Mike Dodd | 8cfa702 | 2010-11-17 11:12:26 -0800 | [diff] [blame] | 1 | /** |
| 2 | * @file debug_line.c |
| 3 | * DWARF 2 debug line info creation helper |
| 4 | * |
| 5 | * @remark Copyright 2007 OProfile authors |
| 6 | * @remark Read the file COPYING |
| 7 | * |
| 8 | * @author Philippe Elie |
| 9 | */ |
| 10 | |
| 11 | #include <stdint.h> |
| 12 | #include <stdlib.h> |
| 13 | #include <string.h> |
| 14 | #include <stddef.h> |
| 15 | #include <stdio.h> |
| 16 | #include <bfd.h> |
| 17 | #include <limits.h> |
| 18 | |
| 19 | #include "opjitconv.h" |
| 20 | #include "jitdump.h" |
| 21 | #include "opagent.h" |
| 22 | #include "op_libiberty.h" |
| 23 | #include "op_growable_buffer.h" |
| 24 | |
| 25 | /* |
| 26 | * Terminology comes from the TIS DWARF Debugging Information Format |
| 27 | * version 2.0 |
| 28 | */ |
| 29 | |
| 30 | typedef uint32_t uword; |
| 31 | typedef uint16_t uhalf; |
| 32 | typedef int32_t sword; |
| 33 | typedef int16_t shalf; |
| 34 | typedef uint8_t ubyte; |
| 35 | typedef int8_t sbyte; |
| 36 | |
| 37 | /* |
| 38 | * Many of the following enum are incomplete and define only the subset |
| 39 | * of DWARF we use. |
| 40 | */ |
| 41 | enum lns_opcode { |
| 42 | DW_LNS_copy=1, |
| 43 | DW_LNS_advance_pc, |
| 44 | DW_LNS_advance_line, |
| 45 | DW_LNS_set_file, |
| 46 | DW_LNS_set_column, |
| 47 | DW_LNS_negate_stmt, |
| 48 | DW_LNS_set_basic_block, |
| 49 | DW_LNS_const_add_pc, |
| 50 | DW_LNS_fixed_advance_pc, |
| 51 | |
| 52 | /* Adding new opcode needs an update of the standard_opcode_length |
| 53 | * array */ |
| 54 | |
| 55 | DW_LNS_max_opcode, |
| 56 | }; |
| 57 | |
| 58 | enum lne_opcode { |
| 59 | DW_LNE_end_sequence = 1, |
| 60 | DW_LNE_set_address, |
| 61 | DW_LNE_define_file |
| 62 | }; |
| 63 | |
| 64 | enum dw_tag { |
| 65 | DW_TAG_compile_unit = 0x11, |
| 66 | }; |
| 67 | |
| 68 | enum dw_at { |
| 69 | DW_AT_name = 0x03, |
| 70 | DW_AT_stmt_list = 0x10, |
| 71 | DW_AT_low_pc, |
| 72 | DW_AT_high_pc, |
| 73 | DW_AT_language, |
| 74 | DW_AT_compdir = 0x1b, |
| 75 | DW_AT_producer = 0x25, |
| 76 | }; |
| 77 | |
| 78 | enum dw_children { |
| 79 | DW_CHILDREN_no, |
| 80 | DW_CHILDREN_yes |
| 81 | }; |
| 82 | |
| 83 | enum dw_form { |
| 84 | DW_FORM_data4 = 0x06, |
| 85 | }; |
| 86 | |
| 87 | struct debug_line_header { |
| 88 | // Not counting this field |
| 89 | uword total_length; |
| 90 | // version number (2 currently) |
| 91 | uhalf version; |
| 92 | // relative offset from next field to |
| 93 | // program statement |
| 94 | uword prolog_length; |
| 95 | ubyte minimum_instruction_length; |
| 96 | ubyte default_is_stmt; |
| 97 | // line_base - see DWARF 2 specs |
| 98 | sbyte line_base; |
| 99 | // line_range - see DWARF 2 specs |
| 100 | ubyte line_range; |
| 101 | // number of opcode + 1 |
| 102 | ubyte opcode_base; |
| 103 | /* follow the array of opcode args nr: ubytes [nr_opcode_base] */ |
