of the next
677 // (as is true for some others, most obviously from 3236->5764).
679 // Case 1: lookup_key has offset == 10. lower_bound returns
680 // offsets[0]. Since it's not an exact match and we're
681 // at the beginning of offsets, we return end() (invalid).
682 // Case 2: lookup_key has offset 10000. lower_bound returns
683 // offset[19] (end()). We return end() (invalid).
684 // Case 3: lookup_key has offset == 3211. lower_bound matches
685 // offsets[0] exactly, and that's the entry we return.
686 // Case 4: lookup_key has offset == 3232. lower_bound returns
687 // offsets[4]. That's an exact match, but indicates
688 // end-of-function. We check if offsets[5] is also an
689 // exact match but not end-of-function. It is, so we
690 // return offsets[5].
691 // Case 5: lookup_key has offset == 3214. lower_bound returns
692 // offsets[1]. Since it's not an exact match, we back
693 // up to the offset that's < lookup_key, offsets[0].
694 // We note offsets[0] is a valid entry (not end-of-function),
695 // so that's the entry we return.
696 // Case 6: lookup_key has offset == 4000. lower_bound returns
697 // offsets[8]. Since it's not an exact match, we back
698 // up to offsets[7]. Since offsets[7] indicates
699 // end-of-function, we know lookup_key is between
700 // functions, so we return end() (not a valid offset).
701 // Case 7: lookup_key has offset == 5794. lower_bound returns
702 // offsets[17]. Since it's not an exact match, we back
703 // up to offsets[15]. Note we back up to the *first*
704 // entry with offset 5793, not just offsets[17-1].
705 // We note offsets[15] is a valid entry, so we return it.
706 // If offsets[15] had had line_num == -1, we would have
707 // checked offsets[16]. The reason for this is that
708 // 15 and 16 can be in an arbitrary order, since we sort
709 // only by offset. (Note it doesn't help to use line_number
710 // as a secondary sort key, since sometimes we want the -1
711 // to be first and sometimes we want it to be last.)
713 // This deals with cases (1) and (2).
714 if ((it == offsets->begin() && offset < it->offset)
715 || it == offsets->end())
716 return offsets->end();
718 // This deals with cases (3) and (4).
719 if (offset == it->offset)
721 while (it != offsets->end()
722 && it->offset == offset
723 && it->line_num == -1)
724 ++it;
725 if (it == offsets->end() || it->offset != offset)
726 return offsets->end();
727 else
728 return it;
731 // This handles the first part of case (7) -- we back up to the
732 // *first* entry that has the offset that's behind us.
733 gold_assert(it != offsets->begin());
734 std::vector<Offset_to_lineno_entry>::const_iterator range_end = it;
735 --it;
736 const off_t range_value = it->offset;
737 while (it != offsets->begin() && (it-1)->offset == range_value)
738 --it;
740 // This handles cases (5), (6), and (7): if any entry in the
741 // equal_range [it, range_end) has a line_num != -1, it's a valid
742 // match. If not, we're not in a function.
743 for (; it != range_end; ++it)
744 if (it->line_num != -1)
745 return it;
746 return offsets->end();
749 // Return a string for a file name and line number.
751 template<int size, bool big_endian>
752 std::string
753 Sized_dwarf_line_info<size, big_endian>::do_addr2line(unsigned int shndx,
754 off_t offset)
756 if (this->data_valid_ == false)
757 return "";
759 const std::vector<Offset_to_lineno_entry>* offsets;
760 // If we do not have reloc information, then our input is a .so or
761 // some similar data structure where all the information is held in
762 // the offset. In that case, we ignore the input shndx.
763 if (this->input_is_relobj())
764 offsets = &this->line_number_map_[shndx];
765 else
766 offsets = &this->line_number_map_[-1U];
767 if (offsets->empty())
768 return "";
770 typename std::vector<Offset_to_lineno_entry>::const_iterator it
771 = offset_to_iterator(offsets, offset);
772 if (it == offsets->end())
773 return "";
775 // Convert the file_num + line_num into a string.
776 std::string ret;
778 gold_assert(it->header_num < static_cast<int>(this->files_.size()));
779 gold_assert(it->file_num
780 < static_cast<int>(this->files_[it->header_num].size()));
781 const std::pair<int, std::string>& filename_pair
782 = this->files_[it->header_num][it->file_num];
783 const std::string& filename = filename_pair.second;
785 gold_assert(it->header_num < static_cast<int>(this->directories_.size()));
786 gold_assert(filename_pair.first
787 < static_cast<int>(this->directories_[it->header_num].size()));
788 const std::string& dirname
789 = this->directories_[it->header_num][filename_pair.first];
791 if (!dirname.empty())
793 ret += dirname;
794 ret += "/";
796 ret += filename;
797 if (ret.empty())
798 ret = "(unknown)";
800 char buffer[64]; // enough to hold a line number
801 snprintf(buffer, sizeof(buffer), "%d", it->line_num);
802 ret += ":";
803 ret += buffer;
805 return ret;
808 // Dwarf_line_info routines.
