Commit | Line | Data |
---|---|---|
5c2c6c95 ILT |
1 | // dwarf_reader.cc -- parse dwarf2/3 debug information |
2 | ||
ebdbb458 | 3 | // Copyright 2007, 2008 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 ILT |
25 | #include <algorithm> |
26 | ||
5c2c6c95 ILT |
27 | #include "elfcpp_swap.h" |
28 | #include "dwarf.h" | |
24badc65 | 29 | #include "object.h" |
a55ce7fe | 30 | #include "parameters.h" |
4c50553d | 31 | #include "reloc.h" |
5c2c6c95 ILT |
32 | #include "dwarf_reader.h" |
33 | ||
34 | namespace { | |
35 | ||
36 | // Read an unsigned LEB128 number. Each byte contains 7 bits of | |
37 | // information, plus one bit saying whether the number continues or | |
38 | // not. | |
39 | ||
40 | uint64_t | |
41 | read_unsigned_LEB_128(const unsigned char* buffer, size_t* len) | |
42 | { | |
43 | uint64_t result = 0; | |
44 | size_t num_read = 0; | |
45 | unsigned int shift = 0; | |
46 | unsigned char byte; | |
47 | ||
48 | do | |
49 | { | |
50 | byte = *buffer++; | |
51 | num_read++; | |
52 | result |= (static_cast<uint64_t>(byte & 0x7f)) << shift; | |
53 | shift += 7; | |
54 | } | |
55 | while (byte & 0x80); | |
56 | ||
57 | *len = num_read; | |
58 | ||
59 | return result; | |
60 | } | |
61 | ||
62 | // Read a signed LEB128 number. These are like regular LEB128 | |
63 | // numbers, except the last byte may have a sign bit set. | |
64 | ||
65 | int64_t | |
66 | read_signed_LEB_128(const unsigned char* buffer, size_t* len) | |
67 | { | |
68 | int64_t result = 0; | |
69 | int shift = 0; | |
70 | size_t num_read = 0; | |
71 | unsigned char byte; | |
72 | ||
73 | do | |
74 | { | |
75 | byte = *buffer++; | |
76 | num_read++; | |
77 | result |= (static_cast<uint64_t>(byte & 0x7f) << shift); | |
78 | shift += 7; | |
79 | } | |
80 | while (byte & 0x80); | |
81 | ||
82 | if ((shift < 8 * static_cast<int>(sizeof(result))) && (byte & 0x40)) | |
83 | result |= -((static_cast<int64_t>(1)) << shift); | |
84 | *len = num_read; | |
85 | return result; | |
86 | } | |
87 | ||
88 | } // End anonymous namespace. | |
89 | ||
90 | ||
91 | namespace gold { | |
92 | ||
93 | // This is the format of a DWARF2/3 line state machine that we process | |
94 | // opcodes using. There is no need for anything outside the lineinfo | |
95 | // processor to know how this works. | |
96 | ||
97 | struct LineStateMachine | |
98 | { | |
99 | int file_num; | |
100 | uint64_t address; | |
101 | int line_num; | |
102 | int column_num; | |
103 | unsigned int shndx; // the section address refers to | |
104 | bool is_stmt; // stmt means statement. | |
105 | bool basic_block; | |
106 | bool end_sequence; | |
107 | }; | |
108 | ||
109 | static void | |
110 | ResetLineStateMachine(struct LineStateMachine* lsm, bool default_is_stmt) | |
111 | { | |
112 | lsm->file_num = 1; | |
113 | lsm->address = 0; | |
114 | lsm->line_num = 1; | |
115 | lsm->column_num = 0; | |
338f2eba | 116 | lsm->shndx = -1U; |
5c2c6c95 ILT |
117 | lsm->is_stmt = default_is_stmt; |
118 | lsm->basic_block = false; | |
119 | lsm->end_sequence = false; | |
120 | } | |
121 | ||
24badc65 | 122 | template<int size, bool big_endian> |
9430daf8 ILT |
123 | Sized_dwarf_line_info<size, big_endian>::Sized_dwarf_line_info(Object* object, |
124 | off_t read_shndx) | |
c261a0be | 125 | : data_valid_(false), buffer_(NULL), symtab_buffer_(NULL), |
af674d1d | 126 | directories_(), files_(), current_header_index_(-1) |
24badc65 ILT |
127 | { |
128 | unsigned int debug_shndx; | |
129 | for (debug_shndx = 0; debug_shndx < object->shnum(); ++debug_shndx) | |
10600224 | 130 | // FIXME: do this more efficiently: section_name() isn't super-fast |
24badc65 ILT |
131 | if (object->section_name(debug_shndx) == ".debug_line") |
132 | { | |
8383303e ILT |
133 | section_size_type buffer_size; |
134 | this->buffer_ = object->section_contents(debug_shndx, &buffer_size, | |
135 | false); | |
24badc65 ILT |
136 | this->buffer_end_ = this->buffer_ + buffer_size; |
137 | break; | |
138 | } | |
139 | if (this->buffer_ == NULL) | |
c261a0be | 140 | return; |
24badc65 ILT |
141 | |
142 | // Find the relocation section for ".debug_line". | |
