1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23 * Use is subject to license terms.
24 */
25
26/*
27 * DWARF to tdata conversion
28 *
29 * For the most part, conversion is straightforward, proceeding in two passes.
30 * On the first pass, we iterate through every die, creating new type nodes as
31 * necessary.  Referenced tdesc_t's are created in an uninitialized state, thus
32 * allowing type reference pointers to be filled in.  If the tdesc_t
33 * corresponding to a given die can be completely filled out (sizes and offsets
34 * calculated, and so forth) without using any referenced types, the tdesc_t is
35 * marked as resolved.  Consider an array type.  If the type corresponding to
36 * the array contents has not yet been processed, we will create a blank tdesc
37 * for the contents type (only the type ID will be filled in, relying upon the
38 * later portion of the first pass to encounter and complete the referenced
39 * type).  We will then attempt to determine the size of the array.  If the
40 * array has a byte size attribute, we will have completely characterized the
41 * array type, and will be able to mark it as resolved.  The lack of a byte
42 * size attribute, on the other hand, will prevent us from fully resolving the
43 * type, as the size will only be calculable with reference to the contents
44 * type, which has not, as yet, been encountered.  The array type will thus be
45 * left without the resolved flag, and the first pass will continue.
46 *
47 * When we begin the second pass, we will have created tdesc_t nodes for every
48 * type in the section.  We will traverse the tree, from the iidescs down,
49 * processing each unresolved node.  As the referenced nodes will have been
50 * populated, the array type used in our example above will be able to use the
51 * size of the referenced types (if available) to determine its own type.  The
52 * traversal will be repeated until all types have been resolved or we have
53 * failed to make progress.  When all tdescs have been resolved, the conversion
54 * is complete.
55 *
56 * There are, as always, a few special cases that are handled during the first
57 * and second passes:
58 *
59 *  1. Empty enums - GCC will occasionally emit an enum without any members.
60 *     Later on in the file, it will emit the same enum type, though this time
61 *     with the full complement of members.  All references to the memberless
62 *     enum need to be redirected to the full definition.  During the first
63 *     pass, each enum is entered in dm_enumhash, along with a pointer to its
64 *     corresponding tdesc_t.  If, during the second pass, we encounter a
65 *     memberless enum, we use the hash to locate the full definition.  All
66 *     tdescs referencing the empty enum are then redirected.
67 *
68 *  2. Forward declarations - If the compiler sees a forward declaration for
69 *     a structure, followed by the definition of that structure, it will emit
70 *     DWARF data for both the forward declaration and the definition.  We need
71 *     to resolve the forward declarations when possible, by redirecting
72 *     forward-referencing tdescs to the actual struct/union definitions.  This
73 *     redirection is done completely within the first pass.  We begin by
74 *     recording all forward declarations in dw_fwdhash.  When we define a
75 *     structure, we check to see if there have been any corresponding forward
76 *     declarations.  If so, we redirect the tdescs which referenced the forward
77 *     declarations to the structure or union definition.
78 *
79 * XXX see if a post traverser will allow the elimination of repeated pass 2
80 * traversals.
81 */
82
83#include <stdio.h>
84#include <stdlib.h>
85#include <string.h>
86#include <strings.h>
87#include <errno.h>
88#include <libelf.h>
89#include <libdwarf.h>
90#include <libgen.h>
91#include <dwarf.h>
92
93#include "ctf_headers.h"
94#include "ctftools.h"
95#include "memory.h"
96#include "list.h"
97#include "traverse.h"
98
99/* The version of DWARF which we support. */
100#define	DWARF_VERSION	2
101
102/*
103 * We need to define a couple of our own intrinsics, to smooth out some of the
104 * differences between the GCC and DevPro DWARF emitters.  See the referenced
105 * routines and the special cases in the file comment for more details.
106 *
107 * Type IDs are 32 bits wide.  We're going to use the top of that field to
108 * indicate types that we've created ourselves.
109 */
110#define	TID_FILEMAX		0x3fffffff	/* highest tid from file */
111#define	TID_VOID		0x40000001	/* see die_void() */
112#define	TID_LONG		0x40000002	/* see die_array() */
113
114#define	TID_MFGTID_BASE		0x40000003	/* first mfg'd tid */
115
116/*
117 * To reduce the staggering amount of error-handling code that would otherwise
118 * be required, the attribute-retrieval routines handle most of their own
119 * errors.  If the following flag is supplied as the value of the `req'
120 * argument, they will also handle the absence of a requested attribute by
121 * terminating the program.
122 */
123#define	DW_ATTR_REQ	1
124
125#define	TDESC_HASH_BUCKETS	511
126
127typedef struct dwarf {
128	Dwarf_Debug dw_dw;		/* for libdwarf */
129	Dwarf_Error dw_err;		/* for libdwarf */
130	Dwarf_Off dw_maxoff;		/* highest legal offset in this cu */
131	tdata_t *dw_td;			/* root of the tdesc/iidesc tree */
132	hash_t *dw_tidhash;		/* hash of tdescs by t_id */
133	hash_t *dw_fwdhash;		/* hash of fwd decls by name */
134	hash_t *dw_enumhash;		/* hash of memberless enums by name */
135	tdesc_t *dw_void;		/* manufactured void type */
136	tdesc_t *dw_long;		/* manufactured long type for arrays */
137	size_t dw_ptrsz;		/* size of a pointer in this file */
138	tid_t dw_mfgtid_last;		/* last mfg'd type ID used */
139	uint_t dw_nunres;		/* count of unresolved types */
140	char *dw_cuname;		/* name of compilation unit */
141} dwarf_t;
142
143static void die_create_one(dwarf_t *, Dwarf_Die);
144static void die_create(dwarf_t *, Dwarf_Die);
145
146static tid_t
147mfgtid_next(dwarf_t *dw)
148{
149	return (++dw->dw_mfgtid_last);
150}
151
152static void
153tdesc_add(dwarf_t *dw, tdesc_t *tdp)
154{
155	hash_add(dw->dw_tidhash, tdp);
156}
157
158static tdesc_t *
159tdesc_lookup(dwarf_t *dw, int tid)
160{
161	tdesc_t tmpl;
162	void *tdp;
163
164	tmpl.t_id = tid;
165
166	if (hash_find(dw->dw_tidhash, &tmpl, &tdp))
167		return (tdp);
168	else
169		return (NULL);
170}
171
172/*
173 * Resolve a tdesc down to a node which should have a size.  Returns the size,
174 * zero if the size hasn't yet been determined.
