Commit | Line | Data |
---|---|---|
5c2c6c95 ILT |
1 | // dwarf_reader.cc -- parse dwarf2/3 debug information |
2 | ||
687a9c39 | 3 | // Copyright 2007, 2008, 2009, 2010, 2011 Free Software Foundation, Inc. |
5c2c6c95 ILT |
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 | ||
04bf7072 | 25 | #include <algorithm> |
e4e5049b | 26 | #include <vector> |
04bf7072 | 27 | |
5c2c6c95 ILT |
28 | #include "elfcpp_swap.h" |
29 | #include "dwarf.h" | |
24badc65 | 30 | #include "object.h" |
a55ce7fe | 31 | #include "parameters.h" |
4c50553d | 32 | #include "reloc.h" |
5c2c6c95 | 33 | #include "dwarf_reader.h" |
4f787271 | 34 | #include "int_encoding.h" |
a2e47362 | 35 | #include "compressed_output.h" |
5c2c6c95 | 36 | |
62b01cb5 | 37 | namespace gold { |
5c2c6c95 | 38 | |
5c2c6c95 ILT |
39 | struct LineStateMachine |
40 | { | |
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; | |
49 | }; | |
50 | ||
51 | static void | |
52 | ResetLineStateMachine(struct LineStateMachine* lsm, bool default_is_stmt) | |
53 | { | |
54 | lsm->file_num = 1; | |
55 | lsm->address = 0; | |
56 | lsm->line_num = 1; | |
57 | lsm->column_num = 0; | |
338f2eba | 58 | lsm->shndx = -1U; |
5c2c6c95 ILT |
59 | lsm->is_stmt = default_is_stmt; |
60 | lsm->basic_block = false; | |
61 | lsm->end_sequence = false; | |
62 | } | |
63 | ||
24badc65 | 64 | template<int size, bool big_endian> |
5dd8762a CC |
65 | Sized_dwarf_line_info<size, big_endian>::Sized_dwarf_line_info( |
66 | Object* object, | |
67 | unsigned int read_shndx) | |
68 | : data_valid_(false), buffer_(NULL), buffer_start_(NULL), | |
69 | symtab_buffer_(NULL), directories_(), files_(), current_header_index_(-1) | |
24badc65 ILT |
70 | { |
71 | unsigned int debug_shndx; | |
5dd8762a | 72 | |
ab8056e0 CC |
73 | for (debug_shndx = 1; debug_shndx < object->shnum(); ++debug_shndx) |
74 | { | |
75 | // FIXME: do this more efficiently: section_name() isn't super-fast | |
76 | std::string name = object->section_name(debug_shndx); | |
77 | if (name == ".debug_line" || name == ".zdebug_line") | |
78 | { | |
79 | section_size_type buffer_size; | |
5dd8762a CC |
80 | bool is_new = false; |
81 | this->buffer_ = object->decompressed_section_contents(debug_shndx, | |
82 | &buffer_size, | |
83 | &is_new); | |
84 | if (is_new) | |
85 | this->buffer_start_ = this->buffer_; | |
ab8056e0 CC |
86 | this->buffer_end_ = this->buffer_ + buffer_size; |
87 | break; | |
88 | } | |
89 | } | |
24badc65 | 90 | if (this->buffer_ == NULL) |
c261a0be | 91 | return; |
24badc65 ILT |
92 | |
93 | // Find the relocation section for ".debug_line". | |
af674d1d | 94 | // We expect these for relobjs (.o's) but not dynobjs (.so's). |
24badc65 ILT |
95 | bool got_relocs = false; |
96 | for (unsigned int reloc_shndx = 0; | |
97 | reloc_shndx < object->shnum(); | |
98 | ++reloc_shndx) | |
99 | { | |
100 | unsigned int reloc_sh_type = object->section_type(reloc_shndx); | |
101 | if ((reloc_sh_type == elfcpp::SHT_REL | |
102 | || reloc_sh_type == elfcpp::SHT_RELA) | |
103 | && object->section_info(reloc_shndx) == debug_shndx) | |
104 | { | |
105 | got_relocs = this->track_relocs_.initialize(object, reloc_shndx, | |
106 | reloc_sh_type); | |
4dbfafcc | 107 | this->track_relocs_type_ = reloc_sh_type; |
24badc65 ILT |
108 | break; |
109 | } | |
110 | } | |
24badc65 ILT |
111 | |
112 | // Finally, we need the symtab section to interpret the relocs. | |
af674d1d ILT |
113 | if (got_relocs) |
114 | { | |
115 | unsigned int symtab_shndx; | |
116 | for (symtab_shndx = 0; symtab_shndx < object->shnum(); ++symtab_shndx) | |
117 | if (object->section_type(symtab_shndx) == elfcpp::SHT_SYMTAB) | |
118 | { | |
119 | this->symtab_buffer_ = object->section_contents( | |
120 | symtab_shndx, &this->symtab_buffer_size_, false); | |
121 | break; | |
122 | } | |
123 | if (this->symtab_buffer_ == NULL) | |
124 | return; | |
125 | } | |
24badc65 ILT |
126 | |
127 | // Now that we have successfully read all the data, parse the debug | |
128 | // info. | |
c261a0be | 129 | this->data_valid_ = true; |
d491d34e | 130 | this->read_line_mappings(object, read_shndx); |
24badc65 ILT |
131 | } |
132 | ||
5c2c6c95 ILT |
133 | // Read the DWARF header. |
134 | ||
135 | template<int size, bool big_endian> | |
136 | const unsigned char* | |
a55ce7fe | 137 | Sized_dwarf_line_info<size, big_endian>::read_header_prolog( |
e43872e9 | 138 | const unsigned char* lineptr) |
5c2c6c95 | 139 | { |
deae2a14 | 140 | uint32_t initial_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr); |
5c2c6c95 ILT |
141 | lineptr += 4; |
142 | ||
143 | // In DWARF2/3, if the initial length is all 1 bits, then the offset | |
144 | // size is 8 and we need to read the next 8 bytes for the real length. | |
