Mike Dodd | 8cfa702 | 2010-11-17 11:12:26 -0800 | [diff] [blame] | 1 | /** |
| 2 | * @file opannotate.cpp |
| 3 | * Implement opannotate utility |
| 4 | * |
| 5 | * @remark Copyright 2003 OProfile authors |
| 6 | * @remark Read the file COPYING |
| 7 | * |
| 8 | * @author John Levon |
| 9 | * @author Philippe Elie |
| 10 | */ |
| 11 | |
| 12 | #include <iostream> |
| 13 | #include <sstream> |
| 14 | #include <algorithm> |
| 15 | #include <iomanip> |
| 16 | #include <fstream> |
| 17 | #include <utility> |
| 18 | |
| 19 | #include "op_exception.h" |
| 20 | #include "op_header.h" |
| 21 | #include "profile.h" |
| 22 | #include "populate.h" |
| 23 | #include "op_sample_file.h" |
| 24 | #include "cverb.h" |
| 25 | #include "string_manip.h" |
| 26 | #include "demangle_symbol.h" |
| 27 | #include "child_reader.h" |
| 28 | #include "op_file.h" |
| 29 | #include "file_manip.h" |
| 30 | #include "arrange_profiles.h" |
| 31 | #include "opannotate_options.h" |
| 32 | #include "profile_container.h" |
| 33 | #include "symbol_sort.h" |
| 34 | #include "image_errors.h" |
| 35 | |
| 36 | using namespace std; |
| 37 | using namespace options; |
| 38 | |
| 39 | namespace { |
| 40 | |
| 41 | size_t nr_events; |
| 42 | |
| 43 | scoped_ptr<profile_container> samples; |
| 44 | |
| 45 | /// how opannotate was invoked |
| 46 | string cmdline; |
| 47 | |
| 48 | /// empty annotation fill string |
| 49 | string annotation_fill; |
| 50 | |
| 51 | /// string used as start / end comment to annotate source |
| 52 | string const begin_comment("/* "); |
| 53 | string const in_comment(" * "); |
| 54 | string const end_comment(" */"); |
| 55 | |
| 56 | /// field width for the sample count |
| 57 | unsigned int const count_width = 6; |
| 58 | |
| 59 | string get_annotation_fill() |
| 60 | { |
| 61 | string str; |
| 62 | |
| 63 | for (size_t i = 0; i < nr_events; ++i) { |
| 64 | str += string(count_width, ' ') + ' '; |
| 65 | str += string(percent_width, ' '); |
| 66 | } |
| 67 | |
| 68 | for (size_t i = 1; i < nr_events; ++i) |
| 69 | str += " "; |
| 70 | |
| 71 | str += " :"; |
| 72 | return str; |
| 73 | } |
| 74 | |
| 75 | |
| 76 | symbol_entry const * find_symbol(string const & image_name, |
| 77 | string const & str_vma) |
| 78 | { |
| 79 | // do not use the bfd equivalent: |
| 80 | // - it does not skip space at begin |
| 81 | // - we does not need cross architecture compile so the native |
| 82 | // strtoull must work, assuming unsigned long long can contain a vma |
| 83 | // and on 32/64 bits box bfd_vma is 64 bits |
| 84 | bfd_vma vma = strtoull(str_vma.c_str(), NULL, 16); |
| 85 | |
| 86 | return samples->find_symbol(image_name, vma); |
| 87 | } |
| 88 | |
| 89 | |
| 90 | void output_info(ostream & out) |
| 91 | { |
| 92 | out << begin_comment << '\n'; |
| 93 | |
| 94 | out << in_comment << "Command line: " << cmdline << '\n' |
| 95 | << in_comment << '\n'; |
| 96 | |
| 97 | out << in_comment << "Interpretation of command line:" << '\n'; |
| 98 | |
| 99 | if (!assembly) { |
| 100 | out << in_comment |
| 101 | << "Output annotated source file with samples" << '\n'; |
| 102 | |
| 103 | if (options::threshold != 0) { |
| 104 | out << in_comment |
| 105 | << "Output files where samples count reach " |
| 106 | << options::threshold << "% of the samples\n"; |
| 107 | } else { |
| 108 | out << in_comment << "Output all files" << '\n'; |
| 109 | } |
| 110 | } else { |
| 111 | out << in_comment |
| 112 | << "Output annotated assembly listing with samples" |
| 113 | << '\n'; |
| 114 | |
| 115 | if (!objdump_params.empty()) { |
| 116 | out << in_comment << "Passing the following " |
| 117 | "additional arguments to objdump ; \""; |
| 118 | for (size_t i = 0 ; i < objdump_params.size() ; ++i) |
| 119 | out << objdump_params[i] << " "; |
