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 /* 27 * DWARF to tdata conversion 28 * 29 * For the most part, conversion is straightforward, proceeding in two passes. 30 * On the first pass, we iterate through every die, creating new type nodes as 31 * necessary. Referenced tdesc_t's are created in an uninitialized state, thus 32 * allowing type reference pointers to be filled in. If the tdesc_t 33 * corresponding to a given die can be completely filled out (sizes and offsets 34 * calculated, and so forth) without using any referenced types, the tdesc_t is 35 * marked as resolved. Consider an array type. If the type corresponding to 36 * the array contents has not yet been processed, we will create a blank tdesc 37 * for the contents type (only the type ID will be filled in, relying upon the 38 * later portion of the first pass to encounter and complete the referenced 39 * type). We will then attempt to determine the size of the array. If the 40 * array has a byte size attribute, we will have completely characterized the 41 * array type, and will be able to mark it as resolved. The lack of a byte 42 * size attribute, on the other hand, will prevent us from fully resolving the 43 * type, as the size will only be calculable with reference to the contents 44 * type, which has not, as yet, been encountered. The array type will thus be 45 * left without the resolved flag, and the first pass will continue. 46 * 47 * When we begin the second pass, we will have created tdesc_t nodes for every 48 * type in the section. We will traverse the tree, from the iidescs down, 49 * processing each unresolved node. As the referenced nodes will have been 50 * populated, the array type used in our example above will be able to use the 51 * size of the referenced types (if available) to determine its own type. The 52 * traversal will be repeated until all types have been resolved or we have 53 * failed to make progress. When all tdescs have been resolved, the conversion 54 * is complete. 55 * 56 * There are, as always, a few special cases that are handled during the first 57 * and second passes: 58 * 59 * 1. Empty enums - GCC will occasionally emit an enum without any members. 60 * Later on in the file, it will emit the same enum type, though this time 61 * with the full complement of members. All references to the memberless 62 * enum need to be redirected to the full definition. During the first 63 * pass, each enum is entered in dm_enumhash, along with a pointer to its 64 * corresponding tdesc_t. If, during the second pass, we encounter a 65 * memberless enum, we use the hash to locate the full definition. All 66 * tdescs referencing the empty enum are then redirected. 67 * 68 * 2. Forward declarations - If the compiler sees a forward declaration for 69 * a structure, followed by the definition of that structure, it will emit 70 * DWARF data for both the forward declaration and the definition. We need 71 * to resolve the forward declarations when possible, by redirecting 72 * forward-referencing tdescs to the actual struct/union definitions. This 73 * redirection is done completely within the first pass. We begin by 74 * recording all forward declarations in dw_fwdhash. When we define a 75 * structure, we check to see if there have been any corresponding forward 76 * declarations. If so, we redirect the tdescs which referenced the forward 77 * declarations to the structure or union definition. 78 * 79 * XXX see if a post traverser will allow the elimination of repeated pass 2 80 * traversals. 81 */ 82 83 #include <stdio.h> 84 #include <stdlib.h> 85 #include <string.h> 86 #include <strings.h> 87 #include <errno.h> 88 #include <libelf.h> 89 #include <libdwarf.h> 90 #include <libgen.h> 91 #include <dwarf.h> 92 93 #include "ctf_headers.h" 94 #include "ctftools.h" 95 #include "memory.h" 96 #include "list.h" 97 #include "traverse.h" 98 99 /* The version of DWARF which we support. */ 100 #define DWARF_VERSION 2 101 102 /* 103 * We need to define a couple of our own intrinsics, to smooth out some of the 104 * differences between the GCC and DevPro DWARF emitters. See the referenced 105 * routines and the special cases in the file comment for more details. 106 * 107 * Type IDs are 32 bits wide. We're going to use the top of that field to 108 * indicate types that we've created ourselves. 109 */ 110 #define TID_FILEMAX 0x3fffffff /* highest tid from file */ 111 #define TID_VOID 0x40000001 /* see die_void() */ 112 #define TID_LONG 0x40000002 /* see die_array() */ 113 114 #define TID_MFGTID_BASE 0x40000003 /* first mfg'd tid */ 115 116 /* 117 * To reduce the staggering amount of error-handling code that would otherwise 118 * be required, the attribute-retrieval routines handle most of their own 119 * errors. If the following flag is supplied as the value of the `req' 120 * argument, they will also handle the absence of a requested attribute by 121 * terminating the program. 122 */ 123 #define DW_ATTR_REQ 1 124 125 #define TDESC_HASH_BUCKETS 511 126 127 typedef struct dwarf { 128 Dwarf_Debug dw_dw; /* for libdwarf */ 129 Dwarf_Error dw_err; /* for libdwarf */ 130 Dwarf_Off dw_maxoff; /* highest legal offset in this cu */ 131 tdata_t *dw_td; /* root of the tdesc/iidesc tree */ 132 hash_t *dw_tidhash; /* hash of tdescs by t_id */ 133 hash_t *dw_fwdhash; /* hash of fwd decls by name */ 134 hash_t *dw_enumhash; /* hash of memberless enums by name */ 135 tdesc_t *dw_void; /* manufactured void type */ 136 tdesc_t *dw_long; /* manufactured long type for arrays */ 137 size_t dw_ptrsz; /* size of a pointer in this file */ 138 tid_t dw_mfgtid_last; /* last mfg'd type ID used */ 139 uint_t dw_nunres; /* count of unresolved types */ 140 char *dw_cuname; /* name of compilation unit */ 141 } dwarf_t; 142 143 static void die_create_one(dwarf_t *, Dwarf_Die); 144 static void die_create(dwarf_t *, Dwarf_Die); 145 146 static tid_t 147 mfgtid_next(dwarf_t *dw) 148 { 149 return (++dw->dw_mfgtid_last); 150 } 151 152 static void 153 tdesc_add(dwarf_t *dw, tdesc_t *tdp) 154 { 155 hash_add(dw->dw_tidhash, tdp); 156 } 157 158 static tdesc_t * 159 tdesc_lookup(dwarf_t *dw, int tid) 160 { 161 tdesc_t tmpl; 162 void *tdp; 163 164 tmpl.t_id = tid; 165 166 if (hash_find(dw->dw_tidhash, &tmpl, &tdp)) 167 return (tdp); 168 else 169 return (NULL); 170 } 171 172 /* 173 * Resolve a tdesc down to a node which should have a size. Returns the size, 174 * zero if the size hasn't yet been determined. 