af674d1d | 143 | // We expect these for relobjs (.o's) but not dynobjs (.so's). |
24badc65 ILT |
144 | bool got_relocs = false; |
145 | for (unsigned int reloc_shndx = 0; | |
146 | reloc_shndx < object->shnum(); | |
147 | ++reloc_shndx) | |
148 | { | |
149 | unsigned int reloc_sh_type = object->section_type(reloc_shndx); | |
150 | if ((reloc_sh_type == elfcpp::SHT_REL | |
151 | || reloc_sh_type == elfcpp::SHT_RELA) | |
152 | && object->section_info(reloc_shndx) == debug_shndx) | |
153 | { | |
154 | got_relocs = this->track_relocs_.initialize(object, reloc_shndx, | |
155 | reloc_sh_type); | |
156 | break; | |
157 | } | |
158 | } | |
24badc65 ILT |
159 | |
160 | // Finally, we need the symtab section to interpret the relocs. | |
af674d1d ILT |
161 | if (got_relocs) |
162 | { | |
163 | unsigned int symtab_shndx; | |
164 | for (symtab_shndx = 0; symtab_shndx < object->shnum(); ++symtab_shndx) | |
165 | if (object->section_type(symtab_shndx) == elfcpp::SHT_SYMTAB) | |
166 | { | |
167 | this->symtab_buffer_ = object->section_contents( | |
168 | symtab_shndx, &this->symtab_buffer_size_, false); | |
169 | break; | |
170 | } | |
171 | if (this->symtab_buffer_ == NULL) | |
172 | return; | |
173 | } | |
24badc65 ILT |
174 | |
175 | // Now that we have successfully read all the data, parse the debug | |
176 | // info. | |
c261a0be | 177 | this->data_valid_ = true; |
d491d34e | 178 | this->read_line_mappings(object, read_shndx); |
24badc65 ILT |
179 | } |
180 | ||
5c2c6c95 ILT |
181 | // Read the DWARF header. |
182 | ||
183 | template<int size, bool big_endian> | |
184 | const unsigned char* | |
a55ce7fe | 185 | Sized_dwarf_line_info<size, big_endian>::read_header_prolog( |
e43872e9 | 186 | const unsigned char* lineptr) |
5c2c6c95 | 187 | { |
deae2a14 | 188 | uint32_t initial_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr); |
5c2c6c95 ILT |
189 | lineptr += 4; |
190 | ||
191 | // In DWARF2/3, if the initial length is all 1 bits, then the offset | |
192 | // size is 8 and we need to read the next 8 bytes for the real length. | |
193 | if (initial_length == 0xffffffff) | |
194 | { | |
195 | header_.offset_size = 8; | |
deae2a14 | 196 | initial_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr); |
5c2c6c95 ILT |
197 | lineptr += 8; |
198 | } | |
199 | else | |
200 | header_.offset_size = 4; | |
201 | ||
202 | header_.total_length = initial_length; | |
203 | ||
204 | gold_assert(lineptr + header_.total_length <= buffer_end_); | |
205 | ||
deae2a14 | 206 | header_.version = elfcpp::Swap_unaligned<16, big_endian>::readval(lineptr); |
5c2c6c95 ILT |
207 | lineptr += 2; |
208 | ||
209 | if (header_.offset_size == 4) | |
deae2a14 | 210 | header_.prologue_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr); |
5c2c6c95 | 211 | else |
deae2a14 | 212 | header_.prologue_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr); |
5c2c6c95 ILT |
213 | lineptr += header_.offset_size; |
214 | ||
215 | header_.min_insn_length = *lineptr; | |
216 | lineptr += 1; | |
217 | ||
218 | header_.default_is_stmt = *lineptr; | |
219 | lineptr += 1; | |
220 | ||
221 | header_.line_base = *reinterpret_cast<const signed char*>(lineptr); | |
222 | lineptr += 1; | |
223 | ||
224 | header_.line_range = *lineptr; | |
225 | lineptr += 1; | |
226 | ||
227 | header_.opcode_base = *lineptr; | |
228 | lineptr += 1; | |
229 | ||
230 | header_.std_opcode_lengths.reserve(header_.opcode_base + 1); | |
231 | header_.std_opcode_lengths[0] = 0; | |
232 | for (int i = 1; i < header_.opcode_base; i++) | |
233 | { | |
234 | header_.std_opcode_lengths[i] = *lineptr; | |
235 | lineptr += 1; | |
236 | } | |
237 | ||
238 | return lineptr; | |
239 | } | |
240 | ||
241 | // The header for a debug_line section is mildly complicated, because | |
242 | // the line info is very tightly encoded. | |
243 | ||
e43872e9 | 244 | template<int size, bool big_endian> |
5c2c6c95 | 245 | const unsigned char* |
a55ce7fe | 246 | Sized_dwarf_line_info<size, big_endian>::read_header_tables( |
e43872e9 | 247 | const unsigned char* lineptr) |
5c2c6c95 | 248 | { |
af674d1d ILT |
249 | ++this->current_header_index_; |
250 | ||
251 | // Create a new directories_ entry and a new files_ entry for our new | |
