Add a testcase for PR gas/11867.
[binutils.git] / gold / dwarf_reader.cc
blobc0188fbd435b5bac060a2017f581abba00bc2274
1 // dwarf_reader.cc -- parse dwarf2/3 debug information
3 // Copyright 2007, 2008, 2009 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"
34 #include "int_encoding.h"
35 #include "compressed_output.h"
37 namespace gold {
39 struct LineStateMachine
41 int file_num;
42 uint64_t address;
43 int line_num;
44 int column_num;
45 unsigned int shndx; // the section address refers to
46 bool is_stmt; // stmt means statement.
47 bool basic_block;
48 bool end_sequence;
51 static void
52 ResetLineStateMachine(struct LineStateMachine* lsm, bool default_is_stmt)
54 lsm->file_num = 1;
55 lsm->address = 0;
56 lsm->line_num = 1;
57 lsm->column_num = 0;
58 lsm->shndx = -1U;
59 lsm->is_stmt = default_is_stmt;
60 lsm->basic_block = false;
61 lsm->end_sequence = false;
64 template<int size, bool big_endian>
65 Sized_dwarf_line_info<size, big_endian>::Sized_dwarf_line_info(Object* object,
66 unsigned int read_shndx)
67 : data_valid_(false), buffer_(NULL), symtab_buffer_(NULL),
68 directories_(), files_(), current_header_index_(-1)
70 unsigned int debug_shndx;
71 for (debug_shndx = 0; debug_shndx < object->shnum(); ++debug_shndx)
72 // FIXME: do this more efficiently: section_name() isn't super-fast
73 if (object->section_name(debug_shndx) == ".debug_line")
75 section_size_type buffer_size;
76 this->buffer_ = object->section_contents(debug_shndx, &buffer_size,
77 false);
78 this->buffer_end_ = this->buffer_ + buffer_size;
79 break;
81 if (this->buffer_ == NULL)
82 return;
84 section_size_type uncompressed_size = 0;
85 unsigned char* uncompressed_data = NULL;
86 if (object->section_is_compressed(debug_shndx, &uncompressed_size))
88 uncompressed_data = new unsigned char[uncompressed_size];
89 if (!decompress_input_section(this->buffer_,
90 this->buffer_end_ - this->buffer_,
91 uncompressed_data,
92 uncompressed_size))
93 object->error(_("could not decompress section %s"),
94 object->section_name(debug_shndx).c_str());
95 this->buffer_ = uncompressed_data;
96 this->buffer_end_ = this->buffer_ + uncompressed_size;
99 // Find the relocation section for ".debug_line".
100 // We expect these for relobjs (.o's) but not dynobjs (.so's).
101 bool got_relocs = false;
102 for (unsigned int reloc_shndx = 0;
103 reloc_shndx < object->shnum();
104 ++reloc_shndx)
106 unsigned int reloc_sh_type = object->section_type(reloc_shndx);
107 if ((reloc_sh_type == elfcpp::SHT_REL
108 || reloc_sh_type == elfcpp::SHT_RELA)
109 && object->section_info(reloc_shndx) == debug_shndx)
111 got_relocs = this->track_relocs_.initialize(object, reloc_shndx,
112 reloc_sh_type);
113 break;
117 // Finally, we need the symtab section to interpret the relocs.
118 if (got_relocs)
120 unsigned int symtab_shndx;
121 for (symtab_shndx = 0; symtab_shndx < object->shnum(); ++symtab_shndx)
122 if (object->section_type(symtab_shndx) == elfcpp::SHT_SYMTAB)
124 this->symtab_buffer_ = object->section_contents(
125 symtab_shndx, &this->symtab_buffer_size_, false);
126 break;
128 if (this->symtab_buffer_ == NULL)
129 return;
132 // Now that we have successfully read all the data, parse the debug
133 // info.
134 this->data_valid_ = true;
135 this->read_line_mappings(object, read_shndx);
138 // Read the DWARF header.
140 template<int size, bool big_endian>
141 const unsigned char*
142 Sized_dwarf_line_info<size, big_endian>::read_header_prolog(
143 const unsigned char* lineptr)
145 uint32_t initial_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr);
146 lineptr += 4;
148 // In DWARF2/3, if the initial length is all 1 bits, then the offset
149 // size is 8 and we need to read the next 8 bytes for the real length.
