dwarf.c revision 263915
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 127typedef 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 143static void die_create_one(dwarf_t *, Dwarf_Die); 144static void die_create(dwarf_t *, Dwarf_Die); 145 146static tid_t 147mfgtid_next(dwarf_t *dw) 148{ 149 return (++dw->dw_mfgtid_last); 150} 151 152static void 153tdesc_add(dwarf_t *dw, tdesc_t *tdp) 154{ 155 hash_add(dw->dw_tidhash, tdp); 156} 157 158static tdesc_t * 159tdesc_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 */ 176static size_t 177tdesc_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 210static size_t 211tdesc_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 246static tdesc_t * 247tdesc_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 265static Dwarf_Off 266die_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 279static Dwarf_Die 280die_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 297static Dwarf_Die 298die_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 314static Dwarf_Half 315die_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 328static Dwarf_Attribute 329die_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 351static int 352die_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) != DWARF_E_NONE) { 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 366static int 367die_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) != DWARF_E_NONE) { 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 381static int 382die_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) != DWARF_E_NONE) { 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 396static int 397die_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) != DWARF_E_NONE || 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 415static Dwarf_Off 416die_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) != DWARF_E_NONE) { 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 428static char * 429die_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 438static int 439die_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 446static int 447die_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 462static tdesc_t * 463die_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 474static tdesc_t * 475die_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 486static int 487die_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 Dwarf_Half form; 494 495 if (name != DW_AT_data_member_location) 496 terminate("die %llu: can only process attribute " 497 "DW_AT_data_member_location\n", die_off(dw, die)); 498 499 if ((at = die_attr(dw, die, name, 0)) == NULL) 500 return (0); 501 502 if (dwarf_whatform(at, &form, &dw->dw_err) != DW_DLV_OK) 503 return (0); 504 505 switch (form) { 506 case DW_FORM_block: 507 case DW_FORM_block1: 508 case DW_FORM_block2: 509 case DW_FORM_block4: 510 /* 511 * GCC in base and Clang (3.3 or below) generates 512 * DW_AT_data_member_location attribute with DW_FORM_block* 513 * form. The attribute contains one DW_OP_plus_uconst 514 * operator. The member offset stores in the operand. 515 */ 516 if (dwarf_locdesc(die, name, &loc, &locnum, &dw->dw_err) != 517 DW_DLV_OK) 518 return (0); 519 if (locnum != 1 || loc->ld_s->lr_atom != DW_OP_plus_uconst) { 520 terminate("die %llu: cannot parse member offset\n", 521 die_off(dw, die)); 522 } 523 *valp = loc->ld_s->lr_number; 524 break; 525 526 case DW_FORM_data1: 527 case DW_FORM_data2: 528 case DW_FORM_data4: 529 case DW_FORM_data8: 530 case DW_FORM_udata: 531 /* 532 * Clang 3.4 generates DW_AT_data_member_location attribute 533 * with DW_FORM_data* form (constant class). The attribute 534 * stores a contant value which is the member offset. 535 */ 536 if (dwarf_attrval_unsigned(die, name, valp, &dw->dw_err) != 537 DW_DLV_OK) 538 return (0); 539 break; 540 541 default: 542 terminate("die %llu: cannot parse member offset with form " 543 "%u\n", die_off(dw, die), form); 544 } 545 546 if (loc != NULL) 547 if (dwarf_locdesc_free(loc, &dw->dw_err) != DW_DLV_OK) 548 terminate("die %llu: cannot free location descriptor: %s\n", 549 die_off(dw, die), dwarf_errmsg(&dw->dw_err)); 550 551 return (1); 552} 553 554static tdesc_t * 555tdesc_intr_common(dwarf_t *dw, int tid, const char *name, size_t sz) 556{ 557 tdesc_t *tdp; 558 intr_t *intr; 559 560 intr = xcalloc(sizeof (intr_t)); 561 intr->intr_type = INTR_INT; 562 intr->intr_signed = 1; 563 intr->intr_nbits = sz * NBBY; 564 565 tdp = xcalloc(sizeof (tdesc_t)); 566 tdp->t_name = xstrdup(name); 567 tdp->t_size = sz; 568 tdp->t_id = tid; 569 tdp->t_type = INTRINSIC; 570 tdp->t_intr = intr; 571 tdp->t_flags = TDESC_F_RESOLVED; 572 573 tdesc_add(dw, tdp); 574 575 return (tdp); 576} 577 578/* 579 * Manufacture a void type. Used for gcc-emitted stabs, where the lack of a 580 * type reference implies a reference to a void type. A void *, for example 581 * will be represented by a pointer die without a DW_AT_type. CTF requires 582 * that pointer nodes point to something, so we'll create a void for use as 583 * the target. Note that the DWARF data may already create a void type. Ours 584 * would then be a duplicate, but it'll be removed in the self-uniquification 585 * merge performed at the completion of DWARF->tdesc conversion. 586 */ 587static tdesc_t * 588tdesc_intr_void(dwarf_t *dw) 589{ 590 if (dw->dw_void == NULL) 591 dw->dw_void = tdesc_intr_common(dw, TID_VOID, "void", 0); 592 593 return (dw->dw_void); 594} 595 596static tdesc_t * 597tdesc_intr_long(dwarf_t *dw) 598{ 599 if (dw->dw_long == NULL) { 600 dw->dw_long = tdesc_intr_common(dw, TID_LONG, "long", 601 dw->dw_ptrsz); 602 } 603 604 return (dw->dw_long); 605} 606 607/* 608 * Used for creating bitfield types. We create a copy of an existing intrinsic, 609 * adjusting the size of the copy to match what the caller requested. The 610 * caller can then use the copy as the type for a bitfield structure member. 