| 120 | out << "\"" << '\n'; |
| 121 | } |
| 122 | } |
| 123 | |
| 124 | out << in_comment << '\n'; |
| 125 | |
| 126 | out << in_comment << classes.cpuinfo << endl; |
| 127 | if (!classes.event.empty()) |
| 128 | out << in_comment << classes.event << endl; |
| 129 | |
| 130 | for (size_t i = 0; i < classes.v.size(); ++i) |
| 131 | out << in_comment << classes.v[i].longname << endl; |
| 132 | |
| 133 | out << end_comment << '\n'; |
| 134 | } |
| 135 | |
| 136 | |
| 137 | string count_str(count_array_t const & count, |
| 138 | count_array_t const & total) |
| 139 | { |
| 140 | ostringstream os; |
| 141 | for (size_t i = 0; i < nr_events; ++i) { |
| 142 | os << setw(count_width) << count[i] << ' '; |
| 143 | |
| 144 | os << format_percent(op_ratio(count[i], total[i]) * 100.0, |
| 145 | percent_int_width, percent_fract_width); |
| 146 | } |
| 147 | return os.str(); |
| 148 | } |
| 149 | |
| 150 | |
| 151 | /// NOTE: This function annotates a list<string> containing output from objdump. |
| 152 | /// It uses a list iterator, and a sample_container iterator which iterates |
| 153 | /// from the beginning to the end, and compare sample address |
| 154 | /// against the instruction address on the asm line. |
| 155 | /// |
| 156 | /// There are 2 cases of annotation: |
| 157 | /// 1. If sample address matches current line address, annotate the current line. |
| 158 | /// 2. If (previous line address < sample address < current line address), |
| 159 | /// then we annotate previous line. This case happens when sample address |
| 160 | /// is not aligned with the instruction address, which is seen when profile |
| 161 | /// using the instruction fetch mode of AMD Instruction-Based Sampling (IBS). |
| 162 | /// |
| 163 | int asm_list_annotation(symbol_entry const * last_symbol, |
| 164 | bfd_vma last_symbol_vma, |
| 165 | list<string>::iterator sit, |
| 166 | sample_container::samples_iterator & samp_it, |
| 167 | list<string> & asm_lines) |
| 168 | { |
| 169 | int ret = 0; |
| 170 | |
| 171 | sample_entry const * sample = NULL; |
| 172 | |
| 173 | if (samp_it != samples->end()) |
| 174 | sample = &samp_it->second; |
| 175 | |
| 176 | // do not use the bfd equivalent: |
| 177 | // - it does not skip space at begin |
| 178 | // - we does not need cross architecture compile so the native |
| 179 | // strtoull must work, assuming unsigned long long can contain a vma |
| 180 | // and on 32/64 bits box bfd_vma is 64 bits |
| 181 | // gcc 2.91.66 workaround |
| 182 | bfd_vma vma = strtoull((*sit).c_str(), NULL, 16); |
| 183 | |
| 184 | if (sample |
| 185 | && ((sample->vma < last_symbol_vma) || (sample->vma > vma))) { |
| 186 | *sit = annotation_fill + *sit; |
| 187 | } else if (sample && sample->vma == vma) { |
| 188 | // Case 1 : Sample address match current line address. |
| 189 | string str = count_str(sample->counts, samples->samples_count()); |
| 190 | |
| 191 | // For each events |
| 192 | for (size_t i = 1; i < nr_events; ++i) |
| 193 | str += " "; |
| 194 | |
| 195 | *sit = str + " :" + *sit; |
| 196 | if (samp_it != samples->end()) |
| 197 | ++samp_it; |
| 198 | |
| 199 | } else if (sample && sample->vma < vma) { |
| 200 | // Case 2 : vma of the current line is greater than vma of the sample |
| 201 | |
| 202 | // Get the string of previous assembly line |
| 203 | list<string>::iterator sit_prev = sit; |
| 204 | string prev_line, prev_vma_str; |
| 205 | string::size_type loc1 = string::npos, loc2 = string::npos; |
| 206 | while (sit_prev != asm_lines.begin()) { |
| 207 | --sit_prev; |
| 208 | prev_line = *sit_prev; |
| 209 | |
| 210 | loc1 = prev_line.find(":", 0); |
| 211 | if (loc1 != string::npos) { |
| 212 | loc2 = prev_line.find(":", loc1+1); |
| 213 | if (loc2 != string::npos) { |