145 | if (initial_length == 0xffffffff) | |
146 | { | |
147 | header_.offset_size = 8; | |
deae2a14 | 148 | initial_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr); |
5c2c6c95 ILT |
149 | lineptr += 8; |
150 | } | |
151 | else | |
152 | header_.offset_size = 4; | |
153 | ||
154 | header_.total_length = initial_length; | |
155 | ||
156 | gold_assert(lineptr + header_.total_length <= buffer_end_); | |
157 | ||
deae2a14 | 158 | header_.version = elfcpp::Swap_unaligned<16, big_endian>::readval(lineptr); |
5c2c6c95 ILT |
159 | lineptr += 2; |
160 | ||
161 | if (header_.offset_size == 4) | |
deae2a14 | 162 | header_.prologue_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr); |
5c2c6c95 | 163 | else |
deae2a14 | 164 | header_.prologue_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr); |
5c2c6c95 ILT |
165 | lineptr += header_.offset_size; |
166 | ||
167 | header_.min_insn_length = *lineptr; | |
168 | lineptr += 1; | |
169 | ||
170 | header_.default_is_stmt = *lineptr; | |
171 | lineptr += 1; | |
172 | ||
173 | header_.line_base = *reinterpret_cast<const signed char*>(lineptr); | |
174 | lineptr += 1; | |
175 | ||
176 | header_.line_range = *lineptr; | |
177 | lineptr += 1; | |
178 | ||
179 | header_.opcode_base = *lineptr; | |
180 | lineptr += 1; | |
181 | ||
a869183f | 182 | header_.std_opcode_lengths.resize(header_.opcode_base + 1); |
5c2c6c95 ILT |
183 | header_.std_opcode_lengths[0] = 0; |
184 | for (int i = 1; i < header_.opcode_base; i++) | |
185 | { | |
186 | header_.std_opcode_lengths[i] = *lineptr; | |
187 | lineptr += 1; | |
188 | } | |
189 | ||
190 | return lineptr; | |
191 | } | |
192 | ||
193 | // The header for a debug_line section is mildly complicated, because | |
194 | // the line info is very tightly encoded. | |
195 | ||
e43872e9 | 196 | template<int size, bool big_endian> |
5c2c6c95 | 197 | const unsigned char* |
a55ce7fe | 198 | Sized_dwarf_line_info<size, big_endian>::read_header_tables( |
e43872e9 | 199 | const unsigned char* lineptr) |
5c2c6c95 | 200 | { |
af674d1d ILT |
201 | ++this->current_header_index_; |
202 | ||
203 | // Create a new directories_ entry and a new files_ entry for our new | |
204 | // header. We initialize each with a single empty element, because | |
205 | // dwarf indexes directory and filenames starting at 1. | |
206 | gold_assert(static_cast<int>(this->directories_.size()) | |
207 | == this->current_header_index_); | |
208 | gold_assert(static_cast<int>(this->files_.size()) | |
209 | == this->current_header_index_); | |
210 | this->directories_.push_back(std::vector<std::string>(1)); | |
211 | this->files_.push_back(std::vector<std::pair<int, std::string> >(1)); | |
212 | ||
5c2c6c95 ILT |
213 | // It is legal for the directory entry table to be empty. |
214 | if (*lineptr) | |
215 | { | |
216 | int dirindex = 1; | |
217 | while (*lineptr) | |
218 | { | |
af674d1d ILT |
219 | const char* dirname = reinterpret_cast<const char*>(lineptr); |
220 | gold_assert(dirindex | |
221 | == static_cast<int>(this->directories_.back().size())); | |
222 | this->directories_.back().push_back(dirname); | |
223 | lineptr += this->directories_.back().back().size() + 1; | |
5c2c6c95 ILT |
224 | dirindex++; |
225 | } | |
226 | } | |
227 | lineptr++; | |
228 | ||
229 | // It is also legal for the file entry table to be empty. | |
230 | if (*lineptr) | |
231 | { | |
232 | int fileindex = 1; | |
233 | size_t len; | |
234 | while (*lineptr) | |
235 | { | |
236 | const char* filename = reinterpret_cast<const char*>(lineptr); | |
237 | lineptr += strlen(filename) + 1; | |
238 | ||
239 | uint64_t dirindex = read_unsigned_LEB_128(lineptr, &len); | |
5c2c6c95 ILT |
240 | lineptr += len; |
241 | ||
af674d1d ILT |
242 | if (dirindex >= this->directories_.back().size()) |
243 | dirindex = 0; | |
244 | int dirindexi = static_cast<int>(dirindex); | |
245 | ||
5c2c6c95 ILT |
246 | read_unsigned_LEB_128(lineptr, &len); // mod_time |
247 | lineptr += len; | |
248 | ||
249 | read_unsigned_LEB_128(lineptr, &len); // filelength | |
250 | lineptr += len; | |
251 | ||
af674d1d ILT |
252 | gold_assert(fileindex |
253 | == static_cast<int>(this->files_.back().size())); | |
254 | this->files_.back().push_back(std::make_pair(dirindexi, filename)); | |
5c2c6c95 ILT |
255 | fileindex++; |
256 | } | |
257 | } | |
258 | lineptr++; | |
259 | ||
260 | return lineptr; | |
261 | } | |
262 | ||
263 | // Process a single opcode in the .debug.line structure. | |
264 | ||
e43872e9 | 265 | template<int size, bool big_endian> |
5c2c6c95 | 266 | bool |
a55ce7fe | 267 | Sized_dwarf_line_info<size, big_endian>::process_one_opcode( |
e43872e9 | 268 | const unsigned char* start, struct LineStateMachine* lsm, size_t* len) |
