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