| 214 | prev_vma_str = prev_line.substr(loc1+1, loc2); |
| 215 | break; |
| 216 | } |
| 217 | } |
| 218 | } |
| 219 | |
| 220 | bfd_vma prev_vma = strtoull(prev_vma_str.c_str(), NULL, 16); |
| 221 | |
| 222 | // Need to check if prev_vma < sample->vma |
| 223 | if (prev_vma != 0 && prev_vma < sample->vma) { |
| 224 | string str; |
| 225 | |
| 226 | // Get sample for previous line. |
| 227 | sample_entry * prev_sample = (sample_entry *)samples-> |
| 228 | find_sample(last_symbol, prev_vma); |
| 229 | if (prev_sample) { |
| 230 | // Aggregate sample with previous line if it already has samples |
| 231 | prev_sample->counts += sample->counts; |
| 232 | str = count_str(prev_sample->counts, samples->samples_count()); |
| 233 | } else { |
| 234 | str = count_str(sample->counts, samples->samples_count()); |
| 235 | } |
| 236 | |
| 237 | // For each events |
| 238 | for (size_t i = 1; i < nr_events; ++i) |
| 239 | str += " "; |
| 240 | |
| 241 | *sit_prev = str + " :" + prev_line.substr(loc1+1); |
| 242 | if (samp_it != samples->end()) |
| 243 | ++samp_it; |
| 244 | ret = -1; |
| 245 | } else { |
| 246 | // Failed to annotate the previous line. Skip sample. |
| 247 | *sit = annotation_fill + *sit; |
| 248 | if (samp_it != samples->end()) |
| 249 | ++samp_it; |
| 250 | } |
| 251 | } else { |
| 252 | // In case sample is NULL |
| 253 | *sit = annotation_fill + *sit; |
| 254 | } |
| 255 | |
| 256 | return ret; |
| 257 | } |
| 258 | |
| 259 | |
| 260 | string symbol_annotation(symbol_entry const * symbol) |
| 261 | { |
| 262 | if (!symbol) |
| 263 | return string(); |
| 264 | |
| 265 | string annot = count_str(symbol->sample.counts, |
| 266 | samples->samples_count()); |
| 267 | |
| 268 | string const & symname = symbol_names.demangle(symbol->name); |
| 269 | |
| 270 | string str = " "; |
| 271 | str += begin_comment + symname + " total: "; |
| 272 | str += count_str(symbol->sample.counts, samples->samples_count()); |
| 273 | str += end_comment; |
| 274 | return str; |
| 275 | } |
| 276 | |
| 277 | |
| 278 | /// return true if this line contains a symbol name in objdump formatting |
| 279 | /// symbol are on the form 08030434 <symbol_name>: we need to be strict |
| 280 | /// here to avoid any interpretation of a source line as a symbol line |
| 281 | bool is_symbol_line(string const & str, string::size_type pos) |
| 282 | { |
| 283 | if (str[pos] != ' ' || str[pos + 1] != '<') |
| 284 | return false; |
| 285 | |
| 286 | return str[str.length() - 1] == ':'; |
| 287 | } |
| 288 | |
| 289 | |
| 290 | void annotate_objdump_str_list(string const & app_name, |
| 291 | symbol_collection const & symbols, |
| 292 | list<string> & asm_lines) |
| 293 | { |
| 294 | symbol_entry const * last_symbol = 0; |
| 295 | bfd_vma last_symbol_vma = 0; |
| 296 | int ret = 0; |
| 297 | |
| 298 | // to filter output of symbols (filter based on command line options) |
| 299 | bool do_output = true; |
| 300 | |
| 301 | // We simultaneously walk the two structures (list and sample_container) |
| 302 | // which are sorted by address. and do address comparision. |
| 303 | list<string>::iterator sit = asm_lines.begin(); |
| 304 | list<string>::iterator send = asm_lines.end(); |
| 305 | sample_container::samples_iterator samp_it = samples->begin(); |
| 306 | |
| 307 | for (; sit != send; (!ret? sit++: sit)) { |
| 308 | // output of objdump is a human readable form and can contain some |
| 309 | // ambiguity so this code is dirty. It is also optimized a little bit |
| 310 | // so it is difficult to simplify it without breaking something ... |
| 311 | |
| 312 | // line of interest are: "[:space:]*[:xdigit:]?[ :]", the last char of |
| 313 | // this regexp dis-ambiguate between a symbol line and an asm line. If |