5c2c6c95 ILT |
269 | { |
270 | size_t oplen = 0; | |
271 | size_t templen; | |
272 | unsigned char opcode = *start; | |
273 | oplen++; | |
274 | start++; | |
275 | ||
276 | // If the opcode is great than the opcode_base, it is a special | |
277 | // opcode. Most line programs consist mainly of special opcodes. | |
278 | if (opcode >= header_.opcode_base) | |
279 | { | |
280 | opcode -= header_.opcode_base; | |
281 | const int advance_address = ((opcode / header_.line_range) | |
282 | * header_.min_insn_length); | |
283 | lsm->address += advance_address; | |
284 | ||
285 | const int advance_line = ((opcode % header_.line_range) | |
286 | + header_.line_base); | |
287 | lsm->line_num += advance_line; | |
288 | lsm->basic_block = true; | |
289 | *len = oplen; | |
290 | return true; | |
291 | } | |
292 | ||
293 | // Otherwise, we have the regular opcodes | |
294 | switch (opcode) | |
295 | { | |
296 | case elfcpp::DW_LNS_copy: | |
297 | lsm->basic_block = false; | |
298 | *len = oplen; | |
299 | return true; | |
300 | ||
301 | case elfcpp::DW_LNS_advance_pc: | |
302 | { | |
303 | const uint64_t advance_address | |
2ea97941 | 304 | = read_unsigned_LEB_128(start, &templen); |
5c2c6c95 ILT |
305 | oplen += templen; |
306 | lsm->address += header_.min_insn_length * advance_address; | |
307 | } | |
308 | break; | |
309 | ||
310 | case elfcpp::DW_LNS_advance_line: | |
311 | { | |
312 | const uint64_t advance_line = read_signed_LEB_128(start, &templen); | |
313 | oplen += templen; | |
314 | lsm->line_num += advance_line; | |
315 | } | |
316 | break; | |
317 | ||
318 | case elfcpp::DW_LNS_set_file: | |
319 | { | |
320 | const uint64_t fileno = read_unsigned_LEB_128(start, &templen); | |
321 | oplen += templen; | |
322 | lsm->file_num = fileno; | |
323 | } | |
324 | break; | |
325 | ||
326 | case elfcpp::DW_LNS_set_column: | |
327 | { | |
328 | const uint64_t colno = read_unsigned_LEB_128(start, &templen); | |
329 | oplen += templen; | |
330 | lsm->column_num = colno; | |
331 | } | |
332 | break; | |
333 | ||
334 | case elfcpp::DW_LNS_negate_stmt: | |
335 | lsm->is_stmt = !lsm->is_stmt; | |
336 | break; | |
337 | ||
338 | case elfcpp::DW_LNS_set_basic_block: | |
339 | lsm->basic_block = true; | |
340 | break; | |
341 | ||
342 | case elfcpp::DW_LNS_fixed_advance_pc: | |
343 | { | |
344 | int advance_address; | |
deae2a14 | 345 | advance_address = elfcpp::Swap_unaligned<16, big_endian>::readval(start); |
5c2c6c95 ILT |
346 | oplen += 2; |
347 | lsm->address += advance_address; | |
348 | } | |
349 | break; | |
350 | ||
351 | case elfcpp::DW_LNS_const_add_pc: | |
352 | { | |
353 | const int advance_address = (header_.min_insn_length | |
354 | * ((255 - header_.opcode_base) | |
355 | / header_.line_range)); | |
356 | lsm->address += advance_address; | |
357 | } | |
358 | break; | |
359 | ||
360 | case elfcpp::DW_LNS_extended_op: | |
361 | { | |
362 | const uint64_t extended_op_len | |
2ea97941 | 363 | = read_unsigned_LEB_128(start, &templen); |
5c2c6c95 ILT |
364 | start += templen; |
365 | oplen += templen + extended_op_len; | |
366 | ||
367 | const unsigned char extended_op = *start; | |
368 | start++; | |
369 | ||
370 | switch (extended_op) | |
371 | { | |
372 | case elfcpp::DW_LNE_end_sequence: | |
124dfc89 ILT |
373 | // This means that the current byte is the one immediately |
374 | // after a set of instructions. Record the current line | |
375 | // for up to one less than the current address. | |
79e052ea | 376 | lsm->line_num = -1; |
5c2c6c95 ILT |
377 | lsm->end_sequence = true; |
378 | *len = oplen; | |
379 | return true; | |
380 | ||
381 | case elfcpp::DW_LNE_set_address: | |
4c50553d | 382 | { |
4dbfafcc ILT |
383 | lsm->address = |
384 | elfcpp::Swap_unaligned<size, big_endian>::readval(start); | |
4c50553d | 385 | typename Reloc_map::const_iterator it |
4dbfafcc | 386 | = this->reloc_map_.find(start - this->buffer_); |
4c50553d ILT |
387 | if (it != reloc_map_.end()) |
388 | { | |
4dbfafcc ILT |
389 | // If this is a SHT_RELA section, then ignore the |
390 | // section contents. This assumes that this is a | |
391 | // straight reloc which just uses the reloc addend. | |
392 | // The reloc addend has already been included in the | |
393 | // symbol value. | |
394 | if (this->track_relocs_type_ == elfcpp::SHT_RELA) | |
395 | lsm->address = 0; | |
396 | // Add in the symbol value. | |
397 | lsm->address += it->second.second; | |
4c50553d ILT |
398 | lsm->shndx = it->second.first; |
399 | } | |
400 | else | |
401 | { | |
af674d1d ILT |
402 | // If we're a normal .o file, with relocs, every |
403 | // set_address should have an associated relocation. | |
404 | if (this->input_is_relobj()) | |
