Commit | Line | Data |
---|---|---|
5c2c6c95 ILT |
1 | // dwarf_reader.cc -- parse dwarf2/3 debug information |
2 | ||
3 | // Copyright 2007 Free Software Foundation, Inc. | |
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 | ||
25 | #include "elfcpp_swap.h" | |
26 | #include "dwarf.h" | |
27 | #include "dwarf_reader.h" | |
28 | ||
29 | namespace { | |
30 | ||
31 | // Read an unsigned LEB128 number. Each byte contains 7 bits of | |
32 | // information, plus one bit saying whether the number continues or | |
33 | // not. | |
34 | ||
35 | uint64_t | |
36 | read_unsigned_LEB_128(const unsigned char* buffer, size_t* len) | |
37 | { | |
38 | uint64_t result = 0; | |
39 | size_t num_read = 0; | |
40 | unsigned int shift = 0; | |
41 | unsigned char byte; | |
42 | ||
43 | do | |
44 | { | |
45 | byte = *buffer++; | |
46 | num_read++; | |
47 | result |= (static_cast<uint64_t>(byte & 0x7f)) << shift; | |
48 | shift += 7; | |
49 | } | |
50 | while (byte & 0x80); | |
51 | ||
52 | *len = num_read; | |
53 | ||
54 | return result; | |
55 | } | |
56 | ||
57 | // Read a signed LEB128 number. These are like regular LEB128 | |
58 | // numbers, except the last byte may have a sign bit set. | |
59 | ||
60 | int64_t | |
61 | read_signed_LEB_128(const unsigned char* buffer, size_t* len) | |
62 | { | |
63 | int64_t result = 0; | |
64 | int shift = 0; | |
65 | size_t num_read = 0; | |
66 | unsigned char byte; | |
67 | ||
68 | do | |
69 | { | |
70 | byte = *buffer++; | |
71 | num_read++; | |
72 | result |= (static_cast<uint64_t>(byte & 0x7f) << shift); | |
73 | shift += 7; | |
74 | } | |
75 | while (byte & 0x80); | |
76 | ||
77 | if ((shift < 8 * static_cast<int>(sizeof(result))) && (byte & 0x40)) | |
78 | result |= -((static_cast<int64_t>(1)) << shift); | |
79 | *len = num_read; | |
80 | return result; | |
81 | } | |
82 | ||
83 | } // End anonymous namespace. | |
84 | ||
85 | ||
86 | namespace gold { | |
87 | ||
88 | // This is the format of a DWARF2/3 line state machine that we process | |
89 | // opcodes using. There is no need for anything outside the lineinfo | |
90 | // processor to know how this works. | |
91 | ||
92 | struct LineStateMachine | |
93 | { | |
94 | int file_num; | |
95 | uint64_t address; | |
96 | int line_num; | |
97 | int column_num; | |
98 | unsigned int shndx; // the section address refers to | |
99 | bool is_stmt; // stmt means statement. | |
100 | bool basic_block; | |
101 | bool end_sequence; | |
102 | }; | |
103 | ||
104 | static void | |
105 | ResetLineStateMachine(struct LineStateMachine* lsm, bool default_is_stmt) | |
106 | { | |
107 | lsm->file_num = 1; | |
108 | lsm->address = 0; | |
109 | lsm->line_num = 1; | |
110 | lsm->column_num = 0; | |
111 | lsm->shndx = -1; | |
112 | lsm->is_stmt = default_is_stmt; | |
113 | lsm->basic_block = false; | |
114 | lsm->end_sequence = false; | |
115 | } | |
116 | ||
117 | // Read the DWARF header. | |
118 | ||
119 | template<int size, bool big_endian> | |
120 | const unsigned char* | |
121 | Dwarf_line_info::read_header_prolog(const unsigned char* lineptr) | |
122 | { | |
123 | uint32_t initial_length = elfcpp::Swap<32, big_endian>::readval(lineptr); | |
124 | lineptr += 4; | |
125 | ||
126 | // In DWARF2/3, if the initial length is all 1 bits, then the offset | |
127 | // size is 8 and we need to read the next 8 bytes for the real length. | |