175 */
176static size_t
177tdesc_size(tdesc_t *tdp)
178{
179	for (;;) {
180		switch (tdp->t_type) {
181		case INTRINSIC:
182		case POINTER:
183		case ARRAY:
184		case FUNCTION:
185		case STRUCT:
186		case UNION:
187		case ENUM:
188			return (tdp->t_size);
189
190		case FORWARD:
191			return (0);
192
193		case TYPEDEF:
194		case VOLATILE:
195		case CONST:
196		case RESTRICT:
197			tdp = tdp->t_tdesc;
198			continue;
199
200		case 0: /* not yet defined */
201			return (0);
202
203		default:
204			terminate("tdp %u: tdesc_size on unknown type %d\n",
205			    tdp->t_id, tdp->t_type);
206		}
207	}
208}
209
210static size_t
211tdesc_bitsize(tdesc_t *tdp)
212{
213	for (;;) {
214		switch (tdp->t_type) {
215		case INTRINSIC:
216			return (tdp->t_intr->intr_nbits);
217
218		case ARRAY:
219		case FUNCTION:
220		case STRUCT:
221		case UNION:
222		case ENUM:
223		case POINTER:
224			return (tdp->t_size * NBBY);
225
226		case FORWARD:
227			return (0);
228
229		case TYPEDEF:
230		case VOLATILE:
231		case RESTRICT:
232		case CONST:
233			tdp = tdp->t_tdesc;
234			continue;
235
236		case 0: /* not yet defined */
237			return (0);
238
239		default:
240			terminate("tdp %u: tdesc_bitsize on unknown type %d\n",
241			    tdp->t_id, tdp->t_type);
242		}
243	}
244}
245
246static tdesc_t *
247tdesc_basetype(tdesc_t *tdp)
248{
249	for (;;) {
250		switch (tdp->t_type) {
251		case TYPEDEF:
252		case VOLATILE:
253		case RESTRICT:
254		case CONST:
255			tdp = tdp->t_tdesc;
256			break;
257		case 0: /* not yet defined */
258			return (NULL);
259		default:
260			return (tdp);
261		}
262	}
263}
264
265static Dwarf_Off
266die_off(dwarf_t *dw, Dwarf_Die die)
267{
268	Dwarf_Off off;
269
270	if (dwarf_dieoffset(die, &off, &dw->dw_err) == DW_DLV_OK)
271		return (off);
272
273	terminate("failed to get offset for die: %s\n",
274	    dwarf_errmsg(&dw->dw_err));
275	/*NOTREACHED*/
276	return (0);
277}
278
279static Dwarf_Die
280die_sibling(dwarf_t *dw, Dwarf_Die die)
281{
282	Dwarf_Die sib;
283	int rc;
284
285	if ((rc = dwarf_siblingof(dw->dw_dw, die, &sib, &dw->dw_err)) ==
286	    DW_DLV_OK)
287		return (sib);
288	else if (rc == DW_DLV_NO_ENTRY)
289		return (NULL);
290
291	terminate("die %llu: failed to find type sibling: %s\n",
292	    die_off(dw, die), dwarf_errmsg(&dw->dw_err));
293	/*NOTREACHED*/
294	return (NULL);
295}
296
297static Dwarf_Die
298die_child(dwarf_t *dw, Dwarf_Die die)
299{
300	Dwarf_Die child;
301	int rc;
302
303	if ((rc = dwarf_child(die, &child, &dw->dw_err)) == DW_DLV_OK)
304		return (child);
305	else if (rc == DW_DLV_NO_ENTRY)
306		return (NULL);
307
308	terminate("die %llu: failed to find type child: %s\n",
309	    die_off(dw, die), dwarf_errmsg(&dw->dw_err));
310	/*NOTREACHED*/
311	return (NULL);
312}
313
314static Dwarf_Half
315die_tag(dwarf_t *dw, Dwarf_Die die)
316{
317	Dwarf_Half tag;
318
319	if (dwarf_tag(die, &tag, &dw->dw_err) == DW_DLV_OK)
320		return (tag);
321
322	terminate("die %llu: failed to get tag for type: %s\n",
323	    die_off(dw, die), dwarf_errmsg(&dw->dw_err));
324	/*NOTREACHED*/
325	return (0);
326}
327
328static Dwarf_Attribute
329die_attr(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name, int req)
330{
331	Dwarf_Attribute attr;
332	int rc;
333
334	if ((rc = dwarf_attr(die, name, &attr, &dw->dw_err)) == DW_DLV_OK) {
335		return (attr);
336	} else if (rc == DW_DLV_NO_ENTRY) {
337		if (req) {
338			terminate("die %llu: no attr 0x%x\n", die_off(dw, die),
339			    name);
340		} else {
341			return (NULL);
342		}
343	}
344
345	terminate("die %llu: failed to get attribute for type: %s\n",
346	    die_off(dw, die), dwarf_errmsg(&dw->dw_err));
347	/*NOTREACHED*/
348	return (NULL);
349}
350
351static int
352die_signed(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name, Dwarf_Signed *valp,
353    int req)
354{
355	*valp = 0;
356	if (dwarf_attrval_signed(die, name, valp, &dw->dw_err) != DWARF_E_NONE) {
357		if (req)
358			terminate("die %llu: failed to get signed: %s\n",
359			    die_off(dw, die), dwarf_errmsg(&dw->dw_err));
360		return (0);
361	}
362
363	return (1);
364}
365
366static int
367die_unsigned(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name, Dwarf_Unsigned *valp,
368    int req)
369{
370	*valp = 0;
371	if (dwarf_attrval_unsigned(die, name, valp, &dw->dw_err) != DWARF_E_NONE) {
372		if (req)
373			terminate("die %llu: failed to get unsigned: %s\n",
374			    die_off(dw, die), dwarf_errmsg(&dw->dw_err));
375		return (0);
376	}
377
378	return (1);
379}
380
381static int
382die_bool(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name, Dwarf_Bool *valp, int req)
383{
384	*valp = 0;
385
386	if (dwarf_attrval_flag(die, name, valp, &dw->dw_err) != DWARF_E_NONE) {
387		if (req)
388			terminate("die %llu: failed to get flag: %s\n",
389			    die_off(dw, die), dwarf_errmsg(&dw->dw_err));
390		return (0);
391	}
392
393	return (1);
394}
395
396static int
397die_string(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name, char **strp, int req)
398{
399	const char *str = NULL;
400
401	if (dwarf_attrval_string(die, name, &str, &dw->dw_err) != DWARF_E_NONE ||
402	    str == NULL) {
403		if (req)
404			terminate("die %llu: failed to get string: %s\n",
405			    die_off(dw, die), dwarf_errmsg(&dw->dw_err));
406		else
407			*strp = NULL;
408		return (0);
409	} else
410		*strp = xstrdup(str);
411
412	return (1);
413}
414
415static Dwarf_Off
416die_attr_ref(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name)
417{
418	Dwarf_Off off;
419
420	if (dwarf_attrval_unsigned(die, name, &off, &dw->dw_err) != DWARF_E_NONE) {
421		terminate("die %llu: failed to get ref: %s\n",
422		    die_off(dw, die), dwarf_errmsg(&dw->dw_err));
423	}
424
425	return (off);
426}
427
428static char *
429die_name(dwarf_t *dw, Dwarf_Die die)
430{
431	char *str = NULL;
432
433	(void) die_string(dw, die, DW_AT_name, &str, 0);
434
435	return (str);
436}
437
438static int
439die_isdecl(dwarf_t *dw, Dwarf_Die die)
440{
441	Dwarf_Bool val;
442
443	return (die_bool(dw, die, DW_AT_declaration, &val, 0) && val);
444}
445
446static int
447die_isglobal(dwarf_t *dw, Dwarf_Die die)
448{
449	Dwarf_Signed vis;
450	Dwarf_Bool ext;
451
452	/*
453	 * Some compilers (gcc) use DW_AT_external to indicate function
454	 * visibility.  Others (Sun) use DW_AT_visibility.
455	 */
456	if (die_signed(dw, die, DW_AT_visibility, &vis, 0))
457		return (vis == DW_VIS_exported);
458	else
459		return (die_bool(dw, die, DW_AT_external, &ext, 0) && ext);
460}
461
462static tdesc_t *
463die_add(dwarf_t *dw, Dwarf_Off off)
464{
465	tdesc_t *tdp = xcalloc(sizeof (tdesc_t));
466
467	tdp->t_id = off;
468
469	tdesc_add(dw, tdp);
470
471	return (tdp);
472}
473
474static tdesc_t *
475die_lookup_pass1(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name)
476{
477	Dwarf_Off ref = die_attr_ref(dw, die, name);
478	tdesc_t *tdp;
479
480	if ((tdp = tdesc_lookup(dw, ref)) != NULL)
481		return (tdp);
482
483	return (die_add(dw, ref));
484}
485
486static int
487die_mem_offset(dwarf_t *dw, Dwarf_Die die, Dwarf_Half name,
488    Dwarf_Unsigned *valp, int req __unused)
489{
490	Dwarf_Locdesc *loc = NULL;
491	Dwarf_Signed locnum = 0;
492
493	if (dwarf_locdesc(die, name, &loc, &locnum, &dw->dw_err) != DW_DLV_OK)
494		return (0);
495
496	if (locnum != 1 || loc->ld_s->lr_atom != DW_OP_plus_uconst) {
497		terminate("die %llu: cannot parse member offset\n",
498		    die_off(dw, die));
499	}
500
501	*valp = loc->ld_s->lr_number;
502
503	if (loc != NULL)
504		if (dwarf_locdesc_free(loc, &dw->dw_err) != DW_DLV_OK)
505			terminate("die %llu: cannot free location descriptor: %s\n",
506			    die_off(dw, die), dwarf_errmsg(&dw->dw_err));
507
508	return (1);
509}
510
511static tdesc_t *
512tdesc_intr_common(dwarf_t *dw, int tid, const char *name, size_t sz)
513{
514	tdesc_t *tdp;
515	intr_t *intr;
516
517	intr = xcalloc(sizeof (intr_t));
518	intr->intr_type = INTR_INT;
519	intr->intr_signed = 1;
520	intr->intr_nbits = sz * NBBY;
521
522	tdp = xcalloc(sizeof (tdesc_t));
523	tdp->t_name = xstrdup(name);
524	tdp->t_size = sz;
525	tdp->t_id = tid;
526	tdp->t_type = INTRINSIC;
527	tdp->t_intr = intr;
528	tdp->t_flags = TDESC_F_RESOLVED;
529
530	tdesc_add(dw, tdp);
531
532	return (tdp);
533}
534
535/*
536 * Manufacture a void type.  Used for gcc-emitted stabs, where the lack of a
537 * type reference implies a reference to a void type.  A void *, for example
538 * will be represented by a pointer die without a DW_AT_type.  CTF requires
539 * that pointer nodes point to something, so we'll create a void for use as
540 * the target.  Note that the DWARF data may already create a void type.  Ours
541 * would then be a duplicate, but it'll be removed in the self-uniquification
542 * merge performed at the completion of DWARF->tdesc conversion.