252 | // header. We initialize each with a single empty element, because | |
253 | // dwarf indexes directory and filenames starting at 1. | |
254 | gold_assert(static_cast<int>(this->directories_.size()) | |
255 | == this->current_header_index_); | |
256 | gold_assert(static_cast<int>(this->files_.size()) | |
257 | == this->current_header_index_); | |
258 | this->directories_.push_back(std::vector<std::string>(1)); | |
259 | this->files_.push_back(std::vector<std::pair<int, std::string> >(1)); | |
260 | ||
5c2c6c95 ILT |
261 | // It is legal for the directory entry table to be empty. |
262 | if (*lineptr) | |
263 | { | |
264 | int dirindex = 1; | |
265 | while (*lineptr) | |
266 | { | |
af674d1d ILT |
267 | const char* dirname = reinterpret_cast<const char*>(lineptr); |
268 | gold_assert(dirindex | |
269 | == static_cast<int>(this->directories_.back().size())); | |
270 | this->directories_.back().push_back(dirname); | |
271 | lineptr += this->directories_.back().back().size() + 1; | |
5c2c6c95 ILT |
272 | dirindex++; |
273 | } | |
274 | } | |
275 | lineptr++; | |
276 | ||
277 | // It is also legal for the file entry table to be empty. | |
278 | if (*lineptr) | |
279 | { | |
280 | int fileindex = 1; | |
281 | size_t len; | |
282 | while (*lineptr) | |
283 | { | |
284 | const char* filename = reinterpret_cast<const char*>(lineptr); | |
285 | lineptr += strlen(filename) + 1; | |
286 | ||
287 | uint64_t dirindex = read_unsigned_LEB_128(lineptr, &len); | |
5c2c6c95 ILT |
288 | lineptr += len; |
289 | ||
af674d1d ILT |
290 | if (dirindex >= this->directories_.back().size()) |
291 | dirindex = 0; | |
292 | int dirindexi = static_cast<int>(dirindex); | |
293 | ||
5c2c6c95 ILT |
294 | read_unsigned_LEB_128(lineptr, &len); // mod_time |
295 | lineptr += len; | |
296 | ||
297 | read_unsigned_LEB_128(lineptr, &len); // filelength | |
298 | lineptr += len; | |
299 | ||
af674d1d ILT |
300 | gold_assert(fileindex |
301 | == static_cast<int>(this->files_.back().size())); | |
302 | this->files_.back().push_back(std::make_pair(dirindexi, filename)); | |
5c2c6c95 ILT |
303 | fileindex++; |
304 | } | |
305 | } | |
306 | lineptr++; | |
307 | ||
308 | return lineptr; | |
309 | } | |
310 | ||
311 | // Process a single opcode in the .debug.line structure. | |
312 | ||
313 | // Templating on size and big_endian would yield more efficient (and | |
314 | // simpler) code, but would bloat the binary. Speed isn't important | |
315 | // here. | |
316 | ||
e43872e9 | 317 | template<int size, bool big_endian> |
5c2c6c95 | 318 | bool |
a55ce7fe | 319 | Sized_dwarf_line_info<size, big_endian>::process_one_opcode( |
e43872e9 | 320 | const unsigned char* start, struct LineStateMachine* lsm, size_t* len) |
5c2c6c95 ILT |
321 | { |
322 | size_t oplen = 0; | |
323 | size_t templen; | |
324 | unsigned char opcode = *start; | |
325 | oplen++; | |
326 | start++; | |
327 | ||
328 | // If the opcode is great than the opcode_base, it is a special | |
329 | // opcode. Most line programs consist mainly of special opcodes. | |
330 | if (opcode >= header_.opcode_base) | |
331 | { | |
332 | opcode -= header_.opcode_base; | |
333 | const int advance_address = ((opcode / header_.line_range) | |
334 | * header_.min_insn_length); | |
335 | lsm->address += advance_address; | |
336 | ||
337 | const int advance_line = ((opcode % header_.line_range) | |
338 | + header_.line_base); | |
339 | lsm->line_num += advance_line; | |
340 | lsm->basic_block = true; | |
341 | *len = oplen; | |
342 | return true; | |
343 | } | |
344 | ||
345 | // Otherwise, we have the regular opcodes | |
346 | switch (opcode) | |
347 | { | |
348 | case elfcpp::DW_LNS_copy: | |
349 | lsm->basic_block = false; | |
350 | *len = oplen; | |
351 | return true; | |
352 | ||
353 | case elfcpp::DW_LNS_advance_pc: | |
354 | { | |
355 | const uint64_t advance_address | |
356 | = read_unsigned_LEB_128(start, &templen); | |
357 | oplen += templen; | |
358 | lsm->address += header_.min_insn_length * advance_address; | |
359 | } | |
360 | break; | |
361 | ||
362 | case elfcpp::DW_LNS_advance_line: | |
363 | { | |
