Print this page
10823 should ignore DW_TAG_subprogram with DW_AT_declaration tags
10824 GCC7-derived CTF can double qualifiers on arrays
10825 ctfdump -c drops last type
10826 ctfdump -c goes off the rails with a missing parent
Reviewed by: Robert Mustacchi <rm@joyent.com>
Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>
Reviewed by: Jason King <jason.king@joyent.com>
Approved by: Jerry Jelinek <jerry.jelinek@joyent.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/lib/libctf/common/ctf_dwarf.c
+++ new/usr/src/lib/libctf/common/ctf_dwarf.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
|
↓ open down ↓ |
20 lines elided |
↑ open up ↑ |
21 21 /*
22 22 * Copyright 2007 Sun Microsystems, Inc. All rights reserved.
23 23 * Use is subject to license terms.
24 24 */
25 25 /*
26 26 * Copyright 2012 Jason King. All rights reserved.
27 27 * Use is subject to license terms.
28 28 */
29 29
30 30 /*
31 - * Copyright 2019 Joyent, Inc.
31 + * Copyright 2019, Joyent, Inc.
32 32 */
33 33
34 34 /*
35 35 * CTF DWARF conversion theory.
36 36 *
37 37 * DWARF data contains a series of compilation units. Each compilation unit
38 38 * generally refers to an object file or what once was, in the case of linked
39 39 * binaries and shared objects. Each compilation unit has a series of what DWARF
40 40 * calls a DIE (Debugging Information Entry). The set of entries that we care
41 41 * about have type information stored in a series of attributes. Each DIE also
42 42 * has a tag that identifies the kind of attributes that it has.
43 43 *
44 44 * A given DIE may itself have children. For example, a DIE that represents a
45 45 * structure has children which represent members. Whenever we encounter a DIE
46 46 * that has children or other values or types associated with it, we recursively
47 47 * process those children first so that way we can then refer to the generated
48 48 * CTF type id while processing its parent. This reduces the amount of unknowns
49 49 * and fixups that we need. It also ensures that we don't accidentally add types
50 50 * that an overzealous compiler might add to the DWARF data but aren't used by
51 51 * anything in the system.
52 52 *
53 53 * Once we do a conversion, we store a mapping in an AVL tree that goes from the
54 54 * DWARF's die offset, which is relative to the given compilation unit, to a
55 55 * ctf_id_t.
56 56 *
57 57 * Unfortunately, some compilers actually will emit duplicate entries for a
58 58 * given type that look similar, but aren't quite. To that end, we go through
59 59 * and do a variant on a merge once we're done processing a single compilation
60 60 * unit which deduplicates all of the types that are in the unit.
61 61 *
62 62 * Finally, if we encounter an object that has multiple compilation units, then
63 63 * we'll convert all of the compilation units separately and then do a merge, so
64 64 * that way we can result in one single ctf_file_t that represents everything
65 65 * for the object.
66 66 *
67 67 * Conversion Steps
68 68 * ----------------
69 69 *
70 70 * Because a given object we've been given to convert may have multiple
71 71 * compilation units, we break the work into two halves. The first half
72 72 * processes each compilation unit (potentially in parallel) and then the second
73 73 * half optionally merges all of the dies in the first half. First, we'll cover
74 74 * what's involved in converting a single ctf_cu_t's dwarf to CTF. This covers
75 75 * the work done in ctf_dwarf_convert_one().
76 76 *
77 77 * An individual ctf_cu_t, which represents a compilation unit, is converted to
78 78 * CTF in a series of multiple passes.
79 79 *
80 80 * Pass 1: During the first pass we walk all of the top-level dies and if we
81 81 * find a function, variable, struct, union, enum or typedef, we recursively
82 82 * transform all of its types. We don't recurse or process everything, because
83 83 * we don't want to add some of the types that compilers may add which are
84 84 * effectively unused.
85 85 *
86 86 * During pass 1, if we encounter any structures or unions we mark them for
87 87 * fixing up later. This is necessary because we may not be able to determine
88 88 * the full size of a structure at the beginning of time. This will happen if
89 89 * the DWARF attribute DW_AT_byte_size is not present for a member. Because of
90 90 * this possibility we defer adding members to structures or even converting
91 91 * them during pass 1 and save that for pass 2. Adding all of the base
92 92 * structures without any of their members helps deal with any circular
93 93 * dependencies that we might encounter.
94 94 *
95 95 * Pass 2: This pass is used to do the first half of fixing up structures and
96 96 * unions. Rather than walk the entire type space again, we actually walk the
97 97 * list of structures and unions that we marked for later fixing up. Here, we
98 98 * iterate over every structure and add members to the underlying ctf_file_t,
99 99 * but not to the structs themselves. One might wonder why we don't, and the
100 100 * main reason is that libctf requires a ctf_update() be done before adding the
101 101 * members to structures or unions.
102 102 *
103 103 * Pass 3: This pass is used to do the second half of fixing up structures and
104 104 * unions. During this part we always go through and add members to structures
105 105 * and unions that we added to the container in the previous pass. In addition,
106 106 * we set the structure and union's actual size, which may have additional
107 107 * padding added by the compiler, it isn't simply the last offset. DWARF always
108 108 * guarantees an attribute exists for this. Importantly no ctf_id_t's change
109 109 * during pass 2.
110 110 *
111 111 * Pass 4: The next phase is to add CTF entries for all of the symbols and
112 112 * variables that are present in this die. During pass 1 we added entries to a
113 113 * map for each variable and function. During this pass, we iterate over the
114 114 * symbol table and when we encounter a symbol that we have in our lists of
115 115 * translated information which matches, we then add it to the ctf_file_t.
116 116 *
117 117 * Pass 5: Here we go and look for any weak symbols and functions and see if
118 118 * they match anything that we recognize. If so, then we add type information
119 119 * for them at this point based on the matching type.
120 120 *
121 121 * Pass 6: This pass is actually a variant on a merge. The traditional merge
122 122 * process expects there to be no duplicate types. As such, at the end of
123 123 * conversion, we do a dedup on all of the types in the system. The
124 124 * deduplication process is described in lib/libctf/common/ctf_merge.c.
125 125 *
126 126 * Once pass 6 is done, we've finished processing the individual compilation
127 127 * unit.
128 128 *
129 129 * The following steps reflect the general process of doing a conversion.
130 130 *
131 131 * 1) Walk the dwarf section and determine the number of compilation units
132 132 * 2) Create a ctf_cu_t for each compilation unit
133 133 * 3) Add all ctf_cu_t's to a workq
134 134 * 4) Have the workq process each die with ctf_dwarf_convert_one. This itself
135 135 * is comprised of several steps, which were already enumerated.
136 136 * 5) If we have multiple cu's, we do a ctf merge of all the dies. The mechanics
137 137 * of the merge are discussed in lib/libctf/common/ctf_merge.c.
138 138 * 6) Free everything up and return a ctf_file_t to the user. If we only had a
139 139 * single compilation unit, then we give that to the user. Otherwise, we
140 140 * return the merged ctf_file_t.
141 141 *
142 142 * Threading
143 143 * ---------
144 144 *
145 145 * The process has been designed to be amenable to threading. Each compilation
146 146 * unit has its own type stream, therefore the logical place to divide and
147 147 * conquer is at the compilation unit. Each ctf_cu_t has been built to be able
148 148 * to be processed independently of the others. It has its own libdwarf handle,
149 149 * as a given libdwarf handle may only be used by a single thread at a time.
150 150 * This allows the various ctf_cu_t's to be processed in parallel by different
151 151 * threads.
152 152 *
153 153 * All of the ctf_cu_t's are loaded into a workq which allows for a number of
154 154 * threads to be specified and used as a thread pool to process all of the
155 155 * queued work. We set the number of threads to use in the workq equal to the
156 156 * number of threads that the user has specified.
157 157 *
158 158 * After all of the compilation units have been drained, we use the same number
159 159 * of threads when performing a merge of multiple compilation units, if they
160 160 * exist.
161 161 *
162 162 * While all of these different parts do support and allow for multiple threads,
163 163 * it's important that when only a single thread is specified, that it be the
164 164 * calling thread. This allows the conversion routines to be used in a context
165 165 * that doesn't allow additional threads, such as rtld.
166 166 *
167 167 * Common DWARF Mechanics and Notes
168 168 * --------------------------------
169 169 *
170 170 * At this time, we really only support DWARFv2, though support for DWARFv4 is
171 171 * mostly there. There is no intent to support DWARFv3.
172 172 *
173 173 * Generally types for something are stored in the DW_AT_type attribute. For
174 174 * example, a function's return type will be stored in the local DW_AT_type
175 175 * attribute while the arguments will be in child DIEs. There are also various
176 176 * times when we don't have any DW_AT_type. In that case, the lack of a type
177 177 * implies, at least for C, that its C type is void. Because DWARF doesn't emit
178 178 * one, we have a synthetic void type that we create and manipulate instead and
179 179 * pass it off to consumers on an as-needed basis. If nothing has a void type,
180 180 * it will not be emitted.
181 181 *
182 182 * Architecture Specific Parts
183 183 * ---------------------------
184 184 *
185 185 * The CTF tooling encodes various information about the various architectures
186 186 * in the system. Importantly, the tool assumes that every architecture has a
187 187 * data model where long and pointer are the same size. This is currently the
188 188 * case, as the two data models illumos supports are ILP32 and LP64.
189 189 *
190 190 * In addition, we encode the mapping of various floating point sizes to various
191 191 * types for each architecture. If a new architecture is being added, it should
192 192 * be added to the list. The general design of the ctf conversion tools is to be
193 193 * architecture independent. eg. any of the tools here should be able to convert
194 194 * any architecture's DWARF into ctf; however, this has not been rigorously
195 195 * tested and more importantly, the ctf routines don't currently write out the
196 196 * data in an endian-aware form, they only use that of the currently running
197 197 * library.
198 198 */
199 199
200 200 #include <libctf_impl.h>
201 201 #include <sys/avl.h>
202 202 #include <sys/debug.h>
203 203 #include <gelf.h>
204 204 #include <libdwarf.h>
205 205 #include <dwarf.h>
206 206 #include <libgen.h>
207 207 #include <workq.h>
208 208 #include <errno.h>
209 209
210 210 #define DWARF_VERSION_TWO 2
211 211 #define DWARF_VARARGS_NAME "..."
212 212
213 213 /*
214 214 * Dwarf may refer recursively to other types that we've already processed. To
215 215 * see if we've already converted them, we look them up in an AVL tree that's
216 216 * sorted by the DWARF id.
217 217 */
218 218 typedef struct ctf_dwmap {
219 219 avl_node_t cdm_avl;
220 220 Dwarf_Off cdm_off;
221 221 Dwarf_Die cdm_die;
222 222 ctf_id_t cdm_id;
223 223 boolean_t cdm_fix;
224 224 } ctf_dwmap_t;
225 225
226 226 typedef struct ctf_dwvar {
227 227 ctf_list_t cdv_list;
228 228 char *cdv_name;
229 229 ctf_id_t cdv_type;
230 230 boolean_t cdv_global;
231 231 } ctf_dwvar_t;
232 232
233 233 typedef struct ctf_dwfunc {
234 234 ctf_list_t cdf_list;
235 235 char *cdf_name;
236 236 ctf_funcinfo_t cdf_fip;
237 237 ctf_id_t *cdf_argv;
238 238 boolean_t cdf_global;
239 239 } ctf_dwfunc_t;
240 240
241 241 typedef struct ctf_dwbitf {
242 242 ctf_list_t cdb_list;
243 243 ctf_id_t cdb_base;
244 244 uint_t cdb_nbits;
245 245 ctf_id_t cdb_id;
246 246 } ctf_dwbitf_t;
247 247
248 248 /*
249 249 * The ctf_cu_t represents a single top-level DWARF die unit. While generally,
250 250 * the typical object file has only a single die, if we're asked to convert
251 251 * something that's been linked from multiple sources, multiple dies will exist.
252 252 */
253 253 typedef struct ctf_die {
254 254 Elf *cu_elf; /* shared libelf handle */
255 255 char *cu_name; /* basename of the DIE */
256 256 ctf_merge_t *cu_cmh; /* merge handle */
257 257 ctf_list_t cu_vars; /* List of variables */
258 258 ctf_list_t cu_funcs; /* List of functions */
259 259 ctf_list_t cu_bitfields; /* Bit field members */
260 260 Dwarf_Debug cu_dwarf; /* libdwarf handle */
261 261 Dwarf_Die cu_cu; /* libdwarf compilation unit */
262 262 Dwarf_Off cu_cuoff; /* cu's offset */
263 263 Dwarf_Off cu_maxoff; /* maximum offset */
264 264 ctf_file_t *cu_ctfp; /* output CTF file */
265 265 avl_tree_t cu_map; /* map die offsets to CTF types */
266 266 char *cu_errbuf; /* error message buffer */
267 267 size_t cu_errlen; /* error message buffer length */
268 268 size_t cu_ptrsz; /* object's pointer size */
269 269 boolean_t cu_bigend; /* is it big endian */
270 270 boolean_t cu_doweaks; /* should we convert weak symbols? */
271 271 uint_t cu_mach; /* machine type */
272 272 ctf_id_t cu_voidtid; /* void pointer */
273 273 ctf_id_t cu_longtid; /* id for a 'long' */
274 274 } ctf_cu_t;
275 275
276 276 static int ctf_dwarf_offset(ctf_cu_t *, Dwarf_Die, Dwarf_Off *);
277 277 static int ctf_dwarf_convert_die(ctf_cu_t *, Dwarf_Die);
278 278 static int ctf_dwarf_convert_type(ctf_cu_t *, Dwarf_Die, ctf_id_t *, int);
279 279
280 280 static int ctf_dwarf_function_count(ctf_cu_t *, Dwarf_Die, ctf_funcinfo_t *,
281 281 boolean_t);
282 282 static int ctf_dwarf_convert_fargs(ctf_cu_t *, Dwarf_Die, ctf_funcinfo_t *,
283 283 ctf_id_t *);
284 284
285 285 /*
286 286 * This is a generic way to set a CTF Conversion backend error depending on what
287 287 * we were doing. Unless it was one of a specific set of errors that don't
288 288 * indicate a programming / translation bug, eg. ENOMEM, then we transform it
289 289 * into a CTF backend error and fill in the error buffer.
290 290 */
291 291 static int
292 292 ctf_dwarf_error(ctf_cu_t *cup, ctf_file_t *cfp, int err, const char *fmt, ...)
293 293 {
294 294 va_list ap;
295 295 int ret;
296 296 size_t off = 0;
297 297 ssize_t rem = cup->cu_errlen;
298 298 if (cfp != NULL)
299 299 err = ctf_errno(cfp);
300 300
301 301 if (err == ENOMEM)
302 302 return (err);
303 303
304 304 ret = snprintf(cup->cu_errbuf, rem, "die %s: ", cup->cu_name);
305 305 if (ret < 0)
306 306 goto err;
307 307 off += ret;
308 308 rem = MAX(rem - ret, 0);
309 309
310 310 va_start(ap, fmt);
311 311 ret = vsnprintf(cup->cu_errbuf + off, rem, fmt, ap);
312 312 va_end(ap);
313 313 if (ret < 0)
314 314 goto err;
315 315
316 316 off += ret;
317 317 rem = MAX(rem - ret, 0);
318 318 if (fmt[strlen(fmt) - 1] != '\n') {
319 319 (void) snprintf(cup->cu_errbuf + off, rem,
320 320 ": %s\n", ctf_errmsg(err));
321 321 }
322 322 va_end(ap);
323 323 return (ECTF_CONVBKERR);
324 324
325 325 err:
326 326 cup->cu_errbuf[0] = '\0';
327 327 return (ECTF_CONVBKERR);
328 328 }
329 329
330 330 /*
331 331 * DWARF often opts to put no explicit type to describe a void type. eg. if we
332 332 * have a reference type whose DW_AT_type member doesn't exist, then we should
333 333 * instead assume it points to void. Because this isn't represented, we
334 334 * instead cause it to come into existence.
335 335 */
336 336 static ctf_id_t
337 337 ctf_dwarf_void(ctf_cu_t *cup)
338 338 {
339 339 if (cup->cu_voidtid == CTF_ERR) {
340 340 ctf_encoding_t enc = { CTF_INT_SIGNED, 0, 0 };
341 341 cup->cu_voidtid = ctf_add_integer(cup->cu_ctfp, CTF_ADD_ROOT,
342 342 "void", &enc);
343 343 if (cup->cu_voidtid == CTF_ERR) {
344 344 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
345 345 "failed to create void type: %s\n",
346 346 ctf_errmsg(ctf_errno(cup->cu_ctfp)));
347 347 }
348 348 }
349 349
350 350 return (cup->cu_voidtid);
351 351 }
352 352
353 353 /*
354 354 * There are many different forms that an array index may take. However, we just
355 355 * always force it to be of a type long no matter what. Therefore we use this to
356 356 * have a single instance of long across everything.