543 */
544static tdesc_t *
545tdesc_intr_void(dwarf_t *dw)
546{
547	if (dw->dw_void == NULL)
548		dw->dw_void = tdesc_intr_common(dw, TID_VOID, "void", 0);
549
550	return (dw->dw_void);
551}
552
553static tdesc_t *
554tdesc_intr_long(dwarf_t *dw)
555{
556	if (dw->dw_long == NULL) {
557		dw->dw_long = tdesc_intr_common(dw, TID_LONG, "long",
558		    dw->dw_ptrsz);
559	}
560
561	return (dw->dw_long);
562}
563
564/*
565 * Used for creating bitfield types.  We create a copy of an existing intrinsic,
566 * adjusting the size of the copy to match what the caller requested.  The
567 * caller can then use the copy as the type for a bitfield structure member.
568 */
569static tdesc_t *
570tdesc_intr_clone(dwarf_t *dw, tdesc_t *old, size_t bitsz)
571{
572	tdesc_t *new = xcalloc(sizeof (tdesc_t));
573
574	if (!(old->t_flags & TDESC_F_RESOLVED)) {
575		terminate("tdp %u: attempt to make a bit field from an "
576		    "unresolved type\n", old->t_id);
577	}
578
579	new->t_name = xstrdup(old->t_name);
580	new->t_size = old->t_size;
581	new->t_id = mfgtid_next(dw);
582	new->t_type = INTRINSIC;
583	new->t_flags = TDESC_F_RESOLVED;
584
585	new->t_intr = xcalloc(sizeof (intr_t));
586	bcopy(old->t_intr, new->t_intr, sizeof (intr_t));
587	new->t_intr->intr_nbits = bitsz;
588
589	tdesc_add(dw, new);
590
591	return (new);
592}
593
594static void
595tdesc_array_create(dwarf_t *dw, Dwarf_Die dim, tdesc_t *arrtdp,
596    tdesc_t *dimtdp)
597{
598	Dwarf_Unsigned uval;
599	Dwarf_Signed sval;
600	tdesc_t *ctdp = NULL;
601	Dwarf_Die dim2;
602	ardef_t *ar;
603
604	if ((dim2 = die_sibling(dw, dim)) == NULL) {
605		ctdp = arrtdp;
606	} else if (die_tag(dw, dim2) == DW_TAG_subrange_type) {
607		ctdp = xcalloc(sizeof (tdesc_t));
608		ctdp->t_id = mfgtid_next(dw);
609		debug(3, "die %llu: creating new type %u for sub-dimension\n",
610		    die_off(dw, dim2), ctdp->t_id);
611		tdesc_array_create(dw, dim2, arrtdp, ctdp);
612	} else {
613		terminate("die %llu: unexpected non-subrange node in array\n",
614		    die_off(dw, dim2));
615	}
616
617	dimtdp->t_type = ARRAY;
618	dimtdp->t_ardef = ar = xcalloc(sizeof (ardef_t));
619
620	/*
621	 * Array bounds can be signed or unsigned, but there are several kinds
622	 * of signless forms (data1, data2, etc) that take their sign from the
623	 * routine that is trying to interpret them.  That is, data1 can be
624	 * either signed or unsigned, depending on whether you use the signed or
625	 * unsigned accessor function.  GCC will use the signless forms to store
626	 * unsigned values which have their high bit set, so we need to try to
627	 * read them first as unsigned to get positive values.  We could also
628	 * try signed first, falling back to unsigned if we got a negative
629	 * value.
630	 */
631	if (die_unsigned(dw, dim, DW_AT_upper_bound, &uval, 0))
632		ar->ad_nelems = uval + 1;
633	else if (die_signed(dw, dim, DW_AT_upper_bound, &sval, 0))
634		ar->ad_nelems = sval + 1;
635	else
636		ar->ad_nelems = 0;
637
638	/*
639	 * Different compilers use different index types.  Force the type to be
640	 * a common, known value (long).
641	 */
642	ar->ad_idxtype = tdesc_intr_long(dw);
643	ar->ad_contents = ctdp;
644
645	if (ar->ad_contents->t_size != 0) {
646		dimtdp->t_size = ar->ad_contents->t_size * ar->ad_nelems;
647		dimtdp->t_flags |= TDESC_F_RESOLVED;
648	}
649}
650
651/*
652 * Create a tdesc from an array node.  Some arrays will come with byte size
653 * attributes, and thus can be resolved immediately.  Others don't, and will
654 * need to wait until the second pass for resolution.
655 */
656static void
657die_array_create(dwarf_t *dw, Dwarf_Die arr, Dwarf_Off off, tdesc_t *tdp)
658{
659	tdesc_t *arrtdp = die_lookup_pass1(dw, arr, DW_AT_type);
660	Dwarf_Unsigned uval;
661	Dwarf_Die dim;
662
663	debug(3, "die %llu <%llx>: creating array\n", off, off);
664
665	if ((dim = die_child(dw, arr)) == NULL ||
666	    die_tag(dw, dim) != DW_TAG_subrange_type)
667		terminate("die %llu: failed to retrieve array bounds\n", off);
668
669	tdesc_array_create(dw, dim, arrtdp, tdp);
670
671	if (die_unsigned(dw, arr, DW_AT_byte_size, &uval, 0)) {
672		tdesc_t *dimtdp;
673		int flags;
674
675		/* Check for bogus gcc DW_AT_byte_size attribute */
676		if (uval == (unsigned)-1) {
677			printf("dwarf.c:%s() working around bogus -1 DW_AT_byte_size\n",
678			    __func__);
679			uval = 0;
680		}
681
682		tdp->t_size = uval;
683
684		/*
685		 * Ensure that sub-dimensions have sizes too before marking
686		 * as resolved.
687		 */
688		flags = TDESC_F_RESOLVED;
689		for (dimtdp = tdp->t_ardef->ad_contents;
690		    dimtdp->t_type == ARRAY;
691		    dimtdp = dimtdp->t_ardef->ad_contents) {
692			if (!(dimtdp->t_flags & TDESC_F_RESOLVED)) {
693				flags = 0;
694				break;
695			}
696		}
697
698		tdp->t_flags |= flags;
699	}
700
701	debug(3, "die %llu <%llx>: array nelems %u size %u\n", off, off,
702	    tdp->t_ardef->ad_nelems, tdp->t_size);
703}
704
705/*ARGSUSED1*/
706static int
707die_array_resolve(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private)
708{
709	dwarf_t *dw = private;
710	size_t sz;
711
712	if (tdp->t_flags & TDESC_F_RESOLVED)
713		return (1);
714
715	debug(3, "trying to resolve array %d (cont %d)\n", tdp->t_id,
716	    tdp->t_ardef->ad_contents->t_id);
717
718	if ((sz = tdesc_size(tdp->t_ardef->ad_contents)) == 0) {
719		debug(3, "unable to resolve array %s (%d) contents %d\n",
720		    tdesc_name(tdp), tdp->t_id,
721		    tdp->t_ardef->ad_contents->t_id);
722
723		dw->dw_nunres++;
724		return (1);
725	}
726
727	tdp->t_size = sz * tdp->t_ardef->ad_nelems;
728	tdp->t_flags |= TDESC_F_RESOLVED;
729
730	debug(3, "resolved array %d: %u bytes\n", tdp->t_id, tdp->t_size);
731
732	return (1);
733}
734
735/*ARGSUSED1*/
736static int
737die_array_failed(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private __unused)
738{
739	tdesc_t *cont = tdp->t_ardef->ad_contents;
740
741	if (tdp->t_flags & TDESC_F_RESOLVED)
742		return (1);
743
744	fprintf(stderr, "Array %d: failed to size contents type %s (%d)\n",
745	    tdp->t_id, tdesc_name(cont), cont->t_id);
746
747	return (1);
748}
749
750/*
751 * Most enums (those with members) will be resolved during this first pass.
752 * Others - those without members (see the file comment) - won't be, and will
753 * need to wait until the second pass when they can be matched with their full
754 * definitions.