405 | this->data_valid_ = false; | |
4c50553d ILT |
406 | } |
407 | break; | |
24badc65 | 408 | } |
5c2c6c95 ILT |
409 | case elfcpp::DW_LNE_define_file: |
410 | { | |
411 | const char* filename = reinterpret_cast<const char*>(start); | |
412 | templen = strlen(filename) + 1; | |
413 | start += templen; | |
414 | ||
415 | uint64_t dirindex = read_unsigned_LEB_128(start, &templen); | |
5c2c6c95 ILT |
416 | oplen += templen; |
417 | ||
af674d1d ILT |
418 | if (dirindex >= this->directories_.back().size()) |
419 | dirindex = 0; | |
420 | int dirindexi = static_cast<int>(dirindex); | |
421 | ||
5c2c6c95 ILT |
422 | read_unsigned_LEB_128(start, &templen); // mod_time |
423 | oplen += templen; | |
424 | ||
425 | read_unsigned_LEB_128(start, &templen); // filelength | |
426 | oplen += templen; | |
427 | ||
af674d1d | 428 | this->files_.back().push_back(std::make_pair(dirindexi, |
5c2c6c95 ILT |
429 | filename)); |
430 | } | |
431 | break; | |
432 | } | |
433 | } | |
434 | break; | |
435 | ||
436 | default: | |
437 | { | |
2ea97941 | 438 | // Ignore unknown opcode silently |
5c2c6c95 ILT |
439 | for (int i = 0; i < header_.std_opcode_lengths[opcode]; i++) |
440 | { | |
2ea97941 | 441 | size_t templen; |
5c2c6c95 ILT |
442 | read_unsigned_LEB_128(start, &templen); |
443 | start += templen; | |
444 | oplen += templen; | |
445 | } | |
446 | } | |
447 | break; | |
2ea97941 | 448 | } |
5c2c6c95 ILT |
449 | *len = oplen; |
450 | return false; | |
451 | } | |
452 | ||
453 | // Read the debug information at LINEPTR and store it in the line | |
454 | // number map. | |
455 | ||
e43872e9 | 456 | template<int size, bool big_endian> |
5c2c6c95 | 457 | unsigned const char* |
9430daf8 | 458 | Sized_dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr, |
75aea3d0 | 459 | unsigned int shndx) |
5c2c6c95 ILT |
460 | { |
461 | struct LineStateMachine lsm; | |
462 | ||
463 | // LENGTHSTART is the place the length field is based on. It is the | |
464 | // point in the header after the initial length field. | |
465 | const unsigned char* lengthstart = buffer_; | |
466 | ||
467 | // In 64 bit dwarf, the initial length is 12 bytes, because of the | |
468 | // 0xffffffff at the start. | |
469 | if (header_.offset_size == 8) | |
470 | lengthstart += 12; | |
471 | else | |
472 | lengthstart += 4; | |
473 | ||
474 | while (lineptr < lengthstart + header_.total_length) | |
475 | { | |
476 | ResetLineStateMachine(&lsm, header_.default_is_stmt); | |
477 | while (!lsm.end_sequence) | |
478 | { | |
479 | size_t oplength; | |
e43872e9 | 480 | bool add_line = this->process_one_opcode(lineptr, &lsm, &oplength); |
9430daf8 ILT |
481 | if (add_line |
482 | && (shndx == -1U || lsm.shndx == -1U || shndx == lsm.shndx)) | |
5c2c6c95 ILT |
483 | { |
484 | Offset_to_lineno_entry entry | |
76677ad0 CC |
485 | = { static_cast<off_t>(lsm.address), |
486 | this->current_header_index_, | |
487 | static_cast<unsigned int>(lsm.file_num), | |
488 | true, lsm.line_num }; | |
7500420b ILT |
489 | std::vector<Offset_to_lineno_entry>& |
490 | map(this->line_number_map_[lsm.shndx]); | |
491 | // If we see two consecutive entries with the same | |
71ff8986 ILT |
492 | // offset and a real line number, then mark the first |
493 | // one as non-canonical. | |
7500420b ILT |
494 | if (!map.empty() |
495 | && (map.back().offset == static_cast<off_t>(lsm.address)) | |
496 | && lsm.line_num != -1 | |
497 | && map.back().line_num != -1) | |
71ff8986 ILT |
498 | map.back().last_line_for_offset = false; |
499 | map.push_back(entry); | |
5c2c6c95 ILT |
500 | } |
501 | lineptr += oplength; | |
502 | } | |
503 | } | |
504 | ||
505 | return lengthstart + header_.total_length; | |
506 | } | |
507 | ||
4c50553d ILT |
508 | // Looks in the symtab to see what section a symbol is in. |
509 | ||
510 | template<int size, bool big_endian> | |
511 | unsigned int | |
a55ce7fe | 512 | Sized_dwarf_line_info<size, big_endian>::symbol_section( |
d491d34e | 513 | Object* object, |
4c50553d | 514 | unsigned int sym, |
d491d34e ILT |
515 | typename elfcpp::Elf_types<size>::Elf_Addr* value, |
516 | bool* is_ordinary) | |
4c50553d ILT |
517 | { |
518 | const int symsize = elfcpp::Elf_sizes<size>::sym_size; | |
af674d1d | 519 | gold_assert(sym * symsize < this->symtab_buffer_size_); |
4c50553d ILT |
520 | elfcpp::Sym<size, big_endian> elfsym(this->symtab_buffer_ + sym * symsize); |
521 | *value = elfsym.get_st_value(); | |
d491d34e | 522 | return object->adjust_sym_shndx(sym, elfsym.get_st_shndx(), is_ordinary); |
4c50553d ILT |
523 | } |
524 | ||
525 | // Read the relocations into a Reloc_map. | |
526 | ||
527 | template<int size, bool big_endian> | |
528 | void | |
d491d34e | 529 | Sized_dwarf_line_info<size, big_endian>::read_relocs(Object* object) |