357 357 */
358 358 static ctf_id_t
359 359 ctf_dwarf_long(ctf_cu_t *cup)
360 360 {
361 361 if (cup->cu_longtid == CTF_ERR) {
362 362 ctf_encoding_t enc;
363 363
364 364 enc.cte_format = CTF_INT_SIGNED;
365 365 enc.cte_offset = 0;
366 366 /* All illumos systems are LP */
367 367 enc.cte_bits = cup->cu_ptrsz * 8;
368 368 cup->cu_longtid = ctf_add_integer(cup->cu_ctfp, CTF_ADD_NONROOT,
369 369 "long", &enc);
370 370 if (cup->cu_longtid == CTF_ERR) {
371 371 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
372 372 "failed to create long type: %s\n",
373 373 ctf_errmsg(ctf_errno(cup->cu_ctfp)));
374 374 }
375 375
376 376 }
377 377
378 378 return (cup->cu_longtid);
379 379 }
380 380
381 381 static int
382 382 ctf_dwmap_comp(const void *a, const void *b)
383 383 {
384 384 const ctf_dwmap_t *ca = a;
385 385 const ctf_dwmap_t *cb = b;
386 386
387 387 if (ca->cdm_off > cb->cdm_off)
388 388 return (1);
389 389 if (ca->cdm_off < cb->cdm_off)
390 390 return (-1);
391 391 return (0);
392 392 }
393 393
394 394 static int
395 395 ctf_dwmap_add(ctf_cu_t *cup, ctf_id_t id, Dwarf_Die die, boolean_t fix)
396 396 {
397 397 int ret;
398 398 avl_index_t index;
399 399 ctf_dwmap_t *dwmap;
400 400 Dwarf_Off off;
401 401
402 402 VERIFY(id > 0 && id < CTF_MAX_TYPE);
403 403
404 404 if ((ret = ctf_dwarf_offset(cup, die, &off)) != 0)
405 405 return (ret);
406 406
407 407 if ((dwmap = ctf_alloc(sizeof (ctf_dwmap_t))) == NULL)
408 408 return (ENOMEM);
409 409
410 410 dwmap->cdm_die = die;
411 411 dwmap->cdm_off = off;
412 412 dwmap->cdm_id = id;
413 413 dwmap->cdm_fix = fix;
414 414
415 415 ctf_dprintf("dwmap: %p %" DW_PR_DUx "->%d\n", dwmap, off, id);
416 416 VERIFY(avl_find(&cup->cu_map, dwmap, &index) == NULL);
417 417 avl_insert(&cup->cu_map, dwmap, index);
418 418 return (0);
419 419 }
420 420
421 421 static int
422 422 ctf_dwarf_attribute(ctf_cu_t *cup, Dwarf_Die die, Dwarf_Half name,
423 423 Dwarf_Attribute *attrp)
424 424 {
425 425 int ret;
426 426 Dwarf_Error derr;
427 427
428 428 if ((ret = dwarf_attr(die, name, attrp, &derr)) == DW_DLV_OK)
429 429 return (0);
430 430 if (ret == DW_DLV_NO_ENTRY) {
431 431 *attrp = NULL;
432 432 return (ENOENT);
433 433 }
434 434 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
435 435 "failed to get attribute for type: %s\n",
436 436 dwarf_errmsg(derr));
437 437 return (ECTF_CONVBKERR);
438 438 }
439 439
440 440 static int
441 441 ctf_dwarf_ref(ctf_cu_t *cup, Dwarf_Die die, Dwarf_Half name, Dwarf_Off *refp)
442 442 {
443 443 int ret;
444 444 Dwarf_Attribute attr;
445 445 Dwarf_Error derr;
446 446
447 447 if ((ret = ctf_dwarf_attribute(cup, die, name, &attr)) != 0)
448 448 return (ret);
449 449
450 450 if (dwarf_formref(attr, refp, &derr) == DW_DLV_OK) {
451 451 dwarf_dealloc(cup->cu_dwarf, attr, DW_DLA_ATTR);
452 452 return (0);
453 453 }
454 454
455 455 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
456 456 "failed to get unsigned attribute for type: %s\n",
457 457 dwarf_errmsg(derr));
458 458 return (ECTF_CONVBKERR);
459 459 }
460 460
461 461 static int
462 462 ctf_dwarf_refdie(ctf_cu_t *cup, Dwarf_Die die, Dwarf_Half name,
463 463 Dwarf_Die *diep)
464 464 {
465 465 int ret;
466 466 Dwarf_Off off;
467 467 Dwarf_Error derr;
468 468
469 469 if ((ret = ctf_dwarf_ref(cup, die, name, &off)) != 0)
470 470 return (ret);
471 471
472 472 off += cup->cu_cuoff;
473 473 if ((ret = dwarf_offdie(cup->cu_dwarf, off, diep, &derr)) !=
474 474 DW_DLV_OK) {
475 475 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
476 476 "failed to get die from offset %" DW_PR_DUu ": %s\n",
477 477 off, dwarf_errmsg(derr));
478 478 return (ECTF_CONVBKERR);
479 479 }
480 480
481 481 return (0);
482 482 }
483 483
484 484 static int
485 485 ctf_dwarf_signed(ctf_cu_t *cup, Dwarf_Die die, Dwarf_Half name,
486 486 Dwarf_Signed *valp)
487 487 {
488 488 int ret;
489 489 Dwarf_Attribute attr;
490 490 Dwarf_Error derr;
491 491
492 492 if ((ret = ctf_dwarf_attribute(cup, die, name, &attr)) != 0)
493 493 return (ret);
494 494
495 495 if (dwarf_formsdata(attr, valp, &derr) == DW_DLV_OK) {
496 496 dwarf_dealloc(cup->cu_dwarf, attr, DW_DLA_ATTR);
497 497 return (0);
498 498 }
499 499
500 500 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
501 501 "failed to get unsigned attribute for type: %s\n",
502 502 dwarf_errmsg(derr));
503 503 return (ECTF_CONVBKERR);
504 504 }
505 505
506 506 static int
507 507 ctf_dwarf_unsigned(ctf_cu_t *cup, Dwarf_Die die, Dwarf_Half name,
508 508 Dwarf_Unsigned *valp)
509 509 {
510 510 int ret;
511 511 Dwarf_Attribute attr;
512 512 Dwarf_Error derr;
513 513
514 514 if ((ret = ctf_dwarf_attribute(cup, die, name, &attr)) != 0)
515 515 return (ret);
516 516
517 517 if (dwarf_formudata(attr, valp, &derr) == DW_DLV_OK) {
518 518 dwarf_dealloc(cup->cu_dwarf, attr, DW_DLA_ATTR);
519 519 return (0);
520 520 }
521 521
522 522 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
523 523 "failed to get unsigned attribute for type: %s\n",
524 524 dwarf_errmsg(derr));
525 525 return (ECTF_CONVBKERR);
526 526 }
527 527
528 528 static int
529 529 ctf_dwarf_boolean(ctf_cu_t *cup, Dwarf_Die die, Dwarf_Half name,
530 530 Dwarf_Bool *val)
531 531 {
532 532 int ret;
533 533 Dwarf_Attribute attr;
534 534 Dwarf_Error derr;
535 535
536 536 if ((ret = ctf_dwarf_attribute(cup, die, name, &attr)) != 0)
537 537 return (ret);
538 538
539 539 if (dwarf_formflag(attr, val, &derr) == DW_DLV_OK) {
540 540 dwarf_dealloc(cup->cu_dwarf, attr, DW_DLA_ATTR);
541 541 return (0);
542 542 }
543 543
544 544 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
545 545 "failed to get boolean attribute for type: %s\n",
546 546 dwarf_errmsg(derr));
547 547
548 548 return (ECTF_CONVBKERR);
549 549 }
550 550
551 551 static int
552 552 ctf_dwarf_string(ctf_cu_t *cup, Dwarf_Die die, Dwarf_Half name, char **strp)
553 553 {
554 554 int ret;
555 555 char *s;
556 556 Dwarf_Attribute attr;
557 557 Dwarf_Error derr;
558 558
559 559 *strp = NULL;
560 560 if ((ret = ctf_dwarf_attribute(cup, die, name, &attr)) != 0)
561 561 return (ret);
562 562
563 563 if (dwarf_formstring(attr, &s, &derr) == DW_DLV_OK) {
564 564 if ((*strp = ctf_strdup(s)) == NULL)
565 565 ret = ENOMEM;
566 566 else
567 567 ret = 0;
568 568 dwarf_dealloc(cup->cu_dwarf, attr, DW_DLA_ATTR);
569 569 return (ret);
570 570 }
571 571
572 572 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
573 573 "failed to get string attribute for type: %s\n",
574 574 dwarf_errmsg(derr));
575 575 return (ECTF_CONVBKERR);
576 576 }
577 577
578 578 static int
579 579 ctf_dwarf_member_location(ctf_cu_t *cup, Dwarf_Die die, Dwarf_Unsigned *valp)
580 580 {
581 581 int ret;
582 582 Dwarf_Error derr;
583 583 Dwarf_Attribute attr;
584 584 Dwarf_Locdesc *loc;
585 585 Dwarf_Signed locnum;
586 586
587 587 if ((ret = ctf_dwarf_attribute(cup, die, DW_AT_data_member_location,
588 588 &attr)) != 0)
589 589 return (ret);
590 590
591 591 if (dwarf_loclist(attr, &loc, &locnum, &derr) != DW_DLV_OK) {
592 592 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
593 593 "failed to obtain location list for member offset: %s",
594 594 dwarf_errmsg(derr));
595 595 dwarf_dealloc(cup->cu_dwarf, attr, DW_DLA_ATTR);
596 596 return (ECTF_CONVBKERR);
597 597 }
598 598 dwarf_dealloc(cup->cu_dwarf, attr, DW_DLA_ATTR);
599 599
600 600 if (locnum != 1 || loc->ld_s->lr_atom != DW_OP_plus_uconst) {
601 601 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
602 602 "failed to parse location structure for member");
603 603 dwarf_dealloc(cup->cu_dwarf, loc->ld_s, DW_DLA_LOC_BLOCK);
604 604 dwarf_dealloc(cup->cu_dwarf, loc, DW_DLA_LOCDESC);
605 605 return (ECTF_CONVBKERR);
606 606 }
607 607
608 608 *valp = loc->ld_s->lr_number;
609 609
610 610 dwarf_dealloc(cup->cu_dwarf, loc->ld_s, DW_DLA_LOC_BLOCK);
611 611 dwarf_dealloc(cup->cu_dwarf, loc, DW_DLA_LOCDESC);
612 612 return (0);
613 613 }
614 614
615 615
616 616 static int
617 617 ctf_dwarf_offset(ctf_cu_t *cup, Dwarf_Die die, Dwarf_Off *offsetp)
618 618 {
619 619 Dwarf_Error derr;
620 620
621 621 if (dwarf_dieoffset(die, offsetp, &derr) == DW_DLV_OK)
622 622 return (0);
623 623
624 624 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
625 625 "failed to get die offset: %s\n",
626 626 dwarf_errmsg(derr));
627 627 return (ECTF_CONVBKERR);
628 628 }
629 629
630 630 /* simpler variant for debugging output */
631 631 static Dwarf_Off
632 632 ctf_die_offset(Dwarf_Die die)
633 633 {
634 634 Dwarf_Off off = -1;
635 635 Dwarf_Error derr;
636 636
637 637 (void) dwarf_dieoffset(die, &off, &derr);
638 638 return (off);
639 639 }
640 640
641 641 static int
642 642 ctf_dwarf_tag(ctf_cu_t *cup, Dwarf_Die die, Dwarf_Half *tagp)
643 643 {
644 644 Dwarf_Error derr;
645 645
646 646 if (dwarf_tag(die, tagp, &derr) == DW_DLV_OK)
647 647 return (0);
648 648
649 649 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
650 650 "failed to get tag type: %s\n",
651 651 dwarf_errmsg(derr));
652 652 return (ECTF_CONVBKERR);
653 653 }
654 654
655 655 static int
656 656 ctf_dwarf_sib(ctf_cu_t *cup, Dwarf_Die base, Dwarf_Die *sibp)
657 657 {
658 658 Dwarf_Error derr;
659 659 int ret;
660 660
661 661 *sibp = NULL;
662 662 ret = dwarf_siblingof(cup->cu_dwarf, base, sibp, &derr);
663 663 if (ret == DW_DLV_OK || ret == DW_DLV_NO_ENTRY)
664 664 return (0);
665 665
666 666 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
667 667 "failed to sibling from die: %s\n",
668 668 dwarf_errmsg(derr));
669 669 return (ECTF_CONVBKERR);
670 670 }
671 671
672 672 static int
673 673 ctf_dwarf_child(ctf_cu_t *cup, Dwarf_Die base, Dwarf_Die *childp)
674 674 {
675 675 Dwarf_Error derr;
676 676 int ret;
677 677
678 678 *childp = NULL;
679 679 ret = dwarf_child(base, childp, &derr);
680 680 if (ret == DW_DLV_OK || ret == DW_DLV_NO_ENTRY)
681 681 return (0);
682 682
683 683 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
684 684 "failed to child from die: %s\n",
685 685 dwarf_errmsg(derr));
686 686 return (ECTF_CONVBKERR);
687 687 }
688 688
689 689 /*
690 690 * Compilers disagree on what to do to determine if something has global
691 691 * visiblity. Traditionally gcc has used DW_AT_external to indicate this while
692 692 * Studio has used DW_AT_visibility. We check DW_AT_visibility first and then
693 693 * fall back to DW_AT_external. Lack of DW_AT_external implies that it is not.
694 694 */
695 695 static int
696 696 ctf_dwarf_isglobal(ctf_cu_t *cup, Dwarf_Die die, boolean_t *igp)
697 697 {
698 698 int ret;
699 699 Dwarf_Signed vis;
700 700 Dwarf_Bool ext;
701 701
702 702 if ((ret = ctf_dwarf_signed(cup, die, DW_AT_visibility, &vis)) == 0) {
703 703 *igp = vis == DW_VIS_exported;
704 704 return (0);
705 705 } else if (ret != ENOENT) {
706 706 return (ret);
707 707 }
708 708
709 709 if ((ret = ctf_dwarf_boolean(cup, die, DW_AT_external, &ext)) != 0) {
710 710 if (ret == ENOENT) {
711 711 *igp = B_FALSE;
712 712 return (0);
713 713 }
714 714 return (ret);
715 715 }
716 716 *igp = ext != 0 ? B_TRUE : B_FALSE;
717 717 return (0);
718 718 }
719 719
720 720 static int
721 721 ctf_dwarf_die_elfenc(Elf *elf, ctf_cu_t *cup, char *errbuf, size_t errlen)
722 722 {
723 723 GElf_Ehdr ehdr;
724 724
725 725 if (gelf_getehdr(elf, &ehdr) == NULL) {
726 726 (void) snprintf(errbuf, errlen,
727 727 "failed to get ELF header: %s\n",
728 728 elf_errmsg(elf_errno()));
729 729 return (ECTF_CONVBKERR);
730 730 }
731 731
732 732 cup->cu_mach = ehdr.e_machine;
733 733
734 734 if (ehdr.e_ident[EI_CLASS] == ELFCLASS32) {
735 735 cup->cu_ptrsz = 4;
736 736 VERIFY(ctf_setmodel(cup->cu_ctfp, CTF_MODEL_ILP32) == 0);
737 737 } else if (ehdr.e_ident[EI_CLASS] == ELFCLASS64) {
738 738 cup->cu_ptrsz = 8;
739 739 VERIFY(ctf_setmodel(cup->cu_ctfp, CTF_MODEL_LP64) == 0);
740 740 } else {
741 741 (void) snprintf(errbuf, errlen,
742 742 "unknown ELF class %d", ehdr.e_ident[EI_CLASS]);
743 743 return (ECTF_CONVBKERR);
744 744 }
745 745
746 746 if (ehdr.e_ident[EI_DATA] == ELFDATA2LSB) {
747 747 cup->cu_bigend = B_FALSE;
748 748 } else if (ehdr.e_ident[EI_DATA] == ELFDATA2MSB) {
749 749 cup->cu_bigend = B_TRUE;
750 750 } else {
751 751 (void) snprintf(errbuf, errlen,
752 752 "unknown ELF data encoding: %hhu", ehdr.e_ident[EI_DATA]);
753 753 return (ECTF_CONVBKERR);
754 754 }
755 755
756 756 return (0);
757 757 }
758 758
759 759 typedef struct ctf_dwarf_fpent {
760 760 size_t cdfe_size;
761 761 uint_t cdfe_enc[3];
762 762 } ctf_dwarf_fpent_t;
763 763
764 764 typedef struct ctf_dwarf_fpmap {
765 765 uint_t cdf_mach;
766 766 ctf_dwarf_fpent_t cdf_ents[4];
767 767 } ctf_dwarf_fpmap_t;
768 768
769 769 static const ctf_dwarf_fpmap_t ctf_dwarf_fpmaps[] = {
770 770 { EM_SPARC, {
771 771 { 4, { CTF_FP_SINGLE, CTF_FP_CPLX, CTF_FP_IMAGRY } },
772 772 { 8, { CTF_FP_DOUBLE, CTF_FP_DCPLX, CTF_FP_DIMAGRY } },
773 773 { 16, { CTF_FP_LDOUBLE, CTF_FP_LDCPLX, CTF_FP_LDIMAGRY } },
774 774 { 0, { 0 } }
775 775 } },
776 776 { EM_SPARC32PLUS, {
777 777 { 4, { CTF_FP_SINGLE, CTF_FP_CPLX, CTF_FP_IMAGRY } },
778 778 { 8, { CTF_FP_DOUBLE, CTF_FP_DCPLX, CTF_FP_DIMAGRY } },
779 779 { 16, { CTF_FP_LDOUBLE, CTF_FP_LDCPLX, CTF_FP_LDIMAGRY } },
780 780 { 0, { 0 } }
781 781 } },
782 782 { EM_SPARCV9, {
783 783 { 4, { CTF_FP_SINGLE, CTF_FP_CPLX, CTF_FP_IMAGRY } },
784 784 { 8, { CTF_FP_DOUBLE, CTF_FP_DCPLX, CTF_FP_DIMAGRY } },
785 785 { 16, { CTF_FP_LDOUBLE, CTF_FP_LDCPLX, CTF_FP_LDIMAGRY } },
786 786 { 0, { 0 } }
787 787 } },
788 788 { EM_386, {
789 789 { 4, { CTF_FP_SINGLE, CTF_FP_CPLX, CTF_FP_IMAGRY } },
790 790 { 8, { CTF_FP_DOUBLE, CTF_FP_DCPLX, CTF_FP_DIMAGRY } },
791 791 { 12, { CTF_FP_LDOUBLE, CTF_FP_LDCPLX, CTF_FP_LDIMAGRY } },
792 792 { 0, { 0 } }
793 793 } },
794 794 { EM_X86_64, {
795 795 { 4, { CTF_FP_SINGLE, CTF_FP_CPLX, CTF_FP_IMAGRY } },
796 796 { 8, { CTF_FP_DOUBLE, CTF_FP_DCPLX, CTF_FP_DIMAGRY } },
797 797 { 16, { CTF_FP_LDOUBLE, CTF_FP_LDCPLX, CTF_FP_LDIMAGRY } },
798 798 { 0, { 0 } }
799 799 } },
800 800 { EM_NONE }
801 801 };
802 802
803 803 static int
804 804 ctf_dwarf_float_base(ctf_cu_t *cup, Dwarf_Signed type, ctf_encoding_t *enc)
805 805 {
806 806 const ctf_dwarf_fpmap_t *map = &ctf_dwarf_fpmaps[0];
807 807 const ctf_dwarf_fpent_t *ent;
808 808 uint_t col = 0, mult = 1;
809 809
810 810 for (map = &ctf_dwarf_fpmaps[0]; map->cdf_mach != EM_NONE; map++) {
811 811 if (map->cdf_mach == cup->cu_mach)
812 812 break;
813 813 }
814 814
815 815 if (map->cdf_mach == EM_NONE) {
816 816 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
817 817 "Unsupported machine type: %d\n", cup->cu_mach);
818 818 return (ENOTSUP);
819 819 }
820 820
821 821 if (type == DW_ATE_complex_float) {
822 822 mult = 2;
823 823 col = 1;
824 824 } else if (type == DW_ATE_imaginary_float ||
825 825 type == DW_ATE_SUN_imaginary_float) {
826 826 col = 2;
827 827 }
828 828
829 829 ent = &map->cdf_ents[0];
830 830 for (ent = &map->cdf_ents[0]; ent->cdfe_size != 0; ent++) {
831 831 if (ent->cdfe_size * mult * 8 == enc->cte_bits) {
832 832 enc->cte_format = ent->cdfe_enc[col];
833 833 return (0);
834 834 }
835 835 }
836 836
837 837 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
838 838 "failed to find valid fp mapping for encoding %d, size %d bits\n",
839 839 type, enc->cte_bits);
840 840 return (EINVAL);
841 841 }
842 842
843 843 static int
844 844 ctf_dwarf_dwarf_base(ctf_cu_t *cup, Dwarf_Die die, int *kindp,
845 845 ctf_encoding_t *enc)
846 846 {
847 847 int ret;
848 848 Dwarf_Signed type;
849 849
850 850 if ((ret = ctf_dwarf_signed(cup, die, DW_AT_encoding, &type)) != 0)
851 851 return (ret);
852 852
853 853 switch (type) {
854 854 case DW_ATE_unsigned:
855 855 case DW_ATE_address:
856 856 *kindp = CTF_K_INTEGER;
857 857 enc->cte_format = 0;
858 858 break;
859 859 case DW_ATE_unsigned_char:
860 860 *kindp = CTF_K_INTEGER;
861 861 enc->cte_format = CTF_INT_CHAR;
862 862 break;
863 863 case DW_ATE_signed:
864 864 *kindp = CTF_K_INTEGER;
865 865 enc->cte_format = CTF_INT_SIGNED;
866 866 break;
867 867 case DW_ATE_signed_char:
868 868 *kindp = CTF_K_INTEGER;
869 869 enc->cte_format = CTF_INT_SIGNED | CTF_INT_CHAR;
870 870 break;
871 871 case DW_ATE_boolean:
872 872 *kindp = CTF_K_INTEGER;
873 873 enc->cte_format = CTF_INT_SIGNED | CTF_INT_BOOL;
874 874 break;
875 875 case DW_ATE_float:
876 876 case DW_ATE_complex_float:
877 877 case DW_ATE_imaginary_float:
878 878 case DW_ATE_SUN_imaginary_float:
879 879 case DW_ATE_SUN_interval_float:
880 880 *kindp = CTF_K_FLOAT;
881 881 if ((ret = ctf_dwarf_float_base(cup, type, enc)) != 0)
882 882 return (ret);
883 883 break;
884 884 default:
885 885 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
886 886 "encountered unknown DWARF encoding: %d", type);
887 887 return (ECTF_CONVBKERR);
888 888 }
889 889
890 890 return (0);
891 891 }
892 892
893 893 /*
894 894 * Different compilers (at least GCC and Studio) use different names for types.