175 */ 176 static size_t 177 tdesc_size(tdesc_t *tdp) 178 { 179 for (;;) { 180 switch (tdp->t_type) { 181 case INTRINSIC: 182 case POINTER: 183 case ARRAY: 184 case FUNCTION: 185 case STRUCT: 186 case UNION: 187 case ENUM: 188 return (tdp->t_size); 189 190 case FORWARD: 191 return (0); 192 193 case TYPEDEF: 194 case VOLATILE: 195 case CONST: 196 case RESTRICT: 197 tdp = tdp->t_tdesc; 198 continue; 199 200 case 0: /* not yet defined */ 201 return (0); 202 203 default: 204 terminate("tdp %u: tdesc_size on unknown type %d\n", 205 tdp->t_id, tdp->t_type); 206 } 207 } 208 } 209 210 static size_t 211 tdesc_bitsize(tdesc_t *tdp) 212 { 213 for (;;) { 214 switch (tdp->t_type) { 215 case INTRINSIC: 216 return (tdp->t_intr->intr_nbits); 217 218 case ARRAY: 219 case FUNCTION: 220 case STRUCT: 221 case UNION: 222 case ENUM: 223 case POINTER: 224 return (tdp->t_size * NBBY); 225 226 case FORWARD: 227 return (0); 228 229 case TYPEDEF: 230 case VOLATILE: 231 case RESTRICT: 232 case CONST: 233 tdp = tdp->t_tdesc; 234 continue; 235 236 case 0: /* not yet defined */ 237 return (0); 238 239 default: 240 terminate("tdp %u: tdesc_bitsize on unknown type %d\n", 241 tdp->t_id, tdp->t_type); 242 } 243 } 244 } 245 246 static tdesc_t * 247 tdesc_basetype(tdesc_t *tdp) 248 { 249 for (;;) { 250 switch (tdp->t_type) { 251 case TYPEDEF: 252 case VOLATILE: 253 case RESTRICT: 254 case CONST: 255 tdp = tdp->t_tdesc; 256 break; 257 case 0: /* not yet defined */ 258 return (NULL); 259 default: 260 return (tdp); 261 } 262 } 263 } 264 265 static Dwarf_Off 266 die_off(dwarf_t *dw, Dwarf_Die die) 267 { 268 Dwarf_Off off; 269 270 if (dwarf_dieoffset(die, &off, &dw->dw_err) == DW_DLV_OK) 271 return (off); 272 273 terminate("failed to get offset for die: %s\n", 274 dwarf_errmsg(dw->dw_err)); 275 /*NOTREACHED*/ 276 return (0); 277 } 278 279 static Dwarf_Die 280 die_sibling(dwarf_t *dw, Dwarf_Die die) 281 { 282 Dwarf_Die sib; 283 int rc; 284 285 if ((rc = dwarf_siblingof(dw->dw_dw, die, &sib, &dw->dw_err)) == 286 DW_DLV_OK) 287 return (sib); 288 else if (rc == DW_DLV_NO_ENTRY) 289 return (NULL); 290 291 terminate("die %llu: failed to find type sibling: %s\n", 292 die_off(dw, die), dwarf_errmsg(dw->dw_err)); 293 /*NOTREACHED*/ 294 return (NULL); 295 } 296 297 static Dwarf_Die 298 die_child(dwarf_t *dw, Dwarf_Die die) 299 { 300 Dwarf_Die child; 301 int rc; 302 303 if ((rc = dwarf_child(die, &child, &dw->dw_err)) == DW_DLV_OK) 304 return (child); 305 else if (rc == DW_DLV_NO_ENTRY) 306 return (NULL); 307 308 terminate("die %llu: failed to find type child: %s\n", 309 die_off(dw, die), dwarf_errmsg(dw->dw_err)); 310 /*NOTREACHED*/ 311 return (NULL); 312 } 313 314 static Dwarf_Half 315 die_tag(dwarf_t *dw, Dwarf_Die die) 316 { 317 Dwarf_Half tag; 318 319 if (dwarf_tag(die, &tag, &dw->dw_err) == DW_DLV_OK) 320 return (tag); 321 322 terminate("die %llu: failed to get tag for type: %s\n", 323 die_off(dw, die), dwarf_errmsg(dw->dw_err)); 324 /*NOTREACHED*/ 325 return (0); 326 } 327 328 static Dwarf_Attribute 329 die_attr(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name, int req) 330 { 331 Dwarf_Attribute attr; 332 int rc; 333 334 if ((rc = dwarf_attr(die, name, &attr, &dw->dw_err)) == DW_DLV_OK) { 335 return (attr); 336 } else if (rc == DW_DLV_NO_ENTRY) { 337 if (req) { 338 terminate("die %llu: no attr 0x%x\n", die_off(dw, die), 339 name); 340 } else { 341 return (NULL); 342 } 343 } 344 345 terminate("die %llu: failed to get attribute for type: %s\n", 346 die_off(dw, die), dwarf_errmsg(dw->dw_err)); 347 /*NOTREACHED*/ 348 return (NULL); 349 } 350 351 static int 352 die_signed(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name, Dwarf_Signed *valp, 353 int req) 354 { 355 *valp = 0; 356 if (dwarf_attrval_signed(die, name, valp, &dw->dw_err) != DW_DLV_OK) { 357 if (req) 358 terminate("die %llu: failed to get signed: %s\n", 359 die_off(dw, die), dwarf_errmsg(dw->dw_err)); 360 return (0); 361 } 362 363 return (1); 364 } 365 366 static int 367 die_unsigned(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name, Dwarf_Unsigned *valp, 368 int req) 369 { 370 *valp = 0; 371 if (dwarf_attrval_unsigned(die, name, valp, &dw->dw_err) != DW_DLV_OK) { 372 if (req) 373 terminate("die %llu: failed to get unsigned: %s\n", 374 die_off(dw, die), dwarf_errmsg(dw->dw_err)); 375 return (0); 376 } 377 378 return (1); 379 } 380 381 static int 382 die_bool(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name, Dwarf_Bool *valp, int req) 383 { 384 *valp = 0; 385 386 if (dwarf_attrval_flag(die, name, valp, &dw->dw_err) != DW_DLV_OK) { 387 if (req) 388 terminate("die %llu: failed to get flag: %s\n", 389 die_off(dw, die), dwarf_errmsg(dw->dw_err)); 390 return (0); 391 } 392 393 return (1); 394 } 395 396 static int 397 die_string(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name, char **strp, int req) 398 { 399 const char *str = NULL; 400 401 if (dwarf_attrval_string(die, name, &str, &dw->dw_err) != DW_DLV_OK || 402 str == NULL) { 403 if (req) 404 terminate("die %llu: failed to get string: %s\n", 405 die_off(dw, die), dwarf_errmsg(dw->dw_err)); 406 else 407 *strp = NULL; 408 return (0); 409 } else 410 *strp = xstrdup(str); 411 412 return (1); 413 } 414 415 static Dwarf_Off 416 die_attr_ref(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name) 417 { 418 Dwarf_Off off; 419 420 if (dwarf_attrval_unsigned(die, name, &off, &dw->dw_err) != DW_DLV_OK) { 421 terminate("die %llu: failed to get ref: %s\n", 422 die_off(dw, die), dwarf_errmsg(dw->dw_err)); 423 } 424 425 return (off); 426 } 427 428 static char * 429 die_name(dwarf_t *dw, Dwarf_Die die) 430 { 431 char *str = NULL; 432 433 (void) die_string(dw, die, DW_AT_name, &str, 0); 434 435 return (str); 436 } 437 438 static int 439 die_isdecl(dwarf_t *dw, Dwarf_Die die) 440 { 441 Dwarf_Bool val; 442 443 return (die_bool(dw, die, DW_AT_declaration, &val, 0) && val); 444 } 445 446 static int 447 die_isglobal(dwarf_t *dw, Dwarf_Die die) 448 { 449 Dwarf_Signed vis; 450 Dwarf_Bool ext; 451 452 /* 453 * Some compilers (gcc) use DW_AT_external to indicate function 454 * visibility. Others (Sun) use DW_AT_visibility. 455 */ 456 if (die_signed(dw, die, DW_AT_visibility, &vis, 0)) 457 return (vis == DW_VIS_exported); 458 else 459 return (die_bool(dw, die, DW_AT_external, &ext, 0) && ext); 460 } 461 462 static tdesc_t * 463 die_add(dwarf_t *dw, Dwarf_Off off) 464 { 465 tdesc_t *tdp = xcalloc(sizeof (tdesc_t)); 466 467 tdp->t_id = off; 468 469 tdesc_add(dw, tdp); 470 471 return (tdp); 472 } 473 474 static tdesc_t * 475 die_lookup_pass1(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name) 476 { 477 Dwarf_Off ref = die_attr_ref(dw, die, name); 478 tdesc_t *tdp; 479 480 if ((tdp = tdesc_lookup(dw, ref)) != NULL) 481 return (tdp); 482 483 return (die_add(dw, ref)); 484 } 485 486 static int 487 die_mem_offset(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name, 488 Dwarf_Unsigned *valp, int req __unused) 489 { 490 Dwarf_Locdesc *loc = NULL; 491 Dwarf_Signed locnum = 0; 492 Dwarf_Attribute at; 493 494 if ((at = die_attr(dw, die, name, 0)) == NULL) 495 return (0); 496 497 if (dwarf_loclist(at, &loc, &locnum, &dw->dw_err) != DW_DLV_OK) 498 return (0); 499 500 if (locnum != 