4c50553d ILT |
530 | { |
531 | if (this->symtab_buffer_ == NULL) | |
532 | return; | |
533 | ||
534 | typename elfcpp::Elf_types<size>::Elf_Addr value; | |
535 | off_t reloc_offset; | |
24badc65 | 536 | while ((reloc_offset = this->track_relocs_.next_offset()) != -1) |
4c50553d | 537 | { |
24badc65 | 538 | const unsigned int sym = this->track_relocs_.next_symndx(); |
d491d34e ILT |
539 | |
540 | bool is_ordinary; | |
541 | const unsigned int shndx = this->symbol_section(object, sym, &value, | |
542 | &is_ordinary); | |
543 | ||
544 | // There is no reason to record non-ordinary section indexes, or | |
545 | // SHN_UNDEF, because they will never match the real section. | |
546 | if (is_ordinary && shndx != elfcpp::SHN_UNDEF) | |
4dbfafcc ILT |
547 | { |
548 | value += this->track_relocs_.next_addend(); | |
549 | this->reloc_map_[reloc_offset] = std::make_pair(shndx, value); | |
550 | } | |
d491d34e | 551 | |
24badc65 | 552 | this->track_relocs_.advance(reloc_offset + 1); |
4c50553d ILT |
553 | } |
554 | } | |
555 | ||
556 | // Read the line number info. | |
557 | ||
e43872e9 | 558 | template<int size, bool big_endian> |
5c2c6c95 | 559 | void |
d491d34e | 560 | Sized_dwarf_line_info<size, big_endian>::read_line_mappings(Object* object, |
75aea3d0 | 561 | unsigned int shndx) |
5c2c6c95 | 562 | { |
c261a0be | 563 | gold_assert(this->data_valid_ == true); |
24badc65 | 564 | |
d491d34e | 565 | this->read_relocs(object); |
4c50553d | 566 | while (this->buffer_ < this->buffer_end_) |
e43872e9 | 567 | { |
4c50553d | 568 | const unsigned char* lineptr = this->buffer_; |
e43872e9 ILT |
569 | lineptr = this->read_header_prolog(lineptr); |
570 | lineptr = this->read_header_tables(lineptr); | |
9430daf8 | 571 | lineptr = this->read_lines(lineptr, shndx); |
4c50553d | 572 | this->buffer_ = lineptr; |
e43872e9 ILT |
573 | } |
574 | ||
575 | // Sort the lines numbers, so addr2line can use binary search. | |
576 | for (typename Lineno_map::iterator it = line_number_map_.begin(); | |
5c2c6c95 ILT |
577 | it != line_number_map_.end(); |
578 | ++it) | |
579 | // Each vector needs to be sorted by offset. | |
4c50553d | 580 | std::sort(it->second.begin(), it->second.end()); |
5c2c6c95 ILT |
581 | } |
582 | ||
af674d1d ILT |
583 | // Some processing depends on whether the input is a .o file or not. |
584 | // For instance, .o files have relocs, and have .debug_lines | |
585 | // information on a per section basis. .so files, on the other hand, | |
586 | // lack relocs, and offsets are unique, so we can ignore the section | |
587 | // information. | |
588 | ||
589 | template<int size, bool big_endian> | |
590 | bool | |
a55ce7fe | 591 | Sized_dwarf_line_info<size, big_endian>::input_is_relobj() |
af674d1d ILT |
592 | { |
593 | // Only .o files have relocs and the symtab buffer that goes with them. | |
594 | return this->symtab_buffer_ != NULL; | |
595 | } | |
596 | ||
79e052ea ILT |
597 | // Given an Offset_to_lineno_entry vector, and an offset, figure out |
598 | // if the offset points into a function according to the vector (see | |
599 | // comments below for the algorithm). If it does, return an iterator | |
600 | // into the vector that points to the line-number that contains that | |
601 | // offset. If not, it returns vector::end(). | |
602 | ||
603 | static std::vector<Offset_to_lineno_entry>::const_iterator | |
604 | offset_to_iterator(const std::vector<Offset_to_lineno_entry>* offsets, | |
605 | off_t offset) | |
606 | { | |
71ff8986 | 607 | const Offset_to_lineno_entry lookup_key = { offset, 0, 0, true, 0 }; |
79e052ea ILT |
608 | |
609 | // lower_bound() returns the smallest offset which is >= lookup_key. | |
610 | // If no offset in offsets is >= lookup_key, returns end(). | |
611 | std::vector<Offset_to_lineno_entry>::const_iterator it | |
612 | = std::lower_bound(offsets->begin(), offsets->end(), lookup_key); | |
613 | ||
614 | // This code is easiest to understand with a concrete example. | |
615 | // Here's a possible offsets array: | |
71ff8986 ILT |
616 | // {{offset = 3211, header_num = 0, file_num = 1, last, line_num = 16}, // 0 |
617 | // {offset = 3224, header_num = 0, file_num = 1, last, line_num = 20}, // 1 | |
618 | // {offset = 3226, header_num = 0, file_num = 1, last, line_num = 22}, // 2 | |
619 | // {offset = 3231, header_num = 0, file_num = 1, last, line_num = 25}, // 3 | |
620 | // {offset = 3232, header_num = 0, file_num = 1, last, line_num = -1}, // 4 | |
621 | // {offset = 3232, header_num = 0, file_num = 1, last, line_num = 65}, // 5 | |
622 | // {offset = 3235, header_num = 0, file_num = 1, last, line_num = 66}, // 6 | |
623 | // {offset = 3236, header_num = 0, file_num = 1, last, line_num = -1}, // 7 | |