895 895 * This parses the types and attempts to unify them. If this fails, we just fall
896 896 * back to using the DWARF itself.
897 897 */
898 898 static int
899 899 ctf_dwarf_parse_base(const char *name, int *kindp, ctf_encoding_t *enc,
900 900 char **newnamep)
901 901 {
902 902 char buf[256];
903 903 char *base, *c, *last;
904 904 int nlong = 0, nshort = 0, nchar = 0, nint = 0;
905 905 int sign = 1;
906 906
907 907 if (strlen(name) + 1 > sizeof (buf))
908 908 return (EINVAL);
909 909
910 910 (void) strlcpy(buf, name, sizeof (buf));
911 911 for (c = strtok_r(buf, " ", &last); c != NULL;
912 912 c = strtok_r(NULL, " ", &last)) {
913 913 if (strcmp(c, "signed") == 0) {
914 914 sign = 1;
915 915 } else if (strcmp(c, "unsigned") == 0) {
916 916 sign = 0;
917 917 } else if (strcmp(c, "long") == 0) {
918 918 nlong++;
919 919 } else if (strcmp(c, "char") == 0) {
920 920 nchar++;
921 921 } else if (strcmp(c, "short") == 0) {
922 922 nshort++;
923 923 } else if (strcmp(c, "int") == 0) {
924 924 nint++;
925 925 } else {
926 926 /*
927 927 * If we don't recognize any of the tokens, we'll tell
928 928 * the caller to fall back to the dwarf-provided
929 929 * encoding information.
930 930 */
931 931 return (EINVAL);
932 932 }
933 933 }
934 934
935 935 if (nchar > 1 || nshort > 1 || nint > 1 || nlong > 2)
936 936 return (EINVAL);
937 937
938 938 if (nchar > 0) {
939 939 if (nlong > 0 || nshort > 0 || nint > 0)
940 940 return (EINVAL);
941 941 base = "char";
942 942 } else if (nshort > 0) {
943 943 if (nlong > 0)
944 944 return (EINVAL);
945 945 base = "short";
946 946 } else if (nlong > 0) {
947 947 base = "long";
948 948 } else {
949 949 base = "int";
950 950 }
951 951
952 952 if (nchar > 0)
953 953 enc->cte_format = CTF_INT_CHAR;
954 954 else
955 955 enc->cte_format = 0;
956 956
957 957 if (sign > 0)
958 958 enc->cte_format |= CTF_INT_SIGNED;
959 959
960 960 (void) snprintf(buf, sizeof (buf), "%s%s%s",
961 961 (sign ? "" : "unsigned "),
962 962 (nlong > 1 ? "long " : ""),
963 963 base);
964 964
965 965 *newnamep = ctf_strdup(buf);
966 966 if (*newnamep == NULL)
967 967 return (ENOMEM);
968 968 *kindp = CTF_K_INTEGER;
969 969 return (0);
970 970 }
971 971
972 972 static int
973 973 ctf_dwarf_create_base(ctf_cu_t *cup, Dwarf_Die die, ctf_id_t *idp, int isroot,
974 974 Dwarf_Off off)
975 975 {
976 976 int ret;
977 977 char *name, *nname;
978 978 Dwarf_Unsigned sz;
979 979 int kind;
980 980 ctf_encoding_t enc;
981 981 ctf_id_t id;
982 982
983 983 if ((ret = ctf_dwarf_string(cup, die, DW_AT_name, &name)) != 0)
984 984 return (ret);
985 985 if ((ret = ctf_dwarf_unsigned(cup, die, DW_AT_byte_size, &sz)) != 0) {
986 986 goto out;
987 987 }
988 988 ctf_dprintf("Creating base type %s from off %llu, size: %d\n", name,
989 989 off, sz);
990 990
991 991 bzero(&enc, sizeof (ctf_encoding_t));
992 992 enc.cte_bits = sz * 8;
993 993 if ((ret = ctf_dwarf_parse_base(name, &kind, &enc, &nname)) == 0) {
994 994 ctf_free(name, strlen(name) + 1);
995 995 name = nname;
996 996 } else {
997 997 if (ret != EINVAL)
998 998 return (ret);
999 999 ctf_dprintf("falling back to dwarf for base type %s\n", name);
1000 1000 if ((ret = ctf_dwarf_dwarf_base(cup, die, &kind, &enc)) != 0)
1001 1001 return (ret);
1002 1002 }
1003 1003
1004 1004 id = ctf_add_encoded(cup->cu_ctfp, isroot, name, &enc, kind);
1005 1005 if (id == CTF_ERR) {
1006 1006 ret = ctf_errno(cup->cu_ctfp);
1007 1007 } else {
1008 1008 *idp = id;
1009 1009 ret = ctf_dwmap_add(cup, id, die, B_FALSE);
1010 1010 }
1011 1011 out:
1012 1012 ctf_free(name, strlen(name) + 1);
1013 1013 return (ret);
1014 1014 }
1015 1015
1016 1016 /*
1017 1017 * Getting a member's offset is a surprisingly intricate dance. It works as
1018 1018 * follows:
1019 1019 *
1020 1020 * 1) If we're in DWARFv4, then we either have a DW_AT_data_bit_offset or we
1021 1021 * have a DW_AT_data_member_location. We won't have both. Thus we check first
1022 1022 * for DW_AT_data_bit_offset, and if it exists, we're set.
1023 1023 *
1024 1024 * Next, if we have a bitfield and we don't have a DW_AT_data_bit_offset, then
1025 1025 * we have to grab the data location and use the following dance:
1026 1026 *
1027 1027 * 2) Gather the set of DW_AT_byte_size, DW_AT_bit_offset, and DW_AT_bit_size.
1028 1028 * Of course, the DW_AT_byte_size may be omitted, even though it isn't always.
1029 1029 * When it's been omitted, we then have to say that the size is that of the
1030 1030 * underlying type, which forces that to be after a ctf_update(). Here, we have
1031 1031 * to do different things based on whether or not we're using big endian or
1032 1032 * little endian to obtain the proper offset.
1033 1033 */
1034 1034 static int
1035 1035 ctf_dwarf_member_offset(ctf_cu_t *cup, Dwarf_Die die, ctf_id_t mid,
1036 1036 ulong_t *offp)
1037 1037 {
1038 1038 int ret;
1039 1039 Dwarf_Unsigned loc, bitsz, bytesz;
1040 1040 Dwarf_Signed bitoff;
1041 1041 size_t off;
1042 1042 ssize_t tsz;
1043 1043
1044 1044 if ((ret = ctf_dwarf_unsigned(cup, die, DW_AT_data_bit_offset,
1045 1045 &loc)) == 0) {
1046 1046 *offp = loc;
1047 1047 return (0);
1048 1048 } else if (ret != ENOENT) {
1049 1049 return (ret);
1050 1050 }
1051 1051
1052 1052 if ((ret = ctf_dwarf_member_location(cup, die, &loc)) != 0)
1053 1053 return (ret);
1054 1054 off = loc * 8;
1055 1055
1056 1056 if ((ret = ctf_dwarf_signed(cup, die, DW_AT_bit_offset,
1057 1057 &bitoff)) != 0) {
1058 1058 if (ret != ENOENT)
1059 1059 return (ret);
1060 1060 *offp = off;
1061 1061 return (0);
1062 1062 }
1063 1063
1064 1064 /* At this point we have to have DW_AT_bit_size */
1065 1065 if ((ret = ctf_dwarf_unsigned(cup, die, DW_AT_bit_size, &bitsz)) != 0)
1066 1066 return (ret);
1067 1067
1068 1068 if ((ret = ctf_dwarf_unsigned(cup, die, DW_AT_byte_size,
1069 1069 &bytesz)) != 0) {
1070 1070 if (ret != ENOENT)
1071 1071 return (ret);
1072 1072 if ((tsz = ctf_type_size(cup->cu_ctfp, mid)) == CTF_ERR) {
1073 1073 int e = ctf_errno(cup->cu_ctfp);
1074 1074 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
1075 1075 "failed to get type size: %s", ctf_errmsg(e));
1076 1076 return (ECTF_CONVBKERR);
1077 1077 }
1078 1078 } else {
1079 1079 tsz = bytesz;
1080 1080 }
1081 1081 tsz *= 8;
1082 1082 if (cup->cu_bigend == B_TRUE) {
1083 1083 *offp = off + bitoff;
1084 1084 } else {
1085 1085 *offp = off + tsz - bitoff - bitsz;
1086 1086 }
1087 1087
1088 1088 return (0);
1089 1089 }
1090 1090
1091 1091 /*
1092 1092 * We need to determine if the member in question is a bitfield. If it is, then
1093 1093 * we need to go through and create a new type that's based on the actual base
1094 1094 * type, but has a different size. We also rename the type as a result to help
1095 1095 * deal with future collisions.
1096 1096 *
1097 1097 * Here we need to look and see if we have a DW_AT_bit_size value. If we have a
1098 1098 * bit size member and it does not equal the byte size member, then we need to
1099 1099 * create a bitfield type based on this.
1100 1100 *
1101 1101 * Note: When we support DWARFv4, there may be a chance that we need to also
1102 1102 * search for the DW_AT_byte_size if we don't have a DW_AT_bit_size member.
1103 1103 */
1104 1104 static int
1105 1105 ctf_dwarf_member_bitfield(ctf_cu_t *cup, Dwarf_Die die, ctf_id_t *idp)
1106 1106 {
1107 1107 int ret;
1108 1108 Dwarf_Unsigned bitsz;
1109 1109 ctf_encoding_t e;
1110 1110 ctf_dwbitf_t *cdb;
1111 1111 ctf_dtdef_t *dtd;
1112 1112 ctf_id_t base = *idp;
1113 1113 int kind;
1114 1114
1115 1115 if ((ret = ctf_dwarf_unsigned(cup, die, DW_AT_bit_size, &bitsz)) != 0) {
1116 1116 if (ret == ENOENT)
1117 1117 return (0);
1118 1118 return (ret);
1119 1119 }
1120 1120
1121 1121 ctf_dprintf("Trying to deal with bitfields on %d:%d\n", base, bitsz);
1122 1122 /*
1123 1123 * Given that we now have a bitsize, time to go do something about it.
1124 1124 * We're going to create a new type based on the current one, but first
1125 1125 * we need to find the base type. This means we need to traverse any
1126 1126 * typedef's, consts, and volatiles until we get to what should be
1127 1127 * something of type integer or enumeration.
1128 1128 */
1129 1129 VERIFY(bitsz < UINT32_MAX);
1130 1130 dtd = ctf_dtd_lookup(cup->cu_ctfp, base);
1131 1131 VERIFY(dtd != NULL);
1132 1132 kind = CTF_INFO_KIND(dtd->dtd_data.ctt_info);
1133 1133 while (kind == CTF_K_TYPEDEF || kind == CTF_K_CONST ||
1134 1134 kind == CTF_K_VOLATILE) {
1135 1135 dtd = ctf_dtd_lookup(cup->cu_ctfp, dtd->dtd_data.ctt_type);
1136 1136 VERIFY(dtd != NULL);
1137 1137 kind = CTF_INFO_KIND(dtd->dtd_data.ctt_info);
1138 1138 }
1139 1139 ctf_dprintf("got kind %d\n", kind);
1140 1140 VERIFY(kind == CTF_K_INTEGER || kind == CTF_K_ENUM);
1141 1141
1142 1142 /*
1143 1143 * As surprising as it may be, it is strictly possible to create a
1144 1144 * bitfield that is based on an enum. Of course, the C standard leaves
1145 1145 * enums sizing as an ABI concern more or less. To that effect, today on
1146 1146 * all illumos platforms the size of an enum is generally that of an
1147 1147 * int as our supported data models and ABIs all agree on that. So what
1148 1148 * we'll do is fake up a CTF encoding here to use. In this case, we'll
1149 1149 * treat it as an unsigned value of whatever size the underlying enum
1150 1150 * currently has (which is in the ctt_size member of its dynamic type
1151 1151 * data).
1152 1152 */
1153 1153 if (kind == CTF_K_INTEGER) {
1154 1154 e = dtd->dtd_u.dtu_enc;
1155 1155 } else {
1156 1156 bzero(&e, sizeof (ctf_encoding_t));
1157 1157 e.cte_bits = dtd->dtd_data.ctt_size * NBBY;
1158 1158 }
1159 1159
1160 1160 for (cdb = ctf_list_next(&cup->cu_bitfields); cdb != NULL;
1161 1161 cdb = ctf_list_next(cdb)) {
1162 1162 if (cdb->cdb_base == base && cdb->cdb_nbits == bitsz)
1163 1163 break;
1164 1164 }
1165 1165
1166 1166 /*
1167 1167 * Create a new type if none exists. We name all types in a way that is
1168 1168 * guaranteed not to conflict with the corresponding C type. We do this
1169 1169 * by using the ':' operator.
1170 1170 */
1171 1171 if (cdb == NULL) {
1172 1172 size_t namesz;
1173 1173 char *name;
1174 1174
1175 1175 e.cte_bits = bitsz;
1176 1176 namesz = snprintf(NULL, 0, "%s:%d", dtd->dtd_name,
1177 1177 (uint32_t)bitsz);
1178 1178 name = ctf_alloc(namesz + 1);
1179 1179 if (name == NULL)
1180 1180 return (ENOMEM);
1181 1181 cdb = ctf_alloc(sizeof (ctf_dwbitf_t));
1182 1182 if (cdb == NULL) {
1183 1183 ctf_free(name, namesz + 1);
1184 1184 return (ENOMEM);
1185 1185 }
1186 1186 (void) snprintf(name, namesz + 1, "%s:%d", dtd->dtd_name,
1187 1187 (uint32_t)bitsz);
1188 1188
1189 1189 cdb->cdb_base = base;
1190 1190 cdb->cdb_nbits = bitsz;
1191 1191 cdb->cdb_id = ctf_add_integer(cup->cu_ctfp, CTF_ADD_NONROOT,
1192 1192 name, &e);
1193 1193 if (cdb->cdb_id == CTF_ERR) {
1194 1194 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
1195 1195 "failed to get add bitfield type %s: %s", name,
1196 1196 ctf_errmsg(ctf_errno(cup->cu_ctfp)));
1197 1197 ctf_free(name, namesz + 1);
1198 1198 ctf_free(cdb, sizeof (ctf_dwbitf_t));
1199 1199 return (ECTF_CONVBKERR);
1200 1200 }
1201 1201 ctf_free(name, namesz + 1);
1202 1202 ctf_list_append(&cup->cu_bitfields, cdb);
1203 1203 }
1204 1204
1205 1205 *idp = cdb->cdb_id;
1206 1206
1207 1207 return (0);
1208 1208 }
1209 1209
1210 1210 static int
1211 1211 ctf_dwarf_fixup_sou(ctf_cu_t *cup, Dwarf_Die die, ctf_id_t base, boolean_t add)
1212 1212 {
1213 1213 int ret, kind;
1214 1214 Dwarf_Die child, memb;
1215 1215 Dwarf_Unsigned size;
1216 1216 ulong_t nsz;
1217 1217
1218 1218 kind = ctf_type_kind(cup->cu_ctfp, base);
1219 1219 VERIFY(kind != CTF_ERR);
1220 1220 VERIFY(kind == CTF_K_STRUCT || kind == CTF_K_UNION);
1221 1221
1222 1222 /*
1223 1223 * Members are in children. However, gcc also allows empty ones.