| 314 | // source contain line of this form an ambiguity occur and we rely on |
| 315 | // the robustness of this code. |
| 316 | string str = *sit; |
| 317 | size_t pos = 0; |
| 318 | while (pos < str.length() && isspace(str[pos])) |
| 319 | ++pos; |
| 320 | |
| 321 | if (pos == str.length() || !isxdigit(str[pos])) { |
| 322 | if (do_output) { |
| 323 | *sit = annotation_fill + str; |
| 324 | continue; |
| 325 | } |
| 326 | } |
| 327 | |
| 328 | while (pos < str.length() && isxdigit(str[pos])) |
| 329 | ++pos; |
| 330 | |
| 331 | if (pos == str.length() || (!isspace(str[pos]) && str[pos] != ':')) { |
| 332 | if (do_output) { |
| 333 | *sit = annotation_fill + str; |
| 334 | continue; |
| 335 | } |
| 336 | } |
| 337 | |
| 338 | if (is_symbol_line(str, pos)) { |
| 339 | |
| 340 | last_symbol = find_symbol(app_name, str); |
| 341 | last_symbol_vma = strtoull(str.c_str(), NULL, 16); |
| 342 | |
| 343 | // ! complexity: linear in number of symbol must use sorted |
| 344 | // by address vector and lower_bound ? |
| 345 | // Note this use a pointer comparison. It work because symbols |
| 346 | // pointer are unique |
| 347 | if (find(symbols.begin(), symbols.end(), last_symbol) |
| 348 | != symbols.end()) |
| 349 | do_output = true; |
| 350 | else |
| 351 | do_output = false; |
| 352 | |
| 353 | if (do_output) { |
| 354 | *sit += symbol_annotation(last_symbol); |
| 355 | |
| 356 | // Realign the sample iterator to |
| 357 | // the beginning of this symbols |
| 358 | samp_it = samples->begin(last_symbol); |
| 359 | } |
| 360 | } else { |
| 361 | // not a symbol, probably an asm line. |
| 362 | if (do_output) |
| 363 | ret = asm_list_annotation(last_symbol, |
| 364 | last_symbol_vma, |
| 365 | sit, samp_it, |
| 366 | asm_lines); |
| 367 | } |
| 368 | |
| 369 | if (!do_output) |
| 370 | *sit = ""; |
| 371 | } |
| 372 | } |
| 373 | |
| 374 | |
| 375 | void output_objdump_str_list(symbol_collection const & symbols, |
| 376 | string const & app_name, |
| 377 | list<string> & asm_lines) |
| 378 | { |
| 379 | |
| 380 | annotate_objdump_str_list(app_name, symbols, asm_lines); |
| 381 | |
| 382 | // Printing objdump output to stdout |
| 383 | list<string>::iterator sit = asm_lines.begin(); |
| 384 | list<string>::iterator send = asm_lines.end(); |
| 385 | sit = asm_lines.begin(); |
| 386 | for (; sit != send; ++sit) { |
| 387 | string str = *sit; |
| 388 | if (str.length() != 0) |
| 389 | cout << str << '\n'; |
| 390 | } |
| 391 | } |
| 392 | |
| 393 | |
| 394 | void do_one_output_objdump(symbol_collection const & symbols, |
| 395 | string const & image_name, string const & app_name, |
| 396 | bfd_vma start, bfd_vma end) |
| 397 | { |
| 398 | vector<string> args; |
| 399 | list<string> asm_lines; |
| 400 | |
| 401 | args.push_back("-d"); |
| 402 | args.push_back("--no-show-raw-insn"); |
| 403 | if (source) |
| 404 | args.push_back("-S"); |
| 405 | |
| 406 | if (start || end != ~(bfd_vma)0) { |
| 407 | ostringstream arg1, arg2; |
| 408 | arg1 << "--start-address=" << start; |
| 409 | arg2 << "--stop-address=" << end; |
| 410 | args.push_back(arg1.str()); |
| 411 | args.push_back(arg2.str()); |
| 412 | } |
| 413 | |
| 414 | if (!objdump_params.empty()) { |
| 415 | for (size_t i = 0 ; i < objdump_params.size() ; ++i) |
| 416 | args.push_back(objdump_params[i]); |
| 417 | } |
| 418 | |
| 419 | args.push_back(image_name); |
Paul Lind | 935b17d | 2013-03-04 17:23:51 -0800 | [diff] [blame] | 420 | #if defined(ANDROID_TARGET_ARM) |
Ben Cheng | e4b944c | 2012-01-06 15:35:28 -0800 | [diff] [blame] | 421 | child_reader reader("arm-eabi-objdump", args); |
Paul Lind | 935b17d | 2013-03-04 17:23:51 -0800 | [diff] [blame] | 422 | #elif defined(ANDROID_TARGET_MIPS) |
| 423 | child_reader reader("mipsel-linux-android-objdump", args); |