755 */
756static void
757die_enum_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp)
758{
759	Dwarf_Die mem;
760	Dwarf_Unsigned uval;
761	Dwarf_Signed sval;
762
763	debug(3, "die %llu: creating enum\n", off);
764
765	tdp->t_type = ENUM;
766
767	(void) die_unsigned(dw, die, DW_AT_byte_size, &uval, DW_ATTR_REQ);
768	/* Check for bogus gcc DW_AT_byte_size attribute */
769	if (uval == (unsigned)-1) {
770		printf("dwarf.c:%s() working around bogus -1 DW_AT_byte_size\n",
771		    __func__);
772		uval = 0;
773	}
774	tdp->t_size = uval;
775
776	if ((mem = die_child(dw, die)) != NULL) {
777		elist_t **elastp = &tdp->t_emem;
778
779		do {
780			elist_t *el;
781
782			if (die_tag(dw, mem) != DW_TAG_enumerator) {
783				/* Nested type declaration */
784				die_create_one(dw, mem);
785				continue;
786			}
787
788			el = xcalloc(sizeof (elist_t));
789			el->el_name = die_name(dw, mem);
790
791			if (die_signed(dw, mem, DW_AT_const_value, &sval, 0)) {
792				el->el_number = sval;
793			} else if (die_unsigned(dw, mem, DW_AT_const_value,
794			    &uval, 0)) {
795				el->el_number = uval;
796			} else {
797				terminate("die %llu: enum %llu: member without "
798				    "value\n", off, die_off(dw, mem));
799			}
800
801			debug(3, "die %llu: enum %llu: created %s = %d\n", off,
802			    die_off(dw, mem), el->el_name, el->el_number);
803
804			*elastp = el;
805			elastp = &el->el_next;
806
807		} while ((mem = die_sibling(dw, mem)) != NULL);
808
809		hash_add(dw->dw_enumhash, tdp);
810
811		tdp->t_flags |= TDESC_F_RESOLVED;
812
813		if (tdp->t_name != NULL) {
814			iidesc_t *ii = xcalloc(sizeof (iidesc_t));
815			ii->ii_type = II_SOU;
816			ii->ii_name = xstrdup(tdp->t_name);
817			ii->ii_dtype = tdp;
818
819			iidesc_add(dw->dw_td->td_iihash, ii);
820		}
821	}
822}
823
824static int
825die_enum_match(void *arg1, void *arg2)
826{
827	tdesc_t *tdp = arg1, **fullp = arg2;
828
829	if (tdp->t_emem != NULL) {
830		*fullp = tdp;
831		return (-1); /* stop the iteration */
832	}
833
834	return (0);
835}
836
837/*ARGSUSED1*/
838static int
839die_enum_resolve(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private)
840{
841	dwarf_t *dw = private;
842	tdesc_t *full = NULL;
843
844	if (tdp->t_flags & TDESC_F_RESOLVED)
845		return (1);
846
847	(void) hash_find_iter(dw->dw_enumhash, tdp, die_enum_match, &full);
848
849	/*
850	 * The answer to this one won't change from iteration to iteration,
851	 * so don't even try.
852	 */
853	if (full == NULL) {
854		terminate("tdp %u: enum %s has no members\n", tdp->t_id,
855		    tdesc_name(tdp));
856	}
857
858	debug(3, "tdp %u: enum %s redirected to %u\n", tdp->t_id,
859	    tdesc_name(tdp), full->t_id);
860
861	tdp->t_flags |= TDESC_F_RESOLVED;
862
863	return (1);
864}
865
866static int
867die_fwd_map(void *arg1, void *arg2)
868{
869	tdesc_t *fwd = arg1, *sou = arg2;
870
871	debug(3, "tdp %u: mapped forward %s to sou %u\n", fwd->t_id,
872	    tdesc_name(fwd), sou->t_id);
873	fwd->t_tdesc = sou;
874
875	return (0);
876}
877
878/*
879 * Structures and unions will never be resolved during the first pass, as we
880 * won't be able to fully determine the member sizes.  The second pass, which
881 * have access to sizing information, will be able to complete the resolution.
882 */
883static void
884die_sou_create(dwarf_t *dw, Dwarf_Die str, Dwarf_Off off, tdesc_t *tdp,
885    int type, const char *typename)
886{
887	Dwarf_Unsigned sz, bitsz, bitoff, maxsz=0;
888	Dwarf_Die mem;
889	mlist_t *ml, **mlastp;
890	iidesc_t *ii;
891
892	tdp->t_type = (die_isdecl(dw, str) ? FORWARD : type);
893
894	debug(3, "die %llu: creating %s %s\n", off,
895	    (tdp->t_type == FORWARD ? "forward decl" : typename),
896	    tdesc_name(tdp));
897
898	if (tdp->t_type == FORWARD) {
899		hash_add(dw->dw_fwdhash, tdp);
900		return;
901	}
902
903	(void) hash_find_iter(dw->dw_fwdhash, tdp, die_fwd_map, tdp);
904
905	(void) die_unsigned(dw, str, DW_AT_byte_size, &sz, DW_ATTR_REQ);
906	tdp->t_size = sz;
907
908	/*
909	 * GCC allows empty SOUs as an extension.
910	 */
911	if ((mem = die_child(dw, str)) == NULL) {
912		goto out;
913	}
914
915	mlastp = &tdp->t_members;
916
917	do {
918		Dwarf_Off memoff = die_off(dw, mem);
919		Dwarf_Half tag = die_tag(dw, mem);
920		Dwarf_Unsigned mloff;
921
922		if (tag != DW_TAG_member) {
923			/* Nested type declaration */
924			die_create_one(dw, mem);
925			continue;
926		}
927
928		debug(3, "die %llu: mem %llu: creating member\n", off, memoff);
929
930		ml = xcalloc(sizeof (mlist_t));
931
932		/*
933		 * This could be a GCC anon struct/union member, so we'll allow
934		 * an empty name, even though nothing can really handle them
935		 * properly.  Note that some versions of GCC miss out debug
936		 * info for anon structs, though recent versions are fixed (gcc
937		 * bug 11816).
938		 */
939		if ((ml->ml_name = die_name(dw, mem)) == NULL)
940			ml->ml_name = NULL;
941
942		ml->ml_type = die_lookup_pass1(dw, mem, DW_AT_type);
943		debug(3, "die_sou_create(): ml_type = %p t_id = %d\n",
944		    ml->ml_type, ml->ml_type->t_id);
945
946		if (die_mem_offset(dw, mem, DW_AT_data_member_location,
947		    &mloff, 0)) {
948			debug(3, "die %llu: got mloff %llx\n", off,
949			    (u_longlong_t)mloff);
950			ml->ml_offset = mloff * 8;
951		}
952
953		if (die_unsigned(dw, mem, DW_AT_bit_size, &bitsz, 0))
954			ml->ml_size = bitsz;
955		else
956			ml->ml_size = tdesc_bitsize(ml->ml_type);
957
958		if (die_unsigned(dw, mem, DW_AT_bit_offset, &bitoff, 0)) {
959#if BYTE_ORDER == _BIG_ENDIAN
960			ml->ml_offset += bitoff;
961#else
962			ml->ml_offset += tdesc_bitsize(ml->ml_type) - bitoff -
963			    ml->ml_size;
964#endif
965		}
966
967		debug(3, "die %llu: mem %llu: created \"%s\" (off %u sz %u)\n",
968		    off, memoff, ml->ml_name, ml->ml_offset, ml->ml_size);
969
970		*mlastp = ml;
971		mlastp = &ml->ml_next;
972
973		/* Find the size of the largest member to work around a gcc
974		 * bug.  See GCC Bugzilla 35998.
975		 */
976		if (maxsz < ml->ml_size)
977			maxsz = ml->ml_size;
978
979	} while ((mem = die_sibling(dw, mem)) != NULL);
980
981	/* See if we got a bogus DW_AT_byte_size.  GCC will sometimes
982	 * emit this.
983	 */
984	if (sz == (unsigned)-1) {
985		 printf("dwarf.c:%s() working around bogus -1 DW_AT_byte_size\n",
986		     __func__);
987		 tdp->t_size = maxsz / 8;  /* maxsz is in bits, t_size is bytes */
988	}
989
990	/*
991	 * GCC will attempt to eliminate unused types, thus decreasing the
992	 * size of the emitted dwarf.  That is, if you declare a foo_t in your
993	 * header, include said header in your source file, and neglect to
994	 * actually use (directly or indirectly) the foo_t in the source file,
995	 * the foo_t won't make it into the emitted DWARF.  So, at least, goes
996	 * the theory.
997	 *
998	 * Occasionally, it'll emit the DW_TAG_structure_type for the foo_t,
999	 * and then neglect to emit the members.  Strangely, the loner struct
1000	 * tag will always be followed by a proper nested declaration of
1001	 * something else.  This is clearly a bug, but we're not going to have
1002	 * time to get it fixed before this goo goes back, so we'll have to work
1003	 * around it.  If we see a no-membered struct with a nested declaration
1004	 * (i.e. die_child of the struct tag won't be null), we'll ignore it.
1005	 * Being paranoid, we won't simply remove it from the hash.  Instead,
1006	 * we'll decline to create an iidesc for it, thus ensuring that this
1007	 * type won't make it into the output file.  To be safe, we'll also
1008	 * change the name.