1224 1224 */
1225 1225 if ((ret = ctf_dwarf_child(cup, die, &child)) != 0)
1226 1226 return (ret);
1227 1227 if (child == NULL)
1228 1228 return (0);
1229 1229
1230 1230 memb = child;
1231 1231 while (memb != NULL) {
1232 1232 Dwarf_Die sib, tdie;
1233 1233 Dwarf_Half tag;
1234 1234 ctf_id_t mid;
1235 1235 char *mname;
1236 1236 ulong_t memboff = 0;
1237 1237
1238 1238 if ((ret = ctf_dwarf_tag(cup, memb, &tag)) != 0)
1239 1239 return (ret);
1240 1240
1241 1241 if (tag != DW_TAG_member)
1242 1242 continue;
1243 1243
1244 1244 if ((ret = ctf_dwarf_refdie(cup, memb, DW_AT_type, &tdie)) != 0)
1245 1245 return (ret);
1246 1246
1247 1247 if ((ret = ctf_dwarf_convert_type(cup, tdie, &mid,
1248 1248 CTF_ADD_NONROOT)) != 0)
1249 1249 return (ret);
1250 1250 ctf_dprintf("Got back type id: %d\n", mid);
1251 1251
1252 1252 /*
1253 1253 * If we're not adding a member, just go ahead and return.
1254 1254 */
1255 1255 if (add == B_FALSE) {
1256 1256 if ((ret = ctf_dwarf_member_bitfield(cup, memb,
1257 1257 &mid)) != 0)
1258 1258 return (ret);
1259 1259 goto next;
1260 1260 }
1261 1261
1262 1262 if ((ret = ctf_dwarf_string(cup, memb, DW_AT_name,
1263 1263 &mname)) != 0 && ret != ENOENT)
1264 1264 return (ret);
1265 1265 if (ret == ENOENT)
1266 1266 mname = NULL;
1267 1267
1268 1268 if (kind == CTF_K_UNION) {
1269 1269 memboff = 0;
1270 1270 } else if ((ret = ctf_dwarf_member_offset(cup, memb, mid,
1271 1271 &memboff)) != 0) {
1272 1272 if (mname != NULL)
1273 1273 ctf_free(mname, strlen(mname) + 1);
1274 1274 return (ret);
1275 1275 }
1276 1276
1277 1277 if ((ret = ctf_dwarf_member_bitfield(cup, memb, &mid)) != 0)
1278 1278 return (ret);
1279 1279
1280 1280 ret = ctf_add_member(cup->cu_ctfp, base, mname, mid, memboff);
1281 1281 if (ret == CTF_ERR) {
1282 1282 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
1283 1283 "failed to add member %s: %s",
1284 1284 mname, ctf_errmsg(ctf_errno(cup->cu_ctfp)));
1285 1285 if (mname != NULL)
1286 1286 ctf_free(mname, strlen(mname) + 1);
1287 1287 return (ECTF_CONVBKERR);
1288 1288 }
1289 1289
1290 1290 if (mname != NULL)
1291 1291 ctf_free(mname, strlen(mname) + 1);
1292 1292
1293 1293 next:
1294 1294 if ((ret = ctf_dwarf_sib(cup, memb, &sib)) != 0)
1295 1295 return (ret);
1296 1296 memb = sib;
1297 1297 }
1298 1298
1299 1299 /*
1300 1300 * If we're not adding members, then we don't know the final size of the
1301 1301 * structure, so end here.
1302 1302 */
1303 1303 if (add == B_FALSE)
1304 1304 return (0);
1305 1305
1306 1306 /* Finally set the size of the structure to the actual byte size */
1307 1307 if ((ret = ctf_dwarf_unsigned(cup, die, DW_AT_byte_size, &size)) != 0)
1308 1308 return (ret);
1309 1309 nsz = size;
1310 1310 if ((ctf_set_size(cup->cu_ctfp, base, nsz)) == CTF_ERR) {
1311 1311 int e = ctf_errno(cup->cu_ctfp);
1312 1312 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
1313 1313 "failed to set type size for %d to 0x%x: %s", base,
1314 1314 (uint32_t)size, ctf_errmsg(e));
1315 1315 return (ECTF_CONVBKERR);
1316 1316 }
1317 1317
1318 1318 return (0);
1319 1319 }
1320 1320
1321 1321 static int
1322 1322 ctf_dwarf_create_sou(ctf_cu_t *cup, Dwarf_Die die, ctf_id_t *idp,
1323 1323 int kind, int isroot)
1324 1324 {
1325 1325 int ret;
1326 1326 char *name;
1327 1327 ctf_id_t base;
1328 1328 Dwarf_Die child;
1329 1329 Dwarf_Bool decl;
1330 1330
1331 1331 /*
1332 1332 * Deal with the terribly annoying case of anonymous structs and unions.
1333 1333 * If they don't have a name, set the name to the empty string.
1334 1334 */
1335 1335 if ((ret = ctf_dwarf_string(cup, die, DW_AT_name, &name)) != 0 &&
1336 1336 ret != ENOENT)
1337 1337 return (ret);
1338 1338 if (ret == ENOENT)
1339 1339 name = NULL;
1340 1340
1341 1341 /*
1342 1342 * We need to check if we just have a declaration here. If we do, then
1343 1343 * instead of creating an actual structure or union, we're just going to
1344 1344 * go ahead and create a forward. During a dedup or merge, the forward
1345 1345 * will be replaced with the real thing.
1346 1346 */
1347 1347 if ((ret = ctf_dwarf_boolean(cup, die, DW_AT_declaration,
1348 1348 &decl)) != 0) {
1349 1349 if (ret != ENOENT)
1350 1350 return (ret);
1351 1351 decl = 0;
1352 1352 }
1353 1353
1354 1354 if (decl != 0) {
1355 1355 base = ctf_add_forward(cup->cu_ctfp, isroot, name, kind);
1356 1356 } else if (kind == CTF_K_STRUCT) {
1357 1357 base = ctf_add_struct(cup->cu_ctfp, isroot, name);
1358 1358 } else {
1359 1359 base = ctf_add_union(cup->cu_ctfp, isroot, name);
1360 1360 }
1361 1361 ctf_dprintf("added sou %s (%d) (%d)\n", name, kind, base);
1362 1362 if (name != NULL)
1363 1363 ctf_free(name, strlen(name) + 1);
1364 1364 if (base == CTF_ERR)
1365 1365 return (ctf_errno(cup->cu_ctfp));
1366 1366 *idp = base;
1367 1367
1368 1368 /*
1369 1369 * If it's just a declaration, we're not going to mark it for fix up or
1370 1370 * do anything else.
1371 1371 */
1372 1372 if (decl == B_TRUE)
1373 1373 return (ctf_dwmap_add(cup, base, die, B_FALSE));
1374 1374 if ((ret = ctf_dwmap_add(cup, base, die, B_TRUE)) != 0)
1375 1375 return (ret);
1376 1376
1377 1377 /*
1378 1378 * Members are in children. However, gcc also allows empty ones.
1379 1379 */
1380 1380 if ((ret = ctf_dwarf_child(cup, die, &child)) != 0)
1381 1381 return (ret);
1382 1382 if (child == NULL)
1383 1383 return (0);
1384 1384
1385 1385 return (0);
1386 1386 }
1387 1387
1388 1388 static int
1389 1389 ctf_dwarf_create_array_range(ctf_cu_t *cup, Dwarf_Die range, ctf_id_t *idp,
1390 1390 ctf_id_t base, int isroot)
1391 1391 {
1392 1392 int ret;
1393 1393 Dwarf_Die sib;
1394 1394 Dwarf_Unsigned val;
1395 1395 Dwarf_Signed sval;
1396 1396 ctf_arinfo_t ar;
1397 1397
1398 1398 ctf_dprintf("creating array range\n");
1399 1399
1400 1400 if ((ret = ctf_dwarf_sib(cup, range, &sib)) != 0)
1401 1401 return (ret);
1402 1402 if (sib != NULL) {
1403 1403 ctf_id_t id;
1404 1404 if ((ret = ctf_dwarf_create_array_range(cup, sib, &id,
1405 1405 base, CTF_ADD_NONROOT)) != 0)
1406 1406 return (ret);
1407 1407 ar.ctr_contents = id;
1408 1408 } else {
1409 1409 ar.ctr_contents = base;
1410 1410 }
1411 1411
1412 1412 if ((ar.ctr_index = ctf_dwarf_long(cup)) == CTF_ERR)
1413 1413 return (ctf_errno(cup->cu_ctfp));
1414 1414
1415 1415 /*
1416 1416 * Array bounds can be signed or unsigned, but there are several kinds
1417 1417 * of signless forms (data1, data2, etc) that take their sign from the
1418 1418 * routine that is trying to interpret them. That is, data1 can be
1419 1419 * either signed or unsigned, depending on whether you use the signed or
1420 1420 * unsigned accessor function. GCC will use the signless forms to store
1421 1421 * unsigned values which have their high bit set, so we need to try to
1422 1422 * read them first as unsigned to get positive values. We could also
1423 1423 * try signed first, falling back to unsigned if we got a negative
1424 1424 * value.
1425 1425 */
1426 1426 if ((ret = ctf_dwarf_unsigned(cup, range, DW_AT_upper_bound,
1427 1427 &val)) == 0) {
1428 1428 ar.ctr_nelems = val + 1;
1429 1429 } else if (ret != ENOENT) {
1430 1430 return (ret);
1431 1431 } else if ((ret = ctf_dwarf_signed(cup, range, DW_AT_upper_bound,
1432 1432 &sval)) == 0) {
1433 1433 ar.ctr_nelems = sval + 1;
1434 1434 } else if (ret != ENOENT) {
1435 1435 return (ret);
1436 1436 } else {
1437 1437 ar.ctr_nelems = 0;
1438 1438 }
1439 1439
1440 1440 if ((*idp = ctf_add_array(cup->cu_ctfp, isroot, &ar)) == CTF_ERR)
1441 1441 return (ctf_errno(cup->cu_ctfp));
1442 1442
1443 1443 return (0);
1444 1444 }
1445 1445
1446 1446 /*
1447 1447 * Try and create an array type. First, the kind of the array is specified in
1448 1448 * the DW_AT_type entry. Next, the number of entries is stored in a more
1449 1449 * complicated form, we should have a child that has the DW_TAG_subrange type.
1450 1450 */
1451 1451 static int
1452 1452 ctf_dwarf_create_array(ctf_cu_t *cup, Dwarf_Die die, ctf_id_t *idp, int isroot)
1453 1453 {
1454 1454 int ret;
1455 1455 Dwarf_Die tdie, rdie;
1456 1456 ctf_id_t tid;
1457 1457 Dwarf_Half rtag;
1458 1458
1459 1459 if ((ret = ctf_dwarf_refdie(cup, die, DW_AT_type, &tdie)) != 0)
1460 1460 return (ret);
1461 1461 if ((ret = ctf_dwarf_convert_type(cup, tdie, &tid,
1462 1462 CTF_ADD_NONROOT)) != 0)
1463 1463 return (ret);
1464 1464
1465 1465 if ((ret = ctf_dwarf_child(cup, die, &rdie)) != 0)
1466 1466 return (ret);
1467 1467 if ((ret = ctf_dwarf_tag(cup, rdie, &rtag)) != 0)
1468 1468 return (ret);
1469 1469 if (rtag != DW_TAG_subrange_type) {
1470 1470 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
1471 1471 "encountered array without DW_TAG_subrange_type child\n");
1472 1472 return (ECTF_CONVBKERR);
1473 1473 }
1474 1474
1475 1475 /*
1476 1476 * The compiler may opt to describe a multi-dimensional array as one
1477 1477 * giant array or it may opt to instead encode it as a series of
|
↓ open down ↓ |
1436 lines elided |
↑ open up ↑ |
1478 1478 * subranges. If it's the latter, then for each subrange we introduce a
1479 1479 * type. We can always use the base type.
1480 1480 */
1481 1481 if ((ret = ctf_dwarf_create_array_range(cup, rdie, idp, tid,
1482 1482 isroot)) != 0)
1483 1483 return (ret);
1484 1484 ctf_dprintf("Got back id %d\n", *idp);
1485 1485 return (ctf_dwmap_add(cup, *idp, die, B_FALSE));
1486 1486 }
1487 1487
1488 +/*
1489 + * Given "const int const_array3[11]", GCC7 at least will create a DIE tree of
1490 + * DW_TAG_const_type:DW_TAG_array_type:DW_Tag_const_type:<member_type>.
1491 + *
1492 + * Given C's syntax, this renders out as "const const int const_array3[11]". To
1493 + * get closer to round-tripping (and make the unit tests work), we'll peek for
1494 + * this case, and avoid adding the extraneous qualifier if we see that the
1495 + * underlying array referent already has the same qualifier.
1496 + *
1497 + * This is unfortunately less trivial than it could be: this issue applies to
1498 + * qualifier sets like "const volatile", as well as multi-dimensional arrays, so
1499 + * we need to descend down those.
1500 + *
1501 + * Returns CTF_ERR on error, or a boolean value otherwise.
1502 + */
1488 1503 static int
1504 +needed_array_qualifier(ctf_cu_t *cup, int kind, ctf_id_t ref_id)
1505 +{
1506 + const ctf_type_t *t;
1507 + ctf_arinfo_t arinfo;
1508 + int akind;
1509 +
1510 + if (kind != CTF_K_CONST && kind != CTF_K_VOLATILE &&
1511 + kind != CTF_K_RESTRICT)
1512 + return (1);
1513 +
1514 + if ((t = ctf_dyn_lookup_by_id(cup->cu_ctfp, ref_id)) == NULL)
1515 + return (CTF_ERR);
1516 +
1517 + if (LCTF_INFO_KIND(cup->cu_ctfp, t->ctt_info) != CTF_K_ARRAY)
1518 + return (1);
1519 +
1520 + if (ctf_dyn_array_info(cup->cu_ctfp, ref_id, &arinfo) != 0)
1521 + return (CTF_ERR);
1522 +
1523 + ctf_id_t id = arinfo.ctr_contents;
1524 +
1525 + for (;;) {
1526 + if ((t = ctf_dyn_lookup_by_id(cup->cu_ctfp, id)) == NULL)
1527 + return (CTF_ERR);
1528 +
1529 + akind = LCTF_INFO_KIND(cup->cu_ctfp, t->ctt_info);
1530 +
1531 + if (akind == kind)
1532 + break;
1533 +
1534 + if (akind == CTF_K_ARRAY) {
1535 + if (ctf_dyn_array_info(cup->cu_ctfp,
1536 + id, &arinfo) != 0)
1537 + return (CTF_ERR);
1538 + id = arinfo.ctr_contents;
1539 + continue;
1540 + }
1541 +
1542 + if (akind != CTF_K_CONST && akind != CTF_K_VOLATILE &&
1543 + akind != CTF_K_RESTRICT)
1544 + break;
1545 +
1546 + id = t->ctt_type;
1547 + }
1548 +
1549 + if (kind == akind) {
1550 + ctf_dprintf("ignoring extraneous %s qualifier for array %d\n",
1551 + ctf_kind_name(cup->cu_ctfp, kind), ref_id);
1552 + }
1553 +
1554 + return (kind != akind);
1555 +}
1556 +
1557 +static int
1489 1558 ctf_dwarf_create_reference(ctf_cu_t *cup, Dwarf_Die die, ctf_id_t *idp,
1490 1559 int kind, int isroot)
1491 1560 {
1492 1561 int ret;
1493 1562 ctf_id_t id;
1494 1563 Dwarf_Die tdie;
1495 1564 char *name;
1496 1565 size_t namelen;
1497 1566
1498 1567 if ((ret = ctf_dwarf_string(cup, die, DW_AT_name, &name)) != 0 &&
1499 1568 ret != ENOENT)
1500 1569 return (ret);
1501 1570 if (ret == ENOENT) {
1502 1571 name = NULL;
1503 1572 namelen = 0;
1504 1573 } else {
1505 1574 namelen = strlen(name);
1506 1575 }
1507 1576
1508 1577 ctf_dprintf("reference kind %d %s\n", kind, name != NULL ? name : "<>");
1509 1578
1510 1579 if ((ret = ctf_dwarf_refdie(cup, die, DW_AT_type, &tdie)) != 0) {
1511 1580 if (ret != ENOENT) {
1512 1581 ctf_free(name, namelen);
1513 1582 return (ret);
1514 1583 }
1515 1584 if ((id = ctf_dwarf_void(cup)) == CTF_ERR) {
1516 1585 ctf_free(name, namelen);
|
↓ open down ↓ |
18 lines elided |
↑ open up ↑ |
1517 1586 return (ctf_errno(cup->cu_ctfp));
1518 1587 }
1519 1588 } else {
1520 1589 if ((ret = ctf_dwarf_convert_type(cup, tdie, &id,
1521 1590 CTF_ADD_NONROOT)) != 0) {
1522 1591 ctf_free(name, namelen);
1523 1592 return (ret);
1524 1593 }
1525 1594 }
1526 1595
1596 + if ((ret = needed_array_qualifier(cup, kind, id)) <= 0) {
1597 + if (ret != 0) {
1598 + ret = (ctf_errno(cup->cu_ctfp));
1599 + } else {
1600 + *idp = id;
1601 + }
1602 +
1603 + ctf_free(name, namelen);
1604 + return (ret);
1605 + }
1606 +
1527 1607 if ((*idp = ctf_add_reftype(cup->cu_ctfp, isroot, name, id, kind)) ==
1528 1608 CTF_ERR) {
1529 1609 ctf_free(name, namelen);
1530 1610 return (ctf_errno(cup->cu_ctfp));
1531 1611 }
1532 1612
1533 1613 ctf_free(name, namelen);
1534 1614 return (ctf_dwmap_add(cup, *idp, die, B_FALSE));
1535 1615 }
1536 1616
1537 1617 static int
1538 1618 ctf_dwarf_create_enum(ctf_cu_t *cup, Dwarf_Die die, ctf_id_t *idp, int isroot)
1539 1619 {
1540 1620 int ret;
1541 1621 ctf_id_t id;
1542 1622 Dwarf_Die child;
1543 1623 char *name;
1544 1624
1545 1625 if ((ret = ctf_dwarf_string(cup, die, DW_AT_name, &name)) != 0 &&
1546 1626 ret != ENOENT)
1547 1627 return (ret);
1548 1628 if (ret == ENOENT)
1549 1629 name = NULL;
1550 1630 id = ctf_add_enum(cup->cu_ctfp, isroot, name);
1551 1631 ctf_dprintf("added enum %s (%d)\n", name, id);
1552 1632 if (name != NULL)
1553 1633 ctf_free(name, strlen(name) + 1);
1554 1634 if (id == CTF_ERR)
1555 1635 return (ctf_errno(cup->cu_ctfp));
1556 1636 *idp = id;
1557 1637 if ((ret = ctf_dwmap_add(cup, id, die, B_FALSE)) != 0)
1558 1638 return (ret);
1559 1639
1560 1640 if ((ret = ctf_dwarf_child(cup, die, &child)) != 0) {
1561 1641 if (ret == ENOENT)
1562 1642 ret = 0;
1563 1643 return (ret);
1564 1644 }
1565 1645
1566 1646 while (child != NULL) {
1567 1647 Dwarf_Half tag;
1568 1648 Dwarf_Signed sval;
1569 1649 Dwarf_Unsigned uval;
1570 1650 Dwarf_Die arg = child;
1571 1651 int eval;
1572 1652
1573 1653 if ((ret = ctf_dwarf_sib(cup, arg, &child)) != 0)
1574 1654 return (ret);
1575 1655
1576 1656 if ((ret = ctf_dwarf_tag(cup, arg, &tag)) != 0)
1577 1657 return (ret);
1578 1658
1579 1659 if (tag != DW_TAG_enumerator) {
1580 1660 if ((ret = ctf_dwarf_convert_type(cup, arg, NULL,
1581 1661 CTF_ADD_NONROOT)) != 0)
1582 1662 return (ret);
1583 1663 continue;
1584 1664 }
1585 1665
1586 1666 /*
1587 1667 * DWARF v4 section 5.7 tells us we'll always have names.