Ben Cheng | e4b944c | 2012-01-06 15:35:28 -0800 | [diff] [blame] | 424 | #else |
Mike Dodd | 8cfa702 | 2010-11-17 11:12:26 -0800 | [diff] [blame] | 425 | child_reader reader("objdump", args); |
Ben Cheng | e4b944c | 2012-01-06 15:35:28 -0800 | [diff] [blame] | 426 | #endif |
Mike Dodd | 8cfa702 | 2010-11-17 11:12:26 -0800 | [diff] [blame] | 427 | if (reader.error()) { |
| 428 | cerr << "An error occur during the execution of objdump:\n\n"; |
| 429 | cerr << reader.error_str() << endl; |
| 430 | return; |
| 431 | } |
| 432 | |
| 433 | // Read each output line from objdump and store in a list. |
| 434 | string str; |
| 435 | while (reader.getline(str)) |
| 436 | asm_lines.push_back(str); |
| 437 | |
| 438 | output_objdump_str_list(symbols, app_name, asm_lines); |
| 439 | |
| 440 | // objdump always returns SUCCESS so we must rely on the stderr state |
| 441 | // of objdump. If objdump error message is cryptic our own error |
| 442 | // message will be probably also cryptic |
| 443 | ostringstream std_err; |
| 444 | ostringstream std_out; |
| 445 | reader.get_data(std_out, std_err); |
| 446 | if (std_err.str().length()) { |
| 447 | cerr << "An error occur during the execution of objdump:\n\n"; |
| 448 | cerr << std_err.str() << endl; |
| 449 | return ; |
| 450 | } |
| 451 | |
| 452 | // force error code to be acquired |
| 453 | reader.terminate_process(); |
| 454 | |
| 455 | // required because if objdump stop by signal all above things suceeed |
| 456 | // (signal error message are not output through stdout/stderr) |
| 457 | if (reader.error()) { |
| 458 | cerr << "An error occur during the execution of objdump:\n\n"; |
| 459 | cerr << reader.error_str() << endl; |
| 460 | return; |
| 461 | } |
| 462 | } |
| 463 | |
| 464 | |
| 465 | void output_objdump_asm(symbol_collection const & symbols, |
| 466 | string const & app_name) |
| 467 | { |
| 468 | image_error error; |
| 469 | string image = |
| 470 | classes.extra_found_images.find_image_path(app_name, error, |
| 471 | true); |
| 472 | |
| 473 | // this is only an optimisation, we can either filter output by |
| 474 | // directly calling objdump and rely on the symbol filtering or |
| 475 | // we can call objdump with the right parameter to just disassemble |
| 476 | // the needed part. This is a real win only when calling objdump |
| 477 | // a medium number of times, I dunno if the used threshold is optimal |
| 478 | // but it is a conservative value. |
| 479 | size_t const max_objdump_exec = 50; |
| 480 | if (symbols.size() <= max_objdump_exec || error != image_ok) { |
| 481 | symbol_collection::const_iterator cit = symbols.begin(); |
| 482 | symbol_collection::const_iterator end = symbols.end(); |
| 483 | for (; cit != end; ++cit) { |
| 484 | bfd_vma start = (*cit)->sample.vma; |
| 485 | bfd_vma end = start + (*cit)->size; |
| 486 | do_one_output_objdump(symbols, image, app_name, |
| 487 | start, end); |
| 488 | } |
| 489 | } else { |
| 490 | do_one_output_objdump(symbols, image, |
| 491 | app_name, 0, ~bfd_vma(0)); |
| 492 | } |
| 493 | } |
| 494 | |
| 495 | |
| 496 | bool output_asm(string const & app_name) |
| 497 | { |
| 498 | profile_container::symbol_choice choice; |
| 499 | choice.threshold = options::threshold; |
| 500 | choice.image_name = app_name; |
| 501 | choice.match_image = true; |
| 502 | symbol_collection symbols = samples->select_symbols(choice); |
| 503 | |
| 504 | if (!symbols.empty()) { |
| 505 | sort_options options; |
| 506 | options.add_sort_option(sort_options::sample); |
| 507 | options.sort(symbols, false, false); |
| 508 | |
| 509 | output_info(cout); |
| 510 | |
| 511 | output_objdump_asm(symbols, app_name); |
| 512 | |
| 513 | return true; |
| 514 | } |
| 515 | |
| 516 | return false; |
| 517 | } |
| 518 | |
