1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 /* 29 * DWARF to tdata conversion 30 * 31 * For the most part, conversion is straightforward, proceeding in two passes. 32 * On the first pass, we iterate through every die, creating new type nodes as 33 * necessary. Referenced tdesc_t's are created in an uninitialized state, thus 34 * allowing type reference pointers to be filled in. If the tdesc_t 35 * corresponding to a given die can be completely filled out (sizes and offsets 36 * calculated, and so forth) without using any referenced types, the tdesc_t is 37 * marked as resolved. Consider an array type. If the type corresponding to 38 * the array contents has not yet been processed, we will create a blank tdesc 39 * for the contents type (only the type ID will be filled in, relying upon the 40 * later portion of the first pass to encounter and complete the referenced 41 * type). We will then attempt to determine the size of the array. If the 42 * array has a byte size attribute, we will have completely characterized the 43 * array type, and will be able to mark it as resolved. The lack of a byte 44 * size attribute, on the other hand, will prevent us from fully resolving the 45 * type, as the size will only be calculable with reference to the contents 46 * type, which has not, as yet, been encountered. The array type will thus be 47 * left without the resolved flag, and the first pass will continue. 48 * 49 * When we begin the second pass, we will have created tdesc_t nodes for every 50 * type in the section. We will traverse the tree, from the iidescs down, 51 * processing each unresolved node. As the referenced nodes will have been 52 * populated, the array type used in our example above will be able to use the 53 * size of the referenced types (if available) to determine its own type. The 54 * traversal will be repeated until all types have been resolved or we have 55 * failed to make progress. When all tdescs have been resolved, the conversion 56 * is complete. 57 * 58 * There are, as always, a few special cases that are handled during the first 59 * and second passes: 60 * 61 * 1. Empty enums - GCC will occasionally emit an enum without any members. 62 * Later on in the file, it will emit the same enum type, though this time 63 * with the full complement of members. All references to the memberless 64 * enum need to be redirected to the full definition. During the first 65 * pass, each enum is entered in dm_enumhash, along with a pointer to its 66 * corresponding tdesc_t. If, during the second pass, we encounter a 67 * memberless enum, we use the hash to locate the full definition. All 68 * tdescs referencing the empty enum are then redirected. 69 * 70 * 2. Forward declarations - If the compiler sees a forward declaration for 71 * a structure, followed by the definition of that structure, it will emit 72 * DWARF data for both the forward declaration and the definition. We need 73 * to resolve the forward declarations when possible, by redirecting 74 * forward-referencing tdescs to the actual struct/union definitions. This 75 * redirection is done completely within the first pass. We begin by 76 * recording all forward declarations in dw_fwdhash. When we define a 77 * structure, we check to see if there have been any corresponding forward 78 * declarations. If so, we redirect the tdescs which referenced the forward 79 * declarations to the structure or union definition. 80 * 81 * XXX see if a post traverser will allow the elimination of repeated pass 2 82 * traversals. 83 */ 84 85 #include <stdio.h> 86 #include <stdlib.h> 87 #include <string.h> 88 #include <strings.h> 89 #include <errno.h> 90 #include <libelf.h> 91 #include <libdwarf.h> 92 #include <libgen.h> 93 #include <dwarf.h> 94 95 #include "ctf_headers.h" 96 #include "ctftools.h" 97 #include "memory.h" 98 #include "list.h" 99 #include "traverse.h" 100 101 /* The version of DWARF which we support. */ 102 #define DWARF_VERSION 2 103 104 /* 105 * We need to define a couple of our own intrinsics, to smooth out some of the 106 * differences between the GCC and DevPro DWARF emitters. See the referenced 107 * routines and the special cases in the file comment for more details. 108 * 109 * Type IDs are 32 bits wide. We're going to use the top of that field to 110 * indicate types that we've created ourselves. 111 */ 112 #define TID_FILEMAX 0x3fffffff /* highest tid from file */ 113 #define TID_VOID 0x40000001 /* see die_void() */ 114 #define TID_LONG 0x40000002 /* see die_array() */ 115 116 #define TID_MFGTID_BASE 0x40000003 /* first mfg'd tid */ 117 118 /* 119 * To reduce the staggering amount of error-handling code that would otherwise 120 * be required, the attribute-retrieval routines handle most of their own 121 * errors. If the following flag is supplied as the value of the `req' 122 * argument, they will also handle the absence of a requested attribute by 123 * terminating the program. 124 */ 125 #define DW_ATTR_REQ 1 126 127 #define TDESC_HASH_BUCKETS 511 128 129 typedef struct dwarf { 130 Dwarf_Debug dw_dw; /* for libdwarf */ 131 Dwarf_Error dw_err; /* for libdwarf */ 132 Dwarf_Off dw_maxoff; /* highest legal offset in this cu */ 133 tdata_t *dw_td; /* root of the tdesc/iidesc tree */ 134 hash_t *dw_tidhash; /* hash of tdescs by t_id */ 135 hash_t *dw_fwdhash; /* hash of fwd decls by name */ 136 hash_t *dw_enumhash; /* hash of memberless enums by name */ 137 tdesc_t *dw_void; /* manufactured void type */ 138 tdesc_t *dw_long; /* manufactured long type for arrays */ 139 size_t dw_ptrsz; /* size of a pointer in this file */ 140 tid_t dw_mfgtid_last; /* last mfg'd type ID used */ 141 uint_t dw_nunres; /* count of unresolved types */ 142 char *dw_cuname; /* name of compilation unit */ 143 } dwarf_t; 144 145 static void die_create_one(dwarf_t *, Dwarf_Die); 146 static void die_create(dwarf_t *, Dwarf_Die); 147 148 static tid_t 149 mfgtid_next(dwarf_t *dw) 150 { 151 return (++dw->dw_mfgtid_last); 152 } 153 154 static void 155 tdesc_add(dwarf_t *dw, tdesc_t *tdp) 156 { 157 hash_add(dw->dw_tidhash, tdp); 158 } 159 160 static tdesc_t * 161 tdesc_lookup(dwarf_t *dw, int tid) 162 { 163 tdesc_t tmpl; 164 void *tdp; 165 166 tmpl.t_id = tid; 167 168 if (hash_find(dw->dw_tidhash, &tmpl, &tdp)) 169 return (tdp); 170 else 171 return (NULL); 172 } 173 174 /* 175 * Resolve a tdesc down to a node which should have a size. Returns the size, 176 * zero if the size hasn't yet been determined. 