364 | const uint64_t advance_line = read_signed_LEB_128(start, &templen); | |
365 | oplen += templen; | |
366 | lsm->line_num += advance_line; | |
367 | } | |
368 | break; | |
369 | ||
370 | case elfcpp::DW_LNS_set_file: | |
371 | { | |
372 | const uint64_t fileno = read_unsigned_LEB_128(start, &templen); | |
373 | oplen += templen; | |
374 | lsm->file_num = fileno; | |
375 | } | |
376 | break; | |
377 | ||
378 | case elfcpp::DW_LNS_set_column: | |
379 | { | |
380 | const uint64_t colno = read_unsigned_LEB_128(start, &templen); | |
381 | oplen += templen; | |
382 | lsm->column_num = colno; | |
383 | } | |
384 | break; | |
385 | ||
386 | case elfcpp::DW_LNS_negate_stmt: | |
387 | lsm->is_stmt = !lsm->is_stmt; | |
388 | break; | |
389 | ||
390 | case elfcpp::DW_LNS_set_basic_block: | |
391 | lsm->basic_block = true; | |
392 | break; | |
393 | ||
394 | case elfcpp::DW_LNS_fixed_advance_pc: | |
395 | { | |
396 | int advance_address; | |
deae2a14 | 397 | advance_address = elfcpp::Swap_unaligned<16, big_endian>::readval(start); |
5c2c6c95 ILT |
398 | oplen += 2; |
399 | lsm->address += advance_address; | |
400 | } | |
401 | break; | |
402 | ||
403 | case elfcpp::DW_LNS_const_add_pc: | |
404 | { | |
405 | const int advance_address = (header_.min_insn_length | |
406 | * ((255 - header_.opcode_base) | |
407 | / header_.line_range)); | |
408 | lsm->address += advance_address; | |
409 | } | |
410 | break; | |
411 | ||
412 | case elfcpp::DW_LNS_extended_op: | |
413 | { | |
414 | const uint64_t extended_op_len | |
415 | = read_unsigned_LEB_128(start, &templen); | |
416 | start += templen; | |
417 | oplen += templen + extended_op_len; | |
418 | ||
419 | const unsigned char extended_op = *start; | |
420 | start++; | |
421 | ||
422 | switch (extended_op) | |
423 | { | |
424 | case elfcpp::DW_LNE_end_sequence: | |
124dfc89 ILT |
425 | // This means that the current byte is the one immediately |
426 | // after a set of instructions. Record the current line | |
427 | // for up to one less than the current address. | |
79e052ea | 428 | lsm->line_num = -1; |
5c2c6c95 ILT |
429 | lsm->end_sequence = true; |
430 | *len = oplen; | |
431 | return true; | |
432 | ||
433 | case elfcpp::DW_LNE_set_address: | |
4c50553d | 434 | { |
deae2a14 | 435 | lsm->address = elfcpp::Swap_unaligned<size, big_endian>::readval(start); |
4c50553d ILT |
436 | typename Reloc_map::const_iterator it |
437 | = reloc_map_.find(start - this->buffer_); | |
438 | if (it != reloc_map_.end()) | |
439 | { | |
440 | // value + addend. | |
af674d1d | 441 | lsm->address += it->second.second; |
4c50553d ILT |
442 | lsm->shndx = it->second.first; |
443 | } | |
444 | else | |
445 | { | |
af674d1d ILT |
446 | // If we're a normal .o file, with relocs, every |
447 | // set_address should have an associated relocation. | |
448 | if (this->input_is_relobj()) | |
449 | this->data_valid_ = false; | |
4c50553d ILT |
450 | } |
451 | break; | |
24badc65 | 452 | } |
5c2c6c95 ILT |
453 | case elfcpp::DW_LNE_define_file: |
454 | { | |
455 | const char* filename = reinterpret_cast<const char*>(start); | |
456 | templen = strlen(filename) + 1; | |
457 | start += templen; | |
458 | ||
459 | uint64_t dirindex = read_unsigned_LEB_128(start, &templen); | |
5c2c6c95 ILT |
460 | oplen += templen; |
461 | ||
af674d1d ILT |
462 | if (dirindex >= this->directories_.back().size()) |
463 | dirindex = 0; | |
464 | int dirindexi = static_cast<int>(dirindex); | |
465 | ||
5c2c6c95 ILT |
466 | read_unsigned_LEB_128(start, &templen); // mod_time |
467 | oplen += templen; | |
468 | ||
469 | read_unsigned_LEB_128(start, &templen); // filelength | |
470 | oplen += templen; | |
471 | ||
af674d1d | 472 | this->files_.back().push_back(std::make_pair(dirindexi, |
5c2c6c95 ILT |
473 | filename)); |
474 | } | |
475 | break; | |
476 | } | |
477 | } | |
478 | break; | |
479 | ||
480 | default: | |
481 | { | |
482 | // Ignore unknown opcode silently | |
483 | for (int i = 0; i < header_.std_opcode_lengths[opcode]; i++) | |
484 | { | |
485 | size_t templen; | |
486 | read_unsigned_LEB_128(start, &templen); | |
487 | start += templen; | |
488 | oplen += templen; | |
489 | } | |
490 | } | |
