* common.cc (Symbol_table::do_allocate_commons_list): For incremental
[binutils.git] / gold / dwarf_reader.cc
blob3dc33e4e106362ce9e47640c74e376f033bbe036
1 // dwarf_reader.cc -- parse dwarf2/3 debug information
3 // Copyright 2007, 2008, 2009, 2010 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 = 1; debug_shndx < object->shnum(); ++debug_shndx)
73 // FIXME: do this more efficiently: section_name() isn't super-fast
74 std::string name = object->section_name(debug_shndx);
75 if (name == ".debug_line" || name == ".zdebug_line")
77 section_size_type buffer_size;
78 this->buffer_ = object->section_contents(debug_shndx, &buffer_size,
79 false);
80 this->buffer_end_ = this->buffer_ + buffer_size;
81 break;
84 if (this->buffer_ == NULL)
85 return;
87 section_size_type uncompressed_size = 0;
88 unsigned char* uncompressed_data = NULL;
89 if (object->section_is_compressed(debug_shndx, &uncompressed_size))
91 uncompressed_data = new unsigned char[uncompressed_size];
92 if (!decompress_input_section(this->buffer_,
93 this->buffer_end_ - this->buffer_,
94 uncompressed_data,
95 uncompressed_size))
96 object->error(_("could not decompress section %s"),
97 object->section_name(debug_shndx).c_str());
98 this->buffer_ = uncompressed_data;
99 this->buffer_end_ = this->buffer_ + uncompressed_size;
102 // Find the relocation section for ".debug_line".
103 // We expect these for relobjs (.o's) but not dynobjs (.so's).
104 bool got_relocs = false;
105 for (unsigned int reloc_shndx = 0;
106 reloc_shndx < object->shnum();
107 ++reloc_shndx)
109 unsigned int reloc_sh_type = object->section_type(reloc_shndx);
110 if ((reloc_sh_type == elfcpp::SHT_REL
111 || reloc_sh_type == elfcpp::SHT_RELA)
112 && object->section_info(reloc_shndx) == debug_shndx)
114 got_relocs = this->track_relocs_.initialize(object, reloc_shndx,
115 reloc_sh_type);
116 this->track_relocs_type_ = reloc_sh_type;
117 break;
121 // Finally, we need the symtab section to interpret the relocs.
122 if (got_relocs)
124 unsigned int symtab_shndx;
125 for (symtab_shndx = 0; symtab_shndx < object->shnum(); ++symtab_shndx)
126 if (object->section_type(symtab_shndx) == elfcpp::SHT_SYMTAB)
128 this->symtab_buffer_ = object->section_contents(
129 symtab_shndx, &this->symtab_buffer_size_, false);
130 break;
132 if (this->symtab_buffer_ == NULL)
133 return;
136 // Now that we have successfully read all the data, parse the debug
137 // info.
138 this->data_valid_ = true;
139 this->read_line_mappings(object, read_shndx);
142 // Read the DWARF header.
144 template<int size, bool big_endian>
145 const unsigned char*
146 Sized_dwarf_line_info<size, big_endian>::read_header_prolog(
147 const unsigned char* lineptr)
149 uint32_t initial_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr);
150 lineptr += 4;
152 // In DWARF2/3, if the initial length is all 1 bits, then the offset
153 // size is 8 and we need to read the next 8 bytes for the real length.