128 | if (initial_length == 0xffffffff) | |
129 | { | |
130 | header_.offset_size = 8; | |
131 | initial_length = elfcpp::Swap<64, big_endian>::readval(lineptr); | |
132 | lineptr += 8; | |
133 | } | |
134 | else | |
135 | header_.offset_size = 4; | |
136 | ||
137 | header_.total_length = initial_length; | |
138 | ||
139 | gold_assert(lineptr + header_.total_length <= buffer_end_); | |
140 | ||
141 | header_.version = elfcpp::Swap<16, big_endian>::readval(lineptr); | |
142 | lineptr += 2; | |
143 | ||
144 | if (header_.offset_size == 4) | |
145 | header_.prologue_length = elfcpp::Swap<32, big_endian>::readval(lineptr); | |
146 | else | |
147 | header_.prologue_length = elfcpp::Swap<64, big_endian>::readval(lineptr); | |
148 | lineptr += header_.offset_size; | |
149 | ||
150 | header_.min_insn_length = *lineptr; | |
151 | lineptr += 1; | |
152 | ||
153 | header_.default_is_stmt = *lineptr; | |
154 | lineptr += 1; | |
155 | ||
156 | header_.line_base = *reinterpret_cast<const signed char*>(lineptr); | |
157 | lineptr += 1; | |
158 | ||
159 | header_.line_range = *lineptr; | |
160 | lineptr += 1; | |
161 | ||
162 | header_.opcode_base = *lineptr; | |
163 | lineptr += 1; | |
164 | ||
165 | header_.std_opcode_lengths.reserve(header_.opcode_base + 1); | |
166 | header_.std_opcode_lengths[0] = 0; | |
167 | for (int i = 1; i < header_.opcode_base; i++) | |
168 | { | |
169 | header_.std_opcode_lengths[i] = *lineptr; | |
170 | lineptr += 1; | |
171 | } | |
172 | ||
173 | return lineptr; | |
174 | } | |
175 | ||
176 | // The header for a debug_line section is mildly complicated, because | |
177 | // the line info is very tightly encoded. | |
178 | ||
179 | const unsigned char* | |
180 | Dwarf_line_info::read_header_tables(const unsigned char* lineptr) | |
181 | { | |
182 | // It is legal for the directory entry table to be empty. | |
183 | if (*lineptr) | |
184 | { | |
185 | int dirindex = 1; | |
186 | while (*lineptr) | |
187 | { | |
188 | const unsigned char* dirname = lineptr; | |
189 | gold_assert(dirindex == static_cast<int>(directories_.size())); | |
190 | directories_.push_back(reinterpret_cast<const char*>(dirname)); | |
191 | lineptr += directories_.back().size() + 1; | |
192 | dirindex++; | |
193 | } | |
194 | } | |
195 | lineptr++; | |
196 | ||
197 | // It is also legal for the file entry table to be empty. | |
198 | if (*lineptr) | |
199 | { | |
200 | int fileindex = 1; | |
201 | size_t len; | |
202 | while (*lineptr) | |
203 | { | |
204 | const char* filename = reinterpret_cast<const char*>(lineptr); | |
205 | lineptr += strlen(filename) + 1; | |
206 | ||
207 | uint64_t dirindex = read_unsigned_LEB_128(lineptr, &len); | |
208 | if (dirindex >= directories_.size()) | |
209 | dirindex = 0; | |
210 | lineptr += len; | |
211 | ||
212 | read_unsigned_LEB_128(lineptr, &len); // mod_time | |
213 | lineptr += len; | |
214 | ||
215 | read_unsigned_LEB_128(lineptr, &len); // filelength | |
216 | lineptr += len; | |
217 | ||
218 | gold_assert(fileindex == static_cast<int>(files_.size())); | |
219 | files_.push_back(std::pair<int, std::string>(dirindex, filename)); | |
220 | fileindex++; | |
221 | } | |
222 | } | |
223 | lineptr++; | |
224 | ||
225 | return lineptr; | |
226 | } | |
227 | ||
228 | // Process a single opcode in the .debug.line structure. | |