150 if (initial_length == 0xffffffff)
152 header_.offset_size = 8;
153 initial_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr);
154 lineptr += 8;
156 else
157 header_.offset_size = 4;
159 header_.total_length = initial_length;
161 gold_assert(lineptr + header_.total_length <= buffer_end_);
163 header_.version = elfcpp::Swap_unaligned<16, big_endian>::readval(lineptr);
164 lineptr += 2;
166 if (header_.offset_size == 4)
167 header_.prologue_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr);
168 else
169 header_.prologue_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr);
170 lineptr += header_.offset_size;
172 header_.min_insn_length = *lineptr;
173 lineptr += 1;
175 header_.default_is_stmt = *lineptr;
176 lineptr += 1;
178 header_.line_base = *reinterpret_cast<const signed char*>(lineptr);
179 lineptr += 1;
181 header_.line_range = *lineptr;
182 lineptr += 1;
184 header_.opcode_base = *lineptr;
185 lineptr += 1;
187 header_.std_opcode_lengths.reserve(header_.opcode_base + 1);
188 header_.std_opcode_lengths[0] = 0;
189 for (int i = 1; i < header_.opcode_base; i++)
191 header_.std_opcode_lengths[i] = *lineptr;
192 lineptr += 1;
195 return lineptr;
198 // The header for a debug_line section is mildly complicated, because
199 // the line info is very tightly encoded.
201 template<int size, bool big_endian>
202 const unsigned char*
203 Sized_dwarf_line_info<size, big_endian>::read_header_tables(
204 const unsigned char* lineptr)
206 ++this->current_header_index_;
208 // Create a new directories_ entry and a new files_ entry for our new
209 // header. We initialize each with a single empty element, because
210 // dwarf indexes directory and filenames starting at 1.
211 gold_assert(static_cast<int>(this->directories_.size())
212 == this->current_header_index_);
213 gold_assert(static_cast<int>(this->files_.size())
214 == this->current_header_index_);
215 this->directories_.push_back(std::vector<std::string>(1));
216 this->files_.push_back(std::vector<std::pair<int, std::string> >(1));
218 // It is legal for the directory entry table to be empty.
219 if (*lineptr)
221 int dirindex = 1;
222 while (*lineptr)
224 const char* dirname = reinterpret_cast<const char*>(lineptr);
225 gold_assert(dirindex
226 == static_cast<int>(this->directories_.back().size()));
227 this->directories_.back().push_back(dirname);
228 lineptr += this->directories_.back().back().size() + 1;
229 dirindex++;
232 lineptr++;
234 // It is also legal for the file entry table to be empty.
235 if (*lineptr)
237 int fileindex = 1;
238 size_t len;
239 while (*lineptr)
241 const char* filename = reinterpret_cast<const char*>(lineptr);
242 lineptr += strlen(filename) + 1;
244 uint64_t dirindex = read_unsigned_LEB_128(lineptr, &len);
245 lineptr += len;
247 if (dirindex >= this->directories_.back().size())
248 dirindex = 0;
249 int dirindexi = static_cast<int>(dirindex);
251 read_unsigned_LEB_128(lineptr, &len); // mod_time
252 lineptr += len;
254 read_unsigned_LEB_128(lineptr, &len); // filelength
255 lineptr += len;
257 gold_assert(fileindex
258 == static_cast<int>(this->files_.back().size()));
259 this->files_.back().push_back(std::make_pair(dirindexi, filename));
260 fileindex++;
263 lineptr++;
265 return lineptr;
268 // Process a single opcode in the .debug.line structure.
270 // Templating on size and big_endian would yield more efficient (and
271 // simpler) code, but would bloat the binary. Speed isn't important
272 // here.
274 template<int size, bool big_endian>
275 bool
276 Sized_dwarf_line_info<size, big_endian>::process_one_opcode(
277 const unsigned char* start, struct LineStateMachine* lsm, size_t* len)
279 size_t oplen = 0;
280 size_t templen;
281 unsigned char opcode = *start;
282 oplen++;
283 start++;
285 // If the opcode is great than the opcode_base, it is a special
286 // opcode. Most line programs consist mainly of special opcodes.