177 */ 178 static size_t 179 tdesc_size(tdesc_t *tdp) 180 { 181 for (;;) { 182 switch (tdp->t_type) { 183 case INTRINSIC: 184 case POINTER: 185 case ARRAY: 186 case FUNCTION: 187 case STRUCT: 188 case UNION: 189 case ENUM: 190 return (tdp->t_size); 191 192 case FORWARD: 193 return (0); 194 195 case TYPEDEF: 196 case VOLATILE: 197 case CONST: 198 case RESTRICT: 199 tdp = tdp->t_tdesc; 200 continue; 201 202 case 0: /* not yet defined */ 203 return (0); 204 205 default: 206 terminate("tdp %u: tdesc_size on unknown type %d\n", 207 tdp->t_id, tdp->t_type); 208 } 209 } 210 } 211 212 static size_t 213 tdesc_bitsize(tdesc_t *tdp) 214 { 215 for (;;) { 216 switch (tdp->t_type) { 217 case INTRINSIC: 218 return (tdp->t_intr->intr_nbits); 219 220 case ARRAY: 221 case FUNCTION: 222 case STRUCT: 223 case UNION: 224 case ENUM: 225 case POINTER: 226 return (tdp->t_size * NBBY); 227 228 case FORWARD: 229 return (0); 230 231 case TYPEDEF: 232 case VOLATILE: 233 case RESTRICT: 234 case CONST: 235 tdp = tdp->t_tdesc; 236 continue; 237 238 case 0: /* not yet defined */ 239 return (0); 240 241 default: 242 terminate("tdp %u: tdesc_bitsize on unknown type %d\n", 243 tdp->t_id, tdp->t_type); 244 } 245 } 246 } 247 248 static tdesc_t * 249 tdesc_basetype(tdesc_t *tdp) 250 { 251 for (;;) { 252 switch (tdp->t_type) { 253 case TYPEDEF: 254 case VOLATILE: 255 case RESTRICT: 256 case CONST: 257 tdp = tdp->t_tdesc; 258 break; 259 case 0: /* not yet defined */ 260 return (NULL); 261 default: 262 return (tdp); 263 } 264 } 265 } 266 267 static Dwarf_Off 268 die_off(dwarf_t *dw, Dwarf_Die die) 269 { 270 Dwarf_Off off; 271 272 if (dwarf_dieoffset(die, &off, &dw->dw_err) == DW_DLV_OK) 273 return (off); 274 275 terminate("failed to get offset for die: %s\n", 276 dwarf_errmsg(&dw->dw_err)); 277 /*NOTREACHED*/ 278 return (0); 279 } 280 281 static Dwarf_Die 282 die_sibling(dwarf_t *dw, Dwarf_Die die) 283 { 284 Dwarf_Die sib; 285 int rc; 286 287 if ((rc = dwarf_siblingof(dw->dw_dw, die, &sib, &dw->dw_err)) == 288 DW_DLV_OK) 289 return (sib); 290 else if (rc == DW_DLV_NO_ENTRY) 291 return (NULL); 292 293 terminate("die %llu: failed to find type sibling: %s\n", 294 die_off(dw, die), dwarf_errmsg(&dw->dw_err)); 295 /*NOTREACHED*/ 296 return (NULL); 297 } 298 299 static Dwarf_Die 300 die_child(dwarf_t *dw, Dwarf_Die die) 301 { 302 Dwarf_Die child; 303 int rc; 304 305 if ((rc = dwarf_child(die, &child, &dw->dw_err)) == DW_DLV_OK) 306 return (child); 307 else if (rc == DW_DLV_NO_ENTRY) 308 return (NULL); 309 310 terminate("die %llu: failed to find type child: %s\n", 311 die_off(dw, die), dwarf_errmsg(&dw->dw_err)); 312 /*NOTREACHED*/ 313 return (NULL); 314 } 315 316 static Dwarf_Half 317 die_tag(dwarf_t *dw, Dwarf_Die die) 318 { 319 Dwarf_Half tag; 320 321 if (dwarf_tag(die, &tag, &dw->dw_err) == DW_DLV_OK) 322 return (tag); 323 324 terminate("die %llu: failed to get tag for type: %s\n", 325 die_off(dw, die), dwarf_errmsg(&dw->dw_err)); 326 /*NOTREACHED*/ 327 return (0); 328 } 329 330 static Dwarf_Attribute 331 die_attr(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name, int req) 332 { 333 Dwarf_Attribute attr; 334 int rc; 335 336 if ((rc = dwarf_attr(die, name, &attr, &dw->dw_err)) == DW_DLV_OK) { 337 return (attr); 338 } else if (rc == DW_DLV_NO_ENTRY) { 339 if (req) { 340 terminate("die %llu: no attr 0x%x\n", die_off(dw, die), 341 name); 342 } else { 343 return (NULL); 344 } 345 } 346 347 terminate("die %llu: failed to get attribute for type: %s\n", 348 die_off(dw, die), dwarf_errmsg(&dw->dw_err)); 349 /*NOTREACHED*/ 350 return (NULL); 351 } 352 353 static int 354 die_signed(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name, Dwarf_Signed *valp, 355 int req) 356 { 357 *valp = 0; 358 if (dwarf_attrval_signed(die, name, valp, &dw->dw_err) != DWARF_E_NONE) { 359 if (req) 360 terminate("die %llu: failed to get signed: %s\n", 361 die_off(dw, die), dwarf_errmsg(&dw->dw_err)); 362 return (0); 363 } 364 365 return (1); 366 } 367 368 static int 369 die_unsigned(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name, Dwarf_Unsigned *valp, 370 int req) 371 { 372 *valp = 0; 373 if (dwarf_attrval_unsigned(die, name, valp, &dw->dw_err) != DWARF_E_NONE) { 374 if (req) 375 terminate("die %llu: failed to get unsigned: %s\n", 376 die_off(dw, die), dwarf_errmsg(&dw->dw_err)); 377 return (0); 378 } 379 380 return (1); 381 } 382 383 static int 384 die_bool(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name, Dwarf_Bool *valp, int req) 385 { 386 *valp = 0; 387 388 if (dwarf_attrval_flag(die, name, valp, &dw->dw_err) != DWARF_E_NONE) { 389 if (req) 390 terminate("die %llu: failed to get flag: %s\n", 391 die_off(dw, die), dwarf_errmsg(&dw->dw_err)); 392 return (0); 393 } 394 395 return (1); 396 } 397 398 static int 399 die_string(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name, char **strp, int req) 400 { 401 const char *str = NULL; 402 403 if (dwarf_attrval_string(die, name, &str, &dw->dw_err) != DWARF_E_NONE || 404 str == NULL) { 405 if (req) 406 terminate("die %llu: failed to get string: %s\n", 407 die_off(dw, die), dwarf_errmsg(&dw->dw_err)); 408 else 409 *strp = NULL; 410 return (0); 411 } else 412 *strp = xstrdup(str); 413 414 return (1); 415 } 416 417 static Dwarf_Off 418 die_attr_ref(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name) 419 { 420 Dwarf_Off off; 421 422 if (dwarf_attrval_unsigned(die, name, &off, &dw->dw_err) != DWARF_E_NONE) { 423 terminate("die %llu: failed to get ref: %s\n", 424 die_off(dw, die), dwarf_errmsg(&dw->dw_err)); 425 } 426 427 return (off); 428 } 429 430 static char * 431 die_name(dwarf_t *dw, Dwarf_Die die) 432 { 433 char *str = NULL; 434 435 (void) die_string(dw, die, DW_AT_name, &str, 0); 436 437 return (str); 438 } 439 440 static int 441 die_isdecl(dwarf_t *dw, Dwarf_Die die) 442 { 443 Dwarf_Bool val; 444 445 return (die_bool(dw, die, DW_AT_declaration, &val, 0) && val); 446 } 447 448 static int 449 die_isglobal(dwarf_t *dw, Dwarf_Die die) 450 { 451 Dwarf_Signed vis; 452 Dwarf_Bool ext; 453 454 /* 455 * Some compilers (gcc) use DW_AT_external to indicate function 456 * visibility. Others (Sun) use DW_AT_visibility. 