491 | break; | |
492 | } | |
493 | *len = oplen; | |
494 | return false; | |
495 | } | |
496 | ||
497 | // Read the debug information at LINEPTR and store it in the line | |
498 | // number map. | |
499 | ||
e43872e9 | 500 | template<int size, bool big_endian> |
5c2c6c95 | 501 | unsigned const char* |
9430daf8 ILT |
502 | Sized_dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr, |
503 | off_t shndx) | |
5c2c6c95 ILT |
504 | { |
505 | struct LineStateMachine lsm; | |
506 | ||
507 | // LENGTHSTART is the place the length field is based on. It is the | |
508 | // point in the header after the initial length field. | |
509 | const unsigned char* lengthstart = buffer_; | |
510 | ||
511 | // In 64 bit dwarf, the initial length is 12 bytes, because of the | |
512 | // 0xffffffff at the start. | |
513 | if (header_.offset_size == 8) | |
514 | lengthstart += 12; | |
515 | else | |
516 | lengthstart += 4; | |
517 | ||
518 | while (lineptr < lengthstart + header_.total_length) | |
519 | { | |
520 | ResetLineStateMachine(&lsm, header_.default_is_stmt); | |
521 | while (!lsm.end_sequence) | |
522 | { | |
523 | size_t oplength; | |
e43872e9 | 524 | bool add_line = this->process_one_opcode(lineptr, &lsm, &oplength); |
9430daf8 ILT |
525 | if (add_line |
526 | && (shndx == -1U || lsm.shndx == -1U || shndx == lsm.shndx)) | |
5c2c6c95 ILT |
527 | { |
528 | Offset_to_lineno_entry entry | |
af674d1d ILT |
529 | = { lsm.address, this->current_header_index_, |
530 | lsm.file_num, lsm.line_num }; | |
5c2c6c95 ILT |
531 | line_number_map_[lsm.shndx].push_back(entry); |
532 | } | |
533 | lineptr += oplength; | |
534 | } | |
535 | } | |
536 | ||
537 | return lengthstart + header_.total_length; | |
538 | } | |
539 | ||
4c50553d ILT |
540 | // Looks in the symtab to see what section a symbol is in. |
541 | ||
542 | template<int size, bool big_endian> | |
543 | unsigned int | |
a55ce7fe | 544 | Sized_dwarf_line_info<size, big_endian>::symbol_section( |
d491d34e | 545 | Object* object, |
4c50553d | 546 | unsigned int sym, |
d491d34e ILT |
547 | typename elfcpp::Elf_types<size>::Elf_Addr* value, |
548 | bool* is_ordinary) | |
4c50553d ILT |
549 | { |
550 | const int symsize = elfcpp::Elf_sizes<size>::sym_size; | |
af674d1d | 551 | gold_assert(sym * symsize < this->symtab_buffer_size_); |
4c50553d ILT |
552 | elfcpp::Sym<size, big_endian> elfsym(this->symtab_buffer_ + sym * symsize); |
553 | *value = elfsym.get_st_value(); | |
d491d34e | 554 | return object->adjust_sym_shndx(sym, elfsym.get_st_shndx(), is_ordinary); |
4c50553d ILT |
555 | } |
556 | ||
557 | // Read the relocations into a Reloc_map. | |
558 | ||
559 | template<int size, bool big_endian> | |
560 | void | |
d491d34e | 561 | Sized_dwarf_line_info<size, big_endian>::read_relocs(Object* object) |
4c50553d ILT |
562 | { |
563 | if (this->symtab_buffer_ == NULL) | |
564 | return; | |
565 | ||
566 | typename elfcpp::Elf_types<size>::Elf_Addr value; | |
567 | off_t reloc_offset; | |
24badc65 | 568 | while ((reloc_offset = this->track_relocs_.next_offset()) != -1) |
4c50553d | 569 | { |
24badc65 | 570 | const unsigned int sym = this->track_relocs_.next_symndx(); |
d491d34e ILT |
571 | |
572 | bool is_ordinary; | |
573 | const unsigned int shndx = this->symbol_section(object, sym, &value, | |
574 | &is_ordinary); | |
575 | ||
576 | // There is no reason to record non-ordinary section indexes, or | |
577 | // SHN_UNDEF, because they will never match the real section. | |
578 | if (is_ordinary && shndx != elfcpp::SHN_UNDEF) | |
579 | this->reloc_map_[reloc_offset] = std::make_pair(shndx, value); | |
580 | ||
24badc65 | 581 | this->track_relocs_.advance(reloc_offset + 1); |
4c50553d ILT |
582 | } |
583 | } | |
584 | ||
585 | // Read the line number info. | |
586 | ||
e43872e9 | 587 | template<int size, bool big_endian> |
5c2c6c95 | 588 | void |
d491d34e ILT |
589 | Sized_dwarf_line_info<size, big_endian>::read_line_mappings(Object* object, |
590 | off_t shndx) | |
5c2c6c95 | 591 | { |
c261a0be | 592 | gold_assert(this->data_valid_ == true); |
24badc65 | 593 | |
d491d34e | 594 | this->read_relocs(object); |
4c50553d | 595 | while (this->buffer_ < this->buffer_end_) |