1009	 */
1010	if (tdp->t_members == NULL) {
1011		const char *old = tdesc_name(tdp);
1012		size_t newsz = 7 + strlen(old) + 1;
1013		char *new = xmalloc(newsz);
1014		(void) snprintf(new, newsz, "orphan %s", old);
1015
1016		debug(3, "die %llu: worked around %s %s\n", off, typename, old);
1017
1018		if (tdp->t_name != NULL)
1019			free(tdp->t_name);
1020		tdp->t_name = new;
1021		return;
1022	}
1023
1024out:
1025	if (tdp->t_name != NULL) {
1026		ii = xcalloc(sizeof (iidesc_t));
1027		ii->ii_type = II_SOU;
1028		ii->ii_name = xstrdup(tdp->t_name);
1029		ii->ii_dtype = tdp;
1030
1031		iidesc_add(dw->dw_td->td_iihash, ii);
1032	}
1033}
1034
1035static void
1036die_struct_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp)
1037{
1038	die_sou_create(dw, die, off, tdp, STRUCT, "struct");
1039}
1040
1041static void
1042die_union_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp)
1043{
1044	die_sou_create(dw, die, off, tdp, UNION, "union");
1045}
1046
1047/*ARGSUSED1*/
1048static int
1049die_sou_resolve(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private)
1050{
1051	dwarf_t *dw = private;
1052	mlist_t *ml;
1053	tdesc_t *mt;
1054
1055	if (tdp->t_flags & TDESC_F_RESOLVED)
1056		return (1);
1057
1058	debug(3, "resolving sou %s\n", tdesc_name(tdp));
1059
1060	for (ml = tdp->t_members; ml != NULL; ml = ml->ml_next) {
1061		if (ml->ml_size == 0) {
1062			mt = tdesc_basetype(ml->ml_type);
1063
1064			if ((ml->ml_size = tdesc_bitsize(mt)) != 0)
1065				continue;
1066
1067			/*
1068			 * For empty members, or GCC/C99 flexible array
1069			 * members, a size of 0 is correct.
1070			 */
1071			if (mt->t_members == NULL)
1072				continue;
1073			if (mt->t_type == ARRAY && mt->t_ardef->ad_nelems == 0)
1074				continue;
1075
1076			dw->dw_nunres++;
1077			return (1);
1078		}
1079
1080		if ((mt = tdesc_basetype(ml->ml_type)) == NULL) {
1081			dw->dw_nunres++;
1082			return (1);
1083		}
1084
1085		if (ml->ml_size != 0 && mt->t_type == INTRINSIC &&
1086		    mt->t_intr->intr_nbits != (int)ml->ml_size) {
1087			/*
1088			 * This member is a bitfield, and needs to reference
1089			 * an intrinsic type with the same width.  If the
1090			 * currently-referenced type isn't of the same width,
1091			 * we'll copy it, adjusting the width of the copy to
1092			 * the size we'd like.
1093			 */
1094			debug(3, "tdp %u: creating bitfield for %d bits\n",
1095			    tdp->t_id, ml->ml_size);
1096
1097			ml->ml_type = tdesc_intr_clone(dw, mt, ml->ml_size);
1098		}
1099	}
1100
1101	tdp->t_flags |= TDESC_F_RESOLVED;
1102
1103	return (1);
1104}
1105
1106/*ARGSUSED1*/
1107static int
1108die_sou_failed(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private __unused)
1109{
1110	const char *typename = (tdp->t_type == STRUCT ? "struct" : "union");
1111	mlist_t *ml;
1112
1113	if (tdp->t_flags & TDESC_F_RESOLVED)
1114		return (1);
1115
1116	for (ml = tdp->t_members; ml != NULL; ml = ml->ml_next) {
1117		if (ml->ml_size == 0) {
1118			fprintf(stderr, "%s %d <%x>: failed to size member \"%s\" "
1119			    "of type %s (%d <%x>)\n", typename, tdp->t_id,
1120			    tdp->t_id,
1121			    ml->ml_name, tdesc_name(ml->ml_type),
1122			    ml->ml_type->t_id, ml->ml_type->t_id);
1123		}
1124	}
1125
1126	return (1);
1127}
1128
1129static void
1130die_funcptr_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp)
1131{
1132	Dwarf_Attribute attr;
1133	Dwarf_Half tag;
1134	Dwarf_Die arg;
1135	fndef_t *fn;
1136	int i;
1137
1138	debug(3, "die %llu <%llx>: creating function pointer\n", off, off);
1139
1140	/*
1141	 * We'll begin by processing any type definition nodes that may be
1142	 * lurking underneath this one.
1143	 */
1144	for (arg = die_child(dw, die); arg != NULL;
1145	    arg = die_sibling(dw, arg)) {
1146		if ((tag = die_tag(dw, arg)) != DW_TAG_formal_parameter &&
1147		    tag != DW_TAG_unspecified_parameters) {
1148			/* Nested type declaration */
1149			die_create_one(dw, arg);
1150		}
1151	}
1152
1153	if (die_isdecl(dw, die)) {
1154		/*
1155		 * This is a prototype.  We don't add prototypes to the
1156		 * tree, so we're going to drop the tdesc.  Unfortunately,
1157		 * it has already been added to the tree.  Nobody will reference
1158		 * it, though, and it will be leaked.
1159		 */
1160		return;
1161	}
1162
1163	fn = xcalloc(sizeof (fndef_t));
1164
1165	tdp->t_type = FUNCTION;
1166
1167	if ((attr = die_attr(dw, die, DW_AT_type, 0)) != NULL) {
1168		fn->fn_ret = die_lookup_pass1(dw, die, DW_AT_type);
1169	} else {
1170		fn->fn_ret = tdesc_intr_void(dw);
1171	}
1172
1173	/*
1174	 * Count the arguments to the function, then read them in.
1175	 */
1176	for (fn->fn_nargs = 0, arg = die_child(dw, die); arg != NULL;
1177	    arg = die_sibling(dw, arg)) {
1178		if ((tag = die_tag(dw, arg)) == DW_TAG_formal_parameter)
1179			fn->fn_nargs++;
1180		else if (tag == DW_TAG_unspecified_parameters &&
1181		    fn->fn_nargs > 0)
1182			fn->fn_vargs = 1;
1183	}
1184
1185	if (fn->fn_nargs != 0) {
1186		debug(3, "die %llu: adding %d argument%s\n", off, fn->fn_nargs,
1187		    (fn->fn_nargs > 1 ? "s" : ""));
1188
1189		fn->fn_args = xcalloc(sizeof (tdesc_t *) * fn->fn_nargs);
1190		for (i = 0, arg = die_child(dw, die);
1191		    arg != NULL && i < (int) fn->fn_nargs;
1192		    arg = die_sibling(dw, arg)) {
1193			if (die_tag(dw, arg) != DW_TAG_formal_parameter)
1194				continue;
1195
1196			fn->fn_args[i++] = die_lookup_pass1(dw, arg,
1197			    DW_AT_type);
1198		}
1199	}
1200
1201	tdp->t_fndef = fn;
1202	tdp->t_flags |= TDESC_F_RESOLVED;
1203}
1204
1205/*
1206 * GCC and DevPro use different names for the base types.  While the terms are
1207 * the same, they are arranged in a different order.  Some terms, such as int,
1208 * are implied in one, and explicitly named in the other.  Given a base type
1209 * as input, this routine will return a common name, along with an intr_t
1210 * that reflects said name.
1211 */
1212static intr_t *
1213die_base_name_parse(const char *name, char **newp)
1214{
1215	char buf[100];
1216	char const *base;
1217	char *c;
1218	int nlong = 0, nshort = 0, nchar = 0, nint = 0;
1219	int sign = 1;
1220	char fmt = '\0';
1221	intr_t *intr;
1222
1223	if (strlen(name) > sizeof (buf) - 1)
1224		terminate("base type name \"%s\" is too long\n", name);
1225
1226	strncpy(buf, name, sizeof (buf));
1227
1228	for (c = strtok(buf, " "); c != NULL; c = strtok(NULL, " ")) {
1229		if (strcmp(c, "signed") == 0)
1230			sign = 1;
1231		else if (strcmp(c, "unsigned") == 0)
1232			sign = 0;
1233		else if (strcmp(c, "long") == 0)
1234			nlong++;
1235		else if (strcmp(c, "char") == 0) {
1236			nchar++;
1237			fmt = 'c';
1238		} else if (strcmp(c, "short") == 0)
1239			nshort++;
1240		else if (strcmp(c, "int") == 0)
1241			nint++;
1242		else {
1243			/*
1244			 * If we don't recognize any of the tokens, we'll tell
1245			 * the caller to fall back to the dwarf-provided
1246			 * encoding information.