| 104 | /* follow the search directories index, zero terminated string |
| 105 | * terminated by an empty string. |
| 106 | */ |
| 107 | /* follow an array of { filename, LEB128, LEB128, LEB128 }, first is |
| 108 | * the directory index entry, 0 means current directory, then mtime |
| 109 | * and filesize, last entry is followed by en empty string. |
| 110 | */ |
| 111 | /* follow the first program statement */ |
| 112 | } __attribute__((packed)); |
| 113 | |
| 114 | /* DWARF 2 spec talk only about one possible compilation unit header while |
| 115 | * binutils can handle two flavours of dwarf 2, 32 and 64 bits, this is not |
| 116 | * related to the used arch, an ELF 32 can hold more than 4 Go of debug |
| 117 | * information. For now we handle only DWARF 2 32 bits comp unit. It'll only |
| 118 | * become a problem if we generate more than 4GB of debug information. |
| 119 | */ |
| 120 | struct compilation_unit_header { |
| 121 | uword total_length; |
| 122 | uhalf version; |
| 123 | uword debug_abbrev_offset; |
| 124 | ubyte pointer_size; |
| 125 | } __attribute__((packed)); |
| 126 | |
| 127 | /* field filled at run time are marked with -1 */ |
| 128 | static struct debug_line_header const default_debug_line_header = { |
| 129 | -1, |
| 130 | 2, |
| 131 | -1, |
| 132 | 1, /* could be better when min instruction size != 1 */ |
| 133 | 1, /* we don't take care about basic block */ |
| 134 | -5, /* sensible value for line base ... */ |
| 135 | 14, /* ... and line range are guessed statically */ |
| 136 | DW_LNS_max_opcode |
| 137 | }; |
| 138 | |
| 139 | static ubyte const standard_opcode_length[DW_LNS_max_opcode - 1] = |
| 140 | { |
| 141 | 0, 1, 1, 1, 1, 0, 0, 0, 1 |
| 142 | }; |
| 143 | |
| 144 | /* field filled at run time are marked with -1 */ |
| 145 | static struct compilation_unit_header const default_comp_unit_header = { |
| 146 | -1, |
| 147 | 2, |
| 148 | 0, /* we reuse the same abbrev entries for all comp unit */ |
| 149 | -1 |
| 150 | }; |
| 151 | |
| 152 | |
| 153 | static void emit_uword(struct growable_buffer * b, uword data) |
| 154 | { |
| 155 | add_data(b, &data, sizeof(uword)); |
| 156 | } |
| 157 | |
| 158 | |
| 159 | static void emit_string(struct growable_buffer * b, char const * s) |
| 160 | { |
| 161 | add_data(b, s, strlen(s) + 1); |
| 162 | } |
| 163 | |
| 164 | |
| 165 | static void emit_unsigned_LEB128(struct growable_buffer * b, |
| 166 | unsigned long data) |
| 167 | { |
| 168 | do { |
| 169 | ubyte cur = data & 0x7F; |
| 170 | data >>= 7; |
| 171 | if (data) |
| 172 | cur |= 0x80; |
| 173 | add_data(b, &cur, 1); |
| 174 | } while (data); |
| 175 | } |
| 176 | |
| 177 | |
| 178 | static void emit_signed_LEB128(struct growable_buffer * b, long data) |
| 179 | { |
| 180 | int more = 1; |
| 181 | int negative = data < 0; |
| 182 | int size = sizeof(long) * CHAR_BIT; |
| 183 | while (more) { |
| 184 | ubyte cur = data & 0x7F; |
| 185 | data >>= 7; |