624 | // {offset = 5764, header_num = 0, file_num = 1, last, line_num = 48}, // 8 | |
625 | // {offset = 5764, header_num = 0, file_num = 1,!last, line_num = 47}, // 9 | |
626 | // {offset = 5765, header_num = 0, file_num = 1, last, line_num = 49}, // 10 | |
627 | // {offset = 5767, header_num = 0, file_num = 1, last, line_num = 50}, // 11 | |
628 | // {offset = 5768, header_num = 0, file_num = 1, last, line_num = 51}, // 12 | |
629 | // {offset = 5773, header_num = 0, file_num = 1, last, line_num = -1}, // 13 | |
630 | // {offset = 5787, header_num = 1, file_num = 1, last, line_num = 19}, // 14 | |
631 | // {offset = 5790, header_num = 1, file_num = 1, last, line_num = 20}, // 15 | |
632 | // {offset = 5793, header_num = 1, file_num = 1, last, line_num = 67}, // 16 | |
633 | // {offset = 5793, header_num = 1, file_num = 1, last, line_num = -1}, // 17 | |
634 | // {offset = 5793, header_num = 1, file_num = 1,!last, line_num = 66}, // 18 | |
635 | // {offset = 5795, header_num = 1, file_num = 1, last, line_num = 68}, // 19 | |
636 | // {offset = 5798, header_num = 1, file_num = 1, last, line_num = -1}, // 20 | |
79e052ea ILT |
637 | // The entries with line_num == -1 mark the end of a function: the |
638 | // associated offset is one past the last instruction in the | |
639 | // function. This can correspond to the beginning of the next | |
640 | // function (as is true for offset 3232); alternately, there can be | |
641 | // a gap between the end of one function and the start of the next | |
ef04e392 | 642 | // (as is true for some others, most obviously from 3236->5764). |
79e052ea ILT |
643 | // |
644 | // Case 1: lookup_key has offset == 10. lower_bound returns | |
645 | // offsets[0]. Since it's not an exact match and we're | |
ef04e392 | 646 | // at the beginning of offsets, we return end() (invalid). |
79e052ea | 647 | // Case 2: lookup_key has offset 10000. lower_bound returns |
71ff8986 | 648 | // offset[21] (end()). We return end() (invalid). |
79e052ea ILT |
649 | // Case 3: lookup_key has offset == 3211. lower_bound matches |
650 | // offsets[0] exactly, and that's the entry we return. | |
651 | // Case 4: lookup_key has offset == 3232. lower_bound returns | |
652 | // offsets[4]. That's an exact match, but indicates | |
653 | // end-of-function. We check if offsets[5] is also an | |
654 | // exact match but not end-of-function. It is, so we | |
655 | // return offsets[5]. | |
656 | // Case 5: lookup_key has offset == 3214. lower_bound returns | |
657 | // offsets[1]. Since it's not an exact match, we back | |
658 | // up to the offset that's < lookup_key, offsets[0]. | |
659 | // We note offsets[0] is a valid entry (not end-of-function), | |
660 | // so that's the entry we return. | |
661 | // Case 6: lookup_key has offset == 4000. lower_bound returns | |
662 | // offsets[8]. Since it's not an exact match, we back | |
663 | // up to offsets[7]. Since offsets[7] indicates | |
664 | // end-of-function, we know lookup_key is between | |
ef04e392 | 665 | // functions, so we return end() (not a valid offset). |
79e052ea | 666 | // Case 7: lookup_key has offset == 5794. lower_bound returns |
71ff8986 ILT |
667 | // offsets[19]. Since it's not an exact match, we back |
668 | // up to offsets[16]. Note we back up to the *first* | |
669 | // entry with offset 5793, not just offsets[19-1]. | |
670 | // We note offsets[16] is a valid entry, so we return it. | |
671 | // If offsets[16] had had line_num == -1, we would have | |
672 | // checked offsets[17]. The reason for this is that | |
673 | // 16 and 17 can be in an arbitrary order, since we sort | |
674 | // only by offset and last_line_for_offset. (Note it | |
675 | // doesn't help to use line_number as a tertiary sort key, | |
676 | // since sometimes we want the -1 to be first and sometimes | |
677 | // we want it to be last.) | |
79e052ea ILT |
678 | |
679 | // This deals with cases (1) and (2). | |
680 | if ((it == offsets->begin() && offset < it->offset) | |
681 | || it == offsets->end()) | |
682 | return offsets->end(); | |
683 | ||
684 | // This deals with cases (3) and (4). | |
685 | if (offset == it->offset) | |
686 | { | |
687 | while (it != offsets->end() | |
688 | && it->offset == offset | |
689 | && it->line_num == -1) | |
690 | ++it; | |
691 | if (it == offsets->end() || it->offset != offset) | |
692 | return offsets->end(); | |
693 | else | |
694 | return it; | |
695 | } | |
696 | ||
697 | // This handles the first part of case (7) -- we back up to the | |
698 | // *first* entry that has the offset that's behind us. | |
699 | gold_assert(it != offsets->begin()); | |
700 | std::vector<Offset_to_lineno_entry>::const_iterator range_end = it; | |