| 519 | |
| 520 | string const source_line_annotation(debug_name_id filename, size_t linenr) |
| 521 | { |
| 522 | string str; |
| 523 | |
| 524 | count_array_t counts = samples->samples_count(filename, linenr); |
| 525 | if (!counts.zero()) { |
| 526 | str += count_str(counts, samples->samples_count()); |
| 527 | for (size_t i = 1; i < nr_events; ++i) |
| 528 | str += " "; |
| 529 | str += " :"; |
| 530 | } else { |
| 531 | str = annotation_fill; |
| 532 | } |
| 533 | |
| 534 | return str; |
| 535 | } |
| 536 | |
| 537 | |
| 538 | string source_symbol_annotation(debug_name_id filename, size_t linenr) |
| 539 | { |
| 540 | symbol_collection const symbols = samples->find_symbol(filename, linenr); |
| 541 | |
| 542 | if (symbols.empty()) |
| 543 | return string(); |
| 544 | |
| 545 | string str = " " + begin_comment; |
| 546 | |
| 547 | count_array_t counts; |
| 548 | for (size_t i = 0; i < symbols.size(); ++i) { |
| 549 | str += symbol_names.demangle(symbols[i]->name); |
| 550 | if (symbols.size() == 1) |
| 551 | str += " total: "; |
| 552 | else |
| 553 | str += " "; |
| 554 | str += count_str(symbols[i]->sample.counts, |
| 555 | samples->samples_count()); |
| 556 | if (symbols.size() != 1) |
| 557 | str += ", "; |
| 558 | |
| 559 | counts += symbols[i]->sample.counts; |
| 560 | } |
| 561 | |
| 562 | if (symbols.size() > 1) |
| 563 | str += "total: " + count_str(counts, samples->samples_count()); |
| 564 | str += end_comment; |
| 565 | |
| 566 | return str; |
| 567 | } |
| 568 | |
| 569 | |
| 570 | void output_per_file_info(ostream & out, debug_name_id filename, |
| 571 | count_array_t const & total_file_count) |
| 572 | { |
| 573 | out << begin_comment << '\n' |
| 574 | << in_comment << "Total samples for file : " |
| 575 | << '"' << debug_names.name(filename) << '"' |
| 576 | << '\n'; |
| 577 | out << in_comment << '\n' << in_comment |
| 578 | << count_str(total_file_count, samples->samples_count()) |
| 579 | << '\n'; |
| 580 | out << end_comment << '\n' << '\n'; |
| 581 | } |
| 582 | |
| 583 | |
| 584 | string const line0_info(debug_name_id filename) |
| 585 | { |
| 586 | string annotation = source_line_annotation(filename, 0); |
| 587 | if (trim(annotation, " \t:").empty()) |
| 588 | return string(); |
| 589 | |
| 590 | string str = "<credited to line zero> "; |
| 591 | str += annotation; |
| 592 | return str; |
| 593 | } |
| 594 | |
| 595 | |
| 596 | void do_output_one_file(ostream & out, istream & in, debug_name_id filename, |
| 597 | bool header) |
| 598 | { |
| 599 | count_array_t count = samples->samples_count(filename); |
| 600 | |
| 601 | if (header) { |
| 602 | output_per_file_info(out, filename, count); |
| 603 | out << line0_info(filename) << '\n'; |
| 604 | } |
| 605 | |
| 606 | |
| 607 | if (in) { |
| 608 | string str; |
| 609 | |
| 610 | for (size_t linenr = 1 ; getline(in, str) ; ++linenr) { |
| 611 | out << source_line_annotation(filename, linenr) << str |
| 612 | << source_symbol_annotation(filename, linenr) |
| 613 | << '\n'; |
| 614 | } |
| 615 | |
| 616 | } else { |
| 617 | // source is not available but we can at least output all the |
| 618 | // symbols belonging to this file. This make more visible the |
| 619 | // problem of having less samples for a given file than the |
| 620 | // sum of all symbols samples for this file due to inlining |
| 621 | symbol_collection const symbols = samples->select_symbols(filename); |
| 622 | for (size_t i = 0; i < symbols.size(); ++i) |
| 623 | out << symbol_annotation(symbols[i]) << endl; |
| 624 | } |
| 625 | |
| 626 | if (!header) { |
| 627 | output_per_file_info(out, filename, count); |
| 628 | out << line0_info(filename) << '\n'; |
| 629 | } |
| 630 | } |
| 631 | |
| 632 | |
| 633 | void output_one_file(istream & in, debug_name_id filename, |