1247			 */
1248			return (NULL);
1249		}
1250	}
1251
1252	if (nchar > 1 || nshort > 1 || nint > 1 || nlong > 2)
1253		return (NULL);
1254
1255	if (nchar > 0) {
1256		if (nlong > 0 || nshort > 0 || nint > 0)
1257			return (NULL);
1258
1259		base = "char";
1260
1261	} else if (nshort > 0) {
1262		if (nlong > 0)
1263			return (NULL);
1264
1265		base = "short";
1266
1267	} else if (nlong > 0) {
1268		base = "long";
1269
1270	} else {
1271		base = "int";
1272	}
1273
1274	intr = xcalloc(sizeof (intr_t));
1275	intr->intr_type = INTR_INT;
1276	intr->intr_signed = sign;
1277	intr->intr_iformat = fmt;
1278
1279	snprintf(buf, sizeof (buf), "%s%s%s",
1280	    (sign ? "" : "unsigned "),
1281	    (nlong > 1 ? "long " : ""),
1282	    base);
1283
1284	*newp = xstrdup(buf);
1285	return (intr);
1286}
1287
1288typedef struct fp_size_map {
1289	size_t fsm_typesz[2];	/* size of {32,64} type */
1290	uint_t fsm_enc[3];	/* CTF_FP_* for {bare,cplx,imagry} type */
1291} fp_size_map_t;
1292
1293static const fp_size_map_t fp_encodings[] = {
1294	{ { 4, 4 }, { CTF_FP_SINGLE, CTF_FP_CPLX, CTF_FP_IMAGRY } },
1295	{ { 8, 8 }, { CTF_FP_DOUBLE, CTF_FP_DCPLX, CTF_FP_DIMAGRY } },
1296#ifdef __sparc
1297	{ { 16, 16 }, { CTF_FP_LDOUBLE, CTF_FP_LDCPLX, CTF_FP_LDIMAGRY } },
1298#else
1299	{ { 12, 16 }, { CTF_FP_LDOUBLE, CTF_FP_LDCPLX, CTF_FP_LDIMAGRY } },
1300#endif
1301	{ { 0, 0 }, { 0, 0, 0 } }
1302};
1303
1304static uint_t
1305die_base_type2enc(dwarf_t *dw, Dwarf_Off off, Dwarf_Signed enc, size_t sz)
1306{
1307	const fp_size_map_t *map = fp_encodings;
1308	uint_t szidx = dw->dw_ptrsz == sizeof (uint64_t);
1309	uint_t mult = 1, col = 0;
1310
1311	if (enc == DW_ATE_complex_float) {
1312		mult = 2;
1313		col = 1;
1314	} else if (enc == DW_ATE_imaginary_float
1315#if defined(sun)
1316	    || enc == DW_ATE_SUN_imaginary_float
1317#endif
1318	    )
1319		col = 2;
1320
1321	while (map->fsm_typesz[szidx] != 0) {
1322		if (map->fsm_typesz[szidx] * mult == sz)
1323			return (map->fsm_enc[col]);
1324		map++;
1325	}
1326
1327	terminate("die %llu: unrecognized real type size %u\n", off, sz);
1328	/*NOTREACHED*/
1329	return (0);
1330}
1331
1332static intr_t *
1333die_base_from_dwarf(dwarf_t *dw, Dwarf_Die base, Dwarf_Off off, size_t sz)
1334{
1335	intr_t *intr = xcalloc(sizeof (intr_t));
1336	Dwarf_Signed enc;
1337
1338	(void) die_signed(dw, base, DW_AT_encoding, &enc, DW_ATTR_REQ);
1339
1340	switch (enc) {
1341	case DW_ATE_unsigned:
1342	case DW_ATE_address:
1343		intr->intr_type = INTR_INT;
1344		break;
1345	case DW_ATE_unsigned_char:
1346		intr->intr_type = INTR_INT;
1347		intr->intr_iformat = 'c';
1348		break;
1349	case DW_ATE_signed:
1350		intr->intr_type = INTR_INT;
1351		intr->intr_signed = 1;
1352		break;
1353	case DW_ATE_signed_char:
1354		intr->intr_type = INTR_INT;
1355		intr->intr_signed = 1;
1356		intr->intr_iformat = 'c';
1357		break;
1358	case DW_ATE_boolean:
1359		intr->intr_type = INTR_INT;
1360		intr->intr_signed = 1;
1361		intr->intr_iformat = 'b';
1362		break;
1363	case DW_ATE_float:
1364	case DW_ATE_complex_float:
1365	case DW_ATE_imaginary_float:
1366#if defined(sun)
1367	case DW_ATE_SUN_imaginary_float:
1368	case DW_ATE_SUN_interval_float:
1369#endif
1370		intr->intr_type = INTR_REAL;
1371		intr->intr_signed = 1;
1372		intr->intr_fformat = die_base_type2enc(dw, off, enc, sz);
1373		break;
1374	default:
1375		terminate("die %llu: unknown base type encoding 0x%llx\n",
1376		    off, enc);
1377	}
1378
1379	return (intr);
1380}
1381
1382static void
1383die_base_create(dwarf_t *dw, Dwarf_Die base, Dwarf_Off off, tdesc_t *tdp)
1384{
1385	Dwarf_Unsigned sz;
1386	intr_t *intr;
1387	char *new;
1388
1389	debug(3, "die %llu: creating base type\n", off);
1390
1391	/*
1392	 * The compilers have their own clever (internally inconsistent) ideas
1393	 * as to what base types should look like.  Some times gcc will, for
1394	 * example, use DW_ATE_signed_char for char.  Other times, however, it
1395	 * will use DW_ATE_signed.  Needless to say, this causes some problems
1396	 * down the road, particularly with merging.  We do, however, use the
1397	 * DWARF idea of type sizes, as this allows us to avoid caring about
1398	 * the data model.
1399	 */
1400	(void) die_unsigned(dw, base, DW_AT_byte_size, &sz, DW_ATTR_REQ);
1401
1402	/* Check for bogus gcc DW_AT_byte_size attribute */
1403	if (sz == (unsigned)-1) {
1404		printf("dwarf.c:%s() working around bogus -1 DW_AT_byte_size\n",
1405		    __func__);
1406		sz = 0;
1407	}
1408
1409	if (tdp->t_name == NULL)
1410		terminate("die %llu: base type without name\n", off);
1411
1412	/* XXX make a name parser for float too */
1413	if ((intr = die_base_name_parse(tdp->t_name, &new)) != NULL) {
1414		/* Found it.  We'll use the parsed version */
1415		debug(3, "die %llu: name \"%s\" remapped to \"%s\"\n", off,
1416		    tdesc_name(tdp), new);
1417
1418		free(tdp->t_name);
1419		tdp->t_name = new;
1420	} else {
1421		/*
1422		 * We didn't recognize the type, so we'll create an intr_t
1423		 * based on the DWARF data.
1424		 */
1425		debug(3, "die %llu: using dwarf data for base \"%s\"\n", off,
1426		    tdesc_name(tdp));
1427
1428		intr = die_base_from_dwarf(dw, base, off, sz);
1429	}
1430
1431	intr->intr_nbits = sz * 8;
1432
1433	tdp->t_type = INTRINSIC;
1434	tdp->t_intr = intr;
1435	tdp->t_size = sz;
1436
1437	tdp->t_flags |= TDESC_F_RESOLVED;
1438}
1439
1440static void
1441die_through_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp,
1442    int type, const char *typename)
1443{
1444	Dwarf_Attribute attr;
1445
1446	debug(3, "die %llu <%llx>: creating %s type %d\n", off, off, typename, type);
1447
1448	tdp->t_type = type;
1449
1450	if ((attr = die_attr(dw, die, DW_AT_type, 0)) != NULL) {
1451		tdp->t_tdesc = die_lookup_pass1(dw, die, DW_AT_type);
1452	} else {
1453		tdp->t_tdesc = tdesc_intr_void(dw);
1454	}
1455
1456	if (type == POINTER)
1457		tdp->t_size = dw->dw_ptrsz;
1458
1459	tdp->t_flags |= TDESC_F_RESOLVED;
1460
1461	if (type == TYPEDEF) {
1462		iidesc_t *ii = xcalloc(sizeof (iidesc_t));
1463		ii->ii_type = II_TYPE;
1464		ii->ii_name = xstrdup(tdp->t_name);
1465		ii->ii_dtype = tdp;
1466
1467		iidesc_add(dw->dw_td->td_iihash, ii);
1468	}
1469}
1470
1471static void
1472die_typedef_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp)
1473{
1474	die_through_create(dw, die, off, tdp, TYPEDEF, "typedef");
1475}
1476
1477static void
1478die_const_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp)
1479{
1480	die_through_create(dw, die, off, tdp, CONST, "const");
1481}
1482
1483static void
1484die_pointer_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp)
1485{
1486	die_through_create(dw, die, off, tdp, POINTER, "pointer");
1487}
1488
1489static void
1490die_restrict_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp)
1491{
1492	die_through_create(dw, die, off, tdp, RESTRICT, "restrict");
1493}
1494
1495static void
1496die_volatile_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp)
1497{
1498	die_through_create(dw, die, off, tdp, VOLATILE, "volatile");
1499}
1500
1501/*ARGSUSED3*/
1502static void
1503die_function_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp __unused)
1504{
1505	Dwarf_Die arg;
1506	Dwarf_Half tag;
1507	iidesc_t *ii;
1508	char *name;
1509
1510	debug(3, "die %llu <%llx>: creating function definition\n", off, off);
1511
1512	/*
1513	 * We'll begin by processing any type definition nodes that may be
1514	 * lurking underneath this one.