1588 1668 */
1589 1669 if ((ret = ctf_dwarf_string(cup, arg, DW_AT_name, &name)) != 0)
1590 1670 return (ret);
1591 1671
1592 1672 /*
1593 1673 * We have to be careful here: newer GCCs generate DWARF where
1594 1674 * an unsigned value will happily pass ctf_dwarf_signed().
1595 1675 * Since negative values will fail ctf_dwarf_unsigned(), we try
1596 1676 * that first to make sure we get the right value.
1597 1677 */
1598 1678 if ((ret = ctf_dwarf_unsigned(cup, arg, DW_AT_const_value,
1599 1679 &uval)) == 0) {
1600 1680 eval = (int)uval;
1601 1681 } else if ((ret = ctf_dwarf_signed(cup, arg, DW_AT_const_value,
1602 1682 &sval)) == 0) {
1603 1683 eval = sval;
1604 1684 }
1605 1685
1606 1686 if (ret != 0) {
1607 1687 if (ret != ENOENT)
1608 1688 return (ret);
1609 1689
1610 1690 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
1611 1691 "encountered enumeration without constant value\n");
1612 1692 return (ECTF_CONVBKERR);
1613 1693 }
1614 1694
1615 1695 ret = ctf_add_enumerator(cup->cu_ctfp, id, name, eval);
1616 1696 if (ret == CTF_ERR) {
1617 1697 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
1618 1698 "failed to add enumarator %s (%d) to %d\n",
1619 1699 name, eval, id);
1620 1700 ctf_free(name, strlen(name) + 1);
1621 1701 return (ctf_errno(cup->cu_ctfp));
1622 1702 }
1623 1703 ctf_free(name, strlen(name) + 1);
1624 1704 }
1625 1705
1626 1706 return (0);
1627 1707 }
1628 1708
1629 1709 /*
1630 1710 * For a function pointer, walk over and process all of its children, unless we
1631 1711 * encounter one that's just a declaration. In which case, we error on it.
1632 1712 */
1633 1713 static int
1634 1714 ctf_dwarf_create_fptr(ctf_cu_t *cup, Dwarf_Die die, ctf_id_t *idp, int isroot)
1635 1715 {
1636 1716 int ret;
1637 1717 Dwarf_Bool b;
1638 1718 ctf_funcinfo_t fi;
1639 1719 Dwarf_Die retdie;
1640 1720 ctf_id_t *argv = NULL;
1641 1721
1642 1722 bzero(&fi, sizeof (ctf_funcinfo_t));
1643 1723
1644 1724 if ((ret = ctf_dwarf_boolean(cup, die, DW_AT_declaration, &b)) != 0) {
1645 1725 if (ret != ENOENT)
1646 1726 return (ret);
1647 1727 } else {
1648 1728 if (b != 0)
1649 1729 return (EPROTOTYPE);
1650 1730 }
1651 1731
1652 1732 /*
1653 1733 * Return type is in DW_AT_type, if none, it returns void.
1654 1734 */
1655 1735 if ((ret = ctf_dwarf_refdie(cup, die, DW_AT_type, &retdie)) != 0) {
1656 1736 if (ret != ENOENT)
1657 1737 return (ret);
1658 1738 if ((fi.ctc_return = ctf_dwarf_void(cup)) == CTF_ERR)
1659 1739 return (ctf_errno(cup->cu_ctfp));
1660 1740 } else {
1661 1741 if ((ret = ctf_dwarf_convert_type(cup, retdie, &fi.ctc_return,
1662 1742 CTF_ADD_NONROOT)) != 0)
1663 1743 return (ret);
1664 1744 }
1665 1745
1666 1746 if ((ret = ctf_dwarf_function_count(cup, die, &fi, B_TRUE)) != 0) {
1667 1747 return (ret);
1668 1748 }
1669 1749
1670 1750 if (fi.ctc_argc != 0) {
1671 1751 argv = ctf_alloc(sizeof (ctf_id_t) * fi.ctc_argc);
1672 1752 if (argv == NULL)
1673 1753 return (ENOMEM);
1674 1754
1675 1755 if ((ret = ctf_dwarf_convert_fargs(cup, die, &fi, argv)) != 0) {
1676 1756 ctf_free(argv, sizeof (ctf_id_t) * fi.ctc_argc);
1677 1757 return (ret);
1678 1758 }
1679 1759 }
1680 1760
1681 1761 if ((*idp = ctf_add_funcptr(cup->cu_ctfp, isroot, &fi, argv)) ==
1682 1762 CTF_ERR) {
1683 1763 ctf_free(argv, sizeof (ctf_id_t) * fi.ctc_argc);
1684 1764 return (ctf_errno(cup->cu_ctfp));
1685 1765 }
1686 1766
1687 1767 ctf_free(argv, sizeof (ctf_id_t) * fi.ctc_argc);
1688 1768 return (ctf_dwmap_add(cup, *idp, die, B_FALSE));
1689 1769 }
1690 1770
1691 1771 static int
1692 1772 ctf_dwarf_convert_type(ctf_cu_t *cup, Dwarf_Die die, ctf_id_t *idp,
1693 1773 int isroot)
1694 1774 {
1695 1775 int ret;
1696 1776 Dwarf_Off offset;
1697 1777 Dwarf_Half tag;
1698 1778 ctf_dwmap_t lookup, *map;
1699 1779 ctf_id_t id;
1700 1780
1701 1781 if (idp == NULL)
1702 1782 idp = &id;
1703 1783
1704 1784 if ((ret = ctf_dwarf_offset(cup, die, &offset)) != 0)
1705 1785 return (ret);
1706 1786
1707 1787 if (offset > cup->cu_maxoff) {
1708 1788 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
1709 1789 "die offset %llu beyond maximum for header %llu\n",
1710 1790 offset, cup->cu_maxoff);
1711 1791 return (ECTF_CONVBKERR);
1712 1792 }
1713 1793
1714 1794 /*
1715 1795 * If we've already added an entry for this offset, then we're done.
1716 1796 */
1717 1797 lookup.cdm_off = offset;
1718 1798 if ((map = avl_find(&cup->cu_map, &lookup, NULL)) != NULL) {
1719 1799 *idp = map->cdm_id;
1720 1800 return (0);
1721 1801 }
1722 1802
1723 1803 if ((ret = ctf_dwarf_tag(cup, die, &tag)) != 0)
1724 1804 return (ret);
1725 1805
1726 1806 ret = ENOTSUP;
1727 1807 switch (tag) {
1728 1808 case DW_TAG_base_type:
1729 1809 ctf_dprintf("base\n");
1730 1810 ret = ctf_dwarf_create_base(cup, die, idp, isroot, offset);
1731 1811 break;
1732 1812 case DW_TAG_array_type:
1733 1813 ctf_dprintf("array\n");
1734 1814 ret = ctf_dwarf_create_array(cup, die, idp, isroot);
1735 1815 break;
1736 1816 case DW_TAG_enumeration_type:
1737 1817 ctf_dprintf("enum\n");
1738 1818 ret = ctf_dwarf_create_enum(cup, die, idp, isroot);
1739 1819 break;
1740 1820 case DW_TAG_pointer_type:
1741 1821 ctf_dprintf("pointer\n");
1742 1822 ret = ctf_dwarf_create_reference(cup, die, idp, CTF_K_POINTER,
1743 1823 isroot);
1744 1824 break;
1745 1825 case DW_TAG_structure_type:
1746 1826 ctf_dprintf("struct\n");
1747 1827 ret = ctf_dwarf_create_sou(cup, die, idp, CTF_K_STRUCT,
1748 1828 isroot);
1749 1829 break;
1750 1830 case DW_TAG_subroutine_type:
1751 1831 ctf_dprintf("fptr\n");
1752 1832 ret = ctf_dwarf_create_fptr(cup, die, idp, isroot);
1753 1833 break;
1754 1834 case DW_TAG_typedef:
1755 1835 ctf_dprintf("typedef\n");
1756 1836 ret = ctf_dwarf_create_reference(cup, die, idp, CTF_K_TYPEDEF,
1757 1837 isroot);
1758 1838 break;
1759 1839 case DW_TAG_union_type:
1760 1840 ctf_dprintf("union\n");
1761 1841 ret = ctf_dwarf_create_sou(cup, die, idp, CTF_K_UNION,
1762 1842 isroot);
1763 1843 break;
1764 1844 case DW_TAG_const_type:
1765 1845 ctf_dprintf("const\n");
1766 1846 ret = ctf_dwarf_create_reference(cup, die, idp, CTF_K_CONST,
1767 1847 isroot);
1768 1848 break;
1769 1849 case DW_TAG_volatile_type:
1770 1850 ctf_dprintf("volatile\n");
1771 1851 ret = ctf_dwarf_create_reference(cup, die, idp, CTF_K_VOLATILE,
1772 1852 isroot);
1773 1853 break;
1774 1854 case DW_TAG_restrict_type:
1775 1855 ctf_dprintf("restrict\n");
1776 1856 ret = ctf_dwarf_create_reference(cup, die, idp, CTF_K_RESTRICT,
1777 1857 isroot);
1778 1858 break;
1779 1859 default:
1780 1860 ctf_dprintf("ignoring tag type %x\n", tag);
1781 1861 *idp = CTF_ERR;
1782 1862 ret = 0;
1783 1863 break;
1784 1864 }
1785 1865 ctf_dprintf("ctf_dwarf_convert_type tag specific handler returned %d\n",
1786 1866 ret);
1787 1867
1788 1868 return (ret);
1789 1869 }
1790 1870
1791 1871 static int
1792 1872 ctf_dwarf_walk_lexical(ctf_cu_t *cup, Dwarf_Die die)
1793 1873 {
1794 1874 int ret;
1795 1875 Dwarf_Die child;
1796 1876
1797 1877 if ((ret = ctf_dwarf_child(cup, die, &child)) != 0)
1798 1878 return (ret);
1799 1879
1800 1880 if (child == NULL)
1801 1881 return (0);
1802 1882
1803 1883 return (ctf_dwarf_convert_die(cup, die));
1804 1884 }
1805 1885
1806 1886 static int
1807 1887 ctf_dwarf_function_count(ctf_cu_t *cup, Dwarf_Die die, ctf_funcinfo_t *fip,
1808 1888 boolean_t fptr)
1809 1889 {
1810 1890 int ret;
1811 1891 Dwarf_Die child, sib, arg;
1812 1892
1813 1893 if ((ret = ctf_dwarf_child(cup, die, &child)) != 0)
|
↓ open down ↓ |
277 lines elided |
↑ open up ↑ |
1814 1894 return (ret);
1815 1895
1816 1896 arg = child;
1817 1897 while (arg != NULL) {
1818 1898 Dwarf_Half tag;
1819 1899
1820 1900 if ((ret = ctf_dwarf_tag(cup, arg, &tag)) != 0)
1821 1901 return (ret);
1822 1902
1823 1903 /*
1824 - * We have to check for a varargs type decleration. This will
1904 + * We have to check for a varargs type declaration. This will
1825 1905 * happen in one of two ways. If we have a function pointer
1826 1906 * type, then it'll be done with a tag of type
1827 1907 * DW_TAG_unspecified_parameters. However, it only means we have
1828 1908 * a variable number of arguments, if we have more than one
1829 1909 * argument found so far. Otherwise, when we have a function
1830 1910 * type, it instead uses a formal parameter whose name is '...'
1831 1911 * to indicate a variable arguments member.
1832 1912 *
1833 1913 * Also, if we have a function pointer, then we have to expect
1834 1914 * that we might not get a name at all.
1835 1915 */
1836 1916 if (tag == DW_TAG_formal_parameter && fptr == B_FALSE) {
1837 1917 char *name;
1838 1918 if ((ret = ctf_dwarf_string(cup, die, DW_AT_name,
1839 1919 &name)) != 0)
1840 1920 return (ret);
1841 1921 if (strcmp(name, DWARF_VARARGS_NAME) == 0)
1842 1922 fip->ctc_flags |= CTF_FUNC_VARARG;
1843 1923 else
1844 1924 fip->ctc_argc++;
1845 1925 ctf_free(name, strlen(name) + 1);
1846 1926 } else if (tag == DW_TAG_formal_parameter) {
1847 1927 fip->ctc_argc++;
1848 1928 } else if (tag == DW_TAG_unspecified_parameters &&
1849 1929 fip->ctc_argc > 0) {
1850 1930 fip->ctc_flags |= CTF_FUNC_VARARG;
1851 1931 }
1852 1932 if ((ret = ctf_dwarf_sib(cup, arg, &sib)) != 0)
1853 1933 return (ret);
1854 1934 arg = sib;
1855 1935 }
1856 1936
1857 1937 return (0);
1858 1938 }
1859 1939
1860 1940 static int
1861 1941 ctf_dwarf_convert_fargs(ctf_cu_t *cup, Dwarf_Die die, ctf_funcinfo_t *fip,
1862 1942 ctf_id_t *argv)
1863 1943 {
1864 1944 int ret;
1865 1945 int i = 0;
1866 1946 Dwarf_Die child, sib, arg;
1867 1947
1868 1948 if ((ret = ctf_dwarf_child(cup, die, &child)) != 0)
1869 1949 return (ret);
1870 1950
1871 1951 arg = child;
1872 1952 while (arg != NULL) {
1873 1953 Dwarf_Half tag;
1874 1954
1875 1955 if ((ret = ctf_dwarf_tag(cup, arg, &tag)) != 0)
1876 1956 return (ret);
1877 1957 if (tag == DW_TAG_formal_parameter) {
1878 1958 Dwarf_Die tdie;
1879 1959
1880 1960 if ((ret = ctf_dwarf_refdie(cup, arg, DW_AT_type,
1881 1961 &tdie)) != 0)
1882 1962 return (ret);
1883 1963
1884 1964 if ((ret = ctf_dwarf_convert_type(cup, tdie, &argv[i],
1885 1965 CTF_ADD_ROOT)) != 0)
1886 1966 return (ret);
1887 1967 i++;
1888 1968
1889 1969 /*
1890 1970 * Once we hit argc entries, we're done. This ensures we
1891 1971 * don't accidentally hit a varargs which should be the
1892 1972 * last entry.
1893 1973 */
1894 1974 if (i == fip->ctc_argc)
1895 1975 break;
1896 1976 }
1897 1977
1898 1978 if ((ret = ctf_dwarf_sib(cup, arg, &sib)) != 0)
|
↓ open down ↓ |
64 lines elided |
↑ open up ↑ |
1899 1979 return (ret);
1900 1980 arg = sib;
1901 1981 }
1902 1982
1903 1983 return (0);
1904 1984 }
1905 1985
1906 1986 static int
1907 1987 ctf_dwarf_convert_function(ctf_cu_t *cup, Dwarf_Die die)
1908 1988 {
1909 - int ret;
1910 - char *name;
1911 1989 ctf_dwfunc_t *cdf;
1912 1990 Dwarf_Die tdie;
1991 + Dwarf_Bool b;
1992 + char *name;
1993 + int ret;
1913 1994
1914 1995 /*
1915 1996 * Functions that don't have a name are generally functions that have
1916 1997 * been inlined and thus most information about them has been lost. If
1917 1998 * we can't get a name, then instead of returning ENOENT, we silently
1918 1999 * swallow the error.
1919 2000 */
1920 2001 if ((ret = ctf_dwarf_string(cup, die, DW_AT_name, &name)) != 0) {
1921 2002 if (ret == ENOENT)
1922 2003 return (0);
1923 2004 return (ret);
1924 2005 }
1925 2006
1926 - ctf_dprintf("beginning work on function %s\n", name);
2007 + ctf_dprintf("beginning work on function %s (die %llx)\n",
2008 + name, ctf_die_offset(die));
2009 +
2010 + if ((ret = ctf_dwarf_boolean(cup, die, DW_AT_declaration, &b)) != 0) {
2011 + if (ret != ENOENT)
2012 + return (ret);
2013 + } else if (b != 0) {
2014 + /*
2015 + * GCC7 at least creates empty DW_AT_declarations for functions
2016 + * defined in headers. As they lack details on the function
2017 + * prototype, we need to ignore them. If we later actually
2018 + * see the relevant function's definition, we will see another
2019 + * DW_TAG_subprogram that is more complete.