154 if (initial_length == 0xffffffff)
156 header_.offset_size = 8;
157 initial_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr);
158 lineptr += 8;
160 else
161 header_.offset_size = 4;
163 header_.total_length = initial_length;
165 gold_assert(lineptr + header_.total_length <= buffer_end_);
167 header_.version = elfcpp::Swap_unaligned<16, big_endian>::readval(lineptr);
168 lineptr += 2;
170 if (header_.offset_size == 4)
171 header_.prologue_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr);
172 else
173 header_.prologue_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr);
174 lineptr += header_.offset_size;
176 header_.min_insn_length = *lineptr;
177 lineptr += 1;
179 header_.default_is_stmt = *lineptr;
180 lineptr += 1;
182 header_.line_base = *reinterpret_cast<const signed char*>(lineptr);
183 lineptr += 1;
185 header_.line_range = *lineptr;
186 lineptr += 1;
188 header_.opcode_base = *lineptr;
189 lineptr += 1;
191 header_.std_opcode_lengths.resize(header_.opcode_base + 1);
192 header_.std_opcode_lengths[0] = 0;
193 for (int i = 1; i < header_.opcode_base; i++)
195 header_.std_opcode_lengths[i] = *lineptr;
196 lineptr += 1;
199 return lineptr;
202 // The header for a debug_line section is mildly complicated, because
203 // the line info is very tightly encoded.
205 template<int size, bool big_endian>
206 const unsigned char*
207 Sized_dwarf_line_info<size, big_endian>::read_header_tables(
208 const unsigned char* lineptr)
210 ++this->current_header_index_;
212 // Create a new directories_ entry and a new files_ entry for our new
213 // header. We initialize each with a single empty element, because
214 // dwarf indexes directory and filenames starting at 1.
215 gold_assert(static_cast<int>(this->directories_.size())
216 == this->current_header_index_);
217 gold_assert(static_cast<int>(this->files_.size())
218 == this->current_header_index_);
219 this->directories_.push_back(std::vector<std::string>(1));
220 this->files_.push_back(std::vector<std::pair<int, std::string> >(1));
222 // It is legal for the directory entry table to be empty.
223 if (*lineptr)
225 int dirindex = 1;
226 while (*lineptr)
228 const char* dirname = reinterpret_cast<const char*>(lineptr);
229 gold_assert(dirindex
230 == static_cast<int>(this->directories_.back().size()));
231 this->directories_.back().push_back(dirname);
232 lineptr += this->directories_.back().back().size() + 1;
233 dirindex++;
236 lineptr++;
238 // It is also legal for the file entry table to be empty.
239 if (*lineptr)
241 int fileindex = 1;
242 size_t len;
243 while (*lineptr)
245 const char* filename = reinterpret_cast<const char*>(lineptr);
246 lineptr += strlen(filename) + 1;
248 uint64_t dirindex = read_unsigned_LEB_128(lineptr, &len);
249 lineptr += len;
251 if (dirindex >= this->directories_.back().size())
252 dirindex = 0;
253 int dirindexi = static_cast<int>(dirindex);
255 read_unsigned_LEB_128(lineptr, &len); // mod_time
256 lineptr += len;
258 read_unsigned_LEB_128(lineptr, &len); // filelength
259 lineptr += len;
261 gold_assert(fileindex
262 == static_cast<int>(this->files_.back().size()));
263 this->files_.back().push_back(std::make_pair(dirindexi, filename));
264 fileindex++;
267 lineptr++;
269 return lineptr;
272 // Process a single opcode in the .debug.line structure.
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 =
393 elfcpp::Swap_unaligned<size, big_endian>::readval(start);
394 typename Reloc_map::const_iterator it
395 = this->reloc_map_.find(start - this->buffer_);
396 if (it != reloc_map_.end())
398 // If this is a SHT_RELA section, then ignore the
399 // section contents. This assumes that this is a
400 // straight reloc which just uses the reloc addend.
401 // The reloc addend has already been included in the
402 // symbol value.
403 if (this->track_relocs_type_ == elfcpp::SHT_RELA)
404 lsm->address = 0;
405 // Add in the symbol value.
406 lsm->address += it->second.second;
407 lsm->shndx = it->second.first;
409 else
411 // If we're a normal .o file, with relocs, every
412 // set_address should have an associated relocation.