229 | ||
230 | // Templating on size and big_endian would yield more efficient (and | |
231 | // simpler) code, but would bloat the binary. Speed isn't important | |
232 | // here. | |
233 | ||
234 | bool | |
235 | Dwarf_line_info::process_one_opcode(int size, bool big_endian, | |
236 | const unsigned char* start, | |
237 | struct LineStateMachine* lsm, | |
238 | size_t* len) | |
239 | { | |
240 | size_t oplen = 0; | |
241 | size_t templen; | |
242 | unsigned char opcode = *start; | |
243 | oplen++; | |
244 | start++; | |
245 | ||
246 | // If the opcode is great than the opcode_base, it is a special | |
247 | // opcode. Most line programs consist mainly of special opcodes. | |
248 | if (opcode >= header_.opcode_base) | |
249 | { | |
250 | opcode -= header_.opcode_base; | |
251 | const int advance_address = ((opcode / header_.line_range) | |
252 | * header_.min_insn_length); | |
253 | lsm->address += advance_address; | |
254 | ||
255 | const int advance_line = ((opcode % header_.line_range) | |
256 | + header_.line_base); | |
257 | lsm->line_num += advance_line; | |
258 | lsm->basic_block = true; | |
259 | *len = oplen; | |
260 | return true; | |
261 | } | |
262 | ||
263 | // Otherwise, we have the regular opcodes | |
264 | switch (opcode) | |
265 | { | |
266 | case elfcpp::DW_LNS_copy: | |
267 | lsm->basic_block = false; | |
268 | *len = oplen; | |
269 | return true; | |
270 | ||
271 | case elfcpp::DW_LNS_advance_pc: | |
272 | { | |
273 | const uint64_t advance_address | |
274 | = read_unsigned_LEB_128(start, &templen); | |
275 | oplen += templen; | |
276 | lsm->address += header_.min_insn_length * advance_address; | |
277 | } | |
278 | break; | |
279 | ||
280 | case elfcpp::DW_LNS_advance_line: | |
281 | { | |
282 | const uint64_t advance_line = read_signed_LEB_128(start, &templen); | |
283 | oplen += templen; | |
284 | lsm->line_num += advance_line; | |
285 | } | |
286 | break; | |
287 | ||
288 | case elfcpp::DW_LNS_set_file: | |
289 | { | |
290 | const uint64_t fileno = read_unsigned_LEB_128(start, &templen); | |
291 | oplen += templen; | |
292 | lsm->file_num = fileno; | |
293 | } | |
294 | break; | |
295 | ||
296 | case elfcpp::DW_LNS_set_column: | |
297 | { | |
298 | const uint64_t colno = read_unsigned_LEB_128(start, &templen); | |
299 | oplen += templen; | |
300 | lsm->column_num = colno; | |
301 | } | |
302 | break; | |
303 | ||
304 | case elfcpp::DW_LNS_negate_stmt: | |
305 | lsm->is_stmt = !lsm->is_stmt; | |
306 | break; | |
307 | ||
308 | case elfcpp::DW_LNS_set_basic_block: | |
309 | lsm->basic_block = true; | |
310 | break; | |
311 | ||
312 | case elfcpp::DW_LNS_fixed_advance_pc: | |
313 | { | |
314 | int advance_address; | |
315 | if (big_endian) | |
316 | advance_address = elfcpp::Swap<16, true>::readval(start); | |
317 | else | |
318 | advance_address = elfcpp::Swap<16, false>::readval(start); | |
319 | oplen += 2; | |
320 | lsm->address += advance_address; | |
321 | } | |
322 | break; | |
323 | ||
324 | case elfcpp::DW_LNS_const_add_pc: | |
325 | { | |
326 | const int advance_address = (header_.min_insn_length | |
327 | * ((255 - header_.opcode_base) | |
328 | / header_.line_range)); | |
329 | lsm->address += advance_address; | |
330 | } | |
331 | break; | |
332 | ||
333 | case elfcpp::DW_LNS_extended_op: | |
334 | { | |