1515	 */
1516	for (arg = die_child(dw, die); arg != NULL;
1517	    arg = die_sibling(dw, arg)) {
1518		if ((tag = die_tag(dw, arg)) != DW_TAG_formal_parameter &&
1519		    tag != DW_TAG_variable) {
1520			/* Nested type declaration */
1521			die_create_one(dw, arg);
1522		}
1523	}
1524
1525	if (die_isdecl(dw, die) || (name = die_name(dw, die)) == NULL) {
1526		/*
1527		 * We process neither prototypes nor subprograms without
1528		 * names.
1529		 */
1530		return;
1531	}
1532
1533	ii = xcalloc(sizeof (iidesc_t));
1534	ii->ii_type = die_isglobal(dw, die) ? II_GFUN : II_SFUN;
1535	ii->ii_name = name;
1536	if (ii->ii_type == II_SFUN)
1537		ii->ii_owner = xstrdup(dw->dw_cuname);
1538
1539	debug(3, "die %llu: function %s is %s\n", off, ii->ii_name,
1540	    (ii->ii_type == II_GFUN ? "global" : "static"));
1541
1542	if (die_attr(dw, die, DW_AT_type, 0) != NULL)
1543		ii->ii_dtype = die_lookup_pass1(dw, die, DW_AT_type);
1544	else
1545		ii->ii_dtype = tdesc_intr_void(dw);
1546
1547	for (arg = die_child(dw, die); arg != NULL;
1548	    arg = die_sibling(dw, arg)) {
1549		char *name1;
1550
1551		debug(3, "die %llu: looking at sub member at %llu\n",
1552		    off, die_off(dw, die));
1553
1554		if (die_tag(dw, arg) != DW_TAG_formal_parameter)
1555			continue;
1556
1557		if ((name1 = die_name(dw, arg)) == NULL) {
1558			terminate("die %llu: func arg %d has no name\n",
1559			    off, ii->ii_nargs + 1);
1560		}
1561
1562		if (strcmp(name1, "...") == 0) {
1563			free(name1);
1564			ii->ii_vargs = 1;
1565			continue;
1566		}
1567
1568		ii->ii_nargs++;
1569	}
1570
1571	if (ii->ii_nargs > 0) {
1572		int i;
1573
1574		debug(3, "die %llu: function has %d argument%s\n", off,
1575		    ii->ii_nargs, (ii->ii_nargs == 1 ? "" : "s"));
1576
1577		ii->ii_args = xcalloc(sizeof (tdesc_t) * ii->ii_nargs);
1578
1579		for (arg = die_child(dw, die), i = 0;
1580		    arg != NULL && i < ii->ii_nargs;
1581		    arg = die_sibling(dw, arg)) {
1582			if (die_tag(dw, arg) != DW_TAG_formal_parameter)
1583				continue;
1584
1585			ii->ii_args[i++] = die_lookup_pass1(dw, arg,
1586			    DW_AT_type);
1587		}
1588	}
1589
1590	iidesc_add(dw->dw_td->td_iihash, ii);
1591}
1592
1593/*ARGSUSED3*/
1594static void
1595die_variable_create(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off, tdesc_t *tdp __unused)
1596{
1597	iidesc_t *ii;
1598	char *name;
1599
1600	debug(3, "die %llu: creating object definition\n", off);
1601
1602	if (die_isdecl(dw, die) || (name = die_name(dw, die)) == NULL)
1603		return; /* skip prototypes and nameless objects */
1604
1605	ii = xcalloc(sizeof (iidesc_t));
1606	ii->ii_type = die_isglobal(dw, die) ? II_GVAR : II_SVAR;
1607	ii->ii_name = name;
1608	ii->ii_dtype = die_lookup_pass1(dw, die, DW_AT_type);
1609	if (ii->ii_type == II_SVAR)
1610		ii->ii_owner = xstrdup(dw->dw_cuname);
1611
1612	iidesc_add(dw->dw_td->td_iihash, ii);
1613}
1614
1615/*ARGSUSED2*/
1616static int
1617die_fwd_resolve(tdesc_t *fwd, tdesc_t **fwdp, void *private __unused)
1618{
1619	if (fwd->t_flags & TDESC_F_RESOLVED)
1620		return (1);
1621
1622	if (fwd->t_tdesc != NULL) {
1623		debug(3, "tdp %u: unforwarded %s\n", fwd->t_id,
1624		    tdesc_name(fwd));
1625		*fwdp = fwd->t_tdesc;
1626	}
1627
1628	fwd->t_flags |= TDESC_F_RESOLVED;
1629
1630	return (1);
1631}
1632
1633/*ARGSUSED*/
1634static void
1635die_lexblk_descend(dwarf_t *dw, Dwarf_Die die, Dwarf_Off off __unused, tdesc_t *tdp __unused)
1636{
1637	Dwarf_Die child = die_child(dw, die);
1638
1639	if (child != NULL)
1640		die_create(dw, child);
1641}
1642
1643/*
1644 * Used to map the die to a routine which can parse it, using the tag to do the
1645 * mapping.  While the processing of most tags entails the creation of a tdesc,
1646 * there are a few which don't - primarily those which result in the creation of
1647 * iidescs which refer to existing tdescs.
1648 */
1649
1650#define	DW_F_NOTDP	0x1	/* Don't create a tdesc for the creator */
1651
1652typedef struct die_creator {
1653	Dwarf_Half dc_tag;
1654	uint16_t dc_flags;
1655	void (*dc_create)(dwarf_t *, Dwarf_Die, Dwarf_Off, tdesc_t *);
1656} die_creator_t;
1657
1658static const die_creator_t die_creators[] = {
1659	{ DW_TAG_array_type,		0,		die_array_create },
1660	{ DW_TAG_enumeration_type,	0,		die_enum_create },
1661	{ DW_TAG_lexical_block,		DW_F_NOTDP,	die_lexblk_descend },
1662	{ DW_TAG_pointer_type,		0,		die_pointer_create },
1663	{ DW_TAG_structure_type,	0,		die_struct_create },
1664	{ DW_TAG_subroutine_type,	0,		die_funcptr_create },
1665	{ DW_TAG_typedef,		0,		die_typedef_create },
1666	{ DW_TAG_union_type,		0,		die_union_create },
1667	{ DW_TAG_base_type,		0,		die_base_create },
1668	{ DW_TAG_const_type,		0,		die_const_create },
1669	{ DW_TAG_subprogram,		DW_F_NOTDP,	die_function_create },
1670	{ DW_TAG_variable,		DW_F_NOTDP,	die_variable_create },
1671	{ DW_TAG_volatile_type,		0,		die_volatile_create },
1672	{ DW_TAG_restrict_type,		0,		die_restrict_create },
1673	{ 0, 0, NULL }
1674};
1675
1676static const die_creator_t *
1677die_tag2ctor(Dwarf_Half tag)
1678{
1679	const die_creator_t *dc;
1680
1681	for (dc = die_creators; dc->dc_create != NULL; dc++) {
1682		if (dc->dc_tag == tag)
1683			return (dc);
1684	}
1685
1686	return (NULL);
1687}
1688
1689static void
1690die_create_one(dwarf_t *dw, Dwarf_Die die)
1691{
1692	Dwarf_Off off = die_off(dw, die);
1693	const die_creator_t *dc;
1694	Dwarf_Half tag;
1695	tdesc_t *tdp;
1696
1697	debug(3, "die %llu <%llx>: create_one\n", off, off);
1698
1699	if (off > dw->dw_maxoff) {
1700		terminate("illegal die offset %llu (max %llu)\n", off,
1701		    dw->dw_maxoff);
1702	}
1703
1704	tag = die_tag(dw, die);
1705
1706	if ((dc = die_tag2ctor(tag)) == NULL) {
1707		debug(2, "die %llu: ignoring tag type %x\n", off, tag);
1708		return;
1709	}
1710
1711	if ((tdp = tdesc_lookup(dw, off)) == NULL &&
1712	    !(dc->dc_flags & DW_F_NOTDP)) {
1713		tdp = xcalloc(sizeof (tdesc_t));
1714		tdp->t_id = off;
1715		tdesc_add(dw, tdp);
1716	}
1717
1718	if (tdp != NULL)
1719		tdp->t_name = die_name(dw, die);
1720
1721	dc->dc_create(dw, die, off, tdp);
1722}
1723
1724static void
1725die_create(dwarf_t *dw, Dwarf_Die die)