| 186 | if (negative) |
| 187 | data |= - (1 << (size - 7)); |
| 188 | if ((data == 0 && !(cur & 0x40)) || |
| 189 | (data == -1l && (cur & 0x40))) |
| 190 | more = 0; |
| 191 | else |
| 192 | cur |= 0x80; |
| 193 | add_data(b, &cur, 1); |
| 194 | } |
| 195 | } |
| 196 | |
| 197 | |
| 198 | static void emit_extended_opcode(struct growable_buffer * b, ubyte opcode, |
| 199 | void * data, size_t data_len) |
| 200 | { |
| 201 | add_data(b, "", 1); |
| 202 | emit_unsigned_LEB128(b, data_len + 1); |
| 203 | add_data(b, &opcode, 1); |
| 204 | add_data(b, data, data_len); |
| 205 | } |
| 206 | |
| 207 | |
| 208 | static void emit_opcode(struct growable_buffer * b, ubyte opcode) |
| 209 | { |
| 210 | add_data(b, &opcode, 1); |
| 211 | } |
| 212 | |
| 213 | |
| 214 | static void emit_opcode_signed(struct growable_buffer * b, |
| 215 | ubyte opcode, long data) |
| 216 | { |
| 217 | add_data(b, &opcode, 1); |
| 218 | emit_signed_LEB128(b, data); |
| 219 | } |
| 220 | |
| 221 | |
| 222 | static void emit_opcode_unsigned(struct growable_buffer * b, ubyte opcode, |
| 223 | unsigned long data) |
| 224 | { |
| 225 | add_data(b, &opcode, 1); |
| 226 | emit_unsigned_LEB128(b, data); |
| 227 | } |
| 228 | |
| 229 | |
| 230 | static void emit_advance_pc(struct growable_buffer * b, unsigned long delta_pc) |
| 231 | { |
| 232 | emit_opcode_unsigned(b, DW_LNS_advance_pc, delta_pc); |
| 233 | } |
| 234 | |
| 235 | |
| 236 | static void emit_advance_lineno(struct growable_buffer * b, long delta_lineno) |
| 237 | { |
| 238 | emit_opcode_signed(b, DW_LNS_advance_line, delta_lineno); |
| 239 | } |
| 240 | |
| 241 | |
| 242 | static void emit_lne_end_of_sequence(struct growable_buffer * b) |
| 243 | { |
| 244 | emit_extended_opcode(b, DW_LNE_end_sequence, NULL, 0); |
| 245 | } |
| 246 | |
| 247 | |
| 248 | static void emit_set_file(struct growable_buffer * b, unsigned long index) |
| 249 | { |
| 250 | emit_opcode_unsigned(b, DW_LNS_set_file, index); |
| 251 | } |
| 252 | |
| 253 | |
| 254 | static void emit_lne_define_filename(struct growable_buffer * b, |
| 255 | char const * filename) |
| 256 | { |
| 257 | /* emit_extended_opcode() can't be used here, we have additional |
| 258 | * data to output and the len field will be miscalculated. */ |
| 259 | add_data(b, "", 1); |
| 260 | /* strlen(filename) + zero terminator + len field + 3 bytes for the dir |
| 261 | * entry, timestamp and filesize */ |
| 262 | emit_unsigned_LEB128(b, strlen(filename) + 5); |
| 263 | emit_opcode(b, DW_LNE_define_file); |
| 264 | emit_string(b, filename); |
| 265 | add_data(b, "\0\0\0", 3); |
| 266 | } |
| 267 | |
| 268 | |
| 269 | static void emit_lne_set_address(struct growable_buffer * b, |
| 270 | void const * address) |
| 271 | { |
| 272 | emit_extended_opcode(b, DW_LNE_set_address, &address, sizeof(address)); |
| 273 | } |
| 274 | |
| 275 | |
| 276 | static ubyte get_special_opcode(struct debug_line_info const * line, |