| 634 | string const & source) |
| 635 | { |
| 636 | if (output_dir.empty()) { |
| 637 | do_output_one_file(cout, in, filename, true); |
| 638 | return; |
| 639 | } |
| 640 | |
| 641 | string const out_file = op_realpath(output_dir + source); |
| 642 | |
| 643 | /* Just because you're paranoid doesn't mean they're not out to |
| 644 | * get you ... |
| 645 | * |
| 646 | * This is just a lame final safety check. If we found the |
| 647 | * source, then "source" should be canonical already, and |
| 648 | * can't escape from the output dir. We can't use op_realpath() |
| 649 | * alone as that needs the file to exist already. |
| 650 | * |
| 651 | * Let's not complain again if we couldn't find the file anyway. |
| 652 | */ |
| 653 | if (out_file.find("/../") != string::npos) { |
| 654 | if (in) { |
| 655 | cerr << "refusing to create non-canonical filename " |
| 656 | << out_file << endl; |
| 657 | } |
| 658 | return; |
| 659 | } else if (!is_prefix(out_file, output_dir)) { |
| 660 | if (in) { |
| 661 | cerr << "refusing to create file " << out_file |
| 662 | << " outside of output directory " << output_dir |
| 663 | << endl; |
| 664 | } |
| 665 | return; |
| 666 | } |
| 667 | |
| 668 | if (is_files_identical(out_file, source)) { |
| 669 | cerr << "input and output files are identical: " |
| 670 | << out_file << endl; |
| 671 | return; |
| 672 | } |
| 673 | |
| 674 | if (create_path(out_file.c_str())) { |
| 675 | cerr << "unable to create file: " |
| 676 | << '"' << op_dirname(out_file) << '"' << endl; |
| 677 | return; |
| 678 | } |
| 679 | |
| 680 | ofstream out(out_file.c_str()); |
| 681 | if (!out) { |
| 682 | cerr << "unable to open output file " |
| 683 | << '"' << out_file << '"' << endl; |
| 684 | } else { |
| 685 | do_output_one_file(out, in, filename, false); |
| 686 | output_info(out); |
| 687 | } |
| 688 | } |
| 689 | |
| 690 | |
| 691 | /* Locate a source file from debug info, which may be relative */ |
| 692 | string const locate_source_file(debug_name_id filename_id) |
| 693 | { |
| 694 | string const origfile = debug_names.name(filename_id); |
| 695 | string file = origfile; |
| 696 | |
| 697 | if (file.empty()) |
| 698 | return file; |
| 699 | |
| 700 | /* Allow absolute paths to be relocated to a different directory */ |
| 701 | if (file[0] == '/') { |
| 702 | vector<string>::const_iterator cit = base_dirs.begin(); |
| 703 | vector<string>::const_iterator end = base_dirs.end(); |
| 704 | for (; cit != end; ++cit) { |
| 705 | string path = op_realpath(*cit); |
| 706 | |
| 707 | if (is_prefix(file, path)) { |
| 708 | file = file.substr(path.length()); |
| 709 | break; |
| 710 | } |
| 711 | } |
| 712 | } |
| 713 | |
| 714 | vector<string>::const_iterator cit = search_dirs.begin(); |
| 715 | vector<string>::const_iterator end = search_dirs.end(); |
| 716 | |
| 717 | for (; cit != end; ++cit) { |
| 718 | string const absfile = op_realpath(*cit + "/" + file); |
| 719 | |
| 720 | if (op_file_readable(absfile)) |
| 721 | return absfile; |
| 722 | } |
| 723 | |
| 724 | /* We didn't find a relocated absolute file, or a relative file, |
| 725 | * assume the original is correct, accounting for the |
| 726 | * possibility it's relative the cwd |
| 727 | */ |
| 728 | return op_realpath(origfile); |
| 729 | } |
| 730 | |
| 731 | |
| 732 | void output_source(path_filter const & filter) |
| 733 | { |
| 734 | bool const separate_file = !output_dir.empty(); |
| 735 | |
| 736 | if (!separate_file) |
| 737 | output_info(cout); |
| 738 | |
| 739 | vector<debug_name_id> filenames = |
| 740 | samples->select_filename(options::threshold); |
| 741 | |
| 742 | for (size_t i = 0 ; i < filenames.size() ; ++i) { |
| 743 | string const & source = locate_source_file(filenames[i]); |
| 744 | |
| 745 | if (!filter.match(source)) |
| 746 | continue; |
| 747 | |
| 748 | ifstream in(source.c_str()); |
| 749 | |
| 750 | // it is common to have empty filename due to the lack |
| 751 | // of debug info (eg _init function) so warn only |
| 752 | // if the filename is non empty. The case: no debug |
| 753 | // info at all has already been checked. |
| 754 | if (!in && source.length()) { |
| 755 | cerr << "opannotate (warning): unable to open for " |
| 756 | "reading: " << source << endl; |
| 757 | } |
| 758 | |
| 759 | if (source.length()) |
| 760 | output_one_file(in, filenames[i], source); |
| 761 | } |
| 762 | } |
| 763 | |
| 764 | |
| 765 | bool annotate_source(list<string> const & images) |
| 766 | { |
| 767 | annotation_fill = get_annotation_fill(); |
| 768 | |
| 769 | if (!output_dir.empty()) { |
| 770 | |
| 771 | if (create_path(output_dir.c_str())) { |
| 772 | cerr << "unable to create " << output_dir |
| 773 | << " directory: " << endl; |
| 774 | return false; |
| 775 | } |
| 776 | |
| 777 | // Make sure we have an absolute path. |
| 778 | output_dir = op_realpath(output_dir); |
| 779 | if (output_dir.length() && |
| 780 | output_dir[output_dir.length() - 1] != '/') |
| 781 | output_dir += '/'; |
| 782 | |
| 783 | /* Don't let the user stomp on their sources */ |
| 784 | if (output_dir == "/") { |
| 785 | cerr << "Output path of / would over-write the " |
| 786 | "source files" << endl; |
| 787 | return false; |
| 788 | } |
| 789 | } |
| 790 | |
| 791 | if (assembly) { |
| 792 | bool some_output = false; |
| 793 | |
| 794 | list<string>::const_iterator it = images.begin(); |
| 795 | list<string>::const_iterator const end = images.end(); |
| 796 | |
| 797 | for (; it != end; ++it) { |
| 798 | if (output_asm(*it)) |
| 799 | some_output = true; |
| 800 | } |
| 801 | |
| 802 | if (!some_output) { |
| 803 | // It's the only case we must care since we know the |
| 804 | // selected image set is not empty |
| 805 | cerr << "selected image set doesn't contain any of " |
| 806 | << "the selected symbol\n"; |
| 807 | } |
| 808 | } else { |
| 809 | output_source(file_filter); |
| 810 | } |
| 811 | |
| 812 | return true; |
| 813 | } |
| 814 | |
| 815 | |
| 816 | int opannotate(options::spec const & spec) |
| 817 | { |
| 818 | handle_options(spec); |
| 819 | |
| 820 | nr_events = classes.v.size(); |
| 821 | |
| 822 | samples.reset(new profile_container(true, true, |
| 823 | classes.extra_found_images)); |
| 824 | |
| 825 | list<string> images; |
| 826 | |
| 827 | list<inverted_profile> iprofiles = invert_profiles(classes); |
| 828 | |
| 829 | report_image_errors(iprofiles, classes.extra_found_images); |
| 830 | |
| 831 | list<inverted_profile>::iterator it = iprofiles.begin(); |
| 832 | list<inverted_profile>::iterator const end = iprofiles.end(); |
| 833 | |
| 834 | bool debug_info = false; |
| 835 | for (; it != end; ++it) { |
| 836 | bool tmp = false; |
| 837 | populate_for_image(*samples, *it, |
| 838 | options::symbol_filter, &tmp); |
| 839 | images.push_back(it->image); |
| 840 | if (tmp) |
| 841 | debug_info = true; |
| 842 | } |
| 843 | |
| 844 | if (!debug_info && !options::assembly) { |
| 845 | cerr << "opannotate (warning): no debug information available for binary " |
| 846 | << it->image << ", and --assembly not requested\n"; |
| 847 | } |
| 848 | |
| 849 | annotate_source(images); |
| 850 | |
| 851 | return 0; |
| 852 | } |
| 853 | |
| 854 | } // anonymous namespace |
| 855 | |
| 856 | |
| 857 | int main(int argc, char const * argv[]) |
| 858 | { |
| 859 | // set the invocation, for the file headers later |
| 860 | for (int i = 0 ; i < argc ; ++i) |
| 861 | cmdline += string(argv[i]) + " "; |
| 862 | |
| 863 | return run_pp_tool(argc, argv, opannotate); |
| 864 | } |