2020 + */
2021 + ctf_dprintf("ignoring declaration of function %s (die %llx)\n",
2022 + name, ctf_die_offset(die));
2023 + return (0);
2024 + }
2025 +
1927 2026 if ((cdf = ctf_alloc(sizeof (ctf_dwfunc_t))) == NULL) {
1928 2027 ctf_free(name, strlen(name) + 1);
1929 2028 return (ENOMEM);
1930 2029 }
1931 2030 bzero(cdf, sizeof (ctf_dwfunc_t));
1932 2031 cdf->cdf_name = name;
1933 2032
1934 2033 if ((ret = ctf_dwarf_refdie(cup, die, DW_AT_type, &tdie)) == 0) {
1935 2034 if ((ret = ctf_dwarf_convert_type(cup, tdie,
1936 2035 &(cdf->cdf_fip.ctc_return), CTF_ADD_ROOT)) != 0) {
1937 2036 ctf_free(name, strlen(name) + 1);
1938 2037 ctf_free(cdf, sizeof (ctf_dwfunc_t));
1939 2038 return (ret);
1940 2039 }
1941 2040 } else if (ret != ENOENT) {
1942 2041 ctf_free(name, strlen(name) + 1);
1943 2042 ctf_free(cdf, sizeof (ctf_dwfunc_t));
1944 2043 return (ret);
1945 2044 } else {
1946 2045 if ((cdf->cdf_fip.ctc_return = ctf_dwarf_void(cup)) ==
1947 2046 CTF_ERR) {
1948 2047 ctf_free(name, strlen(name) + 1);
1949 2048 ctf_free(cdf, sizeof (ctf_dwfunc_t));
1950 2049 return (ctf_errno(cup->cu_ctfp));
1951 2050 }
1952 2051 }
1953 2052
1954 2053 /*
1955 2054 * A function has a number of children, some of which may not be ones we
1956 2055 * care about. Children that we care about have a type of
1957 2056 * DW_TAG_formal_parameter. We're going to do two passes, the first to
1958 2057 * count the arguments, the second to process them. Afterwards, we
1959 2058 * should be good to go ahead and add this function.
1960 2059 *
1961 2060 * Note, we already got the return type by going in and grabbing it out
1962 2061 * of the DW_AT_type.
1963 2062 */
1964 2063 if ((ret = ctf_dwarf_function_count(cup, die, &cdf->cdf_fip,
1965 2064 B_FALSE)) != 0) {
1966 2065 ctf_free(name, strlen(name) + 1);
1967 2066 ctf_free(cdf, sizeof (ctf_dwfunc_t));
1968 2067 return (ret);
1969 2068 }
1970 2069
1971 2070 ctf_dprintf("beginning to convert function arguments %s\n", name);
1972 2071 if (cdf->cdf_fip.ctc_argc != 0) {
1973 2072 uint_t argc = cdf->cdf_fip.ctc_argc;
1974 2073 cdf->cdf_argv = ctf_alloc(sizeof (ctf_id_t) * argc);
1975 2074 if (cdf->cdf_argv == NULL) {
1976 2075 ctf_free(name, strlen(name) + 1);
1977 2076 ctf_free(cdf, sizeof (ctf_dwfunc_t));
1978 2077 return (ENOMEM);
1979 2078 }
1980 2079 if ((ret = ctf_dwarf_convert_fargs(cup, die,
1981 2080 &cdf->cdf_fip, cdf->cdf_argv)) != 0) {
1982 2081 ctf_free(cdf->cdf_argv, sizeof (ctf_id_t) * argc);
1983 2082 ctf_free(name, strlen(name) + 1);
1984 2083 ctf_free(cdf, sizeof (ctf_dwfunc_t));
1985 2084 return (ret);
1986 2085 }
1987 2086 } else {
1988 2087 cdf->cdf_argv = NULL;
1989 2088 }
1990 2089
1991 2090 if ((ret = ctf_dwarf_isglobal(cup, die, &cdf->cdf_global)) != 0) {
1992 2091 ctf_free(cdf->cdf_argv, sizeof (ctf_id_t) *
1993 2092 cdf->cdf_fip.ctc_argc);
1994 2093 ctf_free(name, strlen(name) + 1);
1995 2094 ctf_free(cdf, sizeof (ctf_dwfunc_t));
1996 2095 return (ret);
1997 2096 }
1998 2097
1999 2098 ctf_list_append(&cup->cu_funcs, cdf);
2000 2099 return (ret);
2001 2100 }
2002 2101
2003 2102 /*
2004 2103 * Convert variables, but only if they're not prototypes and have names.
2005 2104 */
2006 2105 static int
2007 2106 ctf_dwarf_convert_variable(ctf_cu_t *cup, Dwarf_Die die)
2008 2107 {
2009 2108 int ret;
2010 2109 char *name;
2011 2110 Dwarf_Bool b;
2012 2111 Dwarf_Die tdie;
2013 2112 ctf_id_t id;
2014 2113 ctf_dwvar_t *cdv;
2015 2114
2016 2115 /* Skip "Non-Defining Declarations" */
2017 2116 if ((ret = ctf_dwarf_boolean(cup, die, DW_AT_declaration, &b)) == 0) {
2018 2117 if (b != 0)
2019 2118 return (0);
2020 2119 } else if (ret != ENOENT) {
2021 2120 return (ret);
2022 2121 }
2023 2122
2024 2123 /*
2025 2124 * If we find a DIE of "Declarations Completing Non-Defining
2026 2125 * Declarations", we will use the referenced type's DIE. This isn't
2027 2126 * quite correct, e.g. DW_AT_decl_line will be the forward declaration
2028 2127 * not this site. It's sufficient for what we need, however: in
2029 2128 * particular, we should find DW_AT_external as needed there.
2030 2129 */
2031 2130 if ((ret = ctf_dwarf_refdie(cup, die, DW_AT_specification,
2032 2131 &tdie)) == 0) {
2033 2132 Dwarf_Off offset;
2034 2133 if ((ret = ctf_dwarf_offset(cup, tdie, &offset)) != 0)
2035 2134 return (ret);
2036 2135 ctf_dprintf("die 0x%llx DW_AT_specification -> die 0x%llx\n",
2037 2136 ctf_die_offset(die), ctf_die_offset(tdie));
2038 2137 die = tdie;
2039 2138 } else if (ret != ENOENT) {
2040 2139 return (ret);
2041 2140 }
2042 2141
2043 2142 if ((ret = ctf_dwarf_string(cup, die, DW_AT_name, &name)) != 0 &&
2044 2143 ret != ENOENT)
2045 2144 return (ret);
2046 2145 if (ret == ENOENT)
2047 2146 return (0);
2048 2147
2049 2148 if ((ret = ctf_dwarf_refdie(cup, die, DW_AT_type, &tdie)) != 0) {
2050 2149 ctf_free(name, strlen(name) + 1);
2051 2150 return (ret);
2052 2151 }
2053 2152
2054 2153 if ((ret = ctf_dwarf_convert_type(cup, tdie, &id,
2055 2154 CTF_ADD_ROOT)) != 0)
2056 2155 return (ret);
2057 2156
2058 2157 if ((cdv = ctf_alloc(sizeof (ctf_dwvar_t))) == NULL) {
2059 2158 ctf_free(name, strlen(name) + 1);
2060 2159 return (ENOMEM);
2061 2160 }
2062 2161
2063 2162 cdv->cdv_name = name;
2064 2163 cdv->cdv_type = id;
2065 2164
2066 2165 if ((ret = ctf_dwarf_isglobal(cup, die, &cdv->cdv_global)) != 0) {
2067 2166 ctf_free(cdv, sizeof (ctf_dwvar_t));
2068 2167 ctf_free(name, strlen(name) + 1);
2069 2168 return (ret);
2070 2169 }
2071 2170
2072 2171 ctf_list_append(&cup->cu_vars, cdv);
2073 2172 return (0);
2074 2173 }
2075 2174
2076 2175 /*
2077 2176 * Walk through our set of top-level types and process them.
2078 2177 */
2079 2178 static int
2080 2179 ctf_dwarf_walk_toplevel(ctf_cu_t *cup, Dwarf_Die die)
2081 2180 {
2082 2181 int ret;
2083 2182 Dwarf_Off offset;
2084 2183 Dwarf_Half tag;
2085 2184
2086 2185 if ((ret = ctf_dwarf_offset(cup, die, &offset)) != 0)
2087 2186 return (ret);
2088 2187
2089 2188 if (offset > cup->cu_maxoff) {
2090 2189 (void) snprintf(cup->cu_errbuf, cup->cu_errlen,
2091 2190 "die offset %llu beyond maximum for header %llu\n",
2092 2191 offset, cup->cu_maxoff);
2093 2192 return (ECTF_CONVBKERR);
2094 2193 }
2095 2194
2096 2195 if ((ret = ctf_dwarf_tag(cup, die, &tag)) != 0)
2097 2196 return (ret);
2098 2197
2099 2198 ret = 0;
2100 2199 switch (tag) {
2101 2200 case DW_TAG_subprogram:
2102 2201 ctf_dprintf("top level func\n");
2103 2202 ret = ctf_dwarf_convert_function(cup, die);
2104 2203 break;
2105 2204 case DW_TAG_variable:
2106 2205 ctf_dprintf("top level var\n");
2107 2206 ret = ctf_dwarf_convert_variable(cup, die);
2108 2207 break;
2109 2208 case DW_TAG_lexical_block:
2110 2209 ctf_dprintf("top level block\n");
2111 2210 ret = ctf_dwarf_walk_lexical(cup, die);
2112 2211 break;
2113 2212 case DW_TAG_enumeration_type:
2114 2213 case DW_TAG_structure_type:
2115 2214 case DW_TAG_typedef:
2116 2215 case DW_TAG_union_type:
2117 2216 ctf_dprintf("top level type\n");
2118 2217 ret = ctf_dwarf_convert_type(cup, die, NULL, B_TRUE);
2119 2218 break;
2120 2219 default:
2121 2220 break;
2122 2221 }
2123 2222
2124 2223 return (ret);
2125 2224 }
2126 2225
2127 2226
2128 2227 /*
2129 2228 * We're given a node. At this node we need to convert it and then proceed to
2130 2229 * convert any siblings that are associaed with this die.
2131 2230 */
2132 2231 static int
2133 2232 ctf_dwarf_convert_die(ctf_cu_t *cup, Dwarf_Die die)
2134 2233 {
2135 2234 while (die != NULL) {
2136 2235 int ret;
2137 2236 Dwarf_Die sib;
2138 2237
2139 2238 if ((ret = ctf_dwarf_walk_toplevel(cup, die)) != 0)
2140 2239 return (ret);
2141 2240
2142 2241 if ((ret = ctf_dwarf_sib(cup, die, &sib)) != 0)
2143 2242 return (ret);
2144 2243 die = sib;
2145 2244 }
2146 2245 return (0);
2147 2246 }
2148 2247
2149 2248 static int
2150 2249 ctf_dwarf_fixup_die(ctf_cu_t *cup, boolean_t addpass)
2151 2250 {
2152 2251 ctf_dwmap_t *map;
2153 2252
2154 2253 for (map = avl_first(&cup->cu_map); map != NULL;
2155 2254 map = AVL_NEXT(&cup->cu_map, map)) {
2156 2255 int ret;
2157 2256 if (map->cdm_fix == B_FALSE)
2158 2257 continue;
2159 2258 if ((ret = ctf_dwarf_fixup_sou(cup, map->cdm_die, map->cdm_id,
2160 2259 addpass)) != 0)
2161 2260 return (ret);
2162 2261 }
2163 2262
2164 2263 return (0);
2165 2264 }
2166 2265
2167 2266 /*
2168 2267 * The DWARF information about a symbol and the information in the symbol table
2169 2268 * may not be the same due to symbol reduction that is performed by ld due to a
2170 2269 * mapfile or other such directive. We process weak symbols at a later time.
2171 2270 *
2172 2271 * The following are the rules that we employ:
2173 2272 *
2174 2273 * 1. A DWARF function that is considered exported matches STB_GLOBAL entries
2175 2274 * with the same name.
2176 2275 *
2177 2276 * 2. A DWARF function that is considered exported matches STB_LOCAL entries
2178 2277 * with the same name and the same file. This case may happen due to mapfile
2179 2278 * reduction.
2180 2279 *
2181 2280 * 3. A DWARF function that is not considered exported matches STB_LOCAL entries
2182 2281 * with the same name and the same file.
2183 2282 *
2184 2283 * 4. A DWARF function that has the same name as the symbol table entry, but the
2185 2284 * files do not match. This is considered a 'fuzzy' match. This may also happen
2186 2285 * due to a mapfile reduction. Fuzzy matching is only used when we know that the
2187 2286 * file in question refers to the primary object. This is because when a symbol
2188 2287 * is reduced in a mapfile, it's always going to be tagged as a local value in
2189 2288 * the generated output and it is considered as to belong to the primary file
2190 2289 * which is the first STT_FILE symbol we see.
2191 2290 */
2192 2291 static boolean_t
2193 2292 ctf_dwarf_symbol_match(const char *symtab_file, const char *symtab_name,
2194 2293 uint_t symtab_bind, const char *dwarf_file, const char *dwarf_name,
2195 2294 boolean_t dwarf_global, boolean_t *is_fuzzy)
2196 2295 {
2197 2296 *is_fuzzy = B_FALSE;
2198 2297
2199 2298 if (symtab_bind != STB_LOCAL && symtab_bind != STB_GLOBAL) {
2200 2299 return (B_FALSE);
2201 2300 }
2202 2301
2203 2302 if (strcmp(symtab_name, dwarf_name) != 0) {
2204 2303 return (B_FALSE);
2205 2304 }
2206 2305
2207 2306 if (symtab_bind == STB_GLOBAL) {
2208 2307 return (dwarf_global);
2209 2308 }
2210 2309
2211 2310 if (strcmp(symtab_file, dwarf_file) == 0) {
2212 2311 return (B_TRUE);
2213 2312 }
2214 2313
2215 2314 if (dwarf_global) {
2216 2315 *is_fuzzy = B_TRUE;
2217 2316 return (B_TRUE);
2218 2317 }
2219 2318
2220 2319 return (B_FALSE);
2221 2320 }
2222 2321
2223 2322 static ctf_dwfunc_t *
2224 2323 ctf_dwarf_match_func(ctf_cu_t *cup, const char *file, const char *name,
2225 2324 uint_t bind, boolean_t primary)
2226 2325 {
2227 2326 ctf_dwfunc_t *cdf, *fuzzy = NULL;
2228 2327
2229 2328 if (bind == STB_WEAK)
2230 2329 return (NULL);
2231 2330
2232 2331 if (bind == STB_LOCAL && (file == NULL || cup->cu_name == NULL))
2233 2332 return (NULL);
2234 2333
2235 2334 for (cdf = ctf_list_next(&cup->cu_funcs); cdf != NULL;
2236 2335 cdf = ctf_list_next(cdf)) {
2237 2336 boolean_t is_fuzzy = B_FALSE;
2238 2337
2239 2338 if (ctf_dwarf_symbol_match(file, name, bind, cup->cu_name,
2240 2339 cdf->cdf_name, cdf->cdf_global, &is_fuzzy)) {
2241 2340 if (is_fuzzy) {
2242 2341 if (primary) {
2243 2342 fuzzy = cdf;
2244 2343 }
2245 2344 continue;
2246 2345 } else {
2247 2346 return (cdf);
2248 2347 }
2249 2348 }
2250 2349 }
2251 2350
2252 2351 return (fuzzy);
2253 2352 }
2254 2353
2255 2354 static ctf_dwvar_t *
2256 2355 ctf_dwarf_match_var(ctf_cu_t *cup, const char *file, const char *name,
2257 2356 uint_t bind, boolean_t primary)
2258 2357 {
2259 2358 ctf_dwvar_t *cdv, *fuzzy = NULL;
2260 2359
2261 2360 if (bind == STB_WEAK)
2262 2361 return (NULL);
2263 2362
2264 2363 if (bind == STB_LOCAL && (file == NULL || cup->cu_name == NULL))
2265 2364 return (NULL);
2266 2365
2267 2366 for (cdv = ctf_list_next(&cup->cu_vars); cdv != NULL;
2268 2367 cdv = ctf_list_next(cdv)) {
2269 2368 boolean_t is_fuzzy = B_FALSE;
2270 2369
2271 2370 if (ctf_dwarf_symbol_match(file, name, bind, cup->cu_name,
2272 2371 cdv->cdv_name, cdv->cdv_global, &is_fuzzy)) {
2273 2372 if (is_fuzzy) {
2274 2373 if (primary) {
2275 2374 fuzzy = cdv;
2276 2375 }
2277 2376 } else {
2278 2377 return (cdv);
2279 2378 }
2280 2379 }
2281 2380 }
2282 2381
2283 2382 return (fuzzy);
2284 2383 }
2285 2384
2286 2385 static int
2287 2386 ctf_dwarf_conv_funcvars_cb(const Elf64_Sym *symp, ulong_t idx,
2288 2387 const char *file, const char *name, boolean_t primary, void *arg)
2289 2388 {
2290 2389 int ret;
2291 2390 uint_t bind, type;
2292 2391 ctf_cu_t *cup = arg;
2293 2392
2294 2393 bind = GELF_ST_BIND(symp->st_info);
2295 2394 type = GELF_ST_TYPE(symp->st_info);
2296 2395
2297 2396 /*
2298 2397 * Come back to weak symbols in another pass
2299 2398 */
2300 2399 if (bind == STB_WEAK)
2301 2400 return (0);
2302 2401
2303 2402 if (type == STT_OBJECT) {
2304 2403 ctf_dwvar_t *cdv = ctf_dwarf_match_var(cup, file, name,
2305 2404 bind, primary);
2306 2405 if (cdv == NULL)
2307 2406 return (0);
2308 2407 ret = ctf_add_object(cup->cu_ctfp, idx, cdv->cdv_type);
2309 2408 ctf_dprintf("added object %s->%ld\n", name, cdv->cdv_type);
2310 2409 } else {
2311 2410 ctf_dwfunc_t *cdf = ctf_dwarf_match_func(cup, file, name,
2312 2411 bind, primary);
2313 2412 if (cdf == NULL)
2314 2413 return (0);
2315 2414 ret = ctf_add_function(cup->cu_ctfp, idx, &cdf->cdf_fip,
2316 2415 cdf->cdf_argv);
2317 2416 ctf_dprintf("added function %s\n", name);
2318 2417 }
2319 2418
2320 2419 if (ret == CTF_ERR) {
2321 2420 return (ctf_errno(cup->cu_ctfp));
2322 2421 }
2323 2422
2324 2423 return (0);
2325 2424 }
2326 2425
2327 2426 static int
2328 2427 ctf_dwarf_conv_funcvars(ctf_cu_t *cup)
2329 2428 {
2330 2429 return (ctf_symtab_iter(cup->cu_ctfp, ctf_dwarf_conv_funcvars_cb, cup));
2331 2430 }
2332 2431
2333 2432 /*
2334 2433 * If we have a weak symbol, attempt to find the strong symbol it will resolve
2335 2434 * to. Note: the code where this actually happens is in sym_process() in
2336 2435 * cmd/sgs/libld/common/syms.c
2337 2436 *
2338 2437 * Finding the matching symbol is unfortunately not trivial. For a symbol to be
2339 2438 * a candidate, it must:
2340 2439 *
2341 2440 * - have the same type (function, object)
2342 2441 * - have the same value (address)
2343 2442 * - have the same size
2344 2443 * - not be another weak symbol
2345 2444 * - belong to the same section (checked via section index)
2346 2445 *
2347 2446 * To perform this check, we first iterate over the symbol table. For each weak
2348 2447 * symbol that we encounter, we then do a second walk over the symbol table,
2349 2448 * calling ctf_dwarf_conv_check_weak(). If a symbol matches the above, then it's
2350 2449 * either a local or global symbol. If we find a global symbol then we go with
2351 2450 * it and stop searching for additional matches.