| 277 | unsigned int last_lineno, unsigned long last_vma) |
| 278 | { |
| 279 | unsigned int temp; |
| 280 | unsigned long delta_addr; |
| 281 | |
| 282 | /* See TIS DWARF Debugging Information Format version 2.0 § 6.2.5.1 */ |
| 283 | |
| 284 | temp = (line->lineno - last_lineno) - |
| 285 | default_debug_line_header.line_base; |
| 286 | if (temp >= default_debug_line_header.line_range) |
| 287 | return 0; |
| 288 | |
| 289 | delta_addr = (line->vma - last_vma) / |
| 290 | default_debug_line_header.minimum_instruction_length; |
| 291 | /* This is not sufficient to ensure opcode will be in [0-256] but |
| 292 | * sufficient to ensure when summing with the delta lineno we will |
| 293 | * not overflow the unsigned long opcode */ |
| 294 | if (delta_addr <= 256 / default_debug_line_header.line_range) { |
| 295 | unsigned long opcode = temp + |
| 296 | (delta_addr * default_debug_line_header.line_range) + |
| 297 | default_debug_line_header.opcode_base; |
| 298 | |
| 299 | return opcode <= 255 ? opcode : 0; |
| 300 | } |
| 301 | |
| 302 | return 0; |
| 303 | } |
| 304 | |
| 305 | |
| 306 | static void emit_lineno_info(struct growable_buffer * b, |
| 307 | struct debug_line_info const * line, size_t nr_entry, |
| 308 | unsigned long code_addr) |
| 309 | { |
| 310 | size_t i; |
| 311 | |
| 312 | /* |
| 313 | * Machine state at start of a statement program |
| 314 | * address = 0 |
| 315 | * file = 1 |
| 316 | * line = 1 |
| 317 | * column = 0 |
| 318 | * is_stmt = default_is_stmt as given in the debug_line_header |
| 319 | * basic block = 0 |
| 320 | * end sequence = 0 |
| 321 | */ |
| 322 | |
| 323 | /* start state of the state machine we take care of */ |
| 324 | unsigned long last_vma = code_addr; |
| 325 | unsigned int last_lineno = 1; |
| 326 | char const * cur_filename = NULL; |
| 327 | unsigned long cur_file_index = 0; |
| 328 | |
| 329 | /* FIXME: relocatable address? */ |
| 330 | emit_lne_set_address(b, (void const *)code_addr); |
| 331 | emit_advance_lineno(b, line[0].lineno - last_lineno); |
| 332 | last_lineno = line[0].lineno; |
| 333 | emit_lne_define_filename(b, line[0].filename); |
| 334 | cur_filename = line[0].filename; |
| 335 | emit_set_file(b, ++cur_file_index); |
| 336 | emit_opcode(b, DW_LNS_copy); |
| 337 | |
| 338 | |
| 339 | for (i = 0; i < nr_entry; i++) { |
| 340 | int need_copy = 0; |
| 341 | ubyte special_opcode; |
| 342 | |
| 343 | if (!cur_filename || strcmp(cur_filename, line[i].filename)) { |
| 344 | emit_lne_define_filename(b, line[i].filename); |
| 345 | cur_filename = line[i].filename; |
| 346 | emit_set_file(b, ++cur_file_index); |
| 347 | need_copy = 1; |
| 348 | } |
| 349 | if ((special_opcode = get_special_opcode(&line[i], |
| 350 | last_lineno, last_vma)) != 0) { |
| 351 | last_lineno = line[i].lineno; |
| 352 | last_vma = line[i].vma; |
| 353 | emit_opcode(b, special_opcode); |
| 354 | } else { |