810 static unsigned int next_generation_count = 0;
812 struct Addr2line_cache_entry
814 Object* object;
815 unsigned int shndx;
816 Dwarf_line_info* dwarf_line_info;
817 unsigned int generation_count;
818 unsigned int access_count;
820 Addr2line_cache_entry(Object* o, unsigned int s, Dwarf_line_info* d)
821 : object(o), shndx(s), dwarf_line_info(d),
822 generation_count(next_generation_count), access_count(0)
824 if (next_generation_count < (1U << 31))
825 ++next_generation_count;
828 // We expect this cache to be small, so don't bother with a hashtable
829 // or priority queue or anything: just use a simple vector.
830 static std::vector<Addr2line_cache_entry> addr2line_cache;
832 std::string
833 Dwarf_line_info::one_addr2line(Object* object,
834 unsigned int shndx, off_t offset,
835 size_t cache_size)
837 Dwarf_line_info* lineinfo = NULL;
838 std::vector<Addr2line_cache_entry>::iterator it;
840 // First, check the cache. If we hit, update the counts.
841 for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it)
843 if (it->object == object && it->shndx == shndx)
845 lineinfo = it->dwarf_line_info;
846 it->generation_count = next_generation_count;
847 // We cap generation_count at 2^31 -1 to avoid overflow.
848 if (next_generation_count < (1U << 31))
849 ++next_generation_count;
850 // We cap access_count at 31 so 2^access_count doesn't overflow
851 if (it->access_count < 31)
852 ++it->access_count;
853 break;
857 // If we don't hit the cache, create a new object and insert into the
858 // cache.
859 if (lineinfo == NULL)
861 switch (parameters->size_and_endianness())
863 #ifdef HAVE_TARGET_32_LITTLE
864 case Parameters::TARGET_32_LITTLE:
865 lineinfo = new Sized_dwarf_line_info<32, false>(object, shndx); break;
866 #endif
867 #ifdef HAVE_TARGET_32_BIG
868 case Parameters::TARGET_32_BIG:
869 lineinfo = new Sized_dwarf_line_info<32, true>(object, shndx); break;
870 #endif
871 #ifdef HAVE_TARGET_64_LITTLE
872 case Parameters::TARGET_64_LITTLE:
873 lineinfo = new Sized_dwarf_line_info<64, false>(object, shndx); break;
874 #endif
875 #ifdef HAVE_TARGET_64_BIG
876 case Parameters::TARGET_64_BIG:
877 lineinfo = new Sized_dwarf_line_info<64, true>(object, shndx); break;
878 #endif
879 default:
880 gold_unreachable();
882 addr2line_cache.push_back(Addr2line_cache_entry(object, shndx, lineinfo));
885 // Now that we have our object, figure out the answer
886 std::string retval = lineinfo->addr2line(shndx, offset);
888 // Finally, if our cache has grown too big, delete old objects. We
889 // assume the common (probably only) case is deleting only one object.
890 // We use a pretty simple scheme to evict: function of LRU and MFU.
891 while (addr2line_cache.size() > cache_size)
893 unsigned int lowest_score = ~0U;
894 std::vector<Addr2line_cache_entry>::iterator lowest
895 = addr2line_cache.end();
896 for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it)
898 const unsigned int score = (it->generation_count
899 + (1U << it->access_count));
900 if (score < lowest_score)
902 lowest_score = score;
903 lowest = it;
906 if (lowest != addr2line_cache.end())
908 delete lowest->dwarf_line_info;
909 addr2line_cache.erase(lowest);
913 return retval;
916 void
917 Dwarf_line_info::clear_addr2line_cache()
919 for (std::vector<Addr2line_cache_entry>::iterator it = addr2line_cache.begin();
920 it != addr2line_cache.end();
921 ++it)
922 delete it->dwarf_line_info;
923 addr2line_cache.clear();
926 #ifdef HAVE_TARGET_32_LITTLE
927 template
928 class Sized_dwarf_line_info<32, false>;
929 #endif
931 #ifdef HAVE_TARGET_32_BIG
932 template
933 class Sized_dwarf_line_info<32, true>;
934 #endif
936 #ifdef HAVE_TARGET_64_LITTLE
937 template
938 class Sized_dwarf_line_info<64, false>;
939 #endif
941 #ifdef HAVE_TARGET_64_BIG
942 template
943 class Sized_dwarf_line_info<64, true>;
944 #endif
946 } // End namespace gold.