335 | const uint64_t extended_op_len | |
336 | = read_unsigned_LEB_128(start, &templen); | |
337 | start += templen; | |
338 | oplen += templen + extended_op_len; | |
339 | ||
340 | const unsigned char extended_op = *start; | |
341 | start++; | |
342 | ||
343 | switch (extended_op) | |
344 | { | |
345 | case elfcpp::DW_LNE_end_sequence: | |
346 | lsm->end_sequence = true; | |
347 | *len = oplen; | |
348 | return true; | |
349 | ||
350 | case elfcpp::DW_LNE_set_address: | |
351 | // FIXME: modify the address based on the reloc | |
352 | if (size == 32 && big_endian == false) | |
353 | lsm->address = elfcpp::Swap<32, false>::readval(start); | |
354 | else if (size == 32 && big_endian == true) | |
355 | lsm->address = elfcpp::Swap<32, true>::readval(start); | |
356 | else if (size == 64 && big_endian == false) | |
357 | lsm->address = elfcpp::Swap<64, false>::readval(start); | |
358 | else if (size == 64 && big_endian == true) | |
359 | lsm->address = elfcpp::Swap<64, true>::readval(start); | |
360 | else | |
361 | gold_assert(false); // We need to implement more cases, then. | |
362 | // FIXME: set lsm->shndx from the reloc | |
363 | lsm->shndx = 1; | |
364 | break; | |
365 | ||
366 | case elfcpp::DW_LNE_define_file: | |
367 | { | |
368 | const char* filename = reinterpret_cast<const char*>(start); | |
369 | templen = strlen(filename) + 1; | |
370 | start += templen; | |
371 | ||
372 | uint64_t dirindex = read_unsigned_LEB_128(start, &templen); | |
373 | if (dirindex >= directories_.size()) | |
374 | dirindex = 0; | |
375 | oplen += templen; | |
376 | ||
377 | read_unsigned_LEB_128(start, &templen); // mod_time | |
378 | oplen += templen; | |
379 | ||
380 | read_unsigned_LEB_128(start, &templen); // filelength | |
381 | oplen += templen; | |
382 | ||
383 | files_.push_back(std::pair<int, std::string>(dirindex, | |
384 | filename)); | |
385 | } | |
386 | break; | |
387 | } | |
388 | } | |
389 | break; | |
390 | ||
391 | default: | |
392 | { | |
393 | // Ignore unknown opcode silently | |
394 | for (int i = 0; i < header_.std_opcode_lengths[opcode]; i++) | |
395 | { | |
396 | size_t templen; | |
397 | read_unsigned_LEB_128(start, &templen); | |
398 | start += templen; | |
399 | oplen += templen; | |
400 | } | |
401 | } | |
402 | break; | |
403 | } | |
404 | *len = oplen; | |
405 | return false; | |
406 | } | |
407 | ||
408 | // Read the debug information at LINEPTR and store it in the line | |
409 | // number map. | |
410 | ||
411 | unsigned const char* | |
412 | Dwarf_line_info::read_lines(int size, bool big_endian, | |
413 | unsigned const char* lineptr) | |
414 | { | |
415 | struct LineStateMachine lsm; | |
416 | ||
417 | // LENGTHSTART is the place the length field is based on. It is the | |
418 | // point in the header after the initial length field. | |
419 | const unsigned char* lengthstart = buffer_; | |
420 | ||
421 | // In 64 bit dwarf, the initial length is 12 bytes, because of the | |
422 | // 0xffffffff at the start. | |
423 | if (header_.offset_size == 8) | |
424 | lengthstart += 12; | |
425 | else | |
426 | lengthstart += 4; | |
427 | ||
428 | while (lineptr < lengthstart + header_.total_length) | |
429 | { | |
430 | ResetLineStateMachine(&lsm, header_.default_is_stmt); | |
431 | while (!lsm.end_sequence) | |
432 | { | |
433 | size_t oplength; | |
434 | bool add_line = this->process_one_opcode(size, big_endian, | |
435 | lineptr, &lsm, &oplength); | |
436 | if (add_line) | |
437 | { | |
438 | Offset_to_lineno_entry entry | |
439 | = { lsm.address, lsm.file_num, lsm.line_num }; | |
440 | line_number_map_[lsm.shndx].push_back(entry); | |
441 | } | |
442 | lineptr += oplength; | |
443 | } | |
444 | } | |
445 | ||
446 | return lengthstart + header_.total_length; | |
447 | } | |
448 | ||
449 | // Called after all line numbers have been read. | |
450 | ||
451 | void | |
452 | Dwarf_line_info::finalize_line_number_map() | |
453 | { | |
454 | for (Lineno_map::iterator it = line_number_map_.begin(); | |
455 | it != line_number_map_.end(); | |
456 | ++it) | |
457 | // Each vector needs to be sorted by offset. | |
458 | sort(it->second.begin(), it->second.end()); | |
459 | } | |
460 | ||
461 | // Return a string for a file name and line number. | |
462 | ||
463 | std::string | |
464 | Dwarf_line_info::addr2line(unsigned int shndx, off_t offset) | |
465 | { | |
466 | const Offset_to_lineno_entry lookup_key = { offset, 0, 0 }; | |
467 | std::vector<Offset_to_lineno_entry>& offsets = line_number_map_[shndx]; | |
468 | std::vector<Offset_to_lineno_entry>::const_iterator it | |
469 | = std::lower_bound(offsets.begin(), offsets.end(), lookup_key); | |
470 | ||
471 | // If we found an exact match, great, otherwise find the last entry | |
472 | // before the passed-in offset. | |
473 | if (it->offset > offset) | |
474 | { | |
475 | if (it == offsets.begin()) | |
476 | return ""; | |
477 | --it; | |
478 | gold_assert(it->offset < offset); | |
479 | } | |
480 | ||
481 | // Convert the file_num + line_num into a string. | |
482 | std::string ret; | |
483 | gold_assert(it->file_num < static_cast<int>(files_.size())); | |
484 | const std::pair<int, std::string>& filename_pair = files_[it->file_num]; | |
485 | gold_assert(filename_pair.first < static_cast<int>(directories_.size())); | |
486 | const std::string& dirname = directories_[filename_pair.first]; | |
487 | const std::string& filename = filename_pair.second; | |
488 | if (!dirname.empty()) | |
489 | { | |
490 | ret += dirname; | |
491 | ret += "/"; | |
492 | } | |
493 | ret += filename; | |
494 | if (ret.empty()) | |
495 | ret = "(unknown)"; | |
496 | ||
497 | char buffer[64]; // enough to hold a line number | |
498 | snprintf(buffer, sizeof(buffer), "%d", it->line_num); | |
499 | ret += ":"; | |
500 | ret += buffer; | |
501 | ||
502 | return ret; | |
503 | } | |
504 | ||
505 | #ifdef HAVE_TARGET_32_LITTLE | |
506 | template | |
507 | const unsigned char* | |
508 | Dwarf_line_info::read_header_prolog<32, false>(const unsigned char* lineptr); | |
509 | #endif | |
510 | ||
511 | #ifdef HAVE_TARGET_32_BIG | |
512 | template | |
513 | const unsigned char* | |
514 | Dwarf_line_info::read_header_prolog<32, true>(const unsigned char* lineptr); | |
515 | #endif | |
516 | ||
517 | #ifdef HAVE_TARGET_64_LITTLE | |
518 | template | |
519 | const unsigned char* | |
520 | Dwarf_line_info::read_header_prolog<64, false>(const unsigned char* lineptr); | |
521 | #endif | |
522 | ||
523 | #ifdef HAVE_TARGET_64_BIG | |
524 | template | |
525 | const unsigned char* | |
526 | Dwarf_line_info::read_header_prolog<64, true>(const unsigned char* lineptr); | |
527 | #endif | |
528 | ||
529 | } // End namespace gold. |