e43872e9 | 596 | { |
4c50553d | 597 | const unsigned char* lineptr = this->buffer_; |
e43872e9 ILT |
598 | lineptr = this->read_header_prolog(lineptr); |
599 | lineptr = this->read_header_tables(lineptr); | |
9430daf8 | 600 | lineptr = this->read_lines(lineptr, shndx); |
4c50553d | 601 | this->buffer_ = lineptr; |
e43872e9 ILT |
602 | } |
603 | ||
604 | // Sort the lines numbers, so addr2line can use binary search. | |
605 | for (typename Lineno_map::iterator it = line_number_map_.begin(); | |
5c2c6c95 ILT |
606 | it != line_number_map_.end(); |
607 | ++it) | |
608 | // Each vector needs to be sorted by offset. | |
4c50553d | 609 | std::sort(it->second.begin(), it->second.end()); |
5c2c6c95 ILT |
610 | } |
611 | ||
af674d1d ILT |
612 | // Some processing depends on whether the input is a .o file or not. |
613 | // For instance, .o files have relocs, and have .debug_lines | |
614 | // information on a per section basis. .so files, on the other hand, | |
615 | // lack relocs, and offsets are unique, so we can ignore the section | |
616 | // information. | |
617 | ||
618 | template<int size, bool big_endian> | |
619 | bool | |
a55ce7fe | 620 | Sized_dwarf_line_info<size, big_endian>::input_is_relobj() |
af674d1d ILT |
621 | { |
622 | // Only .o files have relocs and the symtab buffer that goes with them. | |
623 | return this->symtab_buffer_ != NULL; | |
624 | } | |
625 | ||
79e052ea ILT |
626 | // Given an Offset_to_lineno_entry vector, and an offset, figure out |
627 | // if the offset points into a function according to the vector (see | |
628 | // comments below for the algorithm). If it does, return an iterator | |
629 | // into the vector that points to the line-number that contains that | |
630 | // offset. If not, it returns vector::end(). | |
631 | ||
632 | static std::vector<Offset_to_lineno_entry>::const_iterator | |
633 | offset_to_iterator(const std::vector<Offset_to_lineno_entry>* offsets, | |
634 | off_t offset) | |
635 | { | |
636 | const Offset_to_lineno_entry lookup_key = { offset, 0, 0, 0 }; | |
637 | ||
638 | // lower_bound() returns the smallest offset which is >= lookup_key. | |
639 | // If no offset in offsets is >= lookup_key, returns end(). | |
640 | std::vector<Offset_to_lineno_entry>::const_iterator it | |
641 | = std::lower_bound(offsets->begin(), offsets->end(), lookup_key); | |
642 | ||
643 | // This code is easiest to understand with a concrete example. | |
644 | // Here's a possible offsets array: | |
645 | // {{offset = 3211, header_num = 0, file_num = 1, line_num = 16}, // 0 | |
646 | // {offset = 3224, header_num = 0, file_num = 1, line_num = 20}, // 1 | |
647 | // {offset = 3226, header_num = 0, file_num = 1, line_num = 22}, // 2 | |
648 | // {offset = 3231, header_num = 0, file_num = 1, line_num = 25}, // 3 | |
649 | // {offset = 3232, header_num = 0, file_num = 1, line_num = -1}, // 4 | |
650 | // {offset = 3232, header_num = 0, file_num = 1, line_num = 65}, // 5 | |
651 | // {offset = 3235, header_num = 0, file_num = 1, line_num = 66}, // 6 | |
652 | // {offset = 3236, header_num = 0, file_num = 1, line_num = -1}, // 7 | |
653 | // {offset = 5764, header_num = 0, file_num = 1, line_num = 47}, // 8 | |
654 | // {offset = 5765, header_num = 0, file_num = 1, line_num = 48}, // 9 | |
655 | // {offset = 5767, header_num = 0, file_num = 1, line_num = 49}, // 10 | |
656 | // {offset = 5768, header_num = 0, file_num = 1, line_num = 50}, // 11 | |
657 | // {offset = 5773, header_num = 0, file_num = 1, line_num = -1}, // 12 | |
658 | // {offset = 5787, header_num = 1, file_num = 1, line_num = 19}, // 13 | |
659 | // {offset = 5790, header_num = 1, file_num = 1, line_num = 20}, // 14 | |
660 | // {offset = 5793, header_num = 1, file_num = 1, line_num = 67}, // 15 | |
661 | // {offset = 5793, header_num = 1, file_num = 1, line_num = -1}, // 16 | |
662 | // {offset = 5795, header_num = 1, file_num = 1, line_num = 68}, // 17 | |
ef04e392 | 663 | // {offset = 5798, header_num = 1, file_num = 1, line_num = -1}, // 18 |
79e052ea ILT |
664 | // The entries with line_num == -1 mark the end of a function: the |