701 | --it; | |
702 | const off_t range_value = it->offset; | |
703 | while (it != offsets->begin() && (it-1)->offset == range_value) | |
704 | --it; | |
705 | ||
706 | // This handles cases (5), (6), and (7): if any entry in the | |
707 | // equal_range [it, range_end) has a line_num != -1, it's a valid | |
71ff8986 ILT |
708 | // match. If not, we're not in a function. The line number we saw |
709 | // last for an offset will be sorted first, so it'll get returned if | |
710 | // it's present. | |
79e052ea ILT |
711 | for (; it != range_end; ++it) |
712 | if (it->line_num != -1) | |
713 | return it; | |
714 | return offsets->end(); | |
715 | } | |
af674d1d | 716 | |
71ff8986 ILT |
717 | // Returns the canonical filename:lineno for the address passed in. |
718 | // If other_lines is not NULL, appends the non-canonical lines | |
719 | // assigned to the same address. | |
5c2c6c95 | 720 | |
e43872e9 | 721 | template<int size, bool big_endian> |
5c2c6c95 | 722 | std::string |
71ff8986 ILT |
723 | Sized_dwarf_line_info<size, big_endian>::do_addr2line( |
724 | unsigned int shndx, | |
725 | off_t offset, | |
726 | std::vector<std::string>* other_lines) | |
5c2c6c95 | 727 | { |
4c50553d ILT |
728 | if (this->data_valid_ == false) |
729 | return ""; | |
730 | ||
af674d1d ILT |
731 | const std::vector<Offset_to_lineno_entry>* offsets; |
732 | // If we do not have reloc information, then our input is a .so or | |
733 | // some similar data structure where all the information is held in | |
734 | // the offset. In that case, we ignore the input shndx. | |
735 | if (this->input_is_relobj()) | |
736 | offsets = &this->line_number_map_[shndx]; | |
737 | else | |
738 | offsets = &this->line_number_map_[-1U]; | |
739 | if (offsets->empty()) | |
4c50553d ILT |
740 | return ""; |
741 | ||
e43872e9 | 742 | typename std::vector<Offset_to_lineno_entry>::const_iterator it |
79e052ea ILT |
743 | = offset_to_iterator(offsets, offset); |
744 | if (it == offsets->end()) | |
745 | return ""; | |
5c2c6c95 | 746 | |
71ff8986 ILT |
747 | std::string result = this->format_file_lineno(*it); |
748 | if (other_lines != NULL) | |
749 | for (++it; it != offsets->end() && it->offset == offset; ++it) | |
750 | { | |
751 | if (it->line_num == -1) | |
752 | continue; // The end of a previous function. | |
753 | other_lines->push_back(this->format_file_lineno(*it)); | |
754 | } | |
755 | return result; | |
756 | } | |
757 | ||
758 | // Convert the file_num + line_num into a string. | |
759 | ||
760 | template<int size, bool big_endian> | |
761 | std::string | |
762 | Sized_dwarf_line_info<size, big_endian>::format_file_lineno( | |
763 | const Offset_to_lineno_entry& loc) const | |
764 | { | |
5c2c6c95 | 765 | std::string ret; |
af674d1d | 766 | |
71ff8986 ILT |
767 | gold_assert(loc.header_num < static_cast<int>(this->files_.size())); |
768 | gold_assert(loc.file_num | |
769 | < static_cast<int>(this->files_[loc.header_num].size())); | |
af674d1d | 770 | const std::pair<int, std::string>& filename_pair |
71ff8986 | 771 | = this->files_[loc.header_num][loc.file_num]; |
5c2c6c95 | 772 | const std::string& filename = filename_pair.second; |
af674d1d | 773 | |
71ff8986 | 774 | gold_assert(loc.header_num < static_cast<int>(this->directories_.size())); |
af674d1d | 775 | gold_assert(filename_pair.first |
71ff8986 | 776 | < static_cast<int>(this->directories_[loc.header_num].size())); |
af674d1d | 777 | const std::string& dirname |
71ff8986 | 778 | = this->directories_[loc.header_num][filename_pair.first]; |
af674d1d | 779 | |
5c2c6c95 ILT |
780 | if (!dirname.empty()) |
781 | { | |
782 | ret += dirname; | |
783 | ret += "/"; | |
784 | } | |
785 | ret += filename; | |
786 | if (ret.empty()) | |
787 | ret = "(unknown)"; | |
788 | ||
789 | char buffer[64]; // enough to hold a line number | |
71ff8986 | 790 | snprintf(buffer, sizeof(buffer), "%d", loc.line_num); |
5c2c6c95 ILT |
791 | ret += ":"; |
792 | ret += buffer; | |
793 | ||
794 | return ret; | |
795 | } | |
796 | ||
a55ce7fe ILT |
797 | // Dwarf_line_info routines. |
798 | ||
e4e5049b CS |
799 | static unsigned int next_generation_count = 0; |
800 | ||
801 | struct Addr2line_cache_entry | |
802 | { | |
803 | Object* object; | |
804 | unsigned int shndx; | |
805 | Dwarf_line_info* dwarf_line_info; | |
806 | unsigned int generation_count; | |
807 | unsigned int access_count; | |
808 | ||
809 | Addr2line_cache_entry(Object* o, unsigned int s, Dwarf_line_info* d) | |
810 | : object(o), shndx(s), dwarf_line_info(d), | |
811 | generation_count(next_generation_count), access_count(0) | |
812 | { | |
813 | if (next_generation_count < (1U << 31)) | |
814 | ++next_generation_count; | |
815 | } | |
816 | }; | |
817 | // We expect this cache to be small, so don't bother with a hashtable | |
818 | // or priority queue or anything: just use a simple vector. | |
819 | static std::vector<Addr2line_cache_entry> addr2line_cache; | |
820 | ||
a55ce7fe ILT |
821 | std::string |
822 | Dwarf_line_info::one_addr2line(Object* object, | |
e4e5049b | 823 | unsigned int shndx, off_t offset, |
71ff8986 ILT |
824 | size_t cache_size, |
825 | std::vector<std::string>* other_lines) | |
a55ce7fe | 826 | { |
e4e5049b CS |
827 | Dwarf_line_info* lineinfo = NULL; |
828 | std::vector<Addr2line_cache_entry>::iterator it; | |
829 | ||
830 | // First, check the cache. If we hit, update the counts. | |
831 | for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it) | |
8851ecca | 832 | { |
e4e5049b CS |
833 | if (it->object == object && it->shndx == shndx) |
834 | { | |
835 | lineinfo = it->dwarf_line_info; | |
836 | it->generation_count = next_generation_count; | |
837 | // We cap generation_count at 2^31 -1 to avoid overflow. | |
838 | if (next_generation_count < (1U << 31)) | |
839 | ++next_generation_count; | |
840 | // We cap access_count at 31 so 2^access_count doesn't overflow | |
841 | if (it->access_count < 31) | |
842 | ++it->access_count; | |
843 | break; | |
844 | } | |
845 | } | |
846 | ||
847 | // If we don't hit the cache, create a new object and insert into the | |
848 | // cache. | |
849 | if (lineinfo == NULL) | |
850 | { | |
851 | switch (parameters->size_and_endianness()) | |
852 | { | |
a55ce7fe | 853 | #ifdef HAVE_TARGET_32_LITTLE |
e4e5049b CS |
854 | case Parameters::TARGET_32_LITTLE: |
855 | lineinfo = new Sized_dwarf_line_info<32, false>(object, shndx); break; | |
a55ce7fe | 856 | #endif |
a55ce7fe | 857 | #ifdef HAVE_TARGET_32_BIG |
e4e5049b CS |
858 | case Parameters::TARGET_32_BIG: |
859 | lineinfo = new Sized_dwarf_line_info<32, true>(object, shndx); break; | |
a55ce7fe | 860 | #endif |
a55ce7fe | 861 | #ifdef HAVE_TARGET_64_LITTLE |
e4e5049b CS |
862 | case Parameters::TARGET_64_LITTLE: |
863 | lineinfo = new Sized_dwarf_line_info<64, false>(object, shndx); break; | |
a55ce7fe | 864 | #endif |
8851ecca | 865 | #ifdef HAVE_TARGET_64_BIG |
e4e5049b CS |
866 | case Parameters::TARGET_64_BIG: |
867 | lineinfo = new Sized_dwarf_line_info<64, true>(object, shndx); break; | |
a55ce7fe | 868 | #endif |
e4e5049b CS |
869 | default: |
870 | gold_unreachable(); | |
871 | } | |
872 | addr2line_cache.push_back(Addr2line_cache_entry(object, shndx, lineinfo)); | |
873 | } | |
874 | ||
875 | // Now that we have our object, figure out the answer | |
71ff8986 | 876 | std::string retval = lineinfo->addr2line(shndx, offset, other_lines); |
e4e5049b CS |
877 | |
878 | // Finally, if our cache has grown too big, delete old objects. We | |
879 | // assume the common (probably only) case is deleting only one object. | |
880 | // We use a pretty simple scheme to evict: function of LRU and MFU. | |
881 | while (addr2line_cache.size() > cache_size) | |
882 | { | |
883 | unsigned int lowest_score = ~0U; | |
884 | std::vector<Addr2line_cache_entry>::iterator lowest | |
885 | = addr2line_cache.end(); | |
886 | for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it) | |
887 | { | |
888 | const unsigned int score = (it->generation_count | |
889 | + (1U << it->access_count)); | |
890 | if (score < lowest_score) | |
891 | { | |
892 | lowest_score = score; | |
893 | lowest = it; | |
894 | } | |
895 | } | |
896 | if (lowest != addr2line_cache.end()) | |
897 | { | |
898 | delete lowest->dwarf_line_info; | |
899 | addr2line_cache.erase(lowest); | |
900 | } | |
8851ecca | 901 | } |
e4e5049b CS |
902 | |
903 | return retval; | |
904 | } | |
905 | ||
906 | void | |
907 | Dwarf_line_info::clear_addr2line_cache() | |
908 | { | |
909 | for (std::vector<Addr2line_cache_entry>::iterator it = addr2line_cache.begin(); | |
910 | it != addr2line_cache.end(); | |
911 | ++it) | |
912 | delete it->dwarf_line_info; | |
913 | addr2line_cache.clear(); | |
a55ce7fe ILT |
914 | } |
915 | ||
5c2c6c95 ILT |
916 | #ifdef HAVE_TARGET_32_LITTLE |
917 | template | |
a55ce7fe | 918 | class Sized_dwarf_line_info<32, false>; |
5c2c6c95 ILT |
919 | #endif |
920 | ||
921 | #ifdef HAVE_TARGET_32_BIG | |
922 | template | |
a55ce7fe | 923 | class Sized_dwarf_line_info<32, true>; |
5c2c6c95 ILT |
924 | #endif |
925 | ||
926 | #ifdef HAVE_TARGET_64_LITTLE | |
927 | template | |
a55ce7fe | 928 | class Sized_dwarf_line_info<64, false>; |
5c2c6c95 ILT |
929 | #endif |
930 | ||
931 | #ifdef HAVE_TARGET_64_BIG | |
932 | template | |
a55ce7fe | 933 | class Sized_dwarf_line_info<64, true>; |
5c2c6c95 ILT |
934 | #endif |
935 | ||
936 | } // End namespace gold. |