2352 2451 *
2353 2452 * If instead, we find a local symbol, things are more complicated. The first
2354 2453 * thing we do is to try and see if we have file information about both symbols
2355 2454 * (STT_FILE). If they both have file information and it matches, then we treat
2356 2455 * that as a good match and stop searching for additional matches.
2357 2456 *
2358 2457 * Otherwise, this means we have a non-matching file and a local symbol. We
2359 2458 * treat this as a candidate and if we find a better match (one of the two cases
2360 2459 * above), use that instead. There are two different ways this can happen.
2361 2460 * Either this is a completely different symbol, or it's a once-global symbol
2362 2461 * that was scoped to local via a mapfile. In the former case, curfile is
2363 2462 * likely inaccurate since the linker does not preserve the needed curfile in
2364 2463 * the order of the symbol table (see the comments about locally scoped symbols
2365 2464 * in libld's update_osym()). As we can't tell this case from the former one,
2366 2465 * we use this symbol iff no other matching symbol is found.
2367 2466 *
2368 2467 * What we really need here is a SUNW section containing weak<->strong mappings
2369 2468 * that we can consume.
2370 2469 */
2371 2470 typedef struct ctf_dwarf_weak_arg {
2372 2471 const Elf64_Sym *cweak_symp;
2373 2472 const char *cweak_file;
2374 2473 boolean_t cweak_candidate;
2375 2474 ulong_t cweak_idx;
2376 2475 } ctf_dwarf_weak_arg_t;
2377 2476
2378 2477 static int
2379 2478 ctf_dwarf_conv_check_weak(const Elf64_Sym *symp, ulong_t idx, const char *file,
2380 2479 const char *name, boolean_t primary, void *arg)
2381 2480 {
2382 2481 ctf_dwarf_weak_arg_t *cweak = arg;
2383 2482
2384 2483 const Elf64_Sym *wsymp = cweak->cweak_symp;
2385 2484
2386 2485 ctf_dprintf("comparing weak to %s\n", name);
2387 2486
2388 2487 if (GELF_ST_BIND(symp->st_info) == STB_WEAK) {
2389 2488 return (0);
2390 2489 }
2391 2490
2392 2491 if (GELF_ST_TYPE(wsymp->st_info) != GELF_ST_TYPE(symp->st_info)) {
2393 2492 return (0);
2394 2493 }
2395 2494
2396 2495 if (wsymp->st_value != symp->st_value) {
2397 2496 return (0);
2398 2497 }
2399 2498
2400 2499 if (wsymp->st_size != symp->st_size) {
2401 2500 return (0);
2402 2501 }
2403 2502
2404 2503 if (wsymp->st_shndx != symp->st_shndx) {
2405 2504 return (0);
2406 2505 }
2407 2506
2408 2507 /*
2409 2508 * Check if it's a weak candidate.
2410 2509 */
2411 2510 if (GELF_ST_BIND(symp->st_info) == STB_LOCAL &&
2412 2511 (file == NULL || cweak->cweak_file == NULL ||
2413 2512 strcmp(file, cweak->cweak_file) != 0)) {
2414 2513 cweak->cweak_candidate = B_TRUE;
2415 2514 cweak->cweak_idx = idx;
2416 2515 return (0);
2417 2516 }
2418 2517
2419 2518 /*
2420 2519 * Found a match, break.
2421 2520 */
2422 2521 cweak->cweak_idx = idx;
2423 2522 return (1);
2424 2523 }
2425 2524
2426 2525 static int
2427 2526 ctf_dwarf_duplicate_sym(ctf_cu_t *cup, ulong_t idx, ulong_t matchidx)
2428 2527 {
2429 2528 ctf_id_t id = ctf_lookup_by_symbol(cup->cu_ctfp, matchidx);
2430 2529
2431 2530 /*
2432 2531 * If we matched something that for some reason didn't have type data,
2433 2532 * we don't consider that a fatal error and silently swallow it.
2434 2533 */
2435 2534 if (id == CTF_ERR) {
2436 2535 if (ctf_errno(cup->cu_ctfp) == ECTF_NOTYPEDAT)
2437 2536 return (0);
2438 2537 else
2439 2538 return (ctf_errno(cup->cu_ctfp));
2440 2539 }
2441 2540
2442 2541 if (ctf_add_object(cup->cu_ctfp, idx, id) == CTF_ERR)
2443 2542 return (ctf_errno(cup->cu_ctfp));
2444 2543
2445 2544 return (0);
2446 2545 }
2447 2546
2448 2547 static int
2449 2548 ctf_dwarf_duplicate_func(ctf_cu_t *cup, ulong_t idx, ulong_t matchidx)
2450 2549 {
2451 2550 int ret;
2452 2551 ctf_funcinfo_t fip;
2453 2552 ctf_id_t *args = NULL;
2454 2553
2455 2554 if (ctf_func_info(cup->cu_ctfp, matchidx, &fip) == CTF_ERR) {
2456 2555 if (ctf_errno(cup->cu_ctfp) == ECTF_NOFUNCDAT)
2457 2556 return (0);
2458 2557 else
2459 2558 return (ctf_errno(cup->cu_ctfp));
2460 2559 }
2461 2560
2462 2561 if (fip.ctc_argc != 0) {
2463 2562 args = ctf_alloc(sizeof (ctf_id_t) * fip.ctc_argc);
2464 2563 if (args == NULL)
2465 2564 return (ENOMEM);
2466 2565
2467 2566 if (ctf_func_args(cup->cu_ctfp, matchidx, fip.ctc_argc, args) ==
2468 2567 CTF_ERR) {
2469 2568 ctf_free(args, sizeof (ctf_id_t) * fip.ctc_argc);
2470 2569 return (ctf_errno(cup->cu_ctfp));
2471 2570 }
2472 2571 }
2473 2572
2474 2573 ret = ctf_add_function(cup->cu_ctfp, idx, &fip, args);
2475 2574 if (args != NULL)
2476 2575 ctf_free(args, sizeof (ctf_id_t) * fip.ctc_argc);
2477 2576 if (ret == CTF_ERR)
2478 2577 return (ctf_errno(cup->cu_ctfp));
2479 2578
2480 2579 return (0);
2481 2580 }
2482 2581
2483 2582 static int
2484 2583 ctf_dwarf_conv_weaks_cb(const Elf64_Sym *symp, ulong_t idx, const char *file,
2485 2584 const char *name, boolean_t primary, void *arg)
2486 2585 {
2487 2586 int ret, type;
2488 2587 ctf_dwarf_weak_arg_t cweak;
2489 2588 ctf_cu_t *cup = arg;
2490 2589
2491 2590 /*
2492 2591 * We only care about weak symbols.
2493 2592 */
2494 2593 if (GELF_ST_BIND(symp->st_info) != STB_WEAK)
2495 2594 return (0);
2496 2595
2497 2596 type = GELF_ST_TYPE(symp->st_info);
2498 2597 ASSERT(type == STT_OBJECT || type == STT_FUNC);
2499 2598
2500 2599 /*
2501 2600 * For each weak symbol we encounter, we need to do a second iteration
2502 2601 * to try and find a match. We should probably think about other
2503 2602 * techniques to try and save us time in the future.
2504 2603 */
2505 2604 cweak.cweak_symp = symp;
2506 2605 cweak.cweak_file = file;
2507 2606 cweak.cweak_candidate = B_FALSE;
2508 2607 cweak.cweak_idx = 0;
2509 2608
2510 2609 ctf_dprintf("Trying to find weak equiv for %s\n", name);
2511 2610
2512 2611 ret = ctf_symtab_iter(cup->cu_ctfp, ctf_dwarf_conv_check_weak, &cweak);
2513 2612 VERIFY(ret == 0 || ret == 1);
2514 2613
2515 2614 /*
2516 2615 * Nothing was ever found, we're not going to add anything for this
2517 2616 * entry.
2518 2617 */
2519 2618 if (ret == 0 && cweak.cweak_candidate == B_FALSE) {
2520 2619 ctf_dprintf("found no weak match for %s\n", name);
2521 2620 return (0);
2522 2621 }
2523 2622
2524 2623 /*
2525 2624 * Now, finally go and add the type based on the match.
2526 2625 */
2527 2626 ctf_dprintf("matched weak symbol %lu to %lu\n", idx, cweak.cweak_idx);
2528 2627 if (type == STT_OBJECT) {
2529 2628 ret = ctf_dwarf_duplicate_sym(cup, idx, cweak.cweak_idx);
2530 2629 } else {
2531 2630 ret = ctf_dwarf_duplicate_func(cup, idx, cweak.cweak_idx);
2532 2631 }
2533 2632
2534 2633 return (ret);
2535 2634 }
2536 2635
2537 2636 static int
2538 2637 ctf_dwarf_conv_weaks(ctf_cu_t *cup)
2539 2638 {
2540 2639 return (ctf_symtab_iter(cup->cu_ctfp, ctf_dwarf_conv_weaks_cb, cup));
2541 2640 }
2542 2641
2543 2642 /* ARGSUSED */
2544 2643 static int
2545 2644 ctf_dwarf_convert_one(void *arg, void *unused)
2546 2645 {
2547 2646 int ret;
2548 2647 ctf_file_t *dedup;
2549 2648 ctf_cu_t *cup = arg;
2550 2649
2551 2650 ctf_dprintf("converting die: %s\n", cup->cu_name);
2552 2651 ctf_dprintf("max offset: %x\n", cup->cu_maxoff);
2553 2652 VERIFY(cup != NULL);
2554 2653
2555 2654 ret = ctf_dwarf_convert_die(cup, cup->cu_cu);
2556 2655 ctf_dprintf("ctf_dwarf_convert_die (%s) returned %d\n", cup->cu_name,
2557 2656 ret);
2558 2657 if (ret != 0) {
2559 2658 return (ret);
2560 2659 }
2561 2660 if (ctf_update(cup->cu_ctfp) != 0) {
2562 2661 return (ctf_dwarf_error(cup, cup->cu_ctfp, 0,
2563 2662 "failed to update output ctf container"));
2564 2663 }
2565 2664
2566 2665 ret = ctf_dwarf_fixup_die(cup, B_FALSE);
2567 2666 ctf_dprintf("ctf_dwarf_fixup_die (%s) returned %d\n", cup->cu_name,
2568 2667 ret);
2569 2668 if (ret != 0) {
2570 2669 return (ret);
2571 2670 }
2572 2671 if (ctf_update(cup->cu_ctfp) != 0) {
2573 2672 return (ctf_dwarf_error(cup, cup->cu_ctfp, 0,
2574 2673 "failed to update output ctf container"));
2575 2674 }
2576 2675
2577 2676 ret = ctf_dwarf_fixup_die(cup, B_TRUE);
2578 2677 ctf_dprintf("ctf_dwarf_fixup_die (%s) returned %d\n", cup->cu_name,
2579 2678 ret);
2580 2679 if (ret != 0) {
2581 2680 return (ret);
2582 2681 }
2583 2682 if (ctf_update(cup->cu_ctfp) != 0) {
2584 2683 return (ctf_dwarf_error(cup, cup->cu_ctfp, 0,
2585 2684 "failed to update output ctf container"));
2586 2685 }
2587 2686
2588 2687
2589 2688 if ((ret = ctf_dwarf_conv_funcvars(cup)) != 0) {
2590 2689 return (ctf_dwarf_error(cup, NULL, ret,
2591 2690 "failed to convert strong functions and variables"));
2592 2691 }
2593 2692
2594 2693 if (ctf_update(cup->cu_ctfp) != 0) {
2595 2694 return (ctf_dwarf_error(cup, cup->cu_ctfp, 0,
2596 2695 "failed to update output ctf container"));
2597 2696 }
2598 2697
2599 2698 if (cup->cu_doweaks == B_TRUE) {
2600 2699 if ((ret = ctf_dwarf_conv_weaks(cup)) != 0) {
2601 2700 return (ctf_dwarf_error(cup, NULL, ret,
2602 2701 "failed to convert weak functions and variables"));
2603 2702 }
2604 2703
2605 2704 if (ctf_update(cup->cu_ctfp) != 0) {
2606 2705 return (ctf_dwarf_error(cup, cup->cu_ctfp, 0,
2607 2706 "failed to update output ctf container"));
2608 2707 }
2609 2708 }
2610 2709
2611 2710 ctf_phase_dump(cup->cu_ctfp, "pre-dwarf-dedup", cup->cu_name);
2612 2711 ctf_dprintf("adding inputs for dedup\n");
2613 2712 if ((ret = ctf_merge_add(cup->cu_cmh, cup->cu_ctfp)) != 0) {
2614 2713 return (ctf_dwarf_error(cup, NULL, ret,
2615 2714 "failed to add inputs for merge"));
2616 2715 }
2617 2716
2618 2717 ctf_dprintf("starting dedup of %s\n", cup->cu_name);
2619 2718 if ((ret = ctf_merge_dedup(cup->cu_cmh, &dedup)) != 0) {
2620 2719 return (ctf_dwarf_error(cup, NULL, ret,
2621 2720 "failed to deduplicate die"));
2622 2721 }
2623 2722 ctf_close(cup->cu_ctfp);
2624 2723 cup->cu_ctfp = dedup;
2625 2724 ctf_phase_dump(cup->cu_ctfp, "post-dwarf-dedup", cup->cu_name);
2626 2725
2627 2726 return (0);
2628 2727 }
2629 2728
2630 2729 /*
2631 2730 * Note, we expect that if we're returning a ctf_file_t from one of the dies,
2632 2731 * say in the single node case, it's been saved and the entry here has been set
2633 2732 * to NULL, which ctf_close happily ignores.