1726{
1727	do {
1728		die_create_one(dw, die);
1729	} while ((die = die_sibling(dw, die)) != NULL);
1730}
1731
1732static tdtrav_cb_f die_resolvers[] = {
1733	NULL,
1734	NULL,			/* intrinsic */
1735	NULL,			/* pointer */
1736	die_array_resolve,	/* array */
1737	NULL,			/* function */
1738	die_sou_resolve,	/* struct */
1739	die_sou_resolve,	/* union */
1740	die_enum_resolve,	/* enum */
1741	die_fwd_resolve,	/* forward */
1742	NULL,			/* typedef */
1743	NULL,			/* typedef unres */
1744	NULL,			/* volatile */
1745	NULL,			/* const */
1746	NULL,			/* restrict */
1747};
1748
1749static tdtrav_cb_f die_fail_reporters[] = {
1750	NULL,
1751	NULL,			/* intrinsic */
1752	NULL,			/* pointer */
1753	die_array_failed,	/* array */
1754	NULL,			/* function */
1755	die_sou_failed,		/* struct */
1756	die_sou_failed,		/* union */
1757	NULL,			/* enum */
1758	NULL,			/* forward */
1759	NULL,			/* typedef */
1760	NULL,			/* typedef unres */
1761	NULL,			/* volatile */
1762	NULL,			/* const */
1763	NULL,			/* restrict */
1764};
1765
1766static void
1767die_resolve(dwarf_t *dw)
1768{
1769	int last = -1;
1770	int pass = 0;
1771
1772	do {
1773		pass++;
1774		dw->dw_nunres = 0;
1775
1776		(void) iitraverse_hash(dw->dw_td->td_iihash,
1777		    &dw->dw_td->td_curvgen, NULL, NULL, die_resolvers, dw);
1778
1779		debug(3, "resolve: pass %d, %u left\n", pass, dw->dw_nunres);
1780
1781		if ((int) dw->dw_nunres == last) {
1782			fprintf(stderr, "%s: failed to resolve the following "
1783			    "types:\n", progname);
1784
1785			(void) iitraverse_hash(dw->dw_td->td_iihash,
1786			    &dw->dw_td->td_curvgen, NULL, NULL,
1787			    die_fail_reporters, dw);
1788
1789			terminate("failed to resolve types\n");
1790		}
1791
1792		last = dw->dw_nunres;
1793
1794	} while (dw->dw_nunres != 0);
1795}
1796
1797/*
1798 * Any object containing a function or object symbol at any scope should also
1799 * contain DWARF data.
1800 */
1801static boolean_t
1802should_have_dwarf(Elf *elf)
1803{
1804	Elf_Scn *scn = NULL;
1805	Elf_Data *data = NULL;
1806	GElf_Shdr shdr;
1807	GElf_Sym sym;
1808	uint32_t symdx = 0;
1809	size_t nsyms = 0;
1810	boolean_t found = B_FALSE;
1811
1812	while ((scn = elf_nextscn(elf, scn)) != NULL) {
1813		gelf_getshdr(scn, &shdr);
1814
1815		if (shdr.sh_type == SHT_SYMTAB) {
1816			found = B_TRUE;
1817			break;
1818		}
1819	}
1820
1821	if (!found)
1822		terminate("cannot convert stripped objects\n");
1823
1824	data = elf_getdata(scn, NULL);
1825	nsyms = shdr.sh_size / shdr.sh_entsize;
1826
1827	for (symdx = 0; symdx < nsyms; symdx++) {
1828		gelf_getsym(data, symdx, &sym);
1829
1830		if ((GELF_ST_TYPE(sym.st_info) == STT_FUNC) ||
1831		    (GELF_ST_TYPE(sym.st_info) == STT_TLS) ||
1832		    (GELF_ST_TYPE(sym.st_info) == STT_OBJECT)) {
1833			char *name;
1834
1835			name = elf_strptr(elf, shdr.sh_link, sym.st_name);
1836
1837			/* Studio emits these local symbols regardless */
1838			if ((strcmp(name, "Bbss.bss") != 0) &&
1839			    (strcmp(name, "Ttbss.bss") != 0) &&
1840			    (strcmp(name, "Ddata.data") != 0) &&
1841			    (strcmp(name, "Ttdata.data") != 0) &&
1842			    (strcmp(name, "Drodata.rodata") != 0))
1843				return (B_TRUE);
1844		}
1845	}
1846
1847	return (B_FALSE);
1848}
1849
1850/*ARGSUSED*/
1851int
1852dw_read(tdata_t *td, Elf *elf, char *filename __unused)
1853{
1854	Dwarf_Unsigned abboff, hdrlen, nxthdr;
1855	Dwarf_Half vers, addrsz;
1856	Dwarf_Die cu = 0;
1857	Dwarf_Die child = 0;
1858	dwarf_t dw;
1859	char *prod = NULL;
1860	int rc;
1861
1862	bzero(&dw, sizeof (dwarf_t));
1863	dw.dw_td = td;
1864	dw.dw_ptrsz = elf_ptrsz(elf);
1865	dw.dw_mfgtid_last = TID_MFGTID_BASE;
1866	dw.dw_tidhash = hash_new(TDESC_HASH_BUCKETS, tdesc_idhash, tdesc_idcmp);
1867	dw.dw_fwdhash = hash_new(TDESC_HASH_BUCKETS, tdesc_namehash,
1868	    tdesc_namecmp);
1869	dw.dw_enumhash = hash_new(TDESC_HASH_BUCKETS, tdesc_namehash,
1870	    tdesc_namecmp);
1871
1872	if ((rc = dwarf_elf_init(elf, DW_DLC_READ, &dw.dw_dw,
1873	    &dw.dw_err)) == DW_DLV_NO_ENTRY) {
1874		if (should_have_dwarf(elf)) {
1875			errno = ENOENT;
1876			return (-1);
1877		} else {
1878			return (0);
1879		}
1880	} else if (rc != DW_DLV_OK) {
1881		if (dwarf_errno(&dw.dw_err) == DW_DLE_DEBUG_INFO_NULL) {
1882			/*
1883			 * There's no type data in the DWARF section, but
1884			 * libdwarf is too clever to handle that properly.
1885			 */
1886			return (0);
1887		}
1888
1889		terminate("failed to initialize DWARF: %s\n",
1890		    dwarf_errmsg(&dw.dw_err));
1891	}
1892
1893	if ((rc = dwarf_next_cu_header(dw.dw_dw, &hdrlen, &vers, &abboff,
1894	    &addrsz, &nxthdr, &dw.dw_err)) != DW_DLV_OK)
1895		terminate("rc = %d %s\n", rc, dwarf_errmsg(&dw.dw_err));
1896
1897	if ((cu = die_sibling(&dw, NULL)) == NULL ||
1898	    (((child = die_child(&dw, cu)) == NULL) &&
1899	    should_have_dwarf(elf))) {
1900		terminate("file does not contain dwarf type data "
1901		    "(try compiling with -g)\n");
1902	} else if (child == NULL) {
1903		return (0);
1904	}
1905
1906	dw.dw_maxoff = nxthdr - 1;
1907
1908	if (dw.dw_maxoff > TID_FILEMAX)
1909		terminate("file contains too many types\n");
1910
1911	debug(1, "DWARF version: %d\n", vers);
1912	if (vers != DWARF_VERSION) {
1913		terminate("file contains incompatible version %d DWARF code "
1914		    "(version 2 required)\n", vers);
1915	}
1916
1917	if (die_string(&dw, cu, DW_AT_producer, &prod, 0)) {
1918		debug(1, "DWARF emitter: %s\n", prod);
1919		free(prod);
1920	}
1921
1922	if ((dw.dw_cuname = die_name(&dw, cu)) != NULL) {
1923		char *base = xstrdup(basename(dw.dw_cuname));
1924		free(dw.dw_cuname);
1925		dw.dw_cuname = base;
1926
1927		debug(1, "CU name: %s\n", dw.dw_cuname);
1928	}
1929
1930	if ((child = die_child(&dw, cu)) != NULL)
1931		die_create(&dw, child);
1932
1933	if ((rc = dwarf_next_cu_header(dw.dw_dw, &hdrlen, &vers, &abboff,
1934	    &addrsz, &nxthdr, &dw.dw_err)) != DW_DLV_NO_ENTRY)
1935		terminate("multiple compilation units not supported\n");
1936
1937	(void) dwarf_finish(&dw.dw_dw, &dw.dw_err);
1938
1939	die_resolve(&dw);
1940
1941	cvt_fixups(td, dw.dw_ptrsz);
1942
1943	/* leak the dwarf_t */
1944
1945	return (0);
1946}
1947