1 || loc->ld_s->lr_atom != DW_OP_plus_uconst) { 501 terminate("die %llu: cannot parse member offset\n", 502 die_off(dw, die)); 503 } 504 505 *valp = loc->ld_s->lr_number; 506 507 if (loc != NULL) { 508 dwarf_dealloc(dw->dw_dw, loc->ld_s, DW_DLA_LOC_BLOCK); 509 dwarf_dealloc(dw->dw_dw, loc, DW_DLA_LOCDESC); 510 } 511 512 return (1); 513 } 514 515 static tdesc_t * 516 tdesc_intr_common(dwarf_t *dw, int tid, const char *name, size_t sz) 517 { 518 tdesc_t *tdp; 519 intr_t *intr; 520 521 intr = xcalloc(sizeof (intr_t)); 522 intr->intr_type = INTR_INT; 523 intr->intr_signed = 1; 524 intr->intr_nbits = sz * NBBY; 525 526 tdp = xcalloc(sizeof (tdesc_t)); 527 tdp->t_name = xstrdup(name); 528 tdp->t_size = sz; 529 tdp->t_id = tid; 530 tdp->t_type = INTRINSIC; 531 tdp->t_intr = intr; 532 tdp->t_flags = TDESC_F_RESOLVED; 533 534 tdesc_add(dw, tdp); 535 536 return (tdp); 537 } 538 539 /* 540 * Manufacture a void type. Used for gcc-emitted stabs, where the lack of a 541 * type reference implies a reference to a void type. A void *, for example 542 * will be represented by a pointer die without a DW_AT_type. CTF requires 543 * that pointer nodes point to something, so we'll create a void for use as 544 * the target. Note that the DWARF data may already create a void type. Ours 545 * would then be a duplicate, but it'll be removed in the self-uniquification 546 * merge performed at the completion of DWARF->tdesc conversion. 547 */ 548 static tdesc_t * 549 tdesc_intr_void(dwarf_t *dw) 550 { 551 if (dw->dw_void == NULL) 552 dw->dw_void = tdesc_intr_common(dw, TID_VOID, "void", 0); 553 554 return (dw->dw_void); 555 } 556 557 static tdesc_t * 558 tdesc_intr_long(dwarf_t *dw) 559 { 560 if (dw->dw_long == NULL) { 561 dw->dw_long = tdesc_intr_common(dw, TID_LONG, "long", 562 dw->dw_ptrsz); 563 } 564 565 return (dw->dw_long); 566 } 567 568 /* 569 * Used for creating bitfield types. We create a copy of an existing intrinsic, 570 * adjusting the size of the copy to match what the caller requested. The 571 * caller can then use the copy as the type for a bitfield structure member. 572 */ 573 static tdesc_t * 574 tdesc_intr_clone(dwarf_t *dw, tdesc_t *old, size_t bitsz) 575 { 576 tdesc_t *new = xcalloc(sizeof (tdesc_t)); 577 578 if (!(old->t_flags & TDESC_F_RESOLVED)) { 579 terminate("tdp %u: attempt to make a bit field from an " 580 "unresolved type\n", old->t_id); 581 } 582 583 new->t_name = xstrdup(old->t_name); 584 new->t_size = old->t_size; 585 new->t_id = mfgtid_next(dw); 586 new->t_type = INTRINSIC; 587 new->t_flags = TDESC_F_RESOLVED; 588 589 new->t_intr = xcalloc(sizeof (intr_t)); 590 bcopy(old->t_intr, new->t_intr, sizeof (intr_t)); 591 new->t_intr->intr_nbits = bitsz; 592 593 tdesc_add(dw, new); 594 595 return (new); 596 } 597 598 static void 599 tdesc_array_create(dwarf_t *dw, Dwarf_Die dim, tdesc_t *arrtdp, 600 tdesc_t *dimtdp) 601 { 602 Dwarf_Unsigned uval; 603 Dwarf_Signed sval; 604 tdesc_t *ctdp = NULL; 605 Dwarf_Die dim2; 606 ardef_t *ar; 607 608 if ((dim2 = die_sibling(dw, dim)) == NULL) { 609 ctdp = arrtdp; 610 } else if (die_tag(dw, dim2) == DW_TAG_subrange_type) { 611 ctdp = xcalloc(sizeof (tdesc_t)); 612 ctdp->t_id = mfgtid_next(dw); 613 debug(3, "die %llu: creating new type %u for sub-dimension\n", 614 die_off(dw, dim2), ctdp->t_id); 615 tdesc_array_create(dw, dim2, arrtdp, ctdp); 616 } else { 617 terminate("die %llu: unexpected non-subrange node in array\n", 618 die_off(dw, dim2)); 619 } 620 621 dimtdp->t_type = ARRAY; 622 dimtdp->t_ardef = ar = xcalloc(sizeof (ardef_t)); 623 624 /* 625 * Array bounds can be signed or unsigned, but there are several kinds 626 * of signless forms (data1, data2, etc) that take their sign from the 627 * routine that is trying to interpret them. That is, data1 can be 628 * either signed or unsigned, depending on whether you use the signed or 629 * unsigned accessor function. GCC will use the signless forms to store 630 * unsigned values which have their high bit set, so we need to try to 631 * read them first as unsigned to get positive values. We could also 632 * try signed first, falling back to unsigned if we got a negative 633 * value. 634 */ 635 if (die_unsigned(dw, dim, DW_AT_upper_bound, &uval, 0)) 636 ar->ad_nelems = uval + 1; 637 else if (die_signed(dw, dim, DW_AT_upper_bound, &sval, 0)) 638 ar->ad_nelems = sval + 1; 639 else 640 ar->ad_nelems = 0; 641 642 /* 643 * Different compilers use different index types. Force the type to be 644 * a common, known value (long). 645 */ 646 ar->ad_idxtype = tdesc_intr_long(dw); 647 ar->ad_contents = ctdp; 648 649 if (ar->ad_contents->t_size != 0) { 650 dimtdp->t_size = ar->ad_contents->t_size * ar->ad_nelems; 651 dimtdp->t_flags |= TDESC_F_RESOLVED; 652 } 653 } 654 655 /* 656 * Create a tdesc from an array node. Some arrays will come with byte size 657 * attributes, and thus can be resolved immediately. Others don't, and will 658 * need to wait until the second pass for resolution. 659 */ 660 static void 661 die_array_create(dwarf_t *dw, Dwarf_Die arr, Dwarf_Off off, tdesc_t *tdp) 662 { 663 tdesc_t *arrtdp = die_lookup_pass1(dw, arr, DW_AT_type); 664 Dwarf_Unsigned uval; 665 Dwarf_Die dim; 666 667 debug(3, "die %llu <%llx>: creating array\n", off, off); 668 669 if ((dim = die_child(dw, arr)) == NULL || 670 die_tag(dw, dim) != DW_TAG_subrange_type) 671 terminate("die %llu: failed to retrieve array bounds\n", off); 672 673 tdesc_array_create(dw, dim, arrtdp, tdp); 674 675 if (die_unsigned(dw, arr, DW_AT_byte_size, &uval, 0)) { 676 tdesc_t *dimtdp; 677 int flags; 678 679 /* Check for bogus gcc DW_AT_byte_size attribute */ 680 if (uval == (unsigned)-1) { 681 printf("dwarf.c:%s() working around bogus -1 DW_AT_byte_size\n", 682 __func__); 683 uval = 0; 684 } 685 686 tdp->t_size = uval; 687 688 /* 689 * Ensure that sub-dimensions have sizes too before marking 690 * as resolved. 691 */ 692 flags = TDESC_F_RESOLVED; 693 for (dimtdp = tdp->t_ardef->ad_contents; 694 dimtdp->t_type == ARRAY; 695 dimtdp = dimtdp->t_ardef->ad_contents) { 696 if (!(dimtdp->t_flags & TDESC_F_RESOLVED)) { 697 flags = 0; 698 break; 699 } 700 } 701 702 tdp->t_flags |= flags; 703 } 704 705 debug(3, "die %llu <%llx>: array nelems %u size %u\n", off, off, 706 tdp->t_ardef->ad_nelems, tdp->t_size); 707 } 708 709 /*ARGSUSED1*/ 710 static int 711 die_array_resolve(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private) 712 { 713 dwarf_t *dw = private; 714 size_t sz; 715 716 if (tdp->t_flags & TDESC_F_RESOLVED) 717 return (1); 718 719 debug(3, "trying to resolve array %d (cont %d)\n", tdp->t_id, 720 tdp->t_ardef->ad_contents->t_id); 721 722 if ((sz = tdesc_size(tdp->t_ardef->ad_contents)) == 0) { 723 debug(3, "unable to resolve array %s (%d) contents %d\n", 724 tdesc_name(tdp), tdp->t_id, 725 tdp->t_ardef->ad_contents->t_id); 726 727 dw->dw_nunres++; 728 return (1); 729 } 730 731 tdp->t_size = sz * tdp->t_ardef->ad_nelems; 732 tdp->t_flags |= TDESC_F_RESOLVED; 733 734 debug(3, "resolved array %d: %u bytes\n", tdp->t_id, tdp->t_size); 735 736 return (1); 737 } 738 739 /*ARGSUSED1*/ 740 static int 741 die_array_failed(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private __unused) 742 { 743 tdesc_t *cont = tdp->t_ardef->ad_contents; 744 745 if (tdp->t_flags & TDESC_F_RESOLVED) 746 return (1); 747 748 fprintf(stderr, "Array %d: failed to size contents type %s (%d)\n", 749 tdp->t_id, tdesc_name(cont), cont->t_id); 750 751 return (1); 752 } 753 754 /* 755 * Most enums (those with members) will be resolved during this first pass. 756 * Others - those without members (see the file comment) - won't be, and will 757 * need to wait until the second pass when they can be matched with their full 758 * definitions. 759 */ 760 static void 761 die_enum_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 762 { 763 Dwarf_Die mem; 764 Dwarf_Unsigned uval; 765 Dwarf_Signed sval; 766 767 debug(3, "die %llu: creating enum\n", off); 768 769 tdp->t_type = ENUM; 770 771 (void) die_unsigned(dw, die, DW_AT_byte_size, &uval, DW_ATTR_REQ); 772 /* Check for bogus gcc DW_AT_byte_size attribute */ 773 if (uval == (unsigned)-1) { 774 printf("dwarf.c:%s() working around bogus -1 DW_AT_byte_size\n", 775 __func__); 776 uval = 0; 777 } 778 tdp->t_size = uval; 779 780 if ((mem = die_child(dw, die)) != NULL) { 781 elist_t **elastp = &tdp->t_emem; 782 783 do { 784 elist_t *el; 785 786 if (die_tag(dw, mem) != DW_TAG_enumerator) { 787 /* Nested type declaration */ 788 die_create_one(dw, mem); 789 continue; 790 } 791 792 el = xcalloc(sizeof (elist_t)); 793 el->el_name = die_name(dw, mem); 794 795 if (die_signed(dw, mem, DW_AT_const_value, &sval, 0)) { 796 el->el_number = sval; 797 } else if (die_unsigned(dw, mem, DW_AT_const_value, 798 &uval, 0)) { 799 el->el_number = uval; 800 } else { 801 terminate("die %llu: enum %llu: member without " 802 "value\n", off, die_off(dw, mem)); 803 } 804 805 debug(3, "die %llu: enum %llu: created %s = %d\n", off, 806 die_off(dw, mem), el->el_name, el->el_number); 807 808 *elastp = el; 809 elastp = &el->el_next; 810 811 } while ((mem = die_sibling(dw, mem)) != NULL); 812 813 hash_add(dw->dw_enumhash, tdp); 814 815 tdp->t_flags |= TDESC_F_RESOLVED; 816 817 if (tdp->t_name != NULL) { 818 iidesc_t *ii = xcalloc(sizeof (iidesc_t)); 819 ii->ii_type = II_SOU; 820 ii->ii_name = xstrdup(tdp->t_name); 821 ii->ii_dtype = tdp; 822 823 iidesc_add(dw->dw_td->td_iihash, ii); 824 } 825 } 826 } 827 828 static int 829 die_enum_match(void *arg1, void *arg2) 830 { 831 tdesc_t *tdp = arg1, **fullp = arg2; 832 833 if (tdp->t_emem != NULL) { 834 *fullp = tdp; 835 return (-1); /* stop the iteration */ 836 } 837 838 return (0); 839 } 840 841 /*ARGSUSED1*/ 842 static int 843 die_enum_resolve(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private) 844 { 845 dwarf_t *dw = private; 846 tdesc_t *full = NULL; 847 848 if (tdp->t_flags & TDESC_F_RESOLVED) 849 return (1); 850 851 (void) hash_find_iter(dw->dw_enumhash, tdp, die_enum_match, &full); 852 853 /* 854 * The answer to this one won't change from iteration to iteration, 855 * so don't even try. 856 */ 857 if (full == NULL) { 858 terminate("tdp %u: enum %s has no members\n", tdp->t_id, 859 tdesc_name(tdp)); 860 } 861 862 debug(3, "tdp %u: enum %s redirected to %u\n", tdp->t_id, 863 tdesc_name(tdp), full->t_id); 864 865 tdp->t_flags |= TDESC_F_RESOLVED; 866 867 return (1); 868 } 869 870 static int 871 die_fwd_map(void *arg1, void *arg2) 872 { 873 tdesc_t *fwd = arg1, *sou = arg2; 874 875 debug(3, "tdp %u: mapped forward %s to sou %u\n", fwd->t_id, 876 tdesc_name(fwd), sou->t_id); 877 fwd->t_tdesc = sou; 878 879 return (0); 880 } 881 882 /* 883 * Structures and unions will never be resolved during the first pass, as we 884 * won't be able to fully determine the member sizes. The second pass, which 885 * have access to sizing information, will be able to complete the resolution. 886 */ 887 static void 888 die_sou_create(dwarf_t *dw, Dwarf_Die str, Dwarf_Off off, tdesc_t *tdp, 889 int type, const char *typename) 890 { 891 Dwarf_Unsigned sz, bitsz, bitoff, maxsz=0; 892 Dwarf_Die mem; 893 mlist_t *ml, **mlastp; 894 iidesc_t *ii; 895 896 tdp->t_type = (die_isdecl(dw, str) ? FORWARD : type); 897 898 debug(3, "die %llu: creating %s %s\n", off, 899 (tdp->t_type == FORWARD ? "forward decl" : typename), 900 tdesc_name(tdp)); 901 902 if (tdp->t_type == FORWARD) { 903 hash_add(dw->dw_fwdhash, tdp); 904 return; 905 } 906 907 (void) hash_find_iter(dw->dw_fwdhash, tdp, die_fwd_map, tdp); 908 909 (void) die_unsigned(dw, str, DW_AT_byte_size, &sz, DW_ATTR_REQ); 910 tdp->t_size = sz; 911 912 /* 913 * GCC allows empty SOUs as an extension. 914 */ 915 if ((mem = die_child(dw, str)) == NULL) { 916 goto out; 917 } 918 919 mlastp = &tdp->t_members; 920 921 do { 922 Dwarf_Off memoff = die_off(dw, mem); 923 Dwarf_Half tag = die_tag(dw, mem); 924 Dwarf_Unsigned mloff; 925 926 if (tag != DW_TAG_member) { 927 /* Nested type declaration */ 928 die_create_one(dw, mem); 929 continue; 930 } 931 932 debug(3, "die %llu: mem %llu: creating member\n", off, memoff); 933 934 ml = xcalloc(sizeof (mlist_t)); 935 936 /* 937 * This could be a GCC anon struct/union member, so we'll allow 938 * an empty name, even though nothing can really handle them 939 * properly. Note that some versions of GCC miss out debug 940 * info for anon structs, though recent versions are fixed (gcc 941 * bug 11816). 942 */ 943 if ((ml->ml_name = die_name(dw, mem)) == NULL) 944 ml->ml_name = NULL; 945 946 ml->ml_type = die_lookup_pass1(dw, mem, DW_AT_type); 947 debug(3, "die_sou_create(): ml_type = %p t_id = %d\n", 948 ml->ml_type, ml->ml_type->t_id); 949 950 if (die_mem_offset(dw, mem, DW_AT_data_member_location, 951 &mloff, 0)) { 952 debug(3, "die %llu: got mloff %llx\n", off, 953 (u_longlong_t)mloff); 954 ml->ml_offset = mloff * 8; 955 } 956 957 if (die_unsigned(dw, mem, DW_AT_bit_size, &bitsz, 0)) 958 ml->ml_size = bitsz; 959 else 960 ml->ml_size = tdesc_bitsize(ml->ml_type); 961 962 if (die_unsigned(dw, mem, DW_AT_bit_offset, &bitoff, 0)) { 963 #if BYTE_ORDER == _BIG_ENDIAN 964 ml->ml_offset += bitoff; 965 #else 966 ml->ml_offset += tdesc_bitsize(ml->ml_type) - bitoff - 967 ml->ml_size; 968 #endif 969 } 970 971 debug(3, "die %llu: mem %llu: created \"%s\" (off %u sz %u)\n", 972 off, memoff, ml->ml_name, ml->ml_offset, ml->ml_size); 973 974 *mlastp = ml; 975 mlastp = &ml->ml_next; 976 977 /* Find the size of the largest member to work around a gcc 978 * bug. See GCC Bugzilla 35998. 979 */ 980 if (maxsz < ml->ml_size) 981 maxsz = ml->ml_size; 982 983 } while ((mem = die_sibling(dw, mem)) != NULL); 984 985 /* See if we got a bogus DW_AT_byte_size. GCC will sometimes 986 * emit this. 987 */ 988 if (sz == (unsigned)-1) { 989 printf("dwarf.c:%s() working around bogus -1 DW_AT_byte_size\n", 990 __func__); 991 tdp->t_size = maxsz / 8; /* maxsz is in bits, t_size is bytes */ 992 } 993 994 /* 995 * GCC will attempt to eliminate unused types, thus decreasing the 996 * size of the emitted dwarf. That is, if you declare a foo_t in your 997 * header, include said header in your source file, and neglect to 998 * actually use (directly or indirectly) the foo_t in the source file, 999 * the foo_t won't make it into the emitted DWARF. So, at least, goes 1000 * the theory. 1001 * 1002 * Occasionally, it'll emit the DW_TAG_structure_type for the foo_t, 1003 * and then neglect to emit the members. Strangely, the loner struct 1004 * tag will always be followed by a proper nested declaration of 1005 * something else. This is clearly a bug, but we're not going to have 1006 * time to get it fixed before this goo goes back, so we'll have to work 1007 * around it. If we see a no-membered struct with a nested declaration 1008 * (i.e. die_child of the struct tag won't be null), we'll ignore it. 1009 * Being paranoid, we won't simply remove it from the hash. Instead, 1010 * we'll decline to create an iidesc for it, thus ensuring that this 1011 * type won't make it into the output file. To be safe, we'll also 1012 * change the name. 1013 */ 1014 if (tdp->t_members == NULL) { 1015 const char *old = tdesc_name(tdp); 1016 size_t newsz = 7 + strlen(old) + 1; 1017 char *new = xmalloc(newsz); 1018 (void) snprintf(new, newsz, "orphan %s", old); 1019 1020 debug(3, "die %llu: worked around %s %s\n", off, typename, old); 1021 1022 if (tdp->t_name != NULL) 1023 free(tdp->t_name); 1024 tdp->t_name = new; 1025 return; 1026 } 1027 1028 out: 1029 if (tdp->t_name != NULL) { 1030 ii = xcalloc(sizeof (iidesc_t)); 1031 ii->ii_type = II_SOU; 1032 ii->ii_name = xstrdup(tdp->t_name); 1033 ii->ii_dtype = tdp; 1034 1035 iidesc_add(dw->dw_td->td_iihash, ii); 1036 } 1037 } 1038 1039 static void 1040 die_struct_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1041 { 1042 die_sou_create(dw, die, off, tdp, STRUCT, "struct"); 1043 } 1044 1045 static void 1046 die_union_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1047 { 1048 die_sou_create(dw, die, off, tdp, UNION, "union"); 1049 } 1050 1051 /*ARGSUSED1*/ 1052 static int 1053 die_sou_resolve(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private) 1054 { 1055 dwarf_t *dw = private; 1056 mlist_t *ml; 1057 tdesc_t *mt; 1058 1059 if (tdp->t_flags & TDESC_F_RESOLVED) 1060 return (1); 1061 1062 debug(3, "resolving sou %s\n", tdesc_name(tdp)); 1063 1064 for (ml = tdp->t_members; ml != NULL; ml = ml->ml_next) { 1065 if (ml->ml_size == 0) { 1066 mt = tdesc_basetype(ml->ml_type); 1067 1068 if ((ml->ml_size = tdesc_bitsize(mt)) != 0) 1069 continue; 1070 1071 /* 1072 * For empty members, or GCC/C99 flexible array 1073 * members, a size of 0 is correct. 1074 */ 1075 if (mt->t_members == NULL) 1076 continue; 1077 if (mt->t_type == ARRAY && mt->t_ardef->ad_nelems == 0) 1078 continue; 1079 1080 dw->dw_nunres++; 1081 return (1); 1082 } 1083 1084 if ((mt = tdesc_basetype(ml->ml_type)) == NULL) { 1085 dw->dw_nunres++; 1086 return (1); 1087 } 1088 1089 if (ml->ml_size != 0 && mt->t_type == INTRINSIC && 1090 mt->t_intr->intr_nbits != (int)ml->ml_size) { 1091 /* 1092 * This member is a bitfield, and needs to reference 1093 * an intrinsic type with the same width. If the 1094 * currently-referenced type isn't of the same width, 1095 * we'll copy it, adjusting the width of the copy to 1096 * the size we'd like. 1097 */ 1098 debug(3, "tdp %u: creating bitfield for %d bits\n", 1099 tdp->t_id, ml->ml_size); 1100 1101 ml->ml_type = tdesc_intr_clone(dw, mt, ml->ml_size); 1102 } 1103 } 1104 1105 tdp->t_flags |= TDESC_F_RESOLVED; 1106 1107 return (1); 1108 } 1109 1110 /*ARGSUSED1*/ 1111 static int 1112 die_sou_failed(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private __unused) 1113 { 1114 const char *typename = (tdp->t_type == STRUCT ? "struct" : "union"); 1115 mlist_t *ml; 1116 1117 if (tdp->t_flags & TDESC_F_RESOLVED) 1118 return (1); 1119 1120 for (ml = tdp->t_members; ml != NULL; ml = ml->ml_next) { 1121 if (ml->ml_size == 0) { 1122 fprintf(stderr, "%s %d <%x>: failed to size member \"%s\" " 1123 "of type %s (%d <%x>)\n", typename, tdp->t_id, 1124 tdp->t_id, 1125 ml->ml_name, tdesc_name(ml->ml_type), 1126 ml->ml_type->t_id, ml->ml_type->t_id); 1127 } 1128 } 1129 1130 return (1); 1131 } 1132 1133 static void 1134 die_funcptr_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1135 { 1136 Dwarf_Attribute attr; 1137 Dwarf_Half tag; 1138 Dwarf_Die arg; 1139 fndef_t *fn; 1140 int i; 1141 1142 debug(3, "die %llu <%llx>: creating function pointer\n", off, off); 1143 1144 /* 1145 * We'll begin by processing any type definition nodes that may be 1146 * lurking underneath this one. 1147 */ 1148 for (arg = die_child(dw, die); arg != NULL; 1149 arg = die_sibling(dw, arg)) { 1150 if ((tag = die_tag(dw, arg)) != DW_TAG_formal_parameter && 1151 tag != DW_TAG_unspecified_parameters) { 1152 /* Nested type declaration */ 1153 die_create_one(dw, arg); 1154 } 1155 } 1156 1157 if (die_isdecl(dw, die)) { 1158 /* 1159 * This is a prototype. We don't add prototypes to the 1160 * tree, so we're going to drop the tdesc. Unfortunately, 1161 * it has already been added to the tree. Nobody will reference 1162 * it, though, and it will be leaked. 1163 */ 1164 return; 1165 } 1166 1167 fn = xcalloc(sizeof (fndef_t)); 1168 1169 tdp->t_type = FUNCTION; 1170 1171 if ((attr = die_attr(dw, die, DW_AT_type, 0)) != NULL) { 1172 fn->fn_ret = die_lookup_pass1(dw, die, DW_AT_type); 1173 } else { 1174 fn->fn_ret = tdesc_intr_void(dw); 1175 } 1176 1177 /* 1178 * Count the arguments to the function, then read them in. 1179 */ 1180 for (fn->fn_nargs = 0, arg = die_child(dw, die); arg != NULL; 1181 arg = die_sibling(dw, arg)) { 1182 if ((tag = die_tag(dw, arg)) == DW_TAG_formal_parameter) 1183 fn->fn_nargs++; 1184 else if (tag == DW_TAG_unspecified_parameters && 1185 fn->fn_nargs > 0) 1186 fn->fn_vargs = 1; 1187 } 1188 1189 if (fn->fn_nargs != 0) { 1190 debug(3, "die %llu: adding %d argument%s\n", off, fn->fn_nargs, 1191 (fn->fn_nargs > 1 ? "s" : "")); 1192 1193 fn->fn_args = xcalloc(sizeof (tdesc_t *) * fn->fn_nargs); 1194 for (i = 0, arg = die_child(dw, die); 1195 arg != NULL && i < (int) fn->fn_nargs; 1196 arg = die_sibling(dw, arg)) { 1197 if (die_tag(dw, arg) != DW_TAG_formal_parameter) 1198 continue; 1199 1200 fn->fn_args[i++] = die_lookup_pass1(dw, arg, 1201 DW_AT_type); 1202 } 1203 } 1204 1205 tdp->t_fndef = fn; 1206 tdp->t_flags |= TDESC_F_RESOLVED; 1207 } 1208 1209 /* 1210 * GCC and DevPro use different names for the base types. While the terms are 1211 * the same, they are arranged in a different order. Some terms, such as int, 1212 * are implied in one, and explicitly named in the other. Given a base type 1213 * as input, this routine will return a common name, along with an intr_t 1214 * that reflects said name. 1215 */ 1216 static intr_t * 1217 die_base_name_parse(const char *name, char **newp) 1218 { 1219 char buf[100]; 1220 char const *base; 1221 char *c; 1222 int nlong = 0, nshort = 0, nchar = 0, nint = 0; 1223 int sign = 1; 1224 char fmt = '\0'; 1225 intr_t *intr; 1226 1227 if (strlen(name) > sizeof (buf) - 1) 1228 terminate("base type name \"%s\" is too long\n", name); 1229 1230 strncpy(buf, name, sizeof (buf)); 1231 1232 for (c = strtok(buf, " "); c != NULL; c = strtok(NULL, " ")) { 1233 if (strcmp(c, "signed") == 0) 1234 sign = 1; 1235 else if (strcmp(c, "unsigned") == 0) 1236 sign = 0; 1237 else if (strcmp(c, "long") == 0) 1238 nlong++; 1239 else if (strcmp(c, "char") == 0) { 1240 nchar++; 1241 fmt = 'c'; 1242 } else if (strcmp(c, "short") == 0) 1243 nshort++; 1244 else if (strcmp(c, "int") == 0) 1245 nint++; 1246 else { 1247 /* 1248 * If we don't recognize any of the tokens, we'll tell 1249 * the caller to fall back to the dwarf-provided 1250 * encoding information. 1251 */ 1252 return (NULL); 1253 } 1254 } 1255 1256 if (nchar > 1 || nshort > 1 || nint > 1 || nlong > 2) 1257 return (NULL); 1258 1259 if (nchar > 0) { 1260 if (nlong > 0 || nshort > 0 || nint > 0) 1261 return (NULL); 1262 1263 base = "char"; 1264 1265 } else if (nshort > 0) { 1266 if (nlong > 0) 1267 return (NULL); 1268 1269 base = "short"; 1270 1271 } else if (nlong > 0) { 1272 base = "long"; 1273 1274 } else { 1275 base = "int"; 1276 } 1277 1278 intr = xcalloc(sizeof (intr_t)); 1279 intr->intr_type = INTR_INT; 1280 intr->intr_signed = sign; 1281 intr->intr_iformat = fmt; 1282 1283 snprintf(buf, sizeof (buf), "%s%s%s", 1284 (sign ? "" : "unsigned "), 1285 (nlong > 1 ? "long " : ""), 1286 base); 1287 1288 *newp = xstrdup(buf); 1289 return (intr); 1290 } 1291 1292 typedef struct fp_size_map { 1293 size_t fsm_typesz[2]; /* size of {32,64} type */ 1294 uint_t fsm_enc[3]; /* CTF_FP_* for {bare,cplx,imagry} type */ 1295 } fp_size_map_t; 1296 1297 static const fp_size_map_t fp_encodings[] = { 1298 { { 4, 4 }, { CTF_FP_SINGLE, CTF_FP_CPLX, CTF_FP_IMAGRY } }, 1299 { { 8, 8 }, { CTF_FP_DOUBLE, CTF_FP_DCPLX, CTF_FP_DIMAGRY } }, 1300 #ifdef __sparc 1301 { { 16, 16 }, { CTF_FP_LDOUBLE, CTF_FP_LDCPLX, CTF_FP_LDIMAGRY } }, 1302 #else 1303 { { 12, 16 }, { CTF_FP_LDOUBLE, CTF_FP_LDCPLX, CTF_FP_LDIMAGRY } }, 1304 #endif 1305 { { 0, 0 }, { 0, 0, 0 } } 1306 }; 1307 1308 static uint_t 1309 die_base_type2enc(dwarf_t *dw, Dwarf_Off off, Dwarf_Signed enc, size_t sz) 1310 { 1311 const fp_size_map_t *map = fp_encodings; 1312 uint_t szidx = dw->dw_ptrsz == sizeof (uint64_t); 1313 uint_t mult = 1, col = 0; 1314 1315 if (enc == DW_ATE_complex_float) { 1316 mult = 2; 1317 col = 1; 1318 } else if (enc == DW_ATE_imaginary_float 1319 #if defined(sun) 1320 || enc == DW_ATE_SUN_imaginary_float 1321 #endif 1322 ) 1323 col = 2; 1324 1325 while (map->fsm_typesz[szidx] != 0) { 1326 if (map->fsm_typesz[szidx] * mult == sz) 1327 return (map->fsm_enc[col]); 1328 map++; 1329 } 1330 1331 terminate("die %llu: unrecognized real type size %u\n", off, sz); 1332 /*NOTREACHED*/ 1333 return (0); 1334 } 1335 1336 static intr_t * 1337 die_base_from_dwarf(dwarf_t *dw, Dwarf_Die base, Dwarf_Off off, size_t sz) 1338 { 1339 intr_t *intr = xcalloc(sizeof (intr_t)); 1340 Dwarf_Signed enc; 1341 1342 (void) die_signed(dw, base, DW_AT_encoding, &enc, DW_ATTR_REQ); 1343 1344 switch (enc) { 1345 case DW_ATE_unsigned: 1346 case DW_ATE_address: 1347 intr->intr_type = INTR_INT; 1348 break; 1349 case DW_ATE_unsigned_char: 1350 intr->intr_type = INTR_INT; 1351 intr->intr_iformat = 'c'; 1352 break; 1353 case DW_ATE_signed: 1354 intr->intr_type = INTR_INT; 1355 intr->intr_signed = 1; 1356 break; 1357 case DW_ATE_signed_char: 1358 intr->intr_type = INTR_INT; 1359 intr->intr_signed = 1; 1360 intr->intr_iformat = 'c'; 1361 break; 1362 case DW_ATE_boolean: 1363 intr->intr_type = INTR_INT; 1364 intr->intr_signed = 1; 1365 intr->intr_iformat = 'b'; 1366 break; 1367 case DW_ATE_float: 1368 case DW_ATE_complex_float: 1369 case DW_ATE_imaginary_float: 1370 #if defined(sun) 1371 case DW_ATE_SUN_imaginary_float: 1372 case DW_ATE_SUN_interval_float: 1373 #endif 1374 intr->intr_type = INTR_REAL; 1375 intr->intr_signed = 1; 1376 intr->intr_fformat = die_base_type2enc(dw, off, enc, sz); 1377 break; 1378 default: 1379 terminate("die %llu: unknown base type encoding 0x%llx\n", 1380 off, enc); 1381 } 1382 1383 return (intr); 1384 } 1385 1386 static void 1387 die_base_create(dwarf_t *dw, Dwarf_Die base, Dwarf_Off off, tdesc_t *tdp) 1388 { 1389 Dwarf_Unsigned sz; 1390 intr_t *intr; 1391 char *new; 1392 1393 debug(3, "die %llu: creating base type\n", off); 1394 1395 /* 1396 * The compilers have their own clever (internally inconsistent) ideas 1397 * as to what base types should look like. Some times gcc will, for 1398 * example, use DW_ATE_signed_char for char. Other times, however, it 1399 * will use DW_ATE_signed. Needless to say, this causes some problems 1400 * down the road, particularly with merging. We do, however, use the 1401 * DWARF idea of type sizes, as this allows us to avoid caring about 1402 * the data model. 1403 */ 1404 (void) die_unsigned(dw, base, DW_AT_byte_size, &sz, DW_ATTR_REQ); 1405 1406 /* Check for bogus gcc DW_AT_byte_size attribute */ 1407 if (sz == (unsigned)-1) { 1408 printf("dwarf.c:%s() working around bogus -1 DW_AT_byte_size\n", 1409 __func__); 1410 sz = 0; 1411 } 1412 1413 if (tdp->t_name == NULL) 1414 terminate("die %llu: base type without name\n", off); 1415 1416 /* XXX make a name parser for float too */ 1417 if ((intr = die_base_name_parse(tdp->t_name, &new)) != NULL) { 1418 /* Found it. We'll use the parsed version */ 1419 debug(3, "die %llu: name \"%s\" remapped to \"%s\"\n", off, 1420 tdesc_name(tdp), new); 1421 1422 free(tdp->t_name); 1423 tdp->t_name = new; 1424 } else { 1425 /* 1426 * We didn't recognize the type, so we'll create an intr_t 1427 * based on the DWARF data. 1428 */ 1429 debug(3, "die %llu: using dwarf data for base \"%s\"\n", off, 1430 tdesc_name(tdp)); 1431 1432 intr = die_base_from_dwarf(dw, base, off, sz); 1433 } 1434 1435 intr->intr_nbits = sz * 8; 1436 1437 tdp->t_type = INTRINSIC; 1438 tdp->t_intr = intr; 1439 tdp->t_size = sz; 1440 1441 tdp->t_flags |= TDESC_F_RESOLVED; 1442 } 1443 1444 static void 1445 die_through_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp, 1446 int type, const char *typename) 1447 { 1448 Dwarf_Attribute attr; 1449 1450 debug(3, "die %llu <%llx>: creating %s type %d\n", off, off, typename, type); 1451 1452 tdp->t_type = type; 1453 1454 if ((attr = die_attr(dw, die, DW_AT_type, 0)) != NULL) { 1455 tdp->t_tdesc = die_lookup_pass1(dw, die, DW_AT_type); 1456 } else { 1457 tdp->t_tdesc = tdesc_intr_void(dw); 1458 } 1459 1460 if (type == POINTER) 1461 tdp->t_size = dw->dw_ptrsz; 1462 1463 tdp->t_flags |= TDESC_F_RESOLVED; 1464 1465 if (type == TYPEDEF) { 1466 iidesc_t *ii = xcalloc(sizeof (iidesc_t)); 1467 ii->ii_type = II_TYPE; 1468 ii->ii_name = xstrdup(tdp->t_name); 1469 ii->ii_dtype = tdp; 1470 1471 iidesc_add(dw->dw_td->td_iihash, ii); 1472 } 1473 } 1474 1475 static void 1476 die_typedef_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1477 { 1478 die_through_create(dw, die, off, tdp, TYPEDEF, "typedef"); 1479 } 1480 1481 static void 1482 die_const_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1483 { 1484 die_through_create(dw, die, off, tdp, CONST, "const"); 1485 } 1486 1487 static void 1488 die_pointer_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1489 { 1490 die_through_create(dw, die, off, tdp, POINTER, "pointer"); 1491 } 1492 1493 static void 1494 die_restrict_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1495 { 1496 die_through_create(dw, die, off, tdp, RESTRICT, "restrict"); 1497 } 1498 1499 static void 1500 die_volatile_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1501 { 1502 die_through_create(dw, die, off, tdp, VOLATILE, "volatile"); 1503 } 1504 1505 /*ARGSUSED3*/ 1506 static void 1507 die_function_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp __unused) 1508 { 1509 Dwarf_Die arg; 1510 Dwarf_Half tag; 1511 iidesc_t *ii; 1512 char *name; 1513 1514 debug(3, "die %llu <%llx>: creating function definition\n", off, off); 1515 1516 /* 1517 * We'll begin by processing any type definition nodes that may be 1518 * lurking underneath this one. 1519 */ 1520 for (arg = die_child(dw, die); arg != NULL; 1521 arg = die_sibling(dw, arg)) { 1522 if ((tag = die_tag(dw, arg)) != DW_TAG_formal_parameter && 1523 tag != DW_TAG_variable) { 1524 /* Nested type declaration */ 1525 die_create_one(dw, arg); 1526 } 1527 } 1528 1529 if (die_isdecl(dw, die) || (name = die_name(dw, die)) == NULL) { 1530 /* 1531 * We process neither prototypes nor subprograms without 1532 * names. 1533 */ 1534 return; 1535 } 1536 1537 ii = xcalloc(sizeof (iidesc_t)); 1538 ii->ii_type = die_isglobal(dw, die) ? II_GFUN : II_SFUN; 1539 ii->ii_name = name; 1540 if (ii->ii_type == II_SFUN) 1541 ii->ii_owner = xstrdup(dw->dw_cuname); 1542 1543 debug(3, "die %llu: function %s is %s\n", off, ii->ii_name, 1544 (ii->ii_type == II_GFUN ? "global" : "static")); 1545 1546 if (die_attr(dw, die, DW_AT_type, 0) != NULL) 1547 ii->ii_dtype = die_lookup_pass1(dw, die, DW_AT_type); 1548 else 1549 ii->ii_dtype = tdesc_intr_void(dw); 1550 1551 for (arg = die_child(dw, die); arg != NULL; 1552 arg = die_sibling(dw, arg)) { 1553 char *name1; 1554 1555 debug(3, "die %llu: looking at sub member at %llu\n", 1556 off, die_off(dw, die)); 1557 1558 if (die_tag(dw, arg) != DW_TAG_formal_parameter) 1559 continue; 1560 1561 if ((name1 = die_name(dw, arg)) == NULL) { 1562 terminate("die %llu: func arg %d has no name\n", 1563 off, ii->ii_nargs + 1); 1564 } 1565 1566 if (strcmp(name1, "...") == 0) { 1567 free(name1); 1568 ii->ii_vargs = 1; 1569 continue; 1570 } 1571 1572 ii->ii_nargs++; 1573 } 1574 1575 if (ii->ii_nargs > 0) { 1576 int i; 1577 1578 debug(3, "die %llu: function has %d argument%s\n", off, 1579 ii->ii_nargs, (ii->ii_nargs == 1 ? "" : "s")); 1580 1581 ii->ii_args = xcalloc(sizeof (tdesc_t) * ii->ii_nargs); 1582 1583 for (arg = die_child(dw, die), i = 0; 1584 arg != NULL && i < ii->ii_nargs; 1585 arg = die_sibling(dw, arg)) { 1586 if (die_tag(dw, arg) != DW_TAG_formal_parameter) 1587 continue; 1588 1589 ii->ii_args[i++] = die_lookup_pass1(dw, arg, 1590 DW_AT_type); 1591 } 1592 } 1593 1594 iidesc_add(dw->dw_td->td_iihash, ii); 1595 } 1596 1597 /*ARGSUSED3*/ 1598 static void 1599 die_variable_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp __unused) 1600 { 1601 iidesc_t *ii; 1602 char *name; 1603 1604 debug(3, "die %llu: creating object definition\n", off); 1605 1606 if (die_isdecl(dw, die) || (name = die_name(dw, die)) == NULL) 1607 return; /* skip prototypes and nameless objects */ 1608 1609 ii = xcalloc(sizeof (iidesc_t)); 1610 ii->ii_type = die_isglobal(dw, die) ? II_GVAR : II_SVAR; 1611 ii->ii_name = name; 1612 ii->ii_dtype = die_lookup_pass1(dw, die, DW_AT_type); 1613 if (ii->ii_type == II_SVAR) 1614 ii->ii_owner = xstrdup(dw->dw_cuname); 1615 1616 iidesc_add(dw->dw_td->td_iihash, ii); 1617 } 1618 1619 /*ARGSUSED2*/ 1620 static int 1621 die_fwd_resolve(tdesc_t *fwd, tdesc_t **fwdp, void *private __unused) 1622 { 1623 if (fwd->t_flags & TDESC_F_RESOLVED) 1624 return (1); 1625 1626 if (fwd->t_tdesc != NULL) { 1627 debug(3, "tdp %u: unforwarded %s\n", fwd->t_id, 1628 tdesc_name(fwd)); 1629 *fwdp = fwd->t_tdesc; 1630 } 1631 1632 fwd->t_flags |= TDESC_F_RESOLVED; 1633 1634 return (1); 1635 } 1636 1637 /*ARGSUSED*/ 1638 static void 1639 die_lexblk_descend(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off __unused, tdesc_t *tdp __unused) 1640 { 1641 Dwarf_Die child = die_child(dw, die); 1642 1643 if (child != NULL) 1644 die_create(dw, child); 1645 } 1646 1647 /* 1648 * Used to map the die to a routine which can parse it, using the tag to do the 1649 * mapping. While the processing of most tags entails the creation of a tdesc, 1650 * there are a few which don't - primarily those which result in the creation of 1651 * iidescs which refer to existing tdescs. 1652 */ 1653 1654 #define DW_F_NOTDP 0x1 /* Don't create a tdesc for the creator */ 1655 1656 typedef struct die_creator { 1657 Dwarf_Half dc_tag; 1658 uint16_t dc_flags; 1659 void (*dc_create)(dwarf_t *, Dwarf_Die, Dwarf_Off, tdesc_t *); 1660 } die_creator_t; 1661 1662 static const die_creator_t die_creators[] = { 1663 { DW_TAG_array_type, 0, die_array_create }, 1664 { DW_TAG_enumeration_type, 0, die_enum_create }, 1665 { DW_TAG_lexical_block, DW_F_NOTDP, die_lexblk_descend }, 1666 { DW_TAG_pointer_type, 0, die_pointer_create }, 1667 { DW_TAG_structure_type, 0, die_struct_create }, 1668 { DW_TAG_subroutine_type, 0, die_funcptr_create }, 1669 { DW_TAG_typedef, 0, die_typedef_create }, 1670 { DW_TAG_union_type, 0, die_union_create }, 1671 { DW_TAG_base_type, 0, die_base_create }, 1672 { DW_TAG_const_type, 0, die_const_create }, 1673 { DW_TAG_subprogram, DW_F_NOTDP, die_function_create }, 1674 { DW_TAG_variable, DW_F_NOTDP, die_variable_create }, 1675 { DW_TAG_volatile_type, 0, die_volatile_create }, 1676 { DW_TAG_restrict_type, 0, die_restrict_create }, 1677 { 0, 0, NULL } 1678 }; 1679 1680 static const die_creator_t * 1681 die_tag2ctor(Dwarf_Half tag) 1682 { 1683 const die_creator_t *dc; 1684 1685 for (dc = die_creators; dc->dc_create != NULL; dc++) { 1686 if (dc->dc_tag == tag) 1687 return (dc); 1688 } 1689 1690 return (NULL); 1691 } 1692 1693 static void 1694 die_create_one(dwarf_t *dw, Dwarf_Die die) 1695 { 1696 Dwarf_Off off = die_off(dw, die); 1697 const die_creator_t *dc; 1698 Dwarf_Half tag; 1699 tdesc_t *tdp; 1700 1701 debug(3, "die %llu <%llx>: create_one\n", off, off); 1702 1703 if (off > dw->dw_maxoff) { 1704 terminate("illegal die offset %llu (max %llu)\n", off, 1705 dw->dw_maxoff); 1706 } 1707 1708 tag = die_tag(dw, die); 1709 1710 if ((dc = die_tag2ctor(tag)) == NULL) { 1711 debug(2, "die %llu: ignoring tag type %x\n", off, tag); 1712 return; 1713 } 1714 1715 if ((tdp = tdesc_lookup(dw, off)) == NULL && 1716 !