287 if (opcode >= header_.opcode_base)
289 opcode -= header_.opcode_base;
290 const int advance_address = ((opcode / header_.line_range)
291 * header_.min_insn_length);
292 lsm->address += advance_address;
294 const int advance_line = ((opcode % header_.line_range)
295 + header_.line_base);
296 lsm->line_num += advance_line;
297 lsm->basic_block = true;
298 *len = oplen;
299 return true;
302 // Otherwise, we have the regular opcodes
303 switch (opcode)
305 case elfcpp::DW_LNS_copy:
306 lsm->basic_block = false;
307 *len = oplen;
308 return true;
310 case elfcpp::DW_LNS_advance_pc:
312 const uint64_t advance_address
313 = read_unsigned_LEB_128(start, &templen);
314 oplen += templen;
315 lsm->address += header_.min_insn_length * advance_address;
317 break;
319 case elfcpp::DW_LNS_advance_line:
321 const uint64_t advance_line = read_signed_LEB_128(start, &templen);
322 oplen += templen;
323 lsm->line_num += advance_line;
325 break;
327 case elfcpp::DW_LNS_set_file:
329 const uint64_t fileno = read_unsigned_LEB_128(start, &templen);
330 oplen += templen;
331 lsm->file_num = fileno;
333 break;
335 case elfcpp::DW_LNS_set_column:
337 const uint64_t colno = read_unsigned_LEB_128(start, &templen);
338 oplen += templen;
339 lsm->column_num = colno;
341 break;
343 case elfcpp::DW_LNS_negate_stmt:
344 lsm->is_stmt = !lsm->is_stmt;
345 break;
347 case elfcpp::DW_LNS_set_basic_block:
348 lsm->basic_block = true;
349 break;
351 case elfcpp::DW_LNS_fixed_advance_pc:
353 int advance_address;
354 advance_address = elfcpp::Swap_unaligned<16, big_endian>::readval(start);
355 oplen += 2;
356 lsm->address += advance_address;
358 break;
360 case elfcpp::DW_LNS_const_add_pc:
362 const int advance_address = (header_.min_insn_length
363 * ((255 - header_.opcode_base)
364 / header_.line_range));
365 lsm->address += advance_address;
367 break;
369 case elfcpp::DW_LNS_extended_op:
371 const uint64_t extended_op_len
372 = read_unsigned_LEB_128(start, &templen);
373 start += templen;
374 oplen += templen + extended_op_len;
376 const unsigned char extended_op = *start;
377 start++;
379 switch (extended_op)
381 case elfcpp::DW_LNE_end_sequence:
382 // This means that the current byte is the one immediately
383 // after a set of instructions. Record the current line
384 // for up to one less than the current address.
385 lsm->line_num = -1;
386 lsm->end_sequence = true;
387 *len = oplen;
388 return true;
390 case elfcpp::DW_LNE_set_address:
392 lsm->address = elfcpp::Swap_unaligned<size, big_endian>::readval(start);
393 typename Reloc_map::const_iterator it
394 = reloc_map_.find(start - this->buffer_);
395 if (it != reloc_map_.end())
397 // value + addend.
398 lsm->address += it->second.second;
399 lsm->shndx = it->second.first;
401 else
403 // If we're a normal .o file, with relocs, every
404 // set_address should have an associated relocation.
405 if (this->input_is_relobj())
406 this->data_valid_ = false;
408 break;
410 case elfcpp::DW_LNE_define_file:
412 const char* filename = reinterpret_cast<const char*>(start);
413 templen = strlen(filename) + 1;
414 start += templen;
416 uint64_t dirindex = read_unsigned_LEB_128(start, &templen);
417 oplen += templen;
419 if (dirindex >= this->directories_.back().size())
420 dirindex = 0;
421 int dirindexi = static_cast<int>(dirindex);
423 read_unsigned_LEB_128(start, &templen); // mod_time
424 oplen += templen;
426 read_unsigned_LEB_128(start, &templen); // filelength
427 oplen += templen;
429 this->files_.back().push_back(std::make_pair(dirindexi,
430 filename));
432 break;
435 break;
437 default:
439 // Ignore unknown opcode silently
440 for (int i = 0; i < header_.std_opcode_lengths[opcode]; i++)
442 size_t templen;
443 read_unsigned_LEB_128(start, &templen);
444 start += templen;
445 oplen += templen;
448 break;
450 *len = oplen;
451 return false;
454 // Read the debug information at LINEPTR and store it in the line
455 // number map.
457 template<int size, bool big_endian>
458 unsigned const char*
459 Sized_dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr,
460 unsigned int shndx)
462 struct LineStateMachine lsm;
464 // LENGTHSTART is the place the length field is based on. It is the
465 // point in the header after the initial length field.
466 const unsigned char* lengthstart = buffer_;
468 // In 64 bit dwarf, the initial length is 12 bytes, because of the
469 // 0xffffffff at the start.
470 if (header_.offset_size == 8)
471 lengthstart += 12;
472 else
473 lengthstart += 4;
475 while (lineptr < lengthstart + header_.total_length)
477 ResetLineStateMachine(&lsm, header_.default_is_stmt);
478 while (!lsm.end_sequence)
480 size_t oplength;
481 bool add_line = this->process_one_opcode(lineptr, &lsm, &oplength);
482 if (add_line
483 && (shndx == -1U || lsm.shndx == -1U || shndx == lsm.shndx))
485 Offset_to_lineno_entry entry
486 = { lsm.address, this->current_header_index_,
487 lsm.file_num, lsm.line_num };
488 line_number_map_[lsm.shndx].push_back(entry);
490 lineptr += oplength;
494 return lengthstart + header_.total_length;
497 // Looks in the symtab to see what section a symbol is in.
499 template<int size, bool big_endian>
500 unsigned int
501 Sized_dwarf_line_info<size, big_endian>::symbol_section(
502 Object* object,
503 unsigned int sym,
504 typename elfcpp::Elf_types<size>::Elf_Addr* value,
505 bool* is_ordinary)
507 const int symsize = elfcpp::Elf_sizes<size>::sym_size;
508 gold_assert(sym * symsize < this->symtab_buffer_size_);
509 elfcpp::Sym<size, big_endian> elfsym(this->symtab_buffer_ + sym * symsize);
510 *value = elfsym.get_st_value();
511 return object->adjust_sym_shndx(sym, elfsym.get_st_shndx(), is_ordinary);
514 // Read the relocations into a Reloc_map.
516 template<int size, bool big_endian>
517 void
518 Sized_dwarf_line_info<size, big_endian>::read_relocs(Object* object)
520 if (this->symtab_buffer_ == NULL)
521 return;
523 typename elfcpp::Elf_types<size>::Elf_Addr value;
524 off_t reloc_offset;
525 while ((reloc_offset = this->track_relocs_.next_offset()) != -1)
527 const unsigned int sym = this->track_relocs_.next_symndx();
529 bool is_ordinary;
530 const unsigned int shndx = this->symbol_section(object, sym, &value,
531 &is_ordinary);
533 // There is no reason to record non-ordinary section indexes, or
534 // SHN_UNDEF, because they will never match the real section.
535 if (is_ordinary && shndx != elfcpp::SHN_UNDEF)
536 this->reloc_map_[reloc_offset] = std::make_pair(shndx, value);
538 this->track_relocs_.advance(reloc_offset + 1);
542 // Read the line number info.
544 template<int size, bool big_endian>
545 void
546 Sized_dwarf_line_info<size, big_endian>::read_line_mappings(Object* object,
547 unsigned int shndx)
549 gold_assert(this->data_valid_ == true);
551 this->read_relocs(object);
552 while (this->buffer_ < this->buffer_end_)
554 const unsigned char* lineptr = this->buffer_;
555 lineptr = this->read_header_prolog(lineptr);
556 lineptr = this->read_header_tables(lineptr);
557 lineptr = this->read_lines(lineptr, shndx);
558 this->buffer_ = lineptr;
561 // Sort the lines numbers, so addr2line can use binary search.
562 for (typename Lineno_map::iterator it = line_number_map_.begin();
563 it != line_number_map_.end();
564 ++it)
565 // Each vector needs to be sorted by offset.
566 std::sort(it->second.begin(), it->second.end());
569 // Some processing depends on whether the input is a .o file or not.
570 // For instance, .o files have relocs, and have .debug_lines
571 // information on a per section basis. .so files, on the other hand,
572 // lack relocs, and offsets are unique, so we can ignore the section
573 // information.