457 */ 458 if (die_signed(dw, die, DW_AT_visibility, &vis, 0)) 459 return (vis == DW_VIS_exported); 460 else 461 return (die_bool(dw, die, DW_AT_external, &ext, 0) && ext); 462 } 463 464 static tdesc_t * 465 die_add(dwarf_t *dw, Dwarf_Off off) 466 { 467 tdesc_t *tdp = xcalloc(sizeof (tdesc_t)); 468 469 tdp->t_id = off; 470 471 tdesc_add(dw, tdp); 472 473 return (tdp); 474 } 475 476 static tdesc_t * 477 die_lookup_pass1(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name) 478 { 479 Dwarf_Off ref = die_attr_ref(dw, die, name); 480 tdesc_t *tdp; 481 482 if ((tdp = tdesc_lookup(dw, ref)) != NULL) 483 return (tdp); 484 485 return (die_add(dw, ref)); 486 } 487 488 static int 489 die_mem_offset(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name, 490 Dwarf_Unsigned *valp, int req __unused) 491 { 492 Dwarf_Locdesc *loc = NULL; 493 Dwarf_Signed locnum = 0; 494 495 if (dwarf_locdesc(die, name, &loc, &locnum, &dw->dw_err) != DW_DLV_OK) 496 return (0); 497 498 if (locnum != 1 || loc->ld_s->lr_atom != DW_OP_plus_uconst) { 499 terminate("die %llu: cannot parse member offset\n", 500 die_off(dw, die)); 501 } 502 503 *valp = loc->ld_s->lr_number; 504 505 if (loc != NULL) 506 if (dwarf_locdesc_free(loc, &dw->dw_err) != DW_DLV_OK) 507 terminate("die %llu: cannot free location descriptor: %s\n", 508 die_off(dw, die), dwarf_errmsg(&dw->dw_err)); 509 510 return (1); 511 } 512 513 static tdesc_t * 514 tdesc_intr_common(dwarf_t *dw, int tid, const char *name, size_t sz) 515 { 516 tdesc_t *tdp; 517 intr_t *intr; 518 519 intr = xcalloc(sizeof (intr_t)); 520 intr->intr_type = INTR_INT; 521 intr->intr_signed = 1; 522 intr->intr_nbits = sz * NBBY; 523 524 tdp = xcalloc(sizeof (tdesc_t)); 525 tdp->t_name = xstrdup(name); 526 tdp->t_size = sz; 527 tdp->t_id = tid; 528 tdp->t_type = INTRINSIC; 529 tdp->t_intr = intr; 530 tdp->t_flags = TDESC_F_RESOLVED; 531 532 tdesc_add(dw, tdp); 533 534 return (tdp); 535 } 536 537 /* 538 * Manufacture a void type. Used for gcc-emitted stabs, where the lack of a 539 * type reference implies a reference to a void type. A void *, for example 540 * will be represented by a pointer die without a DW_AT_type. CTF requires 541 * that pointer nodes point to something, so we'll create a void for use as 542 * the target. Note that the DWARF data may already create a void type. Ours 543 * would then be a duplicate, but it'll be removed in the self-uniquification 544 * merge performed at the completion of DWARF->tdesc conversion. 545 */ 546 static tdesc_t * 547 tdesc_intr_void(dwarf_t *dw) 548 { 549 if (dw->dw_void == NULL) 550 dw->dw_void = tdesc_intr_common(dw, TID_VOID, "void", 0); 551 552 return (dw->dw_void); 553 } 554 555 static tdesc_t * 556 tdesc_intr_long(dwarf_t *dw) 557 { 558 if (dw->dw_long == NULL) { 559 dw->dw_long = tdesc_intr_common(dw, TID_LONG, "long", 560 dw->dw_ptrsz); 561 } 562 563 return (dw->dw_long); 564 } 565 566 /* 567 * Used for creating bitfield types. We create a copy of an existing intrinsic, 568 * adjusting the size of the copy to match what the caller requested. The 569 * caller can then use the copy as the type for a bitfield structure member. 570 */ 571 static tdesc_t * 572 tdesc_intr_clone(dwarf_t *dw, tdesc_t *old, size_t bitsz) 573 { 574 tdesc_t *new = xcalloc(sizeof (tdesc_t)); 575 576 if (!(old->t_flags & TDESC_F_RESOLVED)) { 577 terminate("tdp %u: attempt to make a bit field from an " 578 "unresolved type\n", old->t_id); 579 } 580 581 new->t_name = xstrdup(old->t_name); 582 new->t_size = old->t_size; 583 new->t_id = mfgtid_next(dw); 584 new->t_type = INTRINSIC; 585 new->t_flags = TDESC_F_RESOLVED; 586 587 new->t_intr = xcalloc(sizeof (intr_t)); 588 bcopy(old->t_intr, new->t_intr, sizeof (intr_t)); 589 new->t_intr->intr_nbits = bitsz; 590 591 tdesc_add(dw, new); 592 593 return (new); 594 } 595 596 static void 597 tdesc_array_create(dwarf_t *dw, Dwarf_Die dim, tdesc_t *arrtdp, 598 tdesc_t *dimtdp) 599 { 600 Dwarf_Unsigned uval; 601 Dwarf_Signed sval; 602 tdesc_t *ctdp = NULL; 603 Dwarf_Die dim2; 604 ardef_t *ar; 605 606 if ((dim2 = die_sibling(dw, dim)) == NULL) { 607 ctdp = arrtdp; 608 } else if (die_tag(dw, dim2) == DW_TAG_subrange_type) { 609 ctdp = xcalloc(sizeof (tdesc_t)); 610 ctdp->t_id = mfgtid_next(dw); 611 debug(3, "die %llu: creating new type %u for sub-dimension\n", 612 die_off(dw, dim2), ctdp->t_id); 613 tdesc_array_create(dw, dim2, arrtdp, ctdp); 614 } else { 615 terminate("die %llu: unexpected non-subrange node in array\n", 616 die_off(dw, dim2)); 617 } 618 619 dimtdp->t_type = ARRAY; 620 dimtdp->t_ardef = ar = xcalloc(sizeof (ardef_t)); 621 622 /* 623 * Array bounds can be signed or unsigned, but there are several kinds 624 * of signless forms (data1, data2, etc) that take their sign from the 625 * routine that is trying to interpret them. That is, data1 can be 626 * either signed or unsigned, depending on whether you use the signed or 627 * unsigned accessor function. GCC will use the signless forms to store 628 * unsigned values which have their high bit set, so we need to try to 629 * read them first as unsigned to get positive values. We could also 630 * try signed first, falling back to unsigned if we got a negative 631 * value. 632 */ 633 if (die_unsigned(dw, dim, DW_AT_upper_bound, &uval, 0)) 634 ar->ad_nelems = uval + 1; 635 else if (die_signed(dw, dim, DW_AT_upper_bound, &sval, 0)) 636 ar->ad_nelems = sval + 1; 637 else 638 ar->ad_nelems = 0; 639 640 /* 641 * Different compilers use different index types. Force the type to be 642 * a common, known value (long). 643 */ 644 ar->ad_idxtype = tdesc_intr_long(dw); 645 ar->ad_contents = ctdp; 646 647 if (ar->ad_contents->t_size != 0) { 648 dimtdp->t_size = ar->ad_contents->t_size * ar->ad_nelems; 649 dimtdp->t_flags |= TDESC_F_RESOLVED; 650 } 651 } 652 653 /* 654 * Create a tdesc from an array node. Some arrays will come with byte size 655 * attributes, and thus can be resolved immediately. Others don't, and will 656 * need to wait until the second pass for resolution. 657 */ 658 static void 659 die_array_create(dwarf_t *dw, Dwarf_Die arr, Dwarf_Off off, tdesc_t *tdp) 660 { 661 tdesc_t *arrtdp = die_lookup_pass1(dw, arr, DW_AT_type); 662 Dwarf_Unsigned uval; 663 Dwarf_Die dim; 664 665 debug(3, "die %llu <%llx>: creating array\n", off, off); 666 667 if ((dim = die_child(dw, arr)) == NULL || 668 die_tag(dw, dim) != DW_TAG_subrange_type) 669 terminate("die %llu: failed to retrieve array bounds\n", off); 670 671 tdesc_array_create(dw, dim, arrtdp, tdp); 672 673 if (die_unsigned(dw, arr, DW_AT_byte_size, &uval, 0)) { 674 tdesc_t *dimtdp; 675 int flags; 676 677 /* Check for bogus gcc DW_AT_byte_size attribute */ 678 if (uval == (unsigned)-1) { 679 printf("dwarf.c:%s() working around bogus -1 DW_AT_byte_size\n", 680 __func__); 681 uval = 0; 682 } 683 684 tdp->t_size = uval; 685 686 /* 687 * Ensure that sub-dimensions have sizes too before marking 688 * as resolved. 689 */ 690 flags = TDESC_F_RESOLVED; 691 for (dimtdp = tdp->t_ardef->ad_contents; 692 dimtdp->t_type == ARRAY; 693 dimtdp = dimtdp->t_ardef->ad_contents) { 694 if (!