(dc->dc_flags & DW_F_NOTDP)) { 1717 tdp = xcalloc(sizeof (tdesc_t)); 1718 tdp->t_id = off; 1719 tdesc_add(dw, tdp); 1720 } 1721 1722 if (tdp != NULL) 1723 tdp->t_name = die_name(dw, die); 1724 1725 dc->dc_create(dw, die, off, tdp); 1726 } 1727 1728 static void 1729 die_create(dwarf_t *dw, Dwarf_Die die) 1730 { 1731 do { 1732 die_create_one(dw, die); 1733 } while ((die = die_sibling(dw, die)) != NULL); 1734 } 1735 1736 static tdtrav_cb_f die_resolvers[] = { 1737 NULL, 1738 NULL, /* intrinsic */ 1739 NULL, /* pointer */ 1740 die_array_resolve, /* array */ 1741 NULL, /* function */ 1742 die_sou_resolve, /* struct */ 1743 die_sou_resolve, /* union */ 1744 die_enum_resolve, /* enum */ 1745 die_fwd_resolve, /* forward */ 1746 NULL, /* typedef */ 1747 NULL, /* typedef unres */ 1748 NULL, /* volatile */ 1749 NULL, /* const */ 1750 NULL, /* restrict */ 1751 }; 1752 1753 static tdtrav_cb_f die_fail_reporters[] = { 1754 NULL, 1755 NULL, /* intrinsic */ 1756 NULL, /* pointer */ 1757 die_array_failed, /* array */ 1758 NULL, /* function */ 1759 die_sou_failed, /* struct */ 1760 die_sou_failed, /* union */ 1761 NULL, /* enum */ 1762 NULL, /* forward */ 1763 NULL, /* typedef */ 1764 NULL, /* typedef unres */ 1765 NULL, /* volatile */ 1766 NULL, /* const */ 1767 NULL, /* restrict */ 1768 }; 1769 1770 static void 1771 die_resolve(dwarf_t *dw) 1772 { 1773 int last = -1; 1774 int pass = 0; 1775 1776 do { 1777 pass++; 1778 dw->dw_nunres = 0; 1779 1780 (void) iitraverse_hash(dw->dw_td->td_iihash, 1781 &dw->dw_td->td_curvgen, NULL, NULL, die_resolvers, dw); 1782 1783 debug(3, "resolve: pass %d, %u left\n", pass, dw->dw_nunres); 1784 1785 if ((int) dw->dw_nunres == last) { 1786 fprintf(stderr, "%s: failed to resolve the following " 1787 "types:\n", progname); 1788 1789 (void) iitraverse_hash(dw->dw_td->td_iihash, 1790 &dw->dw_td->td_curvgen, NULL, NULL, 1791 die_fail_reporters, dw); 1792 1793 terminate("failed to resolve types\n"); 1794 } 1795 1796 last = dw->dw_nunres; 1797 1798 } while (dw->dw_nunres != 0); 1799 } 1800 1801 /* 1802 * Any object containing a function or object symbol at any scope should also 1803 * contain DWARF data. 1804 */ 1805 static boolean_t 1806 should_have_dwarf(Elf *elf) 1807 { 1808 Elf_Scn *scn = NULL; 1809 Elf_Data *data = NULL; 1810 GElf_Shdr shdr; 1811 GElf_Sym sym; 1812 uint32_t symdx = 0; 1813 size_t nsyms = 0; 1814 boolean_t found = B_FALSE; 1815 1816 while ((scn = elf_nextscn(elf, scn)) != NULL) { 1817 gelf_getshdr(scn, &shdr); 1818 1819 if (shdr.sh_type == SHT_SYMTAB) { 1820 found = B_TRUE; 1821 break; 1822 } 1823 } 1824 1825 if (!found) 1826 terminate("cannot convert stripped objects\n"); 1827 1828 data = elf_getdata(scn, NULL); 1829 nsyms = shdr.sh_size / shdr.sh_entsize; 1830 1831 for (symdx = 0; symdx < nsyms; symdx++) { 1832 gelf_getsym(data, symdx, &sym); 1833 1834 if ((GELF_ST_TYPE(sym.st_info) == STT_FUNC) || 1835 (GELF_ST_TYPE(sym.st_info) == STT_TLS) || 1836 (GELF_ST_TYPE(sym.st_info) == STT_OBJECT)) { 1837 char *name; 1838 1839 name = elf_strptr(elf, shdr.sh_link, sym.st_name); 1840 1841 /* Studio emits these local symbols regardless */ 1842 if ((strcmp(name, "Bbss.bss") != 0) && 1843 (strcmp(name, "Ttbss.bss") != 0) && 1844 (strcmp(name, "Ddata.data") != 0) && 1845 (strcmp(name, "Ttdata.data") != 0) && 1846 (strcmp(name, "Drodata.rodata") != 0)) 1847 return (B_TRUE); 1848 } 1849 } 1850 1851 return (B_FALSE); 1852 } 1853 1854 /*ARGSUSED*/ 1855 int 1856 dw_read(tdata_t *td, Elf *elf, char *filename __unused) 1857 { 1858 Dwarf_Unsigned abboff, hdrlen, nxthdr; 1859 Dwarf_Half vers, addrsz, offsz; 1860 Dwarf_Die cu = 0; 1861 Dwarf_Die child = 0; 1862 dwarf_t dw; 1863 char *prod = NULL; 1864 int rc; 1865 1866 bzero(&dw, sizeof (dwarf_t)); 1867 dw.dw_td = td; 1868 dw.dw_ptrsz = elf_ptrsz(elf); 1869 dw.dw_mfgtid_last = TID_MFGTID_BASE; 1870 dw.dw_tidhash = hash_new(TDESC_HASH_BUCKETS, tdesc_idhash, tdesc_idcmp); 1871 dw.dw_fwdhash = hash_new(TDESC_HASH_BUCKETS, tdesc_namehash, 1872 tdesc_namecmp); 1873 dw.dw_enumhash = hash_new(TDESC_HASH_BUCKETS, tdesc_namehash, 1874 tdesc_namecmp); 1875 1876 if ((rc = dwarf_elf_init(elf, DW_DLC_READ, NULL, NULL, &dw.dw_dw, 1877 &dw.dw_err)) == DW_DLV_NO_ENTRY) { 1878 if (should_have_dwarf(elf)) { 1879 errno = ENOENT; 1880 return (-1); 1881 } else { 1882 return (0); 1883 } 1884 } else if (rc != DW_DLV_OK) { 1885 if (dwarf_errno(dw.dw_err) == DW_DLE_DEBUG_INFO_NULL) { 1886 /* 1887 * There's no type data in the DWARF section, but 1888 * libdwarf is too clever to handle that properly. 1889 */ 1890 return (0); 1891 } 1892 1893 terminate("failed to initialize DWARF: %s\n", 1894 dwarf_errmsg(dw.dw_err)); 1895 } 1896 1897 if ((rc = dwarf_next_cu_header_b(dw.dw_dw, &hdrlen, &vers, &abboff, 1898 &addrsz, &offsz, NULL, &nxthdr, &dw.dw_err)) != DW_DLV_OK) 1899 terminate("rc = %d %s\n", rc, dwarf_errmsg(dw.dw_err)); 1900 1901 if ((cu = die_sibling(&dw, NULL)) == NULL || 1902 (((child = die_child(&dw, cu)) == NULL) && 1903 should_have_dwarf(elf))) { 1904 terminate("file does not contain dwarf type data " 1905 "(try compiling with -g)\n"); 1906 } else if (child == NULL) { 1907 return (0); 1908 } 1909 1910 dw.dw_maxoff = nxthdr - 1; 1911 1912 if (dw.dw_maxoff > TID_FILEMAX) 1913 terminate("file contains too many types\n"); 1914 1915 debug(1, "DWARF version: %d\n", vers); 1916 if (vers != DWARF_VERSION) { 1917 terminate("file contains incompatible version %d DWARF code " 1918 "(version 2 required)\n", vers); 1919 } 1920 1921 if (die_string(&dw, cu, DW_AT_producer, &prod, 0)) { 1922 debug(1, "DWARF emitter: %s\n", prod); 1923 free(prod); 1924 } 1925 1926 if ((dw.dw_cuname = die_name(&dw, cu)) != NULL) { 1927 char *base = xstrdup(basename(dw.dw_cuname)); 1928 free(dw.dw_cuname); 1929 dw.dw_cuname = base; 1930 1931 debug(1, "CU name: %s\n", dw.dw_cuname); 1932 } 1933 1934 if ((child = die_child(&dw, cu)) != NULL) 1935 die_create(&dw, child); 1936 1937 if ((rc = dwarf_next_cu_header_b(dw.dw_dw, &hdrlen, &vers, &abboff, 1938 &addrsz, &offsz, NULL, &nxthdr, &dw.dw_err)) != DW_DLV_NO_ENTRY) 1939 terminate("multiple compilation units not supported\n"); 1940 1941 (void) dwarf_finish(dw.dw_dw, &dw.dw_err); 1942 1943 die_resolve(&dw); 1944 1945 cvt_fixups(td, dw.dw_ptrsz); 1946 1947 /* leak the dwarf_t */ 1948 1949 return (0); 1950 } 1951