575 template<int size, bool big_endian>
576 bool
577 Sized_dwarf_line_info<size, big_endian>::input_is_relobj()
579 // Only .o files have relocs and the symtab buffer that goes with them.
580 return this->symtab_buffer_ != NULL;
583 // Given an Offset_to_lineno_entry vector, and an offset, figure out
584 // if the offset points into a function according to the vector (see
585 // comments below for the algorithm). If it does, return an iterator
586 // into the vector that points to the line-number that contains that
587 // offset. If not, it returns vector::end().
589 static std::vector<Offset_to_lineno_entry>::const_iterator
590 offset_to_iterator(const std::vector<Offset_to_lineno_entry>* offsets,
591 off_t offset)
593 const Offset_to_lineno_entry lookup_key = { offset, 0, 0, 0 };
595 // lower_bound() returns the smallest offset which is >= lookup_key.
596 // If no offset in offsets is >= lookup_key, returns end().
597 std::vector<Offset_to_lineno_entry>::const_iterator it
598 = std::lower_bound(offsets->begin(), offsets->end(), lookup_key);
600 // This code is easiest to understand with a concrete example.
601 // Here's a possible offsets array:
602 // {{offset = 3211, header_num = 0, file_num = 1, line_num = 16}, // 0
603 // {offset = 3224, header_num = 0, file_num = 1, line_num = 20}, // 1
604 // {offset = 3226, header_num = 0, file_num = 1, line_num = 22}, // 2
605 // {offset = 3231, header_num = 0, file_num = 1, line_num = 25}, // 3
606 // {offset = 3232, header_num = 0, file_num = 1, line_num = -1}, // 4
607 // {offset = 3232, header_num = 0, file_num = 1, line_num = 65}, // 5
608 // {offset = 3235, header_num = 0, file_num = 1, line_num = 66}, // 6
609 // {offset = 3236, header_num = 0, file_num = 1, line_num = -1}, // 7
610 // {offset = 5764, header_num = 0, file_num = 1, line_num = 47}, // 8
611 // {offset = 5765, header_num = 0, file_num = 1, line_num = 48}, // 9
612 // {offset = 5767, header_num = 0, file_num = 1, line_num = 49}, // 10
613 // {offset = 5768, header_num = 0, file_num = 1, line_num = 50}, // 11
614 // {offset = 5773, header_num = 0, file_num = 1, line_num = -1}, // 12
615 // {offset = 5787, header_num = 1, file_num = 1, line_num = 19}, // 13
616 // {offset = 5790, header_num = 1, file_num = 1, line_num = 20}, // 14
617 // {offset = 5793, header_num = 1, file_num = 1, line_num = 67}, // 15
618 // {offset = 5793, header_num = 1, file_num = 1, line_num = -1}, // 16
619 // {offset = 5795, header_num = 1, file_num = 1, line_num = 68}, // 17
620 // {offset = 5798, header_num = 1, file_num = 1, line_num = -1}, // 18
621 // The entries with line_num == -1 mark the end of a function: the
622 // associated offset is one past the last instruction in the
623 // function. This can correspond to the beginning of the next
624 // function (as is true for offset 3232); alternately, there can be
625 // a gap between the end of one function and the start of the next
626 // (as is true for some others, most obviously from 3236->5764).
628 // Case 1: lookup_key has offset == 10. lower_bound returns
629 // offsets[0]. Since it's not an exact match and we're
630 // at the beginning of offsets, we return end() (invalid).
631 // Case 2: lookup_key has offset 10000. lower_bound returns
632 // offset[19] (end()). We return end() (invalid).
633 // Case 3: lookup_key has offset == 3211. lower_bound matches
634 // offsets[0] exactly, and that's the entry we return.
635 // Case 4: lookup_key has offset == 3232. lower_bound returns
636 // offsets[4]. That's an exact match, but indicates
637 // end-of-function. We check if offsets[5] is also an
638 // exact match but not end-of-function. It is, so we
639 // return offsets[5].
640 // Case 5: lookup_key has offset == 3214. lower_bound returns
641 // offsets[1]. Since it's not an exact match, we back
642 // up to the offset that's < lookup_key, offsets[0].
643 // We note offsets[0] is a valid entry (not end-of-function),
644 // so that's the entry we return.
645 // Case 6: lookup_key has offset == 4000. lower_bound returns
646 // offsets[8]. Since it's not an exact match, we back
647 // up to offsets[7]. Since offsets[7] indicates
648 // end-of-function, we know lookup_key is between
649 // functions, so we return end() (not a valid offset).
650 // Case 7: lookup_key has offset == 5794. lower_bound returns
651 // offsets[17]. Since it's not an exact match, we back
652 // up to offsets[15]. Note we back up to the *first*
653 // entry with offset 5793, not just offsets[17-1].
654 // We note offsets[15] is a valid entry, so we return it.
655 // If offsets[15] had had line_num == -1, we would have
656 // checked offsets[16]. The reason for this is that
657 // 15 and 16 can be in an arbitrary order, since we sort
658 // only by offset. (Note it doesn't help to use line_number
659 // as a secondary sort key, since sometimes we want the -1
660 // to be first and sometimes we want it to be last.)
662 // This deals with cases (1) and (2).
663 if ((it == offsets->begin() && offset < it->offset)
664 || it == offsets->end())
665 return offsets->end();
667 // This deals with cases (3) and (4).
668 if (offset == it->offset)
670 while (it != offsets->end()
671 && it->offset == offset
672 && it->line_num == -1)
673 ++it;
674 if (it == offsets->end() || it->offset != offset)
675 return offsets->end();
676 else
677 return it;
680 // This handles the first part of case (7) -- we back up to the
681 // *first* entry that has the offset that's behind us.
682 gold_assert(it != offsets->begin());
683 std::vector<Offset_to_lineno_entry>::const_iterator range_end = it;
684 --it;
685 const off_t range_value = it->offset;
686 while (it != offsets->begin() && (it-1)->offset == range_value)
687 --it;
689 // This handles cases (5), (6), and (7): if any entry in the
690 // equal_range [it, range_end) has a line_num != -1, it's a valid
691 // match. If not, we're not in a function.
692 for (; it != range_end; ++it)
693 if (it->line_num != -1)
694 return it;
695 return offsets->end();
698 // Return a string for a file name and line number.
700 template<int size, bool big_endian>
701 std::string
702 Sized_dwarf_line_info<size, big_endian>::do_addr2line(unsigned int shndx,
703 off_t offset)
705 if (this->data_valid_ == false)
706 return "";
708 const std::vector<Offset_to_lineno_entry>* offsets;
709 // If we do not have reloc information, then our input is a .so or
710 // some similar data structure where all the information is held in
711 // the offset. In that case, we ignore the input shndx.
712 if (this->input_is_relobj())
713 offsets = &this->line_number_map_[shndx];
714 else
715 offsets = &this->line_number_map_[-1U];
716 if (offsets->empty())
717 return "";
719 typename std::vector<Offset_to_lineno_entry>::const_iterator it
720 = offset_to_iterator(offsets, offset);
721 if (it == offsets->end())
722 return "";
724 // Convert the file_num + line_num into a string.
725 std::string ret;
727 gold_assert(it->header_num < static_cast<int>(this->files_.size()));
728 gold_assert(it->file_num
729 < static_cast<int>(this->files_[it->header_num].size()));
730 const std::pair<int, std::string>& filename_pair
731 = this->files_[it->header_num][it->file_num];
732 const std::string& filename = filename_pair.second;
734 gold_assert(it->header_num < static_cast<int>(this->directories_.size()));
735 gold_assert(filename_pair.first
736 < static_cast<int>(this->directories_[it->header_num].size()));
737 const std::string& dirname
738 = this->directories_[it->header_num][filename_pair.first];
740 if (!dirname.empty())
742 ret += dirname;
743 ret += "/";
745 ret += filename;
746 if (ret.empty())
747 ret = "(unknown)";
749 char buffer[64]; // enough to hold a line number
750 snprintf(buffer, sizeof(buffer), "%d", it->line_num);
751 ret += ":";
752 ret += buffer;
754 return ret;
757 // Dwarf_line_info routines.