665 | // associated offset is one past the last instruction in the | |
666 | // function. This can correspond to the beginning of the next | |
667 | // function (as is true for offset 3232); alternately, there can be | |
668 | // a gap between the end of one function and the start of the next | |
ef04e392 | 669 | // (as is true for some others, most obviously from 3236->5764). |
79e052ea ILT |
670 | // |
671 | // Case 1: lookup_key has offset == 10. lower_bound returns | |
672 | // offsets[0]. Since it's not an exact match and we're | |
ef04e392 | 673 | // at the beginning of offsets, we return end() (invalid). |
79e052ea | 674 | // Case 2: lookup_key has offset 10000. lower_bound returns |
ef04e392 | 675 | // offset[19] (end()). We return end() (invalid). |
79e052ea ILT |
676 | // Case 3: lookup_key has offset == 3211. lower_bound matches |
677 | // offsets[0] exactly, and that's the entry we return. | |
678 | // Case 4: lookup_key has offset == 3232. lower_bound returns | |
679 | // offsets[4]. That's an exact match, but indicates | |
680 | // end-of-function. We check if offsets[5] is also an | |
681 | // exact match but not end-of-function. It is, so we | |
682 | // return offsets[5]. | |
683 | // Case 5: lookup_key has offset == 3214. lower_bound returns | |
684 | // offsets[1]. Since it's not an exact match, we back | |
685 | // up to the offset that's < lookup_key, offsets[0]. | |
686 | // We note offsets[0] is a valid entry (not end-of-function), | |
687 | // so that's the entry we return. | |
688 | // Case 6: lookup_key has offset == 4000. lower_bound returns | |
689 | // offsets[8]. Since it's not an exact match, we back | |
690 | // up to offsets[7]. Since offsets[7] indicates | |
691 | // end-of-function, we know lookup_key is between | |
ef04e392 | 692 | // functions, so we return end() (not a valid offset). |
79e052ea ILT |
693 | // Case 7: lookup_key has offset == 5794. lower_bound returns |
694 | // offsets[17]. Since it's not an exact match, we back | |
695 | // up to offsets[15]. Note we back up to the *first* | |
696 | // entry with offset 5793, not just offsets[17-1]. | |
697 | // We note offsets[15] is a valid entry, so we return it. | |
698 | // If offsets[15] had had line_num == -1, we would have | |
699 | // checked offsets[16]. The reason for this is that | |
700 | // 15 and 16 can be in an arbitrary order, since we sort | |
701 | // only by offset. (Note it doesn't help to use line_number | |
702 | // as a secondary sort key, since sometimes we want the -1 | |
703 | // to be first and sometimes we want it to be last.) | |
704 | ||
705 | // This deals with cases (1) and (2). | |
706 | if ((it == offsets->begin() && offset < it->offset) | |
707 | || it == offsets->end()) | |
708 | return offsets->end(); | |
709 | ||
710 | // This deals with cases (3) and (4). | |
711 | if (offset == it->offset) | |
712 | { | |
713 | while (it != offsets->end() | |
714 | && it->offset == offset | |
715 | && it->line_num == -1) | |
716 | ++it; | |
717 | if (it == offsets->end() || it->offset != offset) | |
718 | return offsets->end(); | |
719 | else | |
720 | return it; | |
721 | } | |
722 | ||
723 | // This handles the first part of case (7) -- we back up to the | |
724 | // *first* entry that has the offset that's behind us. | |
725 | gold_assert(it != offsets->begin()); | |
726 | std::vector<Offset_to_lineno_entry>::const_iterator range_end = it; | |
727 | --it; | |
728 | const off_t range_value = it->offset; | |
729 | while (it != offsets->begin() && (it-1)->offset == range_value) | |
730 | --it; | |
731 | ||
732 | // This handles cases (5), (6), and (7): if any entry in the | |
733 | // equal_range [it, range_end) has a line_num != -1, it's a valid | |
734 | // match. If not, we're not in a function. | |
735 | for (; it != range_end; ++it) | |
736 | if (it->line_num != -1) | |
737 | return it; | |
738 | return offsets->end(); | |
739 | } | |
af674d1d | 740 | |
5c2c6c95 ILT |
741 | // Return a string for a file name and line number. |
742 | ||
e43872e9 | 743 | template<int size, bool big_endian> |
5c2c6c95 | 744 | std::string |
a55ce7fe ILT |
745 | Sized_dwarf_line_info<size, big_endian>::do_addr2line(unsigned int shndx, |
746 | off_t offset) | |
5c2c6c95 | 747 | { |
4c50553d ILT |
748 | if (this->data_valid_ == false) |
749 | return ""; | |
750 | ||
af674d1d ILT |
751 | const std::vector<Offset_to_lineno_entry>* offsets; |
752 | // If we do not have reloc information, then our input is a .so or | |
753 | // some similar data structure where all the information is held in | |
754 | // the offset. In that case, we ignore the input shndx. | |
755 | if (this->input_is_relobj()) | |
756 | offsets = &this->line_number_map_[shndx]; | |
757 | else | |
758 | offsets = &this->line_number_map_[-1U]; | |
759 | if (offsets->empty()) | |
4c50553d ILT |
760 | return ""; |
761 | ||
e43872e9 | 762 | typename std::vector<Offset_to_lineno_entry>::const_iterator it |
79e052ea ILT |
763 | = offset_to_iterator(offsets, offset); |
764 | if (it == offsets->end()) | |
765 | return ""; | |
5c2c6c95 ILT |
766 | |
767 | // Convert the file_num + line_num into a string. | |
768 | std::string ret; | |
af674d1d ILT |
769 | |
770 | gold_assert(it->header_num < static_cast<int>(this->files_.size())); | |
771 | gold_assert(it->file_num | |
772 | < static_cast<int>(this->files_[it->header_num].size())); | |
773 | const std::pair<int, std::string>& filename_pair | |
774 | = this->files_[it->header_num][it->file_num]; | |
5c2c6c95 | 775 | const std::string& filename = filename_pair.second; |
af674d1d ILT |
776 | |
777 | gold_assert(it->header_num < static_cast<int>(this->directories_.size())); | |
778 | gold_assert(filename_pair.first | |
779 | < static_cast<int>(this->directories_[it->header_num].size())); | |
780 | const std::string& dirname | |
781 | = this->directories_[it->header_num][filename_pair.first]; | |
782 | ||
5c2c6c95 ILT |
783 | if (!dirname.empty()) |
784 | { | |
785 | ret += dirname; | |
786 | ret += "/"; | |
787 | } | |
788 | ret += filename; | |
789 | if (ret.empty()) | |
790 | ret = "(unknown)"; | |
791 | ||
792 | char buffer[64]; // enough to hold a line number | |
793 | snprintf(buffer, sizeof(buffer), "%d", it->line_num); | |
794 | ret += ":"; | |
795 | ret += buffer; | |
796 | ||
797 | return ret; | |
798 | } | |
799 | ||
a55ce7fe ILT |
800 | // Dwarf_line_info routines. |
801 | ||
a55ce7fe ILT |
802 | std::string |
803 | Dwarf_line_info::one_addr2line(Object* object, | |
804 | unsigned int shndx, off_t offset) | |
805 | { | |
8851ecca ILT |
806 | switch (parameters->size_and_endianness()) |
807 | { | |
a55ce7fe | 808 | #ifdef HAVE_TARGET_32_LITTLE |
8851ecca ILT |
809 | case Parameters::TARGET_32_LITTLE: |
810 | return Sized_dwarf_line_info<32, false>(object, shndx).addr2line(shndx, | |
811 | offset); | |
a55ce7fe | 812 | #endif |
a55ce7fe | 813 | #ifdef HAVE_TARGET_32_BIG |
8851ecca ILT |
814 | case Parameters::TARGET_32_BIG: |
815 | return Sized_dwarf_line_info<32, true>(object, shndx).addr2line(shndx, | |
816 | offset); | |
a55ce7fe | 817 | #endif |
a55ce7fe | 818 | #ifdef HAVE_TARGET_64_LITTLE |
8851ecca ILT |
819 | case Parameters::TARGET_64_LITTLE: |
820 | return Sized_dwarf_line_info<64, false>(object, shndx).addr2line(shndx, | |
821 | offset); | |
a55ce7fe | 822 | #endif |
8851ecca ILT |
823 | #ifdef HAVE_TARGET_64_BIG |
824 | case Parameters::TARGET_64_BIG: | |
825 | return Sized_dwarf_line_info<64, true>(object, shndx).addr2line(shndx, | |
826 | offset); | |
a55ce7fe | 827 | #endif |
8851ecca ILT |
828 | default: |
829 | gold_unreachable(); | |
830 | } | |
a55ce7fe ILT |
831 | } |
832 | ||
5c2c6c95 ILT |
833 | #ifdef HAVE_TARGET_32_LITTLE |
834 | template | |
a55ce7fe | 835 | class Sized_dwarf_line_info<32, false>; |
5c2c6c95 ILT |
836 | #endif |
837 | ||
838 | #ifdef HAVE_TARGET_32_BIG | |
839 | template | |
a55ce7fe | 840 | class Sized_dwarf_line_info<32, true>; |
5c2c6c95 ILT |
841 | #endif |
842 | ||
843 | #ifdef HAVE_TARGET_64_LITTLE | |
844 | template | |
a55ce7fe | 845 | class Sized_dwarf_line_info<64, false>; |
5c2c6c95 ILT |
846 | #endif |
847 | ||
848 | #ifdef HAVE_TARGET_64_BIG | |
849 | template | |
a55ce7fe | 850 | class Sized_dwarf_line_info<64, true>; |
5c2c6c95 ILT |
851 | #endif |
852 | ||
853 | } // End namespace gold. |