| 355 | if (last_lineno != line[i].lineno) { |
| 356 | emit_advance_lineno(b, |
| 357 | line[i].lineno - last_lineno); |
| 358 | last_lineno = line[i].lineno; |
| 359 | need_copy = 1; |
| 360 | } |
| 361 | if (last_vma != line[i].vma) { |
| 362 | emit_advance_pc(b, line[i].vma - last_vma); |
| 363 | last_vma = line[i].vma; |
| 364 | need_copy = 1; |
| 365 | } |
| 366 | if (need_copy) |
| 367 | emit_opcode(b, DW_LNS_copy); |
| 368 | } |
| 369 | } |
| 370 | } |
| 371 | |
| 372 | |
| 373 | static void add_debug_line(struct growable_buffer * b, |
| 374 | struct debug_line_info const * line, size_t nr_entry, |
| 375 | unsigned long code_addr) |
| 376 | { |
| 377 | struct debug_line_header * dbg_header; |
| 378 | size_t old_size; |
| 379 | |
| 380 | old_size = b->size; |
| 381 | |
| 382 | add_data(b, &default_debug_line_header, |
| 383 | sizeof(default_debug_line_header)); |
| 384 | add_data(b, &standard_opcode_length, sizeof(standard_opcode_length)); |
| 385 | |
| 386 | // empty directory entry |
| 387 | add_data(b, "", 1); |
| 388 | |
| 389 | // empty filename directory |
| 390 | add_data(b, "", 1); |
| 391 | |
| 392 | dbg_header = b->p + old_size; |
| 393 | dbg_header->prolog_length = (b->size - old_size) - |
| 394 | offsetof(struct debug_line_header, minimum_instruction_length); |
| 395 | |
| 396 | emit_lineno_info(b, line, nr_entry, code_addr); |
| 397 | |
| 398 | emit_lne_end_of_sequence(b); |
| 399 | |
| 400 | dbg_header = b->p + old_size; |
| 401 | dbg_header->total_length = (b->size - old_size) - |
| 402 | offsetof(struct debug_line_header, version); |
| 403 | } |
| 404 | |
| 405 | |
| 406 | static void add_compilation_unit(struct growable_buffer * b, |
| 407 | size_t offset_debug_line) |
| 408 | { |
| 409 | struct compilation_unit_header * comp_unit_header; |
| 410 | |
| 411 | size_t old_size = b->size; |
| 412 | |
| 413 | add_data(b, &default_comp_unit_header, |
| 414 | sizeof(default_comp_unit_header)); |
| 415 | |
| 416 | emit_unsigned_LEB128(b, 1); |
| 417 | emit_uword(b, offset_debug_line); |
| 418 | |
| 419 | comp_unit_header = b->p + old_size; |
| 420 | comp_unit_header->total_length = (b->size - old_size) - |
| 421 | offsetof(struct compilation_unit_header, version); |
| 422 | comp_unit_header->pointer_size = sizeof(void *); |
| 423 | } |
| 424 | |
| 425 | |
| 426 | static void create_debug_abbrev(struct growable_buffer * b) |
| 427 | { |
| 428 | emit_unsigned_LEB128(b, 1); |
| 429 | emit_unsigned_LEB128(b, DW_TAG_compile_unit); |
| 430 | emit_unsigned_LEB128(b, DW_CHILDREN_yes); |
| 431 | emit_unsigned_LEB128(b, DW_AT_stmt_list); |
| 432 | emit_unsigned_LEB128(b, DW_FORM_data4); |
| 433 | emit_unsigned_LEB128(b, 0); |
| 434 | } |
| 435 | |
| 436 | static struct growable_buffer b_line; |
| 437 | static struct growable_buffer b_debug_info; |
| 438 | static struct growable_buffer b_debug_abbrev; |
| 439 | |
| 440 | int init_debug_line_info(bfd * abfd) |
| 441 | { |
| 442 | asection * line_section, * debug_info, * debug_abbrev; |
| 443 | struct jitentry_debug_line * debug_line; |
| 444 | |
| 445 | init_buffer(&b_line); |
| 446 | init_buffer(&b_debug_info); |
| 447 | init_buffer(&b_debug_abbrev); |
| 448 | |
| 449 | for (debug_line = jitentry_debug_line_list; |
| 450 | debug_line; |
| 451 | debug_line = debug_line->next) { |
| 452 | struct jr_code_debug_info const * rec = debug_line->data; |
| 453 | if (rec->nr_entry) { |
| 454 | size_t i; |
| 455 | void const * data = rec + 1; |
| 456 | struct debug_line_info * dbg_line = |
| 457 | xmalloc(rec->nr_entry * |
| 458 | sizeof(struct debug_line_info)); |
| 459 | for (i = 0; i < rec->nr_entry; ++i) { |
| 460 | dbg_line[i].vma = *(unsigned long *)data; |
| 461 | data += sizeof(unsigned long); |
| 462 | dbg_line[i].lineno = *(unsigned int *)data; |
| 463 | data += sizeof(unsigned int); |
| 464 | dbg_line[i].filename = data; |
| 465 | data += strlen(data) + 1; |
| 466 | } |
| 467 | |
| 468 | add_compilation_unit(&b_debug_info, b_line.size); |
| 469 | add_debug_line(&b_line, dbg_line, |
| 470 | rec->nr_entry, rec->code_addr); |
| 471 | create_debug_abbrev(&b_debug_abbrev); |
| 472 | |
| 473 | free(dbg_line); |
| 474 | } |
| 475 | } |
| 476 | |
| 477 | line_section = create_section(abfd, ".debug_line", b_line.size, 0, |
| 478 | SEC_HAS_CONTENTS|SEC_READONLY|SEC_DEBUGGING); |
| 479 | if (!line_section) |
| 480 | return -1; |
| 481 | |
| 482 | debug_info = create_section(abfd, ".debug_info", b_debug_info.size, 0, |
| 483 | SEC_HAS_CONTENTS|SEC_READONLY|SEC_DEBUGGING); |
| 484 | if (!debug_info) |
| 485 | return -1; |
| 486 | |
| 487 | debug_abbrev = create_section(abfd, ".debug_abbrev", |
| 488 | b_debug_abbrev.size, 0, |
| 489 | SEC_HAS_CONTENTS|SEC_READONLY|SEC_DEBUGGING); |
| 490 | if (!debug_abbrev) |
| 491 | return -1; |
| 492 | |
| 493 | return 0; |
| 494 | } |
| 495 | |
| 496 | |
| 497 | int finalize_debug_line_info(bfd * abfd) |
| 498 | { |
| 499 | asection * line_section, * debug_info, * debug_abbrev; |
| 500 | |
| 501 | line_section = bfd_get_section_by_name(abfd, ".debug_line"); |
| 502 | if (!line_section) |
| 503 | return -1; |
| 504 | |
| 505 | debug_info = bfd_get_section_by_name(abfd, ".debug_info"); |
| 506 | if (!debug_info) |
| 507 | return -1; |
| 508 | |
| 509 | |
| 510 | debug_abbrev = bfd_get_section_by_name(abfd, ".debug_abbrev"); |
| 511 | if (!debug_abbrev) |
| 512 | return -1; |
| 513 | |
| 514 | fill_section_content(abfd, line_section, b_line.p, 0, b_line.size); |
| 515 | fill_section_content(abfd, debug_info, b_debug_info.p, |
| 516 | 0, b_debug_info.size); |
| 517 | fill_section_content(abfd, debug_abbrev, b_debug_abbrev.p, 0, |
| 518 | b_debug_abbrev.size); |
| 519 | |
| 520 | |
| 521 | free_buffer(&b_line); |
| 522 | free_buffer(&b_debug_info); |
| 523 | free_buffer(&b_debug_abbrev); |
| 524 | |
| 525 | return 0; |
| 526 | } |