413 if (this->input_is_relobj())
414 this->data_valid_ = false;
416 break;
418 case elfcpp::DW_LNE_define_file:
420 const char* filename = reinterpret_cast<const char*>(start);
421 templen = strlen(filename) + 1;
422 start += templen;
424 uint64_t dirindex = read_unsigned_LEB_128(start, &templen);
425 oplen += templen;
427 if (dirindex >= this->directories_.back().size())
428 dirindex = 0;
429 int dirindexi = static_cast<int>(dirindex);
431 read_unsigned_LEB_128(start, &templen); // mod_time
432 oplen += templen;
434 read_unsigned_LEB_128(start, &templen); // filelength
435 oplen += templen;
437 this->files_.back().push_back(std::make_pair(dirindexi,
438 filename));
440 break;
443 break;
445 default:
447 // Ignore unknown opcode silently
448 for (int i = 0; i < header_.std_opcode_lengths[opcode]; i++)
450 size_t templen;
451 read_unsigned_LEB_128(start, &templen);
452 start += templen;
453 oplen += templen;
456 break;
458 *len = oplen;
459 return false;
462 // Read the debug information at LINEPTR and store it in the line
463 // number map.
465 template<int size, bool big_endian>
466 unsigned const char*
467 Sized_dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr,
468 unsigned int shndx)
470 struct LineStateMachine lsm;
472 // LENGTHSTART is the place the length field is based on. It is the
473 // point in the header after the initial length field.
474 const unsigned char* lengthstart = buffer_;
476 // In 64 bit dwarf, the initial length is 12 bytes, because of the
477 // 0xffffffff at the start.
478 if (header_.offset_size == 8)
479 lengthstart += 12;
480 else
481 lengthstart += 4;
483 while (lineptr < lengthstart + header_.total_length)
485 ResetLineStateMachine(&lsm, header_.default_is_stmt);
486 while (!lsm.end_sequence)
488 size_t oplength;
489 bool add_line = this->process_one_opcode(lineptr, &lsm, &oplength);
490 if (add_line
491 && (shndx == -1U || lsm.shndx == -1U || shndx == lsm.shndx))
493 Offset_to_lineno_entry entry
494 = { lsm.address, this->current_header_index_,
495 lsm.file_num, true, lsm.line_num };
496 std::vector<Offset_to_lineno_entry>&
497 map(this->line_number_map_[lsm.shndx]);
498 // If we see two consecutive entries with the same
499 // offset and a real line number, then mark the first
500 // one as non-canonical.
501 if (!map.empty()
502 && (map.back().offset == static_cast<off_t>(lsm.address))
503 && lsm.line_num != -1
504 && map.back().line_num != -1)
505 map.back().last_line_for_offset = false;
506 map.push_back(entry);
508 lineptr += oplength;
512 return lengthstart + header_.total_length;
515 // Looks in the symtab to see what section a symbol is in.
517 template<int size, bool big_endian>
518 unsigned int
519 Sized_dwarf_line_info<size, big_endian>::symbol_section(
520 Object* object,
521 unsigned int sym,
522 typename elfcpp::Elf_types<size>::Elf_Addr* value,
523 bool* is_ordinary)
525 const int symsize = elfcpp::Elf_sizes<size>::sym_size;
526 gold_assert(sym * symsize < this->symtab_buffer_size_);
527 elfcpp::Sym<size, big_endian> elfsym(this->symtab_buffer_ + sym * symsize);
528 *value = elfsym.get_st_value();
529 return object->adjust_sym_shndx(sym, elfsym.get_st_shndx(), is_ordinary);
532 // Read the relocations into a Reloc_map.
534 template<int size, bool big_endian>
535 void
536 Sized_dwarf_line_info<size, big_endian>::read_relocs(Object* object)
538 if (this->symtab_buffer_ == NULL)
539 return;
541 typename elfcpp::Elf_types<size>::Elf_Addr value;
542 off_t reloc_offset;
543 while ((reloc_offset = this->track_relocs_.next_offset()) != -1)
545 const unsigned int sym = this->track_relocs_.next_symndx();
547 bool is_ordinary;
548 const unsigned int shndx = this->symbol_section(object, sym, &value,
549 &is_ordinary);
551 // There is no reason to record non-ordinary section indexes, or
552 // SHN_UNDEF, because they will never match the real section.
553 if (is_ordinary && shndx != elfcpp::SHN_UNDEF)
555 value += this->track_relocs_.next_addend();
556 this->reloc_map_[reloc_offset] = std::make_pair(shndx, value);
559 this->track_relocs_.advance(reloc_offset + 1);
563 // Read the line number info.
565 template<int size, bool big_endian>
566 void
567 Sized_dwarf_line_info<size, big_endian>::read_line_mappings(Object* object,
568 unsigned int shndx)
570 gold_assert(this->data_valid_ == true);
572 this->read_relocs(object);
573 while (this->buffer_ < this->buffer_end_)
575 const unsigned char* lineptr = this->buffer_;
576 lineptr = this->read_header_prolog(lineptr);
577 lineptr = this->read_header_tables(lineptr);
578 lineptr = this->read_lines(lineptr, shndx);
579 this->buffer_ = lineptr;
582 // Sort the lines numbers, so addr2line can use binary search.
583 for (typename Lineno_map::iterator it = line_number_map_.begin();
584 it != line_number_map_.end();
585 ++it)
586 // Each vector needs to be sorted by offset.
587 std::sort(it->second.begin(), it->second.end());
590 // Some processing depends on whether the input is a .o file or not.
591 // For instance, .o files have relocs, and have .debug_lines
592 // information on a per section basis. .so files, on the other hand,
593 // lack relocs, and offsets are unique, so we can ignore the section
594 // information.
596 template<int size, bool big_endian>
597 bool
598 Sized_dwarf_line_info<size, big_endian>::input_is_relobj()
600 // Only .o files have relocs and the symtab buffer that goes with them.
601 return this->symtab_buffer_ != NULL;
604 // Given an Offset_to_lineno_entry vector, and an offset, figure out
605 // if the offset points into a function according to the vector (see
606 // comments below for the algorithm). If it does, return an iterator
607 // into the vector that points to the line-number that contains that
608 // offset. If not, it returns vector::end().
610 static std::vector<Offset_to_lineno_entry>::const_iterator
611 offset_to_iterator(const std::vector<Offset_to_lineno_entry>* offsets,
612 off_t offset)
614 const Offset_to_lineno_entry lookup_key = { offset, 0, 0, true, 0 };
616 // lower_bound() returns the smallest offset which is >= lookup_key.
617 // If no offset in offsets is >= lookup_key, returns end().
618 std::vector<Offset_to_lineno_entry>::const_iterator it
619 = std::lower_bound(offsets->begin(), offsets->end(), lookup_key);
621 // This code is easiest to understand with a concrete example.
622 // Here's a possible offsets array:
623 // {{offset = 3211, header_num = 0, file_num = 1, last, line_num = 16}, // 0
624 // {offset = 3224, header_num = 0, file_num = 1, last, line_num = 20}, // 1
625 // {offset = 3226, header_num = 0, file_num = 1, last, line_num = 22}, // 2
626 // {offset = 3231, header_num = 0, file_num = 1, last, line_num = 25}, // 3
627 // {offset = 3232, header_num = 0, file_num = 1, last, line_num = -1}, // 4
628 // {offset = 3232, header_num = 0, file_num = 1, last, line_num = 65}, // 5
629 // {offset = 3235, header_num = 0, file_num = 1, last, line_num = 66}, // 6
630 // {offset = 3236, header_num = 0, file_num = 1, last, line_num = -1}, // 7
631 // {offset = 5764, header_num = 0, file_num = 1, last, line_num = 48}, // 8
632 // {offset = 5764, header_num = 0, file_num = 1,!last, line_num = 47}, // 9
633 // {offset = 5765, header_num = 0, file_num = 1, last, line_num = 49}, // 10
634 // {offset = 5767, header_num = 0, file_num = 1, last, line_num = 50}, // 11
635 // {offset = 5768, header_num = 0, file_num = 1, last, line_num = 51}, // 12
636 // {offset = 5773, header_num = 0, file_num = 1, last, line_num = -1}, // 13
637 // {offset = 5787, header_num = 1, file_num = 1, last, line_num = 19}, // 14
638 // {offset = 5790, header_num = 1, file_num = 1, last, line_num = 20}, // 15
639 // {offset = 5793, header_num = 1, file_num = 1, last, line_num = 67}, // 16
640 // {offset = 5793, header_num = 1, file_num = 1, last, line_num = -1}, // 17
641 // {offset = 5793, header_num = 1, file_num = 1,!last, line_num = 66}, // 18
642 // {offset = 5795, header_num = 1, file_num = 1, last, line_num = 68}, // 19
643 // {offset = 5798, header_num = 1, file_num = 1, last, line_num = -1}, // 20
644 // The entries with line_num == -1 mark the end of a function: the
645 // associated offset is one past the last instruction in the
646 // function. This can correspond to the beginning of the next
647 // function (as is true for offset 3232); alternately, there can be
648 // a gap between the end of one function and the start of the next
649 // (as is true for some others, most obviously from 3236->5764).
651 // Case 1: lookup_key has offset == 10. lower_bound returns
652 // offsets[0]. Since it's not an exact match and we're
653 // at the beginning of offsets, we return end() (invalid).
654 // Case 2: lookup_key has offset 10000. lower_bound returns
655 // offset[21] (end()). We return end() (invalid).
656 // Case 3: lookup_key has offset == 3211. lower_bound matches
657 // offsets[0] exactly, and that's the entry we return.
658 // Case 4: lookup_key has offset == 3232. lower_bound returns
659 // offsets[4]. That's an exact match, but indicates
660 // end-of-function. We check if offsets[5] is also an
661 // exact match but not end-of-function. It is, so we
662 // return offsets[5].
663 // Case 5: lookup_key has offset == 3214. lower_bound returns
664 // offsets[1]. Since it's not an exact match, we back
665 // up to the offset that's < lookup_key, offsets[0].
666 // We note offsets[0] is a valid entry (not end-of-function),
667 // so that's the entry we return.
668 // Case 6: lookup_key has offset == 4000. lower_bound returns
669 // offsets[8]. Since it's not an exact match, we back
670 // up to offsets[7]. Since offsets[7] indicates
671 // end-of-function, we know lookup_key is between
672 // functions, so we return end() (not a valid offset).
673 // Case 7: lookup_key has offset == 5794. lower_bound returns
674 // offsets[19]. Since it's not an exact match, we back
675 // up to offsets[16]. Note we back up to the *first*
676 // entry with offset 5793, not just offsets[19-1].
677 // We note offsets[16] is a valid entry, so we return it.
678 // If offsets[16] had had line_num == -1, we would have
679 // checked offsets[17]. The reason for this is that
680 // 16 and 17 can be in an arbitrary order, since we sort
681 // only by offset and last_line_for_offset. (Note it
682 // doesn't help to use line_number as a tertiary sort key,
683 // since sometimes we want the -1 to be first and sometimes
684 // we want it to be last.)
686 // This deals with cases (1) and (2).
687 if ((it == offsets->begin() && offset < it->offset)
688 || it == offsets->end())
689 return offsets->end();
691 // This deals with cases (3) and (4).
692 if (offset == it->offset)
694 while (it != offsets->end()
695 && it->offset == offset
696 && it->line_num == -1)
697 ++it;
698 if (it == offsets->end() || it->offset != offset)
699 return offsets->end();
700 else
701 return it;
704 // This handles the first part of case (7) -- we back up to the
705 // *first* entry that has the offset that's behind us.
706 gold_assert(it != offsets->begin());
707 std::vector<Offset_to_lineno_entry>::const_iterator range_end = it;
708 --it;
709 const off_t range_value = it->offset;
710 while (it != offsets->begin() && (it-1)->offset == range_value)
711 --it;
713 // This handles cases (5), (6), and (7): if any entry in the
714 // equal_range [it, range_end) has a line_num != -1, it's a valid
715 // match. If not, we're not in a function. The line number we saw
716 // last for an offset will be sorted first, so it'll get returned if
717 // it's present.
718 for (; it != range_end; ++it)
719 if (it->line_num != -1)
720 return it;
721 return offsets->end();
724 // Returns the canonical filename:lineno for the address passed in.
725 // If other_lines is not NULL, appends the non-canonical lines
726 // assigned to the same address.
728 template<int size, bool big_endian>
729 std::string
730 Sized_dwarf_line_info<size, big_endian>::do_addr2line(
731 unsigned int shndx,
732 off_t offset,
733 std::vector<std::string>* other_lines)
735 if (this->data_valid_ == false)
736 return "";
738 const std::vector<Offset_to_lineno_entry>* offsets;
739 // If we do not have reloc information, then our input is a .so or
740 // some similar data structure where all the information is held in
741 // the offset. In that case, we ignore the input shndx.
742 if (this->input_is_relobj())
743 offsets = &this->line_number_map_[shndx];
744 else
745 offsets = &this->line_number_map_[-1U];
746 if (offsets->empty())
747 return "";
749 typename std::vector<Offset_to_lineno_entry>::const_iterator it
750 = offset_to_iterator(offsets, offset);
751 if (it == offsets->end())
752 return "";
754 std::string result = this->format_file_lineno(*it);
755 if (other_lines != NULL)
756 for (++it; it != offsets->end() && it->offset == offset; ++it)
758 if (it->line_num == -1)
759 continue; // The end of a previous function.
760 other_lines->push_back(this->format_file_lineno(*it));
762 return result;
765 // Convert the file_num + line_num into a string.
767 template<int size, bool big_endian>
768 std::string
769 Sized_dwarf_line_info<size, big_endian>::format_file_lineno(
770 const Offset_to_lineno_entry& loc) const
772 std::string ret;
774 gold_assert(loc.header_num < static_cast<int>(this->files_.size()));
775 gold_assert(loc.file_num
776 < static_cast<int>(this->files_[loc.header_num].size()));
777 const std::pair<int, std::string>& filename_pair
778 = this->files_[loc.header_num][loc.file_num];
779 const std::string& filename = filename_pair.second;
781 gold_assert(loc.header_num < static_cast<int>(this->directories_.size()));
782 gold_assert(filename_pair.first
783 < static_cast<int>(this->directories_[loc.header_num].size()));
784 const std::string& dirname
785 = this->directories_[loc.header_num][filename_pair.first];
787 if (!dirname.empty())
789 ret += dirname;
790 ret += "/";
792 ret += filename;
793 if (ret.empty())
794 ret = "(unknown)";
796 char buffer[64]; // enough to hold a line number
797 snprintf(buffer, sizeof(buffer), "%d", loc.line_num);
798 ret += ":";
799 ret += buffer;
801 return ret;
804 // Dwarf_line_info routines.
806 static unsigned int next_generation_count = 0;
808 struct Addr2line_cache_entry
810 Object* object;
811 unsigned int shndx;
812 Dwarf_line_info* dwarf_line_info;
813 unsigned int generation_count;
814 unsigned int access_count;
816 Addr2line_cache_entry(Object* o, unsigned int s, Dwarf_line_info* d)
817 : object(o), shndx(s), dwarf_line_info(d),
818 generation_count(next_generation_count), access_count(0)
820 if (next_generation_count < (1U << 31))
821 ++next_generation_count;
824 // We expect this cache to be small, so don't bother with a hashtable
825 // or priority queue or anything: just use a simple vector.
826 static std::vector<Addr2line_cache_entry> addr2line_cache;
828 std::string
829 Dwarf_line_info::one_addr2line(Object* object,
830 unsigned int shndx, off_t offset,
831 size_t cache_size,
832 std::vector<std::string>* other_lines)
834 Dwarf_line_info* lineinfo = NULL;
835 std::vector<Addr2line_cache_entry>::iterator it;
837 // First, check the cache. If we hit, update the counts.
838 for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it)
840 if (it->object == object && it->shndx == shndx)
842 lineinfo = it->dwarf_line_info;
843 it->generation_count = next_generation_count;
844 // We cap generation_count at 2^31 -1 to avoid overflow.
845 if (next_generation_count < (1U << 31))
846 ++next_generation_count;
847 // We cap access_count at 31 so 2^access_count doesn't overflow
848 if (it->access_count < 31)
849 ++it->access_count;
850 break;
854 // If we don't hit the cache, create a new object and insert into the
855 // cache.
856 if (lineinfo == NULL)
858 switch (parameters->size_and_endianness())
860 #ifdef HAVE_TARGET_32_LITTLE
861 case Parameters::TARGET_32_LITTLE:
862 lineinfo = new Sized_dwarf_line_info<32, false>(object, shndx); break;
863 #endif
864 #ifdef HAVE_TARGET_32_BIG
865 case Parameters::TARGET_32_BIG:
866 lineinfo = new Sized_dwarf_line_info<32, true>(object, shndx); break;
867 #endif
868 #ifdef HAVE_TARGET_64_LITTLE
869 case Parameters::TARGET_64_LITTLE:
870 lineinfo = new Sized_dwarf_line_info<64, false>(object, shndx); break;
871 #endif
872 #ifdef HAVE_TARGET_64_BIG
873 case Parameters::TARGET_64_BIG:
874 lineinfo = new Sized_dwarf_line_info<64, true>(object, shndx); break;
875 #endif
876 default:
877 gold_unreachable();
879 addr2line_cache.push_back(Addr2line_cache_entry(object, shndx, lineinfo));
882 // Now that we have our object, figure out the answer
883 std::string retval = lineinfo->addr2line(shndx, offset, other_lines);
885 // Finally, if our cache has grown too big, delete old objects. We
886 // assume the common (probably only) case is deleting only one object.
887 // We use a pretty simple scheme to evict: function of LRU and MFU.
888 while (addr2line_cache.size() > cache_size)
890 unsigned int lowest_score = ~0U;
891 std::vector<Addr2line_cache_entry>::iterator lowest
892 = addr2line_cache.end();
893 for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it)
895 const unsigned int score = (it->generation_count
896 + (1U << it->access_count));
897 if (score < lowest_score)
899 lowest_score = score;
900 lowest = it;
903 if (lowest != addr2line_cache.end())
905 delete lowest->dwarf_line_info;
906 addr2line_cache.erase(lowest);
910 return retval;
913 void
914 Dwarf_line_info::clear_addr2line_cache()
916 for (std::vector<Addr2line_cache_entry>::iterator it = addr2line_cache.begin();
917 it != addr2line_cache.end();
918 ++it)
919 delete it->dwarf_line_info;
920 addr2line_cache.clear();
923 #ifdef HAVE_TARGET_32_LITTLE
924 template
925 class Sized_dwarf_line_info<32, false>;
926 #endif
928 #ifdef HAVE_TARGET_32_BIG
929 template
930 class Sized_dwarf_line_info<32, true>;
931 #endif
933 #ifdef HAVE_TARGET_64_LITTLE
934 template
935 class Sized_dwarf_line_info<64, false>;
936 #endif
938 #ifdef HAVE_TARGET_64_BIG
939 template
940 class Sized_dwarf_line_info<64, true>;
941 #endif
943 } // End namespace gold.