(dimtdp->t_flags & TDESC_F_RESOLVED)) { 695 flags = 0; 696 break; 697 } 698 } 699 700 tdp->t_flags |= flags; 701 } 702 703 debug(3, "die %llu <%llx>: array nelems %u size %u\n", off, off, 704 tdp->t_ardef->ad_nelems, tdp->t_size); 705 } 706 707 /*ARGSUSED1*/ 708 static int 709 die_array_resolve(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private) 710 { 711 dwarf_t *dw = private; 712 size_t sz; 713 714 if (tdp->t_flags & TDESC_F_RESOLVED) 715 return (1); 716 717 debug(3, "trying to resolve array %d (cont %d)\n", tdp->t_id, 718 tdp->t_ardef->ad_contents->t_id); 719 720 if ((sz = tdesc_size(tdp->t_ardef->ad_contents)) == 0) { 721 debug(3, "unable to resolve array %s (%d) contents %d\n", 722 tdesc_name(tdp), tdp->t_id, 723 tdp->t_ardef->ad_contents->t_id); 724 725 dw->dw_nunres++; 726 return (1); 727 } 728 729 tdp->t_size = sz * tdp->t_ardef->ad_nelems; 730 tdp->t_flags |= TDESC_F_RESOLVED; 731 732 debug(3, "resolved array %d: %u bytes\n", tdp->t_id, tdp->t_size); 733 734 return (1); 735 } 736 737 /*ARGSUSED1*/ 738 static int 739 die_array_failed(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private __unused) 740 { 741 tdesc_t *cont = tdp->t_ardef->ad_contents; 742 743 if (tdp->t_flags & TDESC_F_RESOLVED) 744 return (1); 745 746 fprintf(stderr, "Array %d: failed to size contents type %s (%d)\n", 747 tdp->t_id, tdesc_name(cont), cont->t_id); 748 749 return (1); 750 } 751 752 /* 753 * Most enums (those with members) will be resolved during this first pass. 754 * Others - those without members (see the file comment) - won't be, and will 755 * need to wait until the second pass when they can be matched with their full 756 * definitions. 757 */ 758 static void 759 die_enum_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 760 { 761 Dwarf_Die mem; 762 Dwarf_Unsigned uval; 763 Dwarf_Signed sval; 764 765 debug(3, "die %llu: creating enum\n", off); 766 767 tdp->t_type = ENUM; 768 769 (void) die_unsigned(dw, die, DW_AT_byte_size, &uval, DW_ATTR_REQ); 770 /* Check for bogus gcc DW_AT_byte_size attribute */ 771 if (uval == (unsigned)-1) { 772 printf("dwarf.c:%s() working around bogus -1 DW_AT_byte_size\n", 773 __func__); 774 uval = 0; 775 } 776 tdp->t_size = uval; 777 778 if ((mem = die_child(dw, die)) != NULL) { 779 elist_t **elastp = &tdp->t_emem; 780 781 do { 782 elist_t *el; 783 784 if (die_tag(dw, mem) != DW_TAG_enumerator) { 785 /* Nested type declaration */ 786 die_create_one(dw, mem); 787 continue; 788 } 789 790 el = xcalloc(sizeof (elist_t)); 791 el->el_name = die_name(dw, mem); 792 793 if (die_signed(dw, mem, DW_AT_const_value, &sval, 0)) { 794 el->el_number = sval; 795 } else if (die_unsigned(dw, mem, DW_AT_const_value, 796 &uval, 0)) { 797 el->el_number = uval; 798 } else { 799 terminate("die %llu: enum %llu: member without " 800 "value\n", off, die_off(dw, mem)); 801 } 802 803 debug(3, "die %llu: enum %llu: created %s = %d\n", off, 804 die_off(dw, mem), el->el_name, el->el_number); 805 806 *elastp = el; 807 elastp = &el->el_next; 808 809 } while ((mem = die_sibling(dw, mem)) != NULL); 810 811 hash_add(dw->dw_enumhash, tdp); 812 813 tdp->t_flags |= TDESC_F_RESOLVED; 814 815 if (tdp->t_name != NULL) { 816 iidesc_t *ii = xcalloc(sizeof (iidesc_t)); 817 ii->ii_type = II_SOU; 818 ii->ii_name = xstrdup(tdp->t_name); 819 ii->ii_dtype = tdp; 820 821 iidesc_add(dw->dw_td->td_iihash, ii); 822 } 823 } 824 } 825 826 static int 827 die_enum_match(void *arg1, void *arg2) 828 { 829 tdesc_t *tdp = arg1, **fullp = arg2; 830 831 if (tdp->t_emem != NULL) { 832 *fullp = tdp; 833 return (-1); /* stop the iteration */ 834 } 835 836 return (0); 837 } 838 839 /*ARGSUSED1*/ 840 static int 841 die_enum_resolve(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private) 842 { 843 dwarf_t *dw = private; 844 tdesc_t *full = NULL; 845 846 if (tdp->t_flags & TDESC_F_RESOLVED) 847 return (1); 848 849 (void) hash_find_iter(dw->dw_enumhash, tdp, die_enum_match, &full); 850 851 /* 852 * The answer to this one won't change from iteration to iteration, 853 * so don't even try. 854 */ 855 if (full == NULL) { 856 terminate("tdp %u: enum %s has no members\n", tdp->t_id, 857 tdesc_name(tdp)); 858 } 859 860 debug(3, "tdp %u: enum %s redirected to %u\n", tdp->t_id, 861 tdesc_name(tdp), full->t_id); 862 863 tdp->t_flags |= TDESC_F_RESOLVED; 864 865 return (1); 866 } 867 868 static int 869 die_fwd_map(void *arg1, void *arg2) 870 { 871 tdesc_t *fwd = arg1, *sou = arg2; 872 873 debug(3, "tdp %u: mapped forward %s to sou %u\n", fwd->t_id, 874 tdesc_name(fwd), sou->t_id); 875 fwd->t_tdesc = sou; 876 877 return (0); 878 } 879 880 /* 881 * Structures and unions will never be resolved during the first pass, as we 882 * won't be able to fully determine the member sizes. The second pass, which 883 * have access to sizing information, will be able to complete the resolution. 884 */ 885 static void 886 die_sou_create(dwarf_t *dw, Dwarf_Die str, Dwarf_Off off, tdesc_t *tdp, 887 int type, const char *typename) 888 { 889 Dwarf_Unsigned sz, bitsz, bitoff, maxsz=0; 890 Dwarf_Die mem; 891 mlist_t *ml, **mlastp; 892 iidesc_t *ii; 893 894 tdp->t_type = (die_isdecl(dw, str) ? FORWARD : type); 895 896 debug(3, "die %llu: creating %s %s\n", off, 897 (tdp->t_type == FORWARD ? "forward decl" : typename), 898 tdesc_name(tdp)); 899 900 if (tdp->t_type == FORWARD) { 901 hash_add(dw->dw_fwdhash, tdp); 902 return; 903 } 904 905 (void) hash_find_iter(dw->dw_fwdhash, tdp, die_fwd_map, tdp); 906 907 (void) die_unsigned(dw, str, DW_AT_byte_size, &sz, DW_ATTR_REQ); 908 tdp->t_size = sz; 909 910 /* 911 * GCC allows empty SOUs as an extension. 912 */ 913 if ((mem = die_child(dw, str)) == NULL) { 914 goto out; 915 } 916 917 mlastp = &tdp->t_members; 918 919 do { 920 Dwarf_Off memoff = die_off(dw, mem); 921 Dwarf_Half tag = die_tag(dw, mem); 922 Dwarf_Unsigned mloff; 923 924 if (tag != DW_TAG_member) { 925 /* Nested type declaration */ 926 die_create_one(dw, mem); 927 continue; 928 } 929 930 debug(3, "die %llu: mem %llu: creating member\n", off, memoff); 931 932 ml = xcalloc(sizeof (mlist_t)); 933 934 /* 935 * This could be a GCC anon struct/union member, so we'll allow 936 * an empty name, even though nothing can really handle them 937 * properly. Note that some versions of GCC miss out debug 938 * info for anon structs, though recent versions are fixed (gcc 939 * bug 11816). 940 */ 941 if ((ml->ml_name = die_name(dw, mem)) == NULL) 942 ml->ml_name = NULL; 943 944 ml->ml_type = die_lookup_pass1(dw, mem, DW_AT_type); 945 debug(3, "die_sou_create(): ml_type = %p t_id = %d\n", 946 ml->ml_type, ml->ml_type->t_id); 947 948 if (die_mem_offset(dw, mem, DW_AT_data_member_location, 949 &mloff, 0)) { 950 debug(3, "die %llu: got mloff %llx\n", off, 951 (u_longlong_t)mloff); 952 ml->ml_offset = mloff * 8; 953 } 954 955 if (die_unsigned(dw, mem, DW_AT_bit_size, &bitsz, 0)) 956 ml->ml_size = bitsz; 957 else 958 ml->ml_size = tdesc_bitsize(ml->ml_type); 959 960 if (die_unsigned(dw, mem, DW_AT_bit_offset, &bitoff, 0)) { 961 #if BYTE_ORDER == _BIG_ENDIAN 962 ml->ml_offset += bitoff; 963 #else 964 ml->ml_offset += tdesc_bitsize(ml->ml_type) - bitoff - 965 ml->ml_size; 966 #endif 967 } 968 969 debug(3, "die %llu: mem %llu: created \"%s\" (off %u sz %u)\n", 970 off, memoff, ml->ml_name, ml->ml_offset, ml->ml_size); 971 972 *mlastp = ml; 973 mlastp = &ml->ml_next; 974 975 /* Find the size of the largest member to work around a gcc 976 * bug. See GCC Bugzilla 35998. 977 */ 978 if (maxsz < ml->ml_size) 979 maxsz = ml->ml_size; 980 981 } while ((mem = die_sibling(dw, mem)) != NULL); 982 983 /* See if we got a bogus DW_AT_byte_size. GCC will sometimes 984 * emit this. 985 */ 986 if (sz == (unsigned)-1) { 987 printf("dwarf.c:%s() working around bogus -1 DW_AT_byte_size\n", 988 __func__); 989 tdp->t_size = maxsz / 8; /* maxsz is in bits, t_size is bytes */ 990 } 991 992 /* 993 * GCC will attempt to eliminate unused types, thus decreasing the 994 * size of the emitted dwarf. That is, if you declare a foo_t in your 995 * header, include said header in your source file, and neglect to 996 * actually use (directly or indirectly) the foo_t in the source file, 997 * the foo_t won't make it into the emitted DWARF. So, at least, goes 998 * the theory. 999 * 1000 * Occasionally, it'll emit the DW_TAG_structure_type for the foo_t, 1001 * and then neglect to emit the members. Strangely, the loner struct 1002 * tag will always be followed by a proper nested declaration of 1003 * something else. This is clearly a bug, but we're not going to have 1004 * time to get it fixed before this goo goes back, so we'll have to work 1005 * around it. If we see a no-membered struct with a nested declaration 1006 * (i.e. die_child of the struct tag won't be null), we'll ignore it. 1007 * Being paranoid, we won't simply remove it from the hash. Instead, 1008 * we'll decline to create an iidesc for it, thus ensuring that this 1009 * type won't make it into the output file. To be safe, we'll also 1010 * change the name. 1011 */ 1012 if (tdp->t_members == NULL) { 1013 const char *old = tdesc_name(tdp); 1014 size_t newsz = 7 + strlen(old) + 1; 1015 char *new = xmalloc(newsz); 1016 (void) snprintf(new, newsz, "orphan %s", old); 1017 1018 debug(3, "die %llu: worked around %s %s\n", off, typename, old); 1019 1020 if (tdp->t_name != NULL) 1021 free(tdp->t_name); 1022 tdp->t_name = new; 1023 return; 1024 } 1025 1026 out: 1027 if (tdp->t_name != NULL) { 1028 ii = xcalloc(sizeof (iidesc_t)); 1029 ii->ii_type = II_SOU; 1030 ii->ii_name = xstrdup(tdp->t_name); 1031 ii->ii_dtype = tdp; 1032 1033 iidesc_add(dw->dw_td->td_iihash, ii); 1034 } 1035 } 1036 1037 static void 1038 die_struct_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1039 { 1040 die_sou_create(dw, die, off, tdp, STRUCT, "struct"); 1041 } 1042 1043 static void 1044 die_union_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1045 { 1046 die_sou_create(dw, die, off, tdp, UNION, "union"); 1047 } 1048 1049 /*ARGSUSED1*/ 1050 static int 1051 die_sou_resolve(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private) 1052 { 1053 dwarf_t *dw = private; 1054 mlist_t *ml; 1055 tdesc_t *mt; 1056 1057 if (tdp->t_flags & TDESC_F_RESOLVED) 1058 return (1); 1059 1060 debug(3, "resolving sou %s\n", tdesc_name(tdp)); 1061 1062 for (ml = tdp->t_members; ml != NULL; ml = ml->ml_next) { 1063 if (ml->ml_size == 0) { 1064 mt = tdesc_basetype(ml->ml_type); 1065 1066 if ((ml->ml_size = tdesc_bitsize(mt)) != 0) 1067 continue; 1068 1069 /* 1070 * For empty members, or GCC/C99 flexible array 1071 * members, a size of 0 is correct. 1072 */ 1073 if (mt->t_members == NULL) 1074 continue; 1075 if (mt->t_type == ARRAY && mt->t_ardef->ad_nelems == 0) 1076 continue; 1077 1078 dw->dw_nunres++; 1079 return (1); 1080 } 1081 1082 if ((mt = tdesc_basetype(ml->ml_type)) == NULL) { 1083 dw->dw_nunres++; 1084 return (1); 1085 } 1086 1087 if (ml->ml_size != 0 && mt->t_type == INTRINSIC && 1088 mt->t_intr->intr_nbits != (int)ml->ml_size) { 1089 /* 1090 * This member is a bitfield, and needs to reference 1091 * an intrinsic type with the same width. If the 1092 * currently-referenced type isn't of the same width, 1093 * we'll copy it, adjusting the width of the copy to 1094 * the size we'd like. 1095 */ 1096 debug(3, "tdp %u: creating bitfield for %d bits\n", 1097 tdp->t_id, ml->ml_size); 1098 1099 ml->ml_type = tdesc_intr_clone(dw, mt, ml->ml_size); 1100 } 1101 } 1102 1103 tdp->t_flags |= TDESC_F_RESOLVED; 1104 1105 return (1); 1106 } 1107 1108 /*ARGSUSED1*/ 1109 static int 1110 die_sou_failed(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private __unused) 1111 { 1112 const char *typename = (tdp->t_type == STRUCT ? "struct" : "union"); 1113 mlist_t *ml; 1114 1115 if (tdp->t_flags & TDESC_F_RESOLVED) 1116 return (1); 1117 1118 for (ml = tdp->t_members; ml != NULL; ml = ml->ml_next) { 1119 if (ml->ml_size == 0) { 1120 fprintf(stderr, "%s %d <%x>: failed to size member \"%s\" " 1121 "of type %s (%d <%x>)\n", typename, tdp->t_id, 1122 tdp->t_id, 1123 ml->ml_name, tdesc_name(ml->ml_type), 1124 ml->ml_type->t_id, ml->ml_type->t_id); 1125 } 1126 } 1127 1128 return (1); 1129 } 1130 1131 static void 1132 die_funcptr_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1133 { 1134 Dwarf_Attribute attr; 1135 Dwarf_Half tag; 1136 Dwarf_Die arg; 1137 fndef_t *fn; 1138 int i; 1139 1140 debug(3, "die %llu <%llx>: creating function pointer\n", off, off); 1141 1142 /* 1143 * We'll begin by processing any type definition nodes that may be 1144 * lurking underneath this one. 1145 */ 1146 for (arg = die_child(dw, die); arg != NULL; 1147 arg = die_sibling(dw, arg)) { 1148 if ((tag = die_tag(dw, arg)) != DW_TAG_formal_parameter && 1149 tag != DW_TAG_unspecified_parameters) { 1150 /* Nested type declaration */ 1151 die_create_one(dw, arg); 1152 } 1153 } 1154 1155 if (die_isdecl(dw, die)) { 1156 /* 1157 * This is a prototype. We don't add prototypes to the 1158 * tree, so we're going to drop the tdesc. Unfortunately, 1159 * it has already been added to the tree. Nobody will reference 1160 * it, though, and it will be leaked. 1161 */ 1162 return; 1163 } 1164 1165 fn = xcalloc(sizeof (fndef_t)); 1166 1167 tdp->t_type = FUNCTION; 1168 1169 if ((attr = die_attr(dw, die, DW_AT_type, 0)) != NULL) { 1170 fn->fn_ret = die_lookup_pass1(dw, die, DW_AT_type); 1171 } else { 1172 fn->fn_ret = tdesc_intr_void(dw); 1173 } 1174 1175 /* 1176 * Count the arguments to the function, then read them in. 1177 */ 1178 for (fn->fn_nargs = 0, arg = die_child(dw, die); arg != NULL; 1179 arg = die_sibling(dw, arg)) { 1180 if ((tag = die_tag(dw, arg)) == DW_TAG_formal_parameter) 1181 fn->fn_nargs++; 1182 else if (tag == DW_TAG_unspecified_parameters && 1183 fn->fn_nargs > 0) 1184 fn->fn_vargs = 1; 1185 } 1186 1187 if (fn->fn_nargs != 0) { 1188 debug(3, "die %llu: adding %d argument%s\n", off, fn->fn_nargs, 1189 (fn->fn_nargs > 1 ? "s" : "")); 1190 1191 fn->fn_args = xcalloc(sizeof (tdesc_t *) * fn->fn_nargs); 1192 for (i = 0, arg = die_child(dw, die); 1193 arg != NULL && i < (int) fn->fn_nargs; 1194 arg = die_sibling(dw, arg)) { 1195 if (die_tag(dw, arg) != DW_TAG_formal_parameter) 1196 continue; 1197 1198 fn->fn_args[i++] = die_lookup_pass1(dw, arg, 1199 DW_AT_type); 1200 } 1201 } 1202 1203 tdp->t_fndef = fn; 1204 tdp->t_flags |= TDESC_F_RESOLVED; 1205 } 1206 1207 /* 1208 * GCC and DevPro use different names for the base types. While the terms are 1209 * the same, they are arranged in a different order. Some terms, such as int, 1210 * are implied in one, and explicitly named in the other. Given a base type 1211 * as input, this routine will return a common name, along with an intr_t 1212 * that reflects said name. 1213 */ 1214 static intr_t * 1215 die_base_name_parse(const char *name, char **newp) 1216 { 1217 char buf[100]; 1218 char const *base; 1219 char *c; 1220 int nlong = 0, nshort = 0, nchar = 0, nint = 0; 1221 int sign = 1; 1222 char fmt = '\0'; 1223 intr_t *intr; 1224 1225 if (strlen(name) > sizeof (buf) - 1) 1226 terminate("base type name \"%s\" is too long\n", name); 1227 1228 strncpy(buf, name, sizeof (buf)); 1229 1230 for (c = strtok(buf, " "); c != NULL; c = strtok(NULL, " ")) { 1231 if (strcmp(c, "signed") == 0) 1232 sign = 1; 1233 else if (strcmp(c, "unsigned") == 0) 1234 sign = 0; 1235 else if (strcmp(c, "long") == 0) 1236 nlong++; 1237 else if (strcmp(c, "char") == 0) { 1238 nchar++; 1239 fmt = 'c'; 1240 } else if (strcmp(c, "short") == 0) 1241 nshort++; 1242 else if (strcmp(c, "int") == 0) 1243 nint++; 1244 else { 1245 /* 1246 * If we don't recognize any of the tokens, we'll tell 1247 * the caller to fall back to the dwarf-provided 1248 * encoding information. 1249 */ 1250 return (NULL); 1251 } 1252 } 1253 1254 if (nchar > 1 || nshort > 1 || nint > 1 || nlong > 2) 1255 return (NULL); 1256 1257 if (nchar > 0) { 1258 if (nlong > 0 || nshort > 0 || nint > 0) 1259 return (NULL); 1260 1261 base = "char"; 1262 1263 } else if (nshort > 0) { 1264 if (nlong > 0) 1265 return (NULL); 1266 1267 base = "short"; 1268 1269 } else if (nlong > 0) { 1270 base = "long"; 1271 1272 } else { 1273 base = "int"; 1274 } 1275 1276 intr = xcalloc(sizeof (intr_t)); 1277 intr->intr_type = INTR_INT; 1278 intr->intr_signed = sign; 1279 intr->intr_iformat = fmt; 1280 1281 snprintf(buf, sizeof (buf), "%s%s%s", 1282 (sign ? "" : "unsigned "), 1283 (nlong > 1 ? "long " : ""), 1284 base); 1285 1286 *newp = xstrdup(buf); 1287 return (intr); 1288 } 1289 1290 typedef struct fp_size_map { 1291 size_t fsm_typesz[2]; /* size of {32,64} type */ 1292 uint_t fsm_enc[3]; /* CTF_FP_* for {bare,cplx,imagry} type */ 1293 } fp_size_map_t; 1294 1295 static const fp_size_map_t fp_encodings[] = { 1296 { { 4, 4 }, { CTF_FP_SINGLE, CTF_FP_CPLX, CTF_FP_IMAGRY } }, 1297 { { 8, 8 }, { CTF_FP_DOUBLE, CTF_FP_DCPLX, CTF_FP_DIMAGRY } }, 1298 #ifdef __sparc 1299 { { 16, 16 }, { CTF_FP_LDOUBLE, CTF_FP_LDCPLX, CTF_FP_LDIMAGRY } }, 1300 #else 1301 { { 12, 16 }, { CTF_FP_LDOUBLE, CTF_FP_LDCPLX, CTF_FP_LDIMAGRY } }, 1302 #endif 1303 { { 0, 0 }, { 0, 0, 0 } } 1304 }; 1305 1306 static uint_t 1307 die_base_type2enc(dwarf_t *dw, Dwarf_Off off, Dwarf_Signed enc, size_t sz) 1308 { 1309 const fp_size_map_t *map = fp_encodings; 1310 uint_t szidx = dw->dw_ptrsz == sizeof (uint64_t); 1311 uint_t mult = 1, col = 0; 1312 1313 if (enc == DW_ATE_complex_float) { 1314 mult = 2; 1315 col = 1; 1316 } else if (enc == DW_ATE_imaginary_float 1317 #if defined(sun) 1318 || enc == DW_ATE_SUN_imaginary_float 1319 #endif 1320 ) 1321 col = 2; 1322 1323 while (map->fsm_typesz[szidx] != 0) { 1324 if (map->fsm_typesz[szidx] * mult == sz) 1325 return (map->fsm_enc[col]); 1326 map++; 1327 } 1328 1329 terminate("die %llu: unrecognized real type size %u\n", off, sz); 1330 /*NOTREACHED*/ 1331 return (0); 1332 } 1333 1334 static intr_t * 1335 die_base_from_dwarf(dwarf_t *dw, Dwarf_Die base, Dwarf_Off off, size_t sz) 1336 { 1337 intr_t *intr = xcalloc(sizeof (intr_t)); 1338 Dwarf_Signed enc; 1339 1340 (void) die_signed(dw, base, DW_AT_encoding, &enc, DW_ATTR_REQ); 1341 1342 switch (enc) { 1343 case DW_ATE_unsigned: 1344 case DW_ATE_address: 1345 intr->intr_type = INTR_INT; 1346 break; 1347 case DW_ATE_unsigned_char: 1348 intr->intr_type = INTR_INT; 1349 intr->intr_iformat = 'c'; 1350 break; 1351 case DW_ATE_signed: 1352 intr->intr_type = INTR_INT; 1353 intr->intr_signed = 1; 1354 break; 1355 case DW_ATE_signed_char: 1356 intr->intr_type = INTR_INT; 1357 intr->intr_signed = 1; 1358 intr->intr_iformat = 'c'; 1359 break; 1360 case DW_ATE_boolean: 1361 intr->intr_type = INTR_INT; 1362 intr->intr_signed = 1; 1363 intr->intr_iformat = 'b'; 1364 break; 1365 case DW_ATE_float: 1366 case DW_ATE_complex_float: 1367 case DW_ATE_imaginary_float: 1368 #if defined(sun) 1369 case DW_ATE_SUN_imaginary_float: 1370 case DW_ATE_SUN_interval_float: 1371 #endif 1372 intr->intr_type = INTR_REAL; 1373 intr->intr_signed = 1; 1374 intr->intr_fformat = die_base_type2enc(dw, off, enc, sz); 1375 break; 1376 default: 1377 terminate("die %llu: unknown base type encoding 0x%llx\n", 1378 off, enc); 1379 } 1380 1381 return (intr); 1382 } 1383 1384 static void 1385 die_base_create(dwarf_t *dw, Dwarf_Die base, Dwarf_Off off, tdesc_t *tdp) 1386 { 1387 Dwarf_Unsigned sz; 1388 intr_t *intr; 1389 char *new; 1390 1391 debug(3, "die %llu: creating base type\n", off); 1392 1393 /* 1394 * The compilers have their own clever (internally inconsistent) ideas 1395 * as to what base types should look like. Some times gcc will, for 1396 * example, use DW_ATE_signed_char for char. Other times, however, it 1397 * will use DW_ATE_signed. Needless to say, this causes some problems 1398 * down the road, particularly with merging. We do, however, use the 1399 * DWARF idea of type sizes, as this allows us to avoid caring about 1400 * the data model. 1401 */ 1402 (void) die_unsigned(dw, base, DW_AT_byte_size, &sz, DW_ATTR_REQ); 1403 1404 /* Check for bogus gcc DW_AT_byte_size attribute */ 1405 if (sz == (unsigned)-1) { 1406 printf("dwarf.c:%s() working around bogus -1 DW_AT_byte_size\n", 1407 __func__); 1408 sz = 0; 1409 } 1410 1411 if (tdp->t_name == NULL) 1412 terminate("die %llu: base type without name\n", off); 1413 1414 /* XXX make a name parser for float too */ 1415 if ((intr = die_base_name_parse(tdp->t_name, &new)) != NULL) { 1416 /* Found it. We'll use the parsed version */ 1417 debug(3, "die %llu: name \"%s\" remapped to \"%s\"\n", off, 1418 tdesc_name(tdp), new); 1419 1420 free(tdp->t_name); 1421 tdp->t_name = new; 1422 } else { 1423 /* 1424 * We didn't recognize the type, so we'll create an intr_t 1425 * based on the DWARF data. 1426 */ 1427 debug(3, "die %llu: using dwarf data for base \"%s\"\n", off, 1428 tdesc_name(tdp)); 1429 1430 intr = die_base_from_dwarf(dw, base, off, sz); 1431 } 1432 1433 intr->intr_nbits = sz * 8; 1434 1435 tdp->t_type = INTRINSIC; 1436 tdp->t_intr = intr; 1437 tdp->t_size = sz; 1438 1439 tdp->t_flags |= TDESC_F_RESOLVED; 1440 } 1441 1442 static void 1443 die_through_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp, 1444 int type, const char *typename) 1445 { 1446 Dwarf_Attribute attr; 1447 1448 debug(3, "die %llu <%llx>: creating %s type %d\n", off, off, typename, type); 1449 1450 tdp->t_type = type; 1451 1452 if ((attr = die_attr(dw, die, DW_AT_type, 0)) != NULL) { 1453 tdp->t_tdesc = die_lookup_pass1(dw, die, DW_AT_type); 1454 } else { 1455 tdp->t_tdesc = tdesc_intr_void(dw); 1456 } 1457 1458 if (type == POINTER) 1459 tdp->t_size = dw->dw_ptrsz; 1460 1461 tdp->t_flags |= TDESC_F_RESOLVED; 1462 1463 if (type == TYPEDEF) { 1464 iidesc_t *ii = xcalloc(sizeof (iidesc_t)); 1465 ii->ii_type = II_TYPE; 1466 ii->ii_name = xstrdup(tdp->t_name); 1467 ii->ii_dtype = tdp; 1468 1469 iidesc_add(dw->dw_td->td_iihash, ii); 1470 } 1471 } 1472 1473 static void 1474 die_typedef_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1475 { 1476 die_through_create(dw, die, off, tdp, TYPEDEF, "typedef"); 1477 } 1478 1479 static void 1480 die_const_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1481 { 1482 die_through_create(dw, die, off, tdp, CONST, "const"); 1483 } 1484 1485 static void 1486 die_pointer_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1487 { 1488 die_through_create(dw, die, off, tdp, POINTER, "pointer"); 1489 } 1490 1491 static void 1492 die_restrict_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1493 { 1494 die_through_create(dw, die, off, tdp, RESTRICT, "restrict"); 1495 } 1496 1497 static void 1498 die_volatile_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1499 { 1500 die_through_create(dw, die, off, tdp, VOLATILE, "volatile"); 1501 } 1502 1503 /*ARGSUSED3*/ 1504 static void 1505 die_function_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp __unused) 1506 { 1507 Dwarf_Die arg; 1508 Dwarf_Half tag; 1509 iidesc_t *ii; 1510 char *name; 1511 1512 debug(3, "die %llu <%llx>: creating function definition\n", off, off); 1513 1514 /* 1515 * We'll begin by processing any type definition nodes that may be 1516 * lurking underneath this one. 1517 */ 1518 for (arg = die_child(dw, die); arg != NULL; 1519 arg = die_sibling(dw, arg)) { 1520 if ((tag = die_tag(dw, arg)) != DW_TAG_formal_parameter && 1521 tag != DW_TAG_variable) { 1522 /* Nested type declaration */ 1523 die_create_one(dw, arg); 1524 } 1525 } 1526 1527 if (die_isdecl(dw, die) || (name = die_name(dw, die)) == NULL) { 1528 /* 1529 * We process neither prototypes nor subprograms without 1530 * names. 1531 */ 1532 return; 1533 } 1534 1535 ii = xcalloc(sizeof (iidesc_t)); 1536 ii->ii_type = die_isglobal(dw, die) ? II_GFUN : II_SFUN; 1537 ii->ii_name = name; 1538 if (ii->ii_type == II_SFUN) 1539 ii->ii_owner = xstrdup(dw->dw_cuname); 1540 1541 debug(3, "die %llu: function %s is %s\n", off, ii->ii_name, 1542 (ii->ii_type == II_GFUN ? "global" : "static")); 1543 1544 if (die_attr(dw, die, DW_AT_type, 0) != NULL) 1545 ii->ii_dtype = die_lookup_pass1(dw, die, DW_AT_type); 1546 else 1547 ii->ii_dtype = tdesc_intr_void(dw); 1548 1549 for (arg = die_child(dw, die); arg != NULL; 1550 arg = die_sibling(dw, arg)) { 1551 char *name1; 1552 1553 debug(3, "die %llu: looking at sub member at %llu\n", 1554 off, die_off(dw, die)); 1555 1556 if (die_tag(dw, arg) != DW_TAG_formal_parameter) 1557 continue; 1558 1559 if ((name1 = die_name(dw, arg)) == NULL) { 1560 terminate("die %llu: func arg %d has no name\n", 1561 off, ii->ii_nargs + 1); 1562 } 1563 1564 if (strcmp(name1, "...") == 0) { 1565 free(name1); 1566 ii->ii_vargs = 1; 1567 continue; 1568 } 1569 1570 ii->ii_nargs++; 1571 } 1572 1573 if (ii->ii_nargs > 0) { 1574 int i; 1575 1576 debug(3, "die %llu: function has %d argument%s\n", off, 1577 ii->ii_nargs, (ii->ii_nargs == 1 ? "" : "s")); 1578 1579 ii->ii_args = xcalloc(sizeof (tdesc_t) * ii->ii_nargs); 1580 1581 for (arg = die_child(dw, die), i = 0; 1582 arg != NULL && i < ii->ii_nargs; 1583 arg = die_sibling(dw, arg)) { 1584 if (die_tag(dw, arg) != DW_TAG_formal_parameter) 1585 continue; 1586 1587 ii->ii_args[i++] = die_lookup_pass1(dw, arg, 1588 DW_AT_type); 1589 } 1590 } 1591 1592 iidesc_add(dw->dw_td->td_iihash, ii); 1593 } 1594 1595 /*ARGSUSED3*/ 1596 static void 1597 die_variable_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp __unused) 1598 { 1599 iidesc_t *ii; 1600 char *name; 1601 1602 debug(3, "die %llu: creating object definition\n", off); 1603 1604 if (die_isdecl(dw, die) || (name = die_name(dw, die)) == NULL) 1605 return; /* skip prototypes and nameless objects */ 1606 1607 ii = xcalloc(sizeof (iidesc_t)); 1608 ii->ii_type = die_isglobal(dw, die) ? II_GVAR : II_SVAR; 1609 ii->ii_name = name; 1610 ii->ii_dtype = die_lookup_pass1(dw, die, DW_AT_type); 1611 if (ii->ii_type == II_SVAR) 1612 ii->ii_owner = xstrdup(dw->dw_cuname); 1613 1614 iidesc_add(dw->dw_td->td_iihash, ii); 1615 } 1616 1617 /*ARGSUSED2*/ 1618 static int 1619 die_fwd_resolve(tdesc_t *fwd, tdesc_t **fwdp, void *private __unused) 1620 { 1621 if (fwd->t_flags & TDESC_F_RESOLVED) 1622 return (1); 1623 1624 if (fwd->t_tdesc != NULL) { 1625 debug(3, "tdp %u: unforwarded %s\n", fwd->t_id, 1626 tdesc_name(fwd)); 1627 *fwdp = fwd->t_tdesc; 1628 } 1629 1630 fwd->t_flags |= TDESC_F_RESOLVED; 1631 1632 return (1); 1633 } 1634 1635 /*ARGSUSED*/ 1636 static void 1637 die_lexblk_descend(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off __unused, tdesc_t *tdp __unused) 1638 { 1639 Dwarf_Die child = die_child(dw, die); 1640 1641 if (child != NULL) 1642 die_create(dw, child); 1643 } 1644 1645 /* 1646 * Used to map the die to a routine which can parse it, using the tag to do the 1647 * mapping. While the processing of most tags entails the creation of a tdesc, 1648 * there are a few which don't - primarily those which result in the creation of 1649 * iidescs which refer to existing tdescs. 1650 */ 1651 1652 #define DW_F_NOTDP 0x1 /* Don't create a tdesc for the creator */ 1653 1654 typedef struct die_creator { 1655 Dwarf_Half dc_tag; 1656 uint16_t dc_flags; 1657 void (*dc_create)(dwarf_t *, Dwarf_Die, Dwarf_Off, tdesc_t *); 1658 } die_creator_t; 1659 1660 static const die_creator_t die_creators[] = { 1661 { DW_TAG_array_type, 0, die_array_create }, 1662 { DW_TAG_enumeration_type, 0, die_enum_create }, 1663 { DW_TAG_lexical_block, DW_F_NOTDP, die_lexblk_descend }, 1664 { DW_TAG_pointer_type, 0, die_pointer_create }, 1665 { DW_TAG_structure_type, 0, die_struct_create }, 1666 { DW_TAG_subroutine_type, 0, die_funcptr_create }, 1667 { DW_TAG_typedef, 0, die_typedef_create }, 1668 { DW_TAG_union_type, 0, die_union_create }, 1669 { DW_TAG_base_type, 0, die_base_create }, 1670 { DW_TAG_const_type, 0, die_const_create }, 1671 { DW_TAG_subprogram, DW_F_NOTDP, die_function_create }, 1672 { DW_TAG_variable, DW_F_NOTDP, die_variable_create }, 1673 { DW_TAG_volatile_type, 0, die_volatile_create }, 1674 { DW_TAG_restrict_type, 0, die_restrict_create }, 1675 { 0, 0, NULL } 1676 }; 1677 1678 static const die_creator_t * 1679 die_tag2ctor(Dwarf_Half tag) 1680 { 1681 const die_creator_t *dc; 1682 1683 for (dc = die_creators; dc->dc_create != NULL; dc++) { 1684 if (dc->dc_tag == tag) 1685 return (dc); 1686 } 1687 1688 return (NULL); 1689 } 1690 1691 static void 1692 die_create_one(dwarf_t *dw, Dwarf_Die die) 1693 { 1694 Dwarf_Off off = die_off(dw, die); 1695 const die_creator_t *dc; 1696 Dwarf_Half tag; 1697 tdesc_t *tdp; 1698 1699 debug(3, "die %llu <%llx>: create_one\n", off, off); 1700 1701 if (off > dw->dw_maxoff) { 1702 terminate("illegal die offset %llu (max %llu)\n", off, 1703 dw->dw_maxoff); 1704 } 1705 1706 tag = die_tag(dw, die); 1707 1708 if ((dc = die_tag2ctor(tag)) == NULL) { 1709 debug(2, "die %llu: ignoring tag type %x\n", off, tag); 1710 return; 1711 } 1712 1713 if ((tdp = tdesc_lookup(dw, off)) == NULL && 1714 !(dc->dc_flags & DW_F_NOTDP)) { 1715 tdp = xcalloc(sizeof (tdesc_t)); 1716 tdp->t_id = off; 1717 tdesc_add(dw, tdp); 1718 } 1719 1720 if (tdp != NULL) 1721 tdp->t_name = die_name(dw, die); 1722 1723 dc->dc_create(dw, die, off, tdp); 1724 } 1725 1726 static void 1727 die_create(dwarf_t *dw, Dwarf_Die die) 1728 { 1729 do { 1730 die_create_one(dw, die); 1731 } while ((die = die_sibling(dw, die)) != NULL); 1732 } 1733 1734 static tdtrav_cb_f die_resolvers[] = { 1735 NULL, 1736 NULL, /* intrinsic */ 1737 NULL, /* pointer */ 1738 die_array_resolve, /* array */ 1739 NULL, /* function */ 1740 die_sou_resolve, /* struct */ 1741 die_sou_resolve, /* union */ 1742 die_enum_resolve, /* enum */ 1743 die_fwd_resolve, /* forward */ 1744 NULL, /* typedef */ 1745 NULL, /* typedef unres */ 1746 NULL, /* volatile */ 1747 NULL, /* const */ 1748 NULL, /* restrict */ 1749 }; 1750 1751 static tdtrav_cb_f die_fail_reporters[] = { 1752 NULL, 1753 NULL, /* intrinsic */ 1754 NULL, /* pointer */ 1755 die_array_failed, /* array */ 1756 NULL, /* function */ 1757 die_sou_failed, /* struct */ 1758 die_sou_failed, /* union */ 1759 NULL, /* enum */ 1760 NULL, /* forward */ 1761 NULL, /* typedef */ 1762 NULL, /* typedef unres */ 1763 NULL, /* volatile */ 1764 NULL, /* const */ 1765 NULL, /* restrict */ 1766 }; 1767 1768 static void 1769 die_resolve(dwarf_t *dw) 1770 { 1771 int last = -1; 1772 int pass = 0; 1773 1774 do { 1775 pass++; 1776 dw->dw_nunres = 0; 1777 1778 (void) iitraverse_hash(dw->dw_td->td_iihash, 1779 &dw->dw_td->td_curvgen, NULL, NULL, die_resolvers, dw); 1780 1781 debug(3, "resolve: pass %d, %u left\n", pass, dw->dw_nunres); 1782 1783 if ((int) dw->dw_nunres == last) { 1784 fprintf(stderr, "%s: failed to resolve the following " 1785 "types:\n", progname); 1786 1787 (void) iitraverse_hash(dw->dw_td->td_iihash, 1788 &dw->dw_td->td_curvgen, NULL, NULL, 1789 die_fail_reporters, dw); 1790 1791 terminate("failed to resolve types\n"); 1792 } 1793 1794 last = dw->dw_nunres; 1795 1796 } while (dw->dw_nunres != 0); 1797 } 1798 1799 /*ARGSUSED*/ 1800 int 1801 dw_read(tdata_t *td, Elf *elf, char *filename __unused) 1802 { 1803 Dwarf_Unsigned abboff, hdrlen, nxthdr; 1804 Dwarf_Half vers, addrsz; 1805 Dwarf_Die cu = 0; 1806 Dwarf_Die child = 0; 1807 dwarf_t dw; 1808 char *prod = NULL; 1809 int rc; 1810 1811 bzero(&dw, sizeof (dwarf_t)); 1812 dw.dw_td = td; 1813 dw.dw_ptrsz = elf_ptrsz(elf); 1814 dw.dw_mfgtid_last = TID_MFGTID_BASE; 1815 dw.dw_tidhash = hash_new(TDESC_HASH_BUCKETS, tdesc_idhash, tdesc_idcmp); 1816 dw.dw_fwdhash = hash_new(TDESC_HASH_BUCKETS, tdesc_namehash, 1817 tdesc_namecmp); 1818 dw.dw_enumhash = hash_new(TDESC_HASH_BUCKETS, tdesc_namehash, 1819 tdesc_namecmp); 1820 1821 if ((rc = dwarf_elf_init(elf, DW_DLC_READ, &dw.dw_dw, 1822 &dw.dw_err)) == DW_DLV_NO_ENTRY) { 1823 errno = ENOENT; 1824 return (-1); 1825 } else if (rc != DW_DLV_OK) { 1826 if (dwarf_errno(&dw.dw_err) == DW_DLE_DEBUG_INFO_NULL) { 1827 /* 1828 * There's no type data in the DWARF section, but 1829 * libdwarf is too clever to handle that properly. 1830 */ 1831 return (0); 1832 } 1833 1834 terminate("failed to initialize DWARF: %s\n", 1835 dwarf_errmsg(&dw.dw_err)); 1836 } 1837 1838 if ((rc = dwarf_next_cu_header(dw.dw_dw, &hdrlen, &vers, &abboff, 1839 &addrsz, &nxthdr, &dw.dw_err)) != DW_DLV_OK) 1840 terminate("rc = %d %s\n", rc, dwarf_errmsg(&dw.dw_err)); 1841 1842 if ((cu = die_sibling(&dw, NULL)) == NULL) 1843 terminate("file does not contain dwarf type data " 1844 "(try compiling with -g)\n"); 1845 1846 dw.dw_maxoff = nxthdr - 1; 1847 1848 if (dw.dw_maxoff > TID_FILEMAX) 1849 terminate("file contains too many types\n"); 1850 1851 debug(1, "DWARF version: %d\n", vers); 1852 if (vers != DWARF_VERSION) { 1853 terminate("file contains incompatible version %d DWARF code " 1854 "(version 2 required)\n", vers); 1855 } 1856 1857 if (die_string(&dw, cu, DW_AT_producer, &prod, 0)) { 1858 debug(1, "DWARF emitter: %s\n", prod); 1859 free(prod); 1860 } 1861 1862 if ((dw.dw_cuname = die_name(&dw, cu)) != NULL) { 1863 char *base = xstrdup(basename(dw.dw_cuname)); 1864 free(dw.dw_cuname); 1865 dw.dw_cuname = base; 1866 1867 debug(1, "CU name: %s\n", dw.dw_cuname); 1868 } 1869 1870 if ((child = die_child(&dw, cu)) != NULL) 1871 die_create(&dw, child); 1872 1873 if ((rc = dwarf_next_cu_header(dw.dw_dw, &hdrlen, &vers, &abboff, 1874 &addrsz, &nxthdr, &dw.dw_err)) != DW_DLV_NO_ENTRY) 1875 terminate("multiple compilation units not supported\n"); 1876 1877 (void) dwarf_finish(&dw.dw_dw, &dw.dw_err); 1878 1879 die_resolve(&dw); 1880 1881 cvt_fixups(td, dw.dw_ptrsz); 1882 1883 /* leak the dwarf_t */ 1884 1885 return (0); 1886 } 1887