611 */ 612static tdesc_t * 613tdesc_intr_clone(dwarf_t *dw, tdesc_t *old, size_t bitsz) 614{ 615 tdesc_t *new = xcalloc(sizeof (tdesc_t)); 616 617 if (!(old->t_flags & TDESC_F_RESOLVED)) { 618 terminate("tdp %u: attempt to make a bit field from an " 619 "unresolved type\n", old->t_id); 620 } 621 622 new->t_name = xstrdup(old->t_name); 623 new->t_size = old->t_size; 624 new->t_id = mfgtid_next(dw); 625 new->t_type = INTRINSIC; 626 new->t_flags = TDESC_F_RESOLVED; 627 628 new->t_intr = xcalloc(sizeof (intr_t)); 629 bcopy(old->t_intr, new->t_intr, sizeof (intr_t)); 630 new->t_intr->intr_nbits = bitsz; 631 632 tdesc_add(dw, new); 633 634 return (new); 635} 636 637static void 638tdesc_array_create(dwarf_t *dw, Dwarf_Die dim, tdesc_t *arrtdp, 639 tdesc_t *dimtdp) 640{ 641 Dwarf_Unsigned uval; 642 Dwarf_Signed sval; 643 tdesc_t *ctdp = NULL; 644 Dwarf_Die dim2; 645 ardef_t *ar; 646 647 if ((dim2 = die_sibling(dw, dim)) == NULL) { 648 ctdp = arrtdp; 649 } else if (die_tag(dw, dim2) == DW_TAG_subrange_type) { 650 ctdp = xcalloc(sizeof (tdesc_t)); 651 ctdp->t_id = mfgtid_next(dw); 652 debug(3, "die %llu: creating new type %u for sub-dimension\n", 653 die_off(dw, dim2), ctdp->t_id); 654 tdesc_array_create(dw, dim2, arrtdp, ctdp); 655 } else { 656 terminate("die %llu: unexpected non-subrange node in array\n", 657 die_off(dw, dim2)); 658 } 659 660 dimtdp->t_type = ARRAY; 661 dimtdp->t_ardef = ar = xcalloc(sizeof (ardef_t)); 662 663 /* 664 * Array bounds can be signed or unsigned, but there are several kinds 665 * of signless forms (data1, data2, etc) that take their sign from the 666 * routine that is trying to interpret them. That is, data1 can be 667 * either signed or unsigned, depending on whether you use the signed or 668 * unsigned accessor function. GCC will use the signless forms to store 669 * unsigned values which have their high bit set, so we need to try to 670 * read them first as unsigned to get positive values. We could also 671 * try signed first, falling back to unsigned if we got a negative 672 * value. 673 */ 674 if (die_unsigned(dw, dim, DW_AT_upper_bound, &uval, 0)) 675 ar->ad_nelems = uval + 1; 676 else if (die_signed(dw, dim, DW_AT_upper_bound, &sval, 0)) 677 ar->ad_nelems = sval + 1; 678 else 679 ar->ad_nelems = 0; 680 681 /* 682 * Different compilers use different index types. Force the type to be 683 * a common, known value (long). 684 */ 685 ar->ad_idxtype = tdesc_intr_long(dw); 686 ar->ad_contents = ctdp; 687 688 if (ar->ad_contents->t_size != 0) { 689 dimtdp->t_size = ar->ad_contents->t_size * ar->ad_nelems; 690 dimtdp->t_flags |= TDESC_F_RESOLVED; 691 } 692} 693 694/* 695 * Create a tdesc from an array node. Some arrays will come with byte size 696 * attributes, and thus can be resolved immediately. Others don't, and will 697 * need to wait until the second pass for resolution. 698 */ 699static void 700die_array_create(dwarf_t *dw, Dwarf_Die arr, Dwarf_Off off, tdesc_t *tdp) 701{ 702 tdesc_t *arrtdp = die_lookup_pass1(dw, arr, DW_AT_type); 703 Dwarf_Unsigned uval; 704 Dwarf_Die dim; 705 706 debug(3, "die %llu <%llx>: creating array\n", off, off); 707 708 if ((dim = die_child(dw, arr)) == NULL || 709 die_tag(dw, dim) != DW_TAG_subrange_type) 710 terminate("die %llu: failed to retrieve array bounds\n", off); 711 712 tdesc_array_create(dw, dim, arrtdp, tdp); 713 714 if (die_unsigned(dw, arr, DW_AT_byte_size, &uval, 0)) { 715 tdesc_t *dimtdp; 716 int flags; 717 718 /* Check for bogus gcc DW_AT_byte_size attribute */ 719 if (uval == (unsigned)-1) { 720 printf("dwarf.c:%s() working around bogus -1 DW_AT_byte_size\n", 721 __func__); 722 uval = 0; 723 } 724 725 tdp->t_size = uval; 726 727 /* 728 * Ensure that sub-dimensions have sizes too before marking 729 * as resolved. 730 */ 731 flags = TDESC_F_RESOLVED; 732 for (dimtdp = tdp->t_ardef->ad_contents; 733 dimtdp->t_type == ARRAY; 734 dimtdp = dimtdp->t_ardef->ad_contents) { 735 if (!(dimtdp->t_flags & TDESC_F_RESOLVED)) { 736 flags = 0; 737 break; 738 } 739 } 740 741 tdp->t_flags |= flags; 742 } 743 744 debug(3, "die %llu <%llx>: array nelems %u size %u\n", off, off, 745 tdp->t_ardef->ad_nelems, tdp->t_size); 746} 747 748/*ARGSUSED1*/ 749static int 750die_array_resolve(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private) 751{ 752 dwarf_t *dw = private; 753 size_t sz; 754 755 if (tdp->t_flags & TDESC_F_RESOLVED) 756 return (1); 757 758 debug(3, "trying to resolve array %d (cont %d)\n", tdp->t_id, 759 tdp->t_ardef->ad_contents->t_id); 760 761 if ((sz = tdesc_size(tdp->t_ardef->ad_contents)) == 0) { 762 debug(3, "unable to resolve array %s (%d) contents %d\n", 763 tdesc_name(tdp), tdp->t_id, 764 tdp->t_ardef->ad_contents->t_id); 765 766 dw->dw_nunres++; 767 return (1); 768 } 769 770 tdp->t_size = sz * tdp->t_ardef->ad_nelems; 771 tdp->t_flags |= TDESC_F_RESOLVED; 772 773 debug(3, "resolved array %d: %u bytes\n", tdp->t_id, tdp->t_size); 774 775 return (1); 776} 777 778/*ARGSUSED1*/ 779static int 780die_array_failed(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private __unused) 781{ 782 tdesc_t *cont = tdp->t_ardef->ad_contents; 783 784 if (tdp->t_flags & TDESC_F_RESOLVED) 785 return (1); 786 787 fprintf(stderr, "Array %d: failed to size contents type %s (%d)\n", 788 tdp->t_id, tdesc_name(cont), cont->t_id); 789 790 return (1); 791} 792 793/* 794 * Most enums (those with members) will be resolved during this first pass. 795 * Others - those without members (see the file comment) - won't be, and will 796 * need to wait until the second pass when they can be matched with their full 797 * definitions. 798 */ 799static void 800die_enum_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 801{ 802 Dwarf_Die mem; 803 Dwarf_Unsigned uval; 804 Dwarf_Signed sval; 805 806 debug(3, "die %llu: creating enum\n", off); 807 808 tdp->t_type = ENUM; 809 810 (void) die_unsigned(dw, die, DW_AT_byte_size, &uval, DW_ATTR_REQ); 811 /* Check for bogus gcc DW_AT_byte_size attribute */ 812 if (uval == (unsigned)-1) { 813 printf("dwarf.c:%s() working around bogus -1 DW_AT_byte_size\n", 814 __func__); 815 uval = 0; 816 } 817 tdp->t_size = uval; 818 819 if ((mem = die_child(dw, die)) != NULL) { 820 elist_t **elastp = &tdp->t_emem; 821 822 do { 823 elist_t *el; 824 825 if (die_tag(dw, mem) != DW_TAG_enumerator) { 826 /* Nested type declaration */ 827 die_create_one(dw, mem); 828 continue; 829 } 830 831 el = xcalloc(sizeof (elist_t)); 832 el->el_name = die_name(dw, mem); 833 834 if (die_signed(dw, mem, DW_AT_const_value, &sval, 0)) { 835 el->el_number = sval; 836 } else if (die_unsigned(dw, mem, DW_AT_const_value, 837 &uval, 0)) { 838 el->el_number = uval; 839 } else { 840 terminate("die %llu: enum %llu: member without " 841 "value\n", off, die_off(dw, mem)); 842 } 843 844 debug(3, "die %llu: enum %llu: created %s = %d\n", off, 845 die_off(dw, mem), el->el_name, el->el_number); 846 847 *elastp = el; 848 elastp = &el->el_next; 849 850 } while ((mem = die_sibling(dw, mem)) != NULL); 851 852 hash_add(dw->dw_enumhash, tdp); 853 854 tdp->t_flags |= TDESC_F_RESOLVED; 855 856 if (tdp->t_name != NULL) { 857 iidesc_t *ii = xcalloc(sizeof (iidesc_t)); 858 ii->ii_type = II_SOU; 859 ii->ii_name = xstrdup(tdp->t_name); 860 ii->ii_dtype = tdp; 861 862 iidesc_add(dw->dw_td->td_iihash, ii); 863 } 864 } 865} 866 867static int 868die_enum_match(void *arg1, void *arg2) 869{ 870 tdesc_t *tdp = arg1, **fullp = arg2; 871 872 if (tdp->t_emem != NULL) { 873 *fullp = tdp; 874 return (-1); /* stop the iteration */ 875 } 876 877 return (0); 878} 879 880/*ARGSUSED1*/ 881static int 882die_enum_resolve(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private) 883{ 884 dwarf_t *dw = private; 885 tdesc_t *full = NULL; 886 887 if (tdp->t_flags & TDESC_F_RESOLVED) 888 return (1); 889 890 (void) hash_find_iter(dw->dw_enumhash, tdp, die_enum_match, &full); 891 892 /* 893 * The answer to this one won't change from iteration to iteration, 894 * so don't even try. 895 */ 896 if (full == NULL) { 897 terminate("tdp %u: enum %s has no members\n", tdp->t_id, 898 tdesc_name(tdp)); 899 } 900 901 debug(3, "tdp %u: enum %s redirected to %u\n", tdp->t_id, 902 tdesc_name(tdp), full->t_id); 903 904 tdp->t_flags |= TDESC_F_RESOLVED; 905 906 return (1); 907} 908 909static int 910die_fwd_map(void *arg1, void *arg2) 911{ 912 tdesc_t *fwd = arg1, *sou = arg2; 913 914 debug(3, "tdp %u: mapped forward %s to sou %u\n", fwd->t_id, 915 tdesc_name(fwd), sou->t_id); 916 fwd->t_tdesc = sou; 917 918 return (0); 919} 920 921/* 922 * Structures and unions will never be resolved during the first pass, as we 923 * won't be able to fully determine the member sizes. The second pass, which 924 * have access to sizing information, will be able to complete the resolution. 925 */ 926static void 927die_sou_create(dwarf_t *dw, Dwarf_Die str, Dwarf_Off off, tdesc_t *tdp, 928 int type, const char *typename) 929{ 930 Dwarf_Unsigned sz, bitsz, bitoff, maxsz=0; 931#if BYTE_ORDER == _LITTLE_ENDIAN 932 Dwarf_Unsigned bysz; 933#endif 934 Dwarf_Die mem; 935 mlist_t *ml, **mlastp; 936 iidesc_t *ii; 937 938 tdp->t_type = (die_isdecl(dw, str) ? FORWARD : type); 939 940 debug(3, "die %llu: creating %s %s\n", off, 941 (tdp->t_type == FORWARD ? "forward decl" : typename), 942 tdesc_name(tdp)); 943 944 if (tdp->t_type == FORWARD) { 945 hash_add(dw->dw_fwdhash, tdp); 946 return; 947 } 948 949 (void) hash_find_iter(dw->dw_fwdhash, tdp, die_fwd_map, tdp); 950 951 (void) die_unsigned(dw, str, DW_AT_byte_size, &sz, DW_ATTR_REQ); 952 tdp->t_size = sz; 953 954 /* 955 * GCC allows empty SOUs as an extension. 956 */ 957 if ((mem = die_child(dw, str)) == NULL) { 958 goto out; 959 } 960 961 mlastp = &tdp->t_members; 962 963 do { 964 Dwarf_Off memoff = die_off(dw, mem); 965 Dwarf_Half tag = die_tag(dw, mem); 966 Dwarf_Unsigned mloff; 967 968 if (tag != DW_TAG_member) { 969 /* Nested type declaration */ 970 die_create_one(dw, mem); 971 continue; 972 } 973 974 debug(3, "die %llu: mem %llu: creating member\n", off, memoff); 975 976 ml = xcalloc(sizeof (mlist_t)); 977 978 /* 979 * This could be a GCC anon struct/union member, so we'll allow 980 * an empty name, even though nothing can really handle them 981 * properly. Note that some versions of GCC miss out debug 982 * info for anon structs, though recent versions are fixed (gcc 983 * bug 11816). 984 */ 985 if ((ml->ml_name = die_name(dw, mem)) == NULL) 986 ml->ml_name = NULL; 987 988 ml->ml_type = die_lookup_pass1(dw, mem, DW_AT_type); 989 debug(3, "die_sou_create(): ml_type = %p t_id = %d\n", 990 ml->ml_type, ml->ml_type->t_id); 991 992 if (die_mem_offset(dw, mem, DW_AT_data_member_location, 993 &mloff, 0)) { 994 debug(3, "die %llu: got mloff %llx\n", off, 995 (u_longlong_t)mloff); 996 ml->ml_offset = mloff * 8; 997 } 998 999 if (die_unsigned(dw, mem, DW_AT_bit_size, &bitsz, 0)) 1000 ml->ml_size = bitsz; 1001 else 1002 ml->ml_size = tdesc_bitsize(ml->ml_type); 1003 1004 if (die_unsigned(dw, mem, DW_AT_bit_offset, &bitoff, 0)) { 1005#if BYTE_ORDER == _BIG_ENDIAN 1006 ml->ml_offset += bitoff; 1007#else 1008 /* 1009 * Note that Clang 3.4 will sometimes generate 1010 * member DIE before generating the DIE for the 1011 * member's type. The code can not handle this 1012 * properly so that tdesc_bitsize(ml->ml_type) will 1013 * return 0 because ml->ml_type is unknown. As a 1014 * result, a wrong member offset will be calculated. 1015 * To workaround this, we can instead try to 1016 * retrieve the value of DW_AT_byte_size attribute 1017 * which stores the byte size of the space occupied 1018 * by the type. If this attribute exists, its value 1019 * should equal to tdesc_bitsize(ml->ml_type)/NBBY. 1020 */ 1021 if (die_unsigned(dw, mem, DW_AT_byte_size, &bysz, 0) && 1022 bysz > 0) 1023 ml->ml_offset += bysz * NBBY - bitoff - 1024 ml->ml_size; 1025 else 1026 ml->ml_offset += tdesc_bitsize(ml->ml_type) - 1027 bitoff - ml->ml_size; 1028#endif 1029 } 1030 1031 debug(3, "die %llu: mem %llu: created \"%s\" (off %u sz %u)\n", 1032 off, memoff, ml->ml_name, ml->ml_offset, ml->ml_size); 1033 1034 *mlastp = ml; 1035 mlastp = &ml->ml_next; 1036 1037 /* Find the size of the largest member to work around a gcc 1038 * bug. See GCC Bugzilla 35998. 1039 */ 1040 if (maxsz < ml->ml_size) 1041 maxsz = ml->ml_size; 1042 1043 } while ((mem = die_sibling(dw, mem)) != NULL); 1044 1045 /* See if we got a bogus DW_AT_byte_size. GCC will sometimes 1046 * emit this. 1047 */ 1048 if (sz == (unsigned)-1) { 1049 printf("dwarf.c:%s() working around bogus -1 DW_AT_byte_size\n", 1050 __func__); 1051 tdp->t_size = maxsz / 8; /* maxsz is in bits, t_size is bytes */ 1052 } 1053 1054 /* 1055 * GCC will attempt to eliminate unused types, thus decreasing the 1056 * size of the emitted dwarf. That is, if you declare a foo_t in your 1057 * header, include said header in your source file, and neglect to 1058 * actually use (directly or indirectly) the foo_t in the source file, 1059 * the foo_t won't make it into the emitted DWARF. So, at least, goes 1060 * the theory. 1061 * 1062 * Occasionally, it'll emit the DW_TAG_structure_type for the foo_t, 1063 * and then neglect to emit the members. Strangely, the loner struct 1064 * tag will always be followed by a proper nested declaration of 1065 * something else. This is clearly a bug, but we're not going to have 1066 * time to get it fixed before this goo goes back, so we'll have to work 1067 * around it. If we see a no-membered struct with a nested declaration 1068 * (i.e. die_child of the struct tag won't be null), we'll ignore it. 1069 * Being paranoid, we won't simply remove it from the hash. Instead, 1070 * we'll decline to create an iidesc for it, thus ensuring that this 1071 * type won't make it into the output file. To be safe, we'll also 1072 * change the name. 1073 */ 1074 if (tdp->t_members == NULL) { 1075 const char *old = tdesc_name(tdp); 1076 size_t newsz = 7 + strlen(old) + 1; 1077 char *new = xmalloc(newsz); 1078 (void) snprintf(new, newsz, "orphan %s", old); 1079 1080 debug(3, "die %llu: worked around %s %s\n", off, typename, old); 1081 1082 if (tdp->t_name != NULL) 1083 free(tdp->t_name); 1084 tdp->t_name = new; 1085 return; 1086 } 1087 1088out: 1089 if (tdp->t_name != NULL) { 1090 ii = xcalloc(sizeof (iidesc_t)); 1091 ii->ii_type = II_SOU; 1092 ii->ii_name = xstrdup(tdp->t_name); 1093 ii->ii_dtype = tdp; 1094 1095 iidesc_add(dw->dw_td->td_iihash, ii); 1096 } 1097} 1098 1099static void 1100die_struct_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1101{ 1102 die_sou_create(dw, die, off, tdp, STRUCT, "struct"); 1103} 1104 1105static void 1106die_union_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1107{ 1108 die_sou_create(dw, die, off, tdp, UNION, "union"); 1109} 1110 1111/*ARGSUSED1*/ 1112static int 1113die_sou_resolve(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private) 1114{ 1115 dwarf_t *dw = private; 1116 mlist_t *ml; 1117 tdesc_t *mt; 1118 1119 if (tdp->t_flags & TDESC_F_RESOLVED) 1120 return (1); 1121 1122 debug(3, "resolving sou %s\n", tdesc_name(tdp)); 1123 1124 for (ml = tdp->t_members; ml != NULL; ml = ml->ml_next) { 1125 if (ml->ml_size == 0) { 1126 mt = tdesc_basetype(ml->ml_type); 1127 1128 if ((ml->ml_size = tdesc_bitsize(mt)) != 0) 1129 continue; 1130 1131 /* 1132 * For empty members, or GCC/C99 flexible array 1133 * members, a size of 0 is correct. 1134 */ 1135 if (mt->t_members == NULL) 1136 continue; 1137 if (mt->t_type == ARRAY && mt->t_ardef->ad_nelems == 0) 1138 continue; 1139 1140 dw->dw_nunres++; 1141 return (1); 1142 } 1143 1144 if ((mt = tdesc_basetype(ml->ml_type)) == NULL) { 1145 dw->dw_nunres++; 1146 return (1); 1147 } 1148 1149 if (ml->ml_size != 0 && mt->t_type == INTRINSIC && 1150 mt->t_intr->intr_nbits != (int)ml->ml_size) { 1151 /* 1152 * This member is a bitfield, and needs to reference 1153 * an intrinsic type with the same width. If the 1154 * currently-referenced type isn't of the same width, 1155 * we'll copy it, adjusting the width of the copy to 1156 * the size we'd like. 1157 */ 1158 debug(3, "tdp %u: creating bitfield for %d bits\n", 1159 tdp->t_id, ml->ml_size); 1160 1161 ml->ml_type = tdesc_intr_clone(dw, mt, ml->ml_size); 1162 } 1163 } 1164 1165 tdp->t_flags |= TDESC_F_RESOLVED; 1166 1167 return (1); 1168} 1169 1170/*ARGSUSED1*/ 1171static int 1172die_sou_failed(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private __unused) 1173{ 1174 const char *typename = (tdp->t_type == STRUCT ? "struct" : "union"); 1175 mlist_t *ml; 1176 1177 if (tdp->t_flags & TDESC_F_RESOLVED) 1178 return (1); 1179 1180 for (ml = tdp->t_members; ml != NULL; ml = ml->ml_next) { 1181 if (ml->ml_size == 0) { 1182 fprintf(stderr, "%s %d <%x>: failed to size member \"%s\" " 1183 "of type %s (%d <%x>)\n", typename, tdp->t_id, 1184 tdp->t_id, 1185 ml->ml_name, tdesc_name(ml->ml_type), 1186 ml->ml_type->t_id, ml->ml_type->t_id); 1187 } 1188 } 1189 1190 return (1); 1191} 1192 1193static void 1194die_funcptr_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1195{ 1196 Dwarf_Attribute attr; 1197 Dwarf_Half tag; 1198 Dwarf_Die arg; 1199 fndef_t *fn; 1200 int i; 1201 1202 debug(3, "die %llu <%llx>: creating function pointer\n", off, off); 1203 1204 /* 1205 * We'll begin by processing any type definition nodes that may be 1206 * lurking underneath this one. 1207 */ 1208 for (arg = die_child(dw, die); arg != NULL; 1209 arg = die_sibling(dw, arg)) { 1210 if ((tag = die_tag(dw, arg)) != DW_TAG_formal_parameter && 1211 tag != DW_TAG_unspecified_parameters) { 1212 /* Nested type declaration */ 1213 die_create_one(dw, arg); 1214 } 1215 } 1216 1217 if (die_isdecl(dw, die)) { 1218 /* 1219 * This is a prototype. We don't add prototypes to the 1220 * tree, so we're going to drop the tdesc. Unfortunately, 1221 * it has already been added to the tree. Nobody will reference 1222 * it, though, and it will be leaked. 1223 */ 1224 return; 1225 } 1226 1227 fn = xcalloc(sizeof (fndef_t)); 1228 1229 tdp->t_type = FUNCTION; 1230 1231 if ((attr = die_attr(dw, die, DW_AT_type, 0)) != NULL) { 1232 fn->fn_ret = die_lookup_pass1(dw, die, DW_AT_type); 1233 } else { 1234 fn->fn_ret = tdesc_intr_void(dw); 1235 } 1236 1237 /* 1238 * Count the arguments to the function, then read them in. 1239 */ 1240 for (fn->fn_nargs = 0, arg = die_child(dw, die); arg != NULL; 1241 arg = die_sibling(dw, arg)) { 1242 if ((tag = die_tag(dw, arg)) == DW_TAG_formal_parameter) 1243 fn->fn_nargs++; 1244 else if (tag == DW_TAG_unspecified_parameters && 1245 fn->fn_nargs > 0) 1246 fn->fn_vargs = 1; 1247 } 1248 1249 if (fn->fn_nargs != 0) { 1250 debug(3, "die %llu: adding %d argument%s\n", off, fn->fn_nargs, 1251 (fn->fn_nargs > 1 ? "s" : "")); 1252 1253 fn->fn_args = xcalloc(sizeof (tdesc_t *) * fn->fn_nargs); 1254 for (i = 0, arg = die_child(dw, die); 1255 arg != NULL && i < (int) fn->fn_nargs; 1256 arg = die_sibling(dw, arg)) { 1257 if (die_tag(dw, arg) != DW_TAG_formal_parameter) 1258 continue; 1259 1260 fn->fn_args[i++] = die_lookup_pass1(dw, arg, 1261 DW_AT_type); 1262 } 1263 } 1264 1265 tdp->t_fndef = fn; 1266 tdp->t_flags |= TDESC_F_RESOLVED; 1267} 1268 1269/* 1270 * GCC and DevPro use different names for the base types. While the terms are 1271 * the same, they are arranged in a different order. Some terms, such as int, 1272 * are implied in one, and explicitly named in the other. Given a base type 1273 * as input, this routine will return a common name, along with an intr_t 1274 * that reflects said name. 1275 */ 1276static intr_t * 1277die_base_name_parse(const char *name, char **newp) 1278{ 1279 char buf[100]; 1280 char const *base; 1281 char *c; 1282 int nlong = 0, nshort = 0, nchar = 0, nint = 0; 1283 int sign = 1; 1284 char fmt = '\0'; 1285 intr_t *intr; 1286 1287 if (strlen(name) > sizeof (buf) - 1) 1288 terminate("base type name \"%s\" is too long\n", name); 1289 1290 strncpy(buf, name, sizeof (buf)); 1291 1292 for (c = strtok(buf, " "); c != NULL; c = strtok(NULL, " ")) { 1293 if (strcmp(c, "signed") == 0) 1294 sign = 1; 1295 else if (strcmp(c, "unsigned") == 0) 1296 sign = 0; 1297 else if (strcmp(c, "long") == 0) 1298 nlong++; 1299 else if (strcmp(c, "char") == 0) { 1300 nchar++; 1301 fmt = 'c'; 1302 } else if (strcmp(c, "short") == 0) 1303 nshort++; 1304 else if (strcmp(c, "int") == 0) 1305 nint++; 1306 else { 1307 /* 1308 * If we don't recognize any of the tokens, we'll tell 1309 * the caller to fall back to the dwarf-provided 1310 * encoding information. 1311 */ 1312 return (NULL); 1313 } 1314 } 1315 1316 if (nchar > 1 || nshort > 1 || nint > 1 || nlong > 2) 1317 return (NULL); 1318 1319 if (nchar > 0) { 1320 if (nlong > 0 || nshort > 0 || nint > 0) 1321 return (NULL); 1322 1323 base = "char"; 1324 1325 } else if (nshort > 0) { 1326 if (nlong > 0) 1327 return (NULL); 1328 1329 base = "short"; 1330 1331 } else if (nlong > 0) { 1332 base = "long"; 1333 1334 } else { 1335 base = "int"; 1336 } 1337 1338 intr = xcalloc(sizeof (intr_t)); 1339 intr->intr_type = INTR_INT; 1340 intr->intr_signed = sign; 1341 intr->intr_iformat = fmt; 1342 1343 snprintf(buf, sizeof (buf), "%s%s%s", 1344 (sign ? "" : "unsigned "), 1345 (nlong > 1 ? "long " : ""), 1346 base); 1347 1348 *newp = xstrdup(buf); 1349 return (intr); 1350} 1351 1352typedef struct fp_size_map { 1353 size_t fsm_typesz[2]; /* size of {32,64} type */ 1354 uint_t fsm_enc[3]; /* CTF_FP_* for {bare,cplx,imagry} type */ 1355} fp_size_map_t; 1356 1357static const fp_size_map_t fp_encodings[] = { 1358 { { 4, 4 }, { CTF_FP_SINGLE, CTF_FP_CPLX, CTF_FP_IMAGRY } }, 1359 { { 8, 8 }, { CTF_FP_DOUBLE, CTF_FP_DCPLX, CTF_FP_DIMAGRY } }, 1360#ifdef __sparc 1361 { { 16, 16 }, { CTF_FP_LDOUBLE, CTF_FP_LDCPLX, CTF_FP_LDIMAGRY } }, 1362#else 1363 { { 12, 16 }, { CTF_FP_LDOUBLE, CTF_FP_LDCPLX, CTF_FP_LDIMAGRY } }, 1364#endif 1365 { { 0, 0 }, { 0, 0, 0 } } 1366}; 1367 1368static uint_t 1369die_base_type2enc(dwarf_t *dw, Dwarf_Off off, Dwarf_Signed enc, size_t sz) 1370{ 1371 const fp_size_map_t *map = fp_encodings; 1372 uint_t szidx = dw->dw_ptrsz == sizeof (uint64_t); 1373 uint_t mult = 1, col = 0; 1374 1375 if (enc == DW_ATE_complex_float) { 1376 mult = 2; 1377 col = 1; 1378 } else if (enc == DW_ATE_imaginary_float 1379#if defined(sun) 1380 || enc == DW_ATE_SUN_imaginary_float 1381#endif 1382 ) 1383 col = 2; 1384 1385 while (map->fsm_typesz[szidx] != 0) { 1386 if (map->fsm_typesz[szidx] * mult == sz) 1387 return (map->fsm_enc[col]); 1388 map++; 1389 } 1390 1391 terminate("die %llu: unrecognized real type size %u\n", off, sz); 1392 /*NOTREACHED*/ 1393 return (0); 1394} 1395 1396static intr_t * 1397die_base_from_dwarf(dwarf_t *dw, Dwarf_Die base, Dwarf_Off off, size_t sz) 1398{ 1399 intr_t *intr = xcalloc(sizeof (intr_t)); 1400 Dwarf_Signed enc; 1401 1402 (void) die_signed(dw, base, DW_AT_encoding, &enc, DW_ATTR_REQ); 1403 1404 switch (enc) { 1405 case DW_ATE_unsigned: 1406 case DW_ATE_address: 1407 intr->intr_type = INTR_INT; 1408 break; 1409 case DW_ATE_unsigned_char: 1410 intr->intr_type = INTR_INT; 1411 intr->intr_iformat = 'c'; 1412 break; 1413 case DW_ATE_signed: 1414 intr->intr_type = INTR_INT; 1415 intr->intr_signed = 1; 1416 break; 1417 case DW_ATE_signed_char: 1418 intr->intr_type = INTR_INT; 1419 intr->intr_signed = 1; 1420 intr->intr_iformat = 'c'; 1421 break; 1422 case DW_ATE_boolean: 1423 intr->intr_type = INTR_INT; 1424 intr->intr_signed = 1; 1425 intr->intr_iformat = 'b'; 1426 break; 1427 case DW_ATE_float: 1428 case DW_ATE_complex_float: 1429 case DW_ATE_imaginary_float: 1430#if defined(sun) 1431 case DW_ATE_SUN_imaginary_float: 1432 case DW_ATE_SUN_interval_float: 1433#endif 1434 intr->intr_type = INTR_REAL; 1435 intr->intr_signed = 1; 1436 intr->intr_fformat = die_base_type2enc(dw, off, enc, sz); 1437 break; 1438 default: 1439 terminate("die %llu: unknown base type encoding 0x%llx\n", 1440 off, enc); 1441 } 1442 1443 return (intr); 1444} 1445 1446static void 1447die_base_create(dwarf_t *dw, Dwarf_Die base, Dwarf_Off off, tdesc_t *tdp) 1448{ 1449 Dwarf_Unsigned sz; 1450 intr_t *intr; 1451 char *new; 1452 1453 debug(3, "die %llu: creating base type\n", off); 1454 1455 /* 1456 * The compilers have their own clever (internally inconsistent) ideas 1457 * as to what base types should look like. Some times gcc will, for 1458 * example, use DW_ATE_signed_char for char. Other times, however, it 1459 * will use DW_ATE_signed. Needless to say, this causes some problems 1460 * down the road, particularly with merging. We do, however, use the 1461 * DWARF idea of type sizes, as this allows us to avoid caring about 1462 * the data model. 1463 */ 1464 (void) die_unsigned(dw, base, DW_AT_byte_size, &sz, DW_ATTR_REQ); 1465 1466 /* Check for bogus gcc DW_AT_byte_size attribute */ 1467 if (sz == (unsigned)-1) { 1468 printf("dwarf.c:%s() working around bogus -1 DW_AT_byte_size\n", 1469 __func__); 1470 sz = 0; 1471 } 1472 1473 if (tdp->t_name == NULL) 1474 terminate("die %llu: base type without name\n", off); 1475 1476 /* XXX make a name parser for float too */ 1477 if ((intr = die_base_name_parse(tdp->t_name, &new)) != NULL) { 1478 /* Found it. We'll use the parsed version */ 1479 debug(3, "die %llu: name \"%s\" remapped to \"%s\"\n", off, 1480 tdesc_name(tdp), new); 1481 1482 free(tdp->t_name); 1483 tdp->t_name = new; 1484 } else { 1485 /* 1486 * We didn't recognize the type, so we'll create an intr_t 1487 * based on the DWARF data. 1488 */ 1489 debug(3, "die %llu: using dwarf data for base \"%s\"\n", off, 1490 tdesc_name(tdp)); 1491 1492 intr = die_base_from_dwarf(dw, base, off, sz); 1493 } 1494 1495 intr->intr_nbits = sz * 8; 1496 1497 tdp->t_type = INTRINSIC; 1498 tdp->t_intr = intr; 1499 tdp->t_size = sz; 1500 1501 tdp->t_flags |= TDESC_F_RESOLVED; 1502} 1503 1504static void 1505die_through_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp, 1506 int type, const char *typename) 1507{ 1508 Dwarf_Attribute attr; 1509 1510 debug(3, "die %llu <%llx>: creating %s type %d\n", off, off, typename, type); 1511 1512 tdp->t_type = type; 1513 1514 if ((attr = die_attr(dw, die, DW_AT_type, 0)) != NULL) { 1515 tdp->t_tdesc = die_lookup_pass1(dw, die, DW_AT_type); 1516 } else { 1517 tdp->t_tdesc = tdesc_intr_void(dw); 1518 } 1519 1520 if (type == POINTER) 1521 tdp->t_size = dw->dw_ptrsz; 1522 1523 tdp->t_flags |= TDESC_F_RESOLVED; 1524 1525 if (type == TYPEDEF) { 1526 iidesc_t *ii = xcalloc(sizeof (iidesc_t)); 1527 ii->ii_type = II_TYPE; 1528 ii->ii_name = xstrdup(tdp->t_name); 1529 ii->ii_dtype = tdp; 1530 1531 iidesc_add(dw->dw_td->td_iihash, ii); 1532 } 1533} 1534 1535static void 1536die_typedef_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1537{ 1538 die_through_create(dw, die, off, tdp, TYPEDEF, "typedef"); 1539} 1540 1541static void 1542die_const_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1543{ 1544 die_through_create(dw, die, off, tdp, CONST, "const"); 1545} 1546 1547static void 1548die_pointer_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1549{ 1550 die_through_create(dw, die, off, tdp, POINTER, "pointer"); 1551} 1552 1553static void 1554die_restrict_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1555{ 1556 die_through_create(dw, die, off, tdp, RESTRICT, "restrict"); 1557} 1558 1559static void 1560die_volatile_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp) 1561{ 1562 die_through_create(dw, die, off, tdp, VOLATILE, "volatile"); 1563} 1564 1565/*ARGSUSED3*/ 1566static void 1567die_function_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp __unused) 1568{ 1569 Dwarf_Die arg; 1570 Dwarf_Half tag; 1571 iidesc_t *ii; 1572 char *name; 1573 1574 debug(3, "die %llu <%llx>: creating function definition\n", off, off); 1575 1576 /* 1577 * We'll begin by processing any type definition nodes that may be 1578 * lurking underneath this one. 1579 */ 1580 for (arg = die_child(dw, die); arg != NULL; 1581 arg = die_sibling(dw, arg)) { 1582 if ((tag = die_tag(dw, arg)) != DW_TAG_formal_parameter && 1583 tag != DW_TAG_variable) { 1584 /* Nested type declaration */ 1585 die_create_one(dw, arg); 1586 } 1587 } 1588 1589 if (die_isdecl(dw, die) || (name = die_name(dw, die)) == NULL) { 1590 /* 1591 * We process neither prototypes nor subprograms without 1592 * names. 1593 */ 1594 return; 1595 } 1596 1597 ii = xcalloc(sizeof (iidesc_t)); 1598 ii->ii_type = die_isglobal(dw, die) ? II_GFUN : II_SFUN; 1599 ii->ii_name = name; 1600 if (ii->ii_type == II_SFUN) 1601 ii->ii_owner = xstrdup(dw->dw_cuname); 1602 1603 debug(3, "die %llu: function %s is %s\n", off, ii->ii_name, 1604 (ii->ii_type == II_GFUN ? "global" : "static")); 1605 1606 if (die_attr(dw, die, DW_AT_type, 0) != NULL) 1607 ii->ii_dtype = die_lookup_pass1(dw, die, DW_AT_type); 1608 else 1609 ii->ii_dtype = tdesc_intr_void(dw); 1610 1611 for (arg = die_child(dw, die); arg != NULL; 1612 arg = die_sibling(dw, arg)) { 1613 char *name1; 1614 1615 debug(3, "die %llu: looking at sub member at %llu\n", 1616 off, die_off(dw, die)); 1617 1618 if (die_tag(dw, arg) != DW_TAG_formal_parameter) 1619 continue; 1620 1621 if ((name1 = die_name(dw, arg)) == NULL) { 1622 terminate("die %llu: func arg %d has no name\n", 1623 off, ii->ii_nargs + 1); 1624 } 1625 1626 if (strcmp(name1, "...") == 0) { 1627 free(name1); 1628 ii->ii_vargs = 1; 1629 continue; 1630 } 1631 1632 ii->ii_nargs++; 1633 } 1634 1635 if (ii->ii_nargs > 0) { 1636 int i; 1637 1638 debug(3, "die %llu: function has %d argument%s\n", off, 1639 ii->ii_nargs, (ii->ii_nargs == 1 ? "" : "s")); 1640 1641 ii->ii_args = xcalloc(sizeof (tdesc_t) * ii->ii_nargs); 1642 1643 for (arg = die_child(dw, die), i = 0; 1644 arg != NULL && i < ii->ii_nargs; 1645 arg = die_sibling(dw, arg)) { 1646 if (die_tag(dw, arg) != DW_TAG_formal_parameter) 1647 continue; 1648 1649 ii->ii_args[i++] = die_lookup_pass1(dw, arg, 1650 DW_AT_type); 1651 } 1652 } 1653 1654 iidesc_add(dw->dw_td->td_iihash, ii); 1655} 1656 1657/*ARGSUSED3*/ 1658static void 1659die_variable_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp __unused) 1660{ 1661 iidesc_t *ii; 1662 char *name; 1663 1664 debug(3, "die %llu: creating object definition\n", off); 1665 1666 if (die_isdecl(dw, die) || (name = die_name(dw, die)) == NULL) 1667 return; /* skip prototypes and nameless objects */ 1668 1669 ii = xcalloc(sizeof (iidesc_t)); 1670 ii->ii_type = die_isglobal(dw, die) ? II_GVAR : II_SVAR; 1671 ii->ii_name = name; 1672 ii->ii_dtype = die_lookup_pass1(dw, die, DW_AT_type); 1673 if (ii->ii_type == II_SVAR) 1674 ii->ii_owner = xstrdup(dw->dw_cuname); 1675 1676 iidesc_add(dw->dw_td->td_iihash, ii); 1677} 1678 1679/*ARGSUSED2*/ 1680static int 1681die_fwd_resolve(tdesc_t *fwd, tdesc_t **fwdp, void *private __unused) 1682{ 1683 if (fwd->t_flags & TDESC_F_RESOLVED) 1684 return (1); 1685 1686 if (fwd->t_tdesc != NULL) { 1687 debug(3, "tdp %u: unforwarded %s\n", fwd->t_id, 1688 tdesc_name(fwd)); 1689 *fwdp = fwd->t_tdesc; 1690 } 1691 1692 fwd->t_flags |= TDESC_F_RESOLVED; 1693 1694 return (1); 1695} 1696 1697/*ARGSUSED*/ 1698static void 1699die_lexblk_descend(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off __unused, tdesc_t *tdp __unused) 1700{ 1701 Dwarf_Die child = die_child(dw, die); 1702 1703 if (child != NULL) 1704 die_create(dw, child); 1705} 1706 1707/* 1708 * Used to map the die to a routine which can parse it, using the tag to do the 1709 * mapping. While the processing of most tags entails the creation of a tdesc, 1710 * there are a few which don't - primarily those which result in the creation of 1711 * iidescs which refer to existing tdescs. 1712 */ 1713 1714#define DW_F_NOTDP 0x1 /* Don't create a tdesc for the creator */ 1715 1716typedef struct die_creator { 1717 Dwarf_Half dc_tag; 1718 uint16_t dc_flags; 1719 void (*dc_create)(dwarf_t *, Dwarf_Die, Dwarf_Off, tdesc_t *); 1720} die_creator_t; 1721 1722static const die_creator_t die_creators[] = { 1723 { DW_TAG_array_type, 0, die_array_create }, 1724 { DW_TAG_enumeration_type, 0, die_enum_create }, 1725 { DW_TAG_lexical_block, DW_F_NOTDP, die_lexblk_descend }, 1726 { DW_TAG_pointer_type, 0, die_pointer_create }, 1727 { DW_TAG_structure_type, 0, die_struct_create }, 1728 { DW_TAG_subroutine_type, 0, die_funcptr_create }, 1729 { DW_TAG_typedef, 0, die_typedef_create }, 1730 { DW_TAG_union_type, 0, die_union_create }, 1731 { DW_TAG_base_type, 0, die_base_create }, 1732 { DW_TAG_const_type, 0, die_const_create }, 1733 { DW_TAG_subprogram, DW_F_NOTDP, die_function_create }, 1734 { DW_TAG_variable, DW_F_NOTDP, die_variable_create }, 1735 { DW_TAG_volatile_type, 0, die_volatile_create }, 1736 { DW_TAG_restrict_type, 0, die_restrict_create }, 1737 { 0, 0, NULL } 1738}; 1739 1740static const die_creator_t * 1741die_tag2ctor(Dwarf_Half tag) 1742{ 1743 const die_creator_t *dc; 1744 1745 for (dc = die_creators; dc->dc_create != NULL; dc++) { 1746 if (dc->dc_tag == tag) 1747 return (dc); 1748 } 1749 1750 return (NULL); 1751} 1752 1753static void 1754die_create_one(dwarf_t *dw, Dwarf_Die die) 1755{ 1756 Dwarf_Off off = die_off(dw, die); 1757 const die_creator_t *dc; 1758 Dwarf_Half tag; 1759 tdesc_t *tdp; 1760 1761 debug(3, "die %llu <%llx>: create_one\n", off, off); 1762 1763 if (off > dw->dw_maxoff) { 1764 terminate("illegal die offset %llu (max %llu)\n", off, 1765 dw->dw_maxoff); 1766 } 1767 1768 tag = die_tag(dw, die); 1769 1770 if ((dc = die_tag2ctor(tag)) == NULL) { 1771 debug(2, "die %llu: ignoring tag type %x\n", off, tag); 1772 return; 1773 } 1774 1775 if ((tdp = tdesc_lookup(dw, off)) == NULL && 1776 !(dc->dc_flags & DW_F_NOTDP)) { 1777 tdp = xcalloc(sizeof (tdesc_t)); 1778 tdp->t_id = off; 1779 tdesc_add(dw, tdp); 1780 } 1781 1782 if (tdp != NULL) 1783 tdp->t_name = die_name(dw, die); 1784 1785 dc->dc_create(dw, die, off, tdp); 1786} 1787 1788static void 1789die_create(dwarf_t *dw, Dwarf_Die die) 1790{ 1791 do { 1792 die_create_one(dw, die); 1793 } while ((die = die_sibling(dw, die)) != NULL); 1794} 1795 1796static tdtrav_cb_f die_resolvers[] = { 1797 NULL, 1798 NULL, /* intrinsic */ 1799 NULL, /* pointer */ 1800 die_array_resolve, /* array */ 1801 NULL, /* function */ 1802 die_sou_resolve, /* struct */ 1803 die_sou_resolve, /* union */ 1804 die_enum_resolve, /* enum */ 1805 die_fwd_resolve, /* forward */ 1806 NULL, /* typedef */ 1807 NULL, /* typedef unres */ 1808 NULL, /* volatile */ 1809 NULL, /* const */ 1810 NULL, /* restrict */ 1811}; 1812 1813static tdtrav_cb_f die_fail_reporters[] = { 1814 NULL, 1815 NULL, /* intrinsic */ 1816 NULL, /* pointer */ 1817 die_array_failed, /* array */ 1818 NULL, /* function */ 1819 die_sou_failed, /* struct */ 1820 die_sou_failed, /* union */ 1821 NULL, /* enum */ 1822 NULL, /* forward */ 1823 NULL, /* typedef */ 1824 NULL, /* typedef unres */ 1825 NULL, /* volatile */ 1826 NULL, /* const */ 1827 NULL, /* restrict */ 1828}; 1829 1830static void 1831die_resolve(dwarf_t *dw) 1832{ 1833 int last = -1; 1834 int pass = 0; 1835 1836 do { 1837 pass++; 1838 dw->dw_nunres = 0; 1839 1840 (void) iitraverse_hash(dw->dw_td->td_iihash, 1841 &dw->dw_td->td_curvgen, NULL, NULL, die_resolvers, dw); 1842 1843 debug(3, "resolve: pass %d, %u left\n", pass, dw->dw_nunres); 1844 1845 if ((int) dw->dw_nunres == last) { 1846 fprintf(stderr, "%s: failed to resolve the following " 1847 "types:\n", progname); 1848 1849 (void) iitraverse_hash(dw->dw_td->td_iihash, 1850 &dw->dw_td->td_curvgen, NULL, NULL, 1851 die_fail_reporters, dw); 1852 1853 terminate("failed to resolve types\n"); 1854 } 1855 1856 last = dw->dw_nunres; 1857 1858 } while (dw->dw_nunres != 0); 1859} 1860 1861/* 1862 * Any object containing a function or object symbol at any scope should also 1863 * contain DWARF data. 1864 */ 1865static boolean_t 1866should_have_dwarf(Elf *elf) 1867{ 1868 Elf_Scn *scn = NULL; 1869 Elf_Data *data = NULL; 1870 GElf_Shdr shdr; 1871 GElf_Sym sym; 1872 uint32_t symdx = 0; 1873 size_t nsyms = 0; 1874 boolean_t found = B_FALSE; 1875 1876 while ((scn = elf_nextscn(elf, scn)) != NULL) { 1877 gelf_getshdr(scn, &shdr); 1878 1879 if (shdr.sh_type == SHT_SYMTAB) { 1880 found = B_TRUE; 1881 break; 1882 } 1883 } 1884 1885 if (!found) 1886 terminate("cannot convert stripped objects\n"); 1887 1888 data = elf_getdata(scn, NULL); 1889 nsyms = shdr.sh_size / shdr.sh_entsize; 1890 1891 for (symdx = 0; symdx < nsyms; symdx++) { 1892 gelf_getsym(data, symdx, &sym); 1893 1894 if ((GELF_ST_TYPE(sym.st_info) == STT_FUNC) || 1895 (GELF_ST_TYPE(sym.st_info) == STT_TLS) || 1896 (GELF_ST_TYPE(sym.st_info) == STT_OBJECT)) { 1897 char *name; 1898 1899 name = elf_strptr(elf, shdr.sh_link, sym.st_name); 1900 1901 /* Studio emits these local symbols regardless */ 1902 if ((strcmp(name, "Bbss.bss") != 0) && 1903 (strcmp(name, "Ttbss.bss") != 0) && 1904 (strcmp(name, "Ddata.data") != 0) && 1905 (strcmp(name, "Ttdata.data") != 0) && 1906 (strcmp(name, "Drodata.rodata") != 0)) 1907 return (B_TRUE); 1908 } 1909 } 1910 1911 return (B_FALSE); 1912} 1913 1914/*ARGSUSED*/ 1915int 1916dw_read(tdata_t *td, Elf *elf, char *filename __unused) 1917{ 1918 Dwarf_Unsigned abboff, hdrlen, nxthdr; 1919 Dwarf_Half vers, addrsz; 1920 Dwarf_Die cu = 0; 1921 Dwarf_Die child = 0; 1922 dwarf_t dw; 1923 char *prod = NULL; 1924 int rc; 1925 1926 bzero(&dw, sizeof (dwarf_t)); 1927 dw.dw_td = td; 1928 dw.dw_ptrsz = elf_ptrsz(elf); 1929 dw.dw_mfgtid_last = TID_MFGTID_BASE; 1930 dw.dw_tidhash = hash_new(TDESC_HASH_BUCKETS, tdesc_idhash, tdesc_idcmp); 1931 dw.dw_fwdhash = hash_new(TDESC_HASH_BUCKETS, tdesc_namehash, 1932 tdesc_namecmp); 1933 dw.dw_enumhash = hash_new(TDESC_HASH_BUCKETS, tdesc_namehash, 1934 tdesc_namecmp); 1935 1936 if ((rc = dwarf_elf_init(elf, DW_DLC_READ, &dw.dw_dw, 1937 &dw.dw_err)) == DW_DLV_NO_ENTRY) { 1938 if (should_have_dwarf(elf)) { 1939 errno = ENOENT; 1940 return (-1); 1941 } else { 1942 return (0); 1943 } 1944 } else if (rc != DW_DLV_OK) { 1945 if (dwarf_errno(&dw.dw_err) == DW_DLE_DEBUG_INFO_NULL) { 1946 /* 1947 * There's no type data in the DWARF section, but 1948 * libdwarf is too clever to handle that properly. 1949 */ 1950 return (0); 1951 } 1952 1953 terminate("failed to initialize DWARF: %s\n", 1954 dwarf_errmsg(&dw.dw_err)); 1955 } 1956 1957 if ((rc = dwarf_next_cu_header(dw.dw_dw, &hdrlen, &vers, &abboff, 1958 &addrsz, &nxthdr, &dw.dw_err)) != DW_DLV_OK) 1959 terminate("rc = %d %s\n", rc, dwarf_errmsg(&dw.dw_err)); 1960 1961 if ((cu = die_sibling(&dw, NULL)) == NULL || 1962 (((child = die_child(&dw, cu)) == NULL) && 1963 should_have_dwarf(elf))) { 1964 terminate("file does not contain dwarf type data " 1965 "(try compiling with -g)\n"); 1966 } else if (child == NULL) { 1967 return (0); 1968 } 1969 1970 dw.dw_maxoff = nxthdr - 1; 1971 1972 if (dw.dw_maxoff > TID_FILEMAX) 1973 terminate("file contains too many types\n"); 1974 1975 debug(1, "DWARF version: %d\n", vers); 1976 if (vers != DWARF_VERSION) { 1977 terminate("file contains incompatible version %d DWARF code " 1978 "(version 2 required)\n", vers); 1979 } 1980 1981 if (die_string(&dw, cu, DW_AT_producer, &prod, 0)) { 1982 debug(1, "DWARF emitter: %s\n", prod); 1983 free(prod); 1984 } 1985 1986 if ((dw.dw_cuname = die_name(&dw, cu)) != NULL) { 1987 char *base = xstrdup(basename(dw.dw_cuname)); 1988 free(dw.dw_cuname); 1989 dw.dw_cuname = base; 1990 1991 debug(1, "CU name: %s\n", dw.dw_cuname); 1992 } 1993 1994 if ((child = die_child(&dw, cu)) != NULL) 1995 die_create(&dw, child); 1996 1997 if ((rc = dwarf_next_cu_header(dw.dw_dw, &hdrlen, &vers, &abboff, 1998 &addrsz, &nxthdr, &dw.dw_err)) != DW_DLV_NO_ENTRY) 1999 terminate("multiple compilation units not supported\n"); 2000 2001 (void) dwarf_finish(&dw.dw_dw, &dw.dw_err); 2002 2003 die_resolve(&dw); 2004 2005 cvt_fixups(td, dw.dw_ptrsz); 2006 2007 /* leak the dwarf_t */ 2008 2009 return (0); 2010} 2011