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