759 static unsigned int next_generation_count = 0;
761 struct Addr2line_cache_entry
763 Object* object;
764 unsigned int shndx;
765 Dwarf_line_info* dwarf_line_info;
766 unsigned int generation_count;
767 unsigned int access_count;
769 Addr2line_cache_entry(Object* o, unsigned int s, Dwarf_line_info* d)
770 : object(o), shndx(s), dwarf_line_info(d),
771 generation_count(next_generation_count), access_count(0)
773 if (next_generation_count < (1U << 31))
774 ++next_generation_count;
777 // We expect this cache to be small, so don't bother with a hashtable
778 // or priority queue or anything: just use a simple vector.
779 static std::vector<Addr2line_cache_entry> addr2line_cache;
781 std::string
782 Dwarf_line_info::one_addr2line(Object* object,
783 unsigned int shndx, off_t offset,
784 size_t cache_size)
786 Dwarf_line_info* lineinfo = NULL;
787 std::vector<Addr2line_cache_entry>::iterator it;
789 // First, check the cache. If we hit, update the counts.
790 for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it)
792 if (it->object == object && it->shndx == shndx)
794 lineinfo = it->dwarf_line_info;
795 it->generation_count = next_generation_count;
796 // We cap generation_count at 2^31 -1 to avoid overflow.
797 if (next_generation_count < (1U << 31))
798 ++next_generation_count;
799 // We cap access_count at 31 so 2^access_count doesn't overflow
800 if (it->access_count < 31)
801 ++it->access_count;
802 break;
806 // If we don't hit the cache, create a new object and insert into the
807 // cache.
808 if (lineinfo == NULL)
810 switch (parameters->size_and_endianness())
812 #ifdef HAVE_TARGET_32_LITTLE
813 case Parameters::TARGET_32_LITTLE:
814 lineinfo = new Sized_dwarf_line_info<32, false>(object, shndx); break;
815 #endif
816 #ifdef HAVE_TARGET_32_BIG
817 case Parameters::TARGET_32_BIG:
818 lineinfo = new Sized_dwarf_line_info<32, true>(object, shndx); break;
819 #endif
820 #ifdef HAVE_TARGET_64_LITTLE
821 case Parameters::TARGET_64_LITTLE:
822 lineinfo = new Sized_dwarf_line_info<64, false>(object, shndx); break;
823 #endif
824 #ifdef HAVE_TARGET_64_BIG
825 case Parameters::TARGET_64_BIG:
826 lineinfo = new Sized_dwarf_line_info<64, true>(object, shndx); break;
827 #endif
828 default:
829 gold_unreachable();
831 addr2line_cache.push_back(Addr2line_cache_entry(object, shndx, lineinfo));
834 // Now that we have our object, figure out the answer
835 std::string retval = lineinfo->addr2line(shndx, offset);
837 // Finally, if our cache has grown too big, delete old objects. We
838 // assume the common (probably only) case is deleting only one object.
839 // We use a pretty simple scheme to evict: function of LRU and MFU.
840 while (addr2line_cache.size() > cache_size)
842 unsigned int lowest_score = ~0U;
843 std::vector<Addr2line_cache_entry>::iterator lowest
844 = addr2line_cache.end();
845 for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it)
847 const unsigned int score = (it->generation_count
848 + (1U << it->access_count));
849 if (score < lowest_score)
851 lowest_score = score;
852 lowest = it;
855 if (lowest != addr2line_cache.end())
857 delete lowest->dwarf_line_info;
858 addr2line_cache.erase(lowest);
862 return retval;
865 void
866 Dwarf_line_info::clear_addr2line_cache()
868 for (std::vector<Addr2line_cache_entry>::iterator it = addr2line_cache.begin();
869 it != addr2line_cache.end();
870 ++it)
871 delete it->dwarf_line_info;
872 addr2line_cache.clear();
875 #ifdef HAVE_TARGET_32_LITTLE
876 template
877 class Sized_dwarf_line_info<32, false>;
878 #endif
880 #ifdef HAVE_TARGET_32_BIG
881 template
882 class Sized_dwarf_line_info<32, true>;
883 #endif
885 #ifdef HAVE_TARGET_64_LITTLE
886 template
887 class Sized_dwarf_line_info<64, false>;
888 #endif
890 #ifdef HAVE_TARGET_64_BIG
891 template
892 class Sized_dwarf_line_info<64, true>;
893 #endif
895 } // End namespace gold.