2634 2733 */
2635 2734 static void
2636 2735 ctf_dwarf_free_die(ctf_cu_t *cup)
2637 2736 {
2638 2737 ctf_dwfunc_t *cdf, *ndf;
2639 2738 ctf_dwvar_t *cdv, *ndv;
2640 2739 ctf_dwbitf_t *cdb, *ndb;
2641 2740 ctf_dwmap_t *map;
2642 2741 void *cookie;
2643 2742 Dwarf_Error derr;
2644 2743
2645 2744 ctf_dprintf("Beginning to free die: %p\n", cup);
2646 2745 cup->cu_elf = NULL;
2647 2746 ctf_dprintf("Trying to free name: %p\n", cup->cu_name);
2648 2747 if (cup->cu_name != NULL)
2649 2748 ctf_free(cup->cu_name, strlen(cup->cu_name) + 1);
2650 2749 ctf_dprintf("Trying to free merge handle: %p\n", cup->cu_cmh);
2651 2750 if (cup->cu_cmh != NULL) {
2652 2751 ctf_merge_fini(cup->cu_cmh);
2653 2752 cup->cu_cmh = NULL;
2654 2753 }
2655 2754
2656 2755 ctf_dprintf("Trying to free functions\n");
2657 2756 for (cdf = ctf_list_next(&cup->cu_funcs); cdf != NULL; cdf = ndf) {
2658 2757 ndf = ctf_list_next(cdf);
2659 2758 ctf_free(cdf->cdf_name, strlen(cdf->cdf_name) + 1);
2660 2759 if (cdf->cdf_fip.ctc_argc != 0) {
2661 2760 ctf_free(cdf->cdf_argv,
2662 2761 sizeof (ctf_id_t) * cdf->cdf_fip.ctc_argc);
2663 2762 }
2664 2763 ctf_free(cdf, sizeof (ctf_dwfunc_t));
2665 2764 }
2666 2765
2667 2766 ctf_dprintf("Trying to free variables\n");
2668 2767 for (cdv = ctf_list_next(&cup->cu_vars); cdv != NULL; cdv = ndv) {
2669 2768 ndv = ctf_list_next(cdv);
2670 2769 ctf_free(cdv->cdv_name, strlen(cdv->cdv_name) + 1);
2671 2770 ctf_free(cdv, sizeof (ctf_dwvar_t));
2672 2771 }
2673 2772
2674 2773 ctf_dprintf("Trying to free bitfields\n");
2675 2774 for (cdb = ctf_list_next(&cup->cu_bitfields); cdb != NULL; cdb = ndb) {
2676 2775 ndb = ctf_list_next(cdb);
2677 2776 ctf_free(cdb, sizeof (ctf_dwbitf_t));
2678 2777 }
2679 2778
2680 2779 ctf_dprintf("Trying to clean up dwarf_t: %p\n", cup->cu_dwarf);
2681 2780 if (cup->cu_dwarf != NULL)
2682 2781 (void) dwarf_finish(cup->cu_dwarf, &derr);
2683 2782 cup->cu_dwarf = NULL;
2684 2783 ctf_close(cup->cu_ctfp);
2685 2784
2686 2785 cookie = NULL;
2687 2786 while ((map = avl_destroy_nodes(&cup->cu_map, &cookie)) != NULL) {
2688 2787 ctf_free(map, sizeof (ctf_dwmap_t));
2689 2788 }
2690 2789 avl_destroy(&cup->cu_map);
2691 2790 cup->cu_errbuf = NULL;
2692 2791 }
2693 2792
2694 2793 static void
2695 2794 ctf_dwarf_free_dies(ctf_cu_t *cdies, int ndies)
2696 2795 {
2697 2796 int i;
2698 2797
2699 2798 ctf_dprintf("Beginning to free dies\n");
2700 2799 for (i = 0; i < ndies; i++) {
2701 2800 ctf_dwarf_free_die(&cdies[i]);
2702 2801 }
2703 2802
2704 2803 ctf_free(cdies, sizeof (ctf_cu_t) * ndies);
2705 2804 }
2706 2805
2707 2806 static int
2708 2807 ctf_dwarf_count_dies(Dwarf_Debug dw, Dwarf_Error *derr, int *ndies,
2709 2808 char *errbuf, size_t errlen)
2710 2809 {
2711 2810 int ret;
2712 2811 Dwarf_Half vers;
2713 2812 Dwarf_Unsigned nexthdr;
2714 2813
2715 2814 while ((ret = dwarf_next_cu_header(dw, NULL, &vers, NULL, NULL,
2716 2815 &nexthdr, derr)) != DW_DLV_NO_ENTRY) {
2717 2816 if (ret != DW_DLV_OK) {
2718 2817 (void) snprintf(errbuf, errlen,
2719 2818 "file does not contain valid DWARF data: %s\n",
2720 2819 dwarf_errmsg(*derr));
2721 2820 return (ECTF_CONVBKERR);
2722 2821 }
2723 2822
2724 2823 if (vers != DWARF_VERSION_TWO) {
2725 2824 (void) snprintf(errbuf, errlen,
2726 2825 "unsupported DWARF version: %d\n", vers);
2727 2826 return (ECTF_CONVBKERR);
2728 2827 }
2729 2828 *ndies = *ndies + 1;
2730 2829 }
2731 2830
2732 2831 return (0);
2733 2832 }
2734 2833
2735 2834 static int
2736 2835 ctf_dwarf_init_die(int fd, Elf *elf, ctf_cu_t *cup, int ndie, char *errbuf,
2737 2836 size_t errlen)
2738 2837 {
2739 2838 int ret;
2740 2839 Dwarf_Unsigned hdrlen, abboff, nexthdr;
2741 2840 Dwarf_Half addrsz;
2742 2841 Dwarf_Unsigned offset = 0;
2743 2842 Dwarf_Error derr;
2744 2843
2745 2844 while ((ret = dwarf_next_cu_header(cup->cu_dwarf, &hdrlen, NULL,
2746 2845 &abboff, &addrsz, &nexthdr, &derr)) != DW_DLV_NO_ENTRY) {
2747 2846 char *name;
2748 2847 Dwarf_Die cu, child;
2749 2848
2750 2849 /* Based on the counting above, we should be good to go */
2751 2850 VERIFY(ret == DW_DLV_OK);
2752 2851 if (ndie > 0) {
2753 2852 ndie--;
2754 2853 offset = nexthdr;
2755 2854 continue;
2756 2855 }
2757 2856
2758 2857 /*
2759 2858 * Compilers are apparently inconsistent. Some emit no DWARF for
2760 2859 * empty files and others emit empty compilation unit.
2761 2860 */
2762 2861 cup->cu_voidtid = CTF_ERR;
2763 2862 cup->cu_longtid = CTF_ERR;
2764 2863 cup->cu_elf = elf;
2765 2864 cup->cu_maxoff = nexthdr - 1;
2766 2865 cup->cu_ctfp = ctf_fdcreate(fd, &ret);
2767 2866 if (cup->cu_ctfp == NULL)
2768 2867 return (ret);
2769 2868
2770 2869 avl_create(&cup->cu_map, ctf_dwmap_comp, sizeof (ctf_dwmap_t),
2771 2870 offsetof(ctf_dwmap_t, cdm_avl));
2772 2871 cup->cu_errbuf = errbuf;
2773 2872 cup->cu_errlen = errlen;
2774 2873 bzero(&cup->cu_vars, sizeof (ctf_list_t));
2775 2874 bzero(&cup->cu_funcs, sizeof (ctf_list_t));
2776 2875 bzero(&cup->cu_bitfields, sizeof (ctf_list_t));
2777 2876
2778 2877 if ((ret = ctf_dwarf_die_elfenc(elf, cup, errbuf,
2779 2878 errlen)) != 0)
2780 2879 return (ret);
2781 2880
2782 2881 if ((ret = ctf_dwarf_sib(cup, NULL, &cu)) != 0)
2783 2882 return (ret);
2784 2883
2785 2884 if (cu == NULL) {
2786 2885 (void) snprintf(errbuf, errlen,
2787 2886 "file does not contain DWARF data");
2788 2887 return (ECTF_CONVNODEBUG);
2789 2888 }
2790 2889
2791 2890 if ((ret = ctf_dwarf_child(cup, cu, &child)) != 0)
2792 2891 return (ret);
2793 2892
2794 2893 if (child == NULL) {
2795 2894 (void) snprintf(errbuf, errlen,
2796 2895 "file does not contain DWARF data");
2797 2896 return (ECTF_CONVNODEBUG);
2798 2897 }
2799 2898
2800 2899 cup->cu_cuoff = offset;
2801 2900 cup->cu_cu = child;
2802 2901
2803 2902 if ((cup->cu_cmh = ctf_merge_init(fd, &ret)) == NULL)
2804 2903 return (ret);
2805 2904
2806 2905 if (ctf_dwarf_string(cup, cu, DW_AT_name, &name) == 0) {
2807 2906 size_t len = strlen(name) + 1;
2808 2907 char *b = basename(name);
2809 2908 cup->cu_name = strdup(b);
2810 2909 ctf_free(name, len);
2811 2910 }
2812 2911 break;
2813 2912 }
2814 2913
2815 2914 return (0);
2816 2915 }
2817 2916
2818 2917 /*
2819 2918 * This is our only recourse to identify a C source file that is missing debug
2820 2919 * info: it will be mentioned as an STT_FILE, but not have a compile unit entry.
2821 2920 * (A traditional ctfmerge works on individual files, so can identify missing
2822 2921 * DWARF more directly, via ctf_has_c_source() on the .o file.)
2823 2922 *
2824 2923 * As we operate on basenames, this can of course miss some cases, but it's
2825 2924 * better than not checking at all.
2826 2925 *
2827 2926 * We explicitly whitelist some CRT components. Failing that, there's always
2828 2927 * the -m option.
2829 2928 */
2830 2929 static boolean_t
2831 2930 c_source_has_debug(const char *file, ctf_cu_t *cus, size_t nr_cus)
2832 2931 {
2833 2932 const char *basename = strrchr(file, '/');
2834 2933
2835 2934 if (basename == NULL)
2836 2935 basename = file;
2837 2936 else
2838 2937 basename++;
2839 2938
2840 2939 if (strcmp(basename, "common-crt.c") == 0 ||
2841 2940 strcmp(basename, "gmon.c") == 0 ||
2842 2941 strcmp(basename, "dlink_init.c") == 0 ||
2843 2942 strcmp(basename, "dlink_common.c") == 0 ||
2844 2943 strncmp(basename, "crt", strlen("crt")) == 0 ||
2845 2944 strncmp(basename, "values-", strlen("values-")) == 0)
2846 2945 return (B_TRUE);
2847 2946
2848 2947 for (size_t i = 0; i < nr_cus; i++) {
2849 2948 if (strcmp(basename, cus[i].cu_name) == 0)
2850 2949 return (B_TRUE);
2851 2950 }
2852 2951
2853 2952 return (B_FALSE);
2854 2953 }
2855 2954
2856 2955 static int
2857 2956 ctf_dwarf_check_missing(ctf_cu_t *cus, size_t nr_cus, Elf *elf,
2858 2957 char *errmsg, size_t errlen)
2859 2958 {
2860 2959 Elf_Scn *scn, *strscn;
2861 2960 Elf_Data *data, *strdata;
2862 2961 GElf_Shdr shdr;
2863 2962 ulong_t i;
2864 2963
2865 2964 scn = NULL;
2866 2965 while ((scn = elf_nextscn(elf, scn)) != NULL) {
2867 2966 if (gelf_getshdr(scn, &shdr) == NULL) {
2868 2967 (void) snprintf(errmsg, errlen,
2869 2968 "failed to get section header: %s\n",
2870 2969 elf_errmsg(elf_errno()));
2871 2970 return (EINVAL);
2872 2971 }
2873 2972
2874 2973 if (shdr.sh_type == SHT_SYMTAB)
2875 2974 break;
2876 2975 }
2877 2976
2878 2977 if (scn == NULL)
2879 2978 return (0);
2880 2979
2881 2980 if ((strscn = elf_getscn(elf, shdr.sh_link)) == NULL) {
2882 2981 (void) snprintf(errmsg, errlen,
2883 2982 "failed to get str section: %s\n",
2884 2983 elf_errmsg(elf_errno()));
2885 2984 return (EINVAL);
2886 2985 }
2887 2986
2888 2987 if ((data = elf_getdata(scn, NULL)) == NULL) {
2889 2988 (void) snprintf(errmsg, errlen, "failed to read section: %s\n",
2890 2989 elf_errmsg(elf_errno()));
2891 2990 return (EINVAL);
2892 2991 }
2893 2992
2894 2993 if ((strdata = elf_getdata(strscn, NULL)) == NULL) {
2895 2994 (void) snprintf(errmsg, errlen,
2896 2995 "failed to read string table: %s\n",
2897 2996 elf_errmsg(elf_errno()));
2898 2997 return (EINVAL);
2899 2998 }
2900 2999
2901 3000 for (i = 0; i < shdr.sh_size / shdr.sh_entsize; i++) {
2902 3001 GElf_Sym sym;
2903 3002 const char *file;
2904 3003 size_t len;
2905 3004
2906 3005 if (gelf_getsym(data, i, &sym) == NULL) {
2907 3006 (void) snprintf(errmsg, errlen,
2908 3007 "failed to read sym %lu: %s\n",
2909 3008 i, elf_errmsg(elf_errno()));
2910 3009 return (EINVAL);
2911 3010 }
2912 3011
2913 3012 if (GELF_ST_TYPE(sym.st_info) != STT_FILE)
2914 3013 continue;
2915 3014
2916 3015 file = (const char *)((uintptr_t)strdata->d_buf + sym.st_name);
2917 3016 len = strlen(file);
2918 3017 if (len < 2 || strncmp(".c", &file[len - 2], 2) != 0)
2919 3018 continue;
2920 3019
2921 3020 if (!c_source_has_debug(file, cus, nr_cus)) {
2922 3021 (void) snprintf(errmsg, errlen,
2923 3022 "file %s is missing debug info\n", file);
2924 3023 return (ECTF_CONVNODEBUG);
2925 3024 }
2926 3025 }
2927 3026
2928 3027 return (0);
2929 3028 }
2930 3029
2931 3030 int
2932 3031 ctf_dwarf_convert(int fd, Elf *elf, uint_t nthrs, uint_t flags,
2933 3032 ctf_file_t **fpp, char *errbuf, size_t errlen)
2934 3033 {
2935 3034 int err, ret, ndies, i;
2936 3035 Dwarf_Debug dw;
2937 3036 Dwarf_Error derr;
2938 3037 ctf_cu_t *cdies = NULL, *cup;
2939 3038 workq_t *wqp = NULL;
2940 3039
2941 3040 *fpp = NULL;
2942 3041
2943 3042 ret = dwarf_elf_init(elf, DW_DLC_READ, NULL, NULL, &dw, &derr);
2944 3043 if (ret != DW_DLV_OK) {
2945 3044 if (ret == DW_DLV_NO_ENTRY ||
2946 3045 dwarf_errno(derr) == DW_DLE_DEBUG_INFO_NULL) {
2947 3046 (void) snprintf(errbuf, errlen,
2948 3047 "file does not contain DWARF data\n");
2949 3048 return (ECTF_CONVNODEBUG);
2950 3049 }
2951 3050
2952 3051 (void) snprintf(errbuf, errlen,
2953 3052 "dwarf_elf_init() failed: %s\n", dwarf_errmsg(derr));
2954 3053 return (ECTF_CONVBKERR);
2955 3054 }
2956 3055
2957 3056 /*
2958 3057 * Iterate over all of the compilation units and create a ctf_cu_t for
2959 3058 * each of them. This is used to determine if we have zero, one, or
2960 3059 * multiple dies to convert. If we have zero, that's an error. If
2961 3060 * there's only one die, that's the simple case. No merge needed and
2962 3061 * only a single Dwarf_Debug as well.
2963 3062 */
2964 3063 ndies = 0;
2965 3064 err = ctf_dwarf_count_dies(dw, &derr, &ndies, errbuf, errlen);
2966 3065
2967 3066 ctf_dprintf("found %d DWARF CUs\n", ndies);
2968 3067
2969 3068 if (ndies == 0) {
2970 3069 (void) snprintf(errbuf, errlen,
2971 3070 "file does not contain DWARF data\n");
2972 3071 return (ECTF_CONVNODEBUG);
2973 3072 }
2974 3073
2975 3074 (void) dwarf_finish(dw, &derr);
2976 3075 cdies = ctf_alloc(sizeof (ctf_cu_t) * ndies);
2977 3076 if (cdies == NULL) {
2978 3077 return (ENOMEM);
2979 3078 }
2980 3079
2981 3080 bzero(cdies, sizeof (ctf_cu_t) * ndies);
2982 3081
2983 3082 for (i = 0; i < ndies; i++) {
2984 3083 cup = &cdies[i];
2985 3084 ret = dwarf_elf_init(elf, DW_DLC_READ, NULL, NULL,
2986 3085 &cup->cu_dwarf, &derr);
2987 3086 if (ret != 0) {
2988 3087 ctf_free(cdies, sizeof (ctf_cu_t) * ndies);
2989 3088 (void) snprintf(errbuf, errlen,
2990 3089 "failed to initialize DWARF: %s\n",
2991 3090 dwarf_errmsg(derr));
2992 3091 return (ECTF_CONVBKERR);
2993 3092 }
2994 3093
2995 3094 err = ctf_dwarf_init_die(fd, elf, cup, i, errbuf, errlen);
2996 3095 if (err != 0)
2997 3096 goto out;
2998 3097
2999 3098 cup->cu_doweaks = ndies > 1 ? B_FALSE : B_TRUE;
3000 3099 }
3001 3100
3002 3101 if (!(flags & CTF_ALLOW_MISSING_DEBUG) &&
3003 3102 (err = ctf_dwarf_check_missing(cdies, ndies,
3004 3103 elf, errbuf, errlen)) != 0)
3005 3104 goto out;
3006 3105
3007 3106 /*
3008 3107 * If we only have one compilation unit, there's no reason to use
3009 3108 * multiple threads, even if the user requested them. After all, they
3010 3109 * just gave us an upper bound.
3011 3110 */
3012 3111 if (ndies == 1)
3013 3112 nthrs = 1;
3014 3113
3015 3114 if (workq_init(&wqp, nthrs) == -1) {
3016 3115 err = errno;
3017 3116 goto out;
3018 3117 }
3019 3118
3020 3119 for (i = 0; i < ndies; i++) {
3021 3120 cup = &cdies[i];
3022 3121 ctf_dprintf("adding cu %s: %p, %x %x\n", cup->cu_name,
3023 3122 cup->cu_cu, cup->cu_cuoff, cup->cu_maxoff);
3024 3123 if (workq_add(wqp, cup) == -1) {
3025 3124 err = errno;
3026 3125 goto out;
3027 3126 }
3028 3127 }
3029 3128
3030 3129 ret = workq_work(wqp, ctf_dwarf_convert_one, NULL, &err);
3031 3130 if (ret == WORKQ_ERROR) {
3032 3131 err = errno;
3033 3132 goto out;
3034 3133 } else if (ret == WORKQ_UERROR) {
3035 3134 ctf_dprintf("internal convert failed: %s\n",
3036 3135 ctf_errmsg(err));
3037 3136 goto out;
3038 3137 }
3039 3138
3040 3139 ctf_dprintf("Determining next phase: have %d CUs\n", ndies);
3041 3140 if (ndies != 1) {
3042 3141 ctf_merge_t *cmp;
3043 3142
3044 3143 cmp = ctf_merge_init(fd, &err);
3045 3144 if (cmp == NULL)
3046 3145 goto out;
3047 3146
3048 3147 ctf_dprintf("setting threads\n");
3049 3148 if ((err = ctf_merge_set_nthreads(cmp, nthrs)) != 0) {
3050 3149 ctf_merge_fini(cmp);
3051 3150 goto out;
3052 3151 }
3053 3152
3054 3153 for (i = 0; i < ndies; i++) {
3055 3154 cup = &cdies[i];
3056 3155 if ((err = ctf_merge_add(cmp, cup->cu_ctfp)) != 0) {
3057 3156 ctf_merge_fini(cmp);
3058 3157 goto out;
3059 3158 }
3060 3159 }
3061 3160
3062 3161 ctf_dprintf("performing merge\n");
3063 3162 err = ctf_merge_merge(cmp, fpp);
3064 3163 if (err != 0) {
3065 3164 ctf_dprintf("failed merge!\n");
3066 3165 *fpp = NULL;
3067 3166 ctf_merge_fini(cmp);
3068 3167 goto out;
3069 3168 }
3070 3169 ctf_merge_fini(cmp);
3071 3170 err = 0;
3072 3171 ctf_dprintf("successfully converted!\n");
3073 3172 } else {
3074 3173 err = 0;
3075 3174 *fpp = cdies->cu_ctfp;
3076 3175 cdies->cu_ctfp = NULL;
3077 3176 ctf_dprintf("successfully converted!\n");
3078 3177 }
3079 3178
3080 3179 out:
3081 3180 workq_fini(wqp);
3082 3181 ctf_dwarf_free_dies(cdies, ndies);
3083 3182 return (err);
3084 3183 }
|
↓ open down ↓ |
1148 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX