]> git.ipfire.org Git - thirdparty/binutils-gdb.git/blob - bfd/dwarf2.c
Re: PR29785, memory bloat after b43771b045fb
[thirdparty/binutils-gdb.git] / bfd / dwarf2.c
1 /* DWARF 2 support.
2 Copyright (C) 1994-2024 Free Software Foundation, Inc.
3
4 Adapted from gdb/dwarf2read.c by Gavin Koch of Cygnus Solutions
5 (gavin@cygnus.com).
6
7 From the dwarf2read.c header:
8 Adapted by Gary Funck (gary@intrepid.com), Intrepid Technology,
9 Inc. with support from Florida State University (under contract
10 with the Ada Joint Program Office), and Silicon Graphics, Inc.
11 Initial contribution by Brent Benson, Harris Computer Systems, Inc.,
12 based on Fred Fish's (Cygnus Support) implementation of DWARF 1
13 support in dwarfread.c
14
15 This file is part of BFD.
16
17 This program is free software; you can redistribute it and/or modify
18 it under the terms of the GNU General Public License as published by
19 the Free Software Foundation; either version 3 of the License, or (at
20 your option) any later version.
21
22 This program is distributed in the hope that it will be useful, but
23 WITHOUT ANY WARRANTY; without even the implied warranty of
24 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
25 General Public License for more details.
26
27 You should have received a copy of the GNU General Public License
28 along with this program; if not, write to the Free Software
29 Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston,
30 MA 02110-1301, USA. */
31
32 #include "sysdep.h"
33 #include "bfd.h"
34 #include "libiberty.h"
35 #include "demangle.h"
36 #include "libbfd.h"
37 #include "elf-bfd.h"
38 #include "dwarf2.h"
39 #include "hashtab.h"
40 #include "splay-tree.h"
41
42 /* The data in the .debug_line statement prologue looks like this. */
43
44 struct line_head
45 {
46 bfd_vma total_length;
47 unsigned short version;
48 bfd_vma prologue_length;
49 unsigned char minimum_instruction_length;
50 unsigned char maximum_ops_per_insn;
51 unsigned char default_is_stmt;
52 int line_base;
53 unsigned char line_range;
54 unsigned char opcode_base;
55 unsigned char *standard_opcode_lengths;
56 };
57
58 /* Attributes have a name and a value. */
59
60 struct attribute
61 {
62 enum dwarf_attribute name;
63 enum dwarf_form form;
64 union
65 {
66 char *str;
67 struct dwarf_block *blk;
68 uint64_t val;
69 int64_t sval;
70 }
71 u;
72 };
73
74 /* Blocks are a bunch of untyped bytes. */
75 struct dwarf_block
76 {
77 unsigned int size;
78 bfd_byte *data;
79 };
80
81 struct adjusted_section
82 {
83 asection *section;
84 bfd_vma adj_vma;
85 bfd_vma orig_vma;
86 };
87
88 /* A trie to map quickly from address range to compilation unit.
89
90 This is a fairly standard radix-256 trie, used to quickly locate which
91 compilation unit any given address belongs to. Given that each compilation
92 unit may register hundreds of very small and unaligned ranges (which may
93 potentially overlap, due to inlining and other concerns), and a large
94 program may end up containing hundreds of thousands of such ranges, we cannot
95 scan through them linearly without undue slowdown.
96
97 We use a hybrid trie to avoid memory explosion: There are two types of trie
98 nodes, leaves and interior nodes. (Almost all nodes are leaves, so they
99 take up the bulk of the memory usage.) Leaves contain a simple array of
100 ranges (high/low address) and which compilation unit contains those ranges,
101 and when we get to a leaf, we scan through it linearly. Interior nodes
102 contain pointers to 256 other nodes, keyed by the next byte of the address.
103 So for a 64-bit address like 0x1234567abcd, we would start at the root and go
104 down child[0x00]->child[0x00]->child[0x01]->child[0x23]->child[0x45] etc.,
105 until we hit a leaf. (Nodes are, in general, leaves until they exceed the
106 default allocation of 16 elements, at which point they are converted to
107 interior node if possible.) This gives us near-constant lookup times;
108 the only thing that can be costly is if there are lots of overlapping ranges
109 within a single 256-byte segment of the binary, in which case we have to
110 scan through them all to find the best match.
111
112 For a binary with few ranges, we will in practice only have a single leaf
113 node at the root, containing a simple array. Thus, the scheme is efficient
114 for both small and large binaries.
115 */
116
117 /* Experiments have shown 16 to be a memory-efficient default leaf size.
118 The only case where a leaf will hold more memory than this, is at the
119 bottomost level (covering 256 bytes in the binary), where we'll expand
120 the leaf to be able to hold more ranges if needed.
121 */
122 #define TRIE_LEAF_SIZE 16
123
124 /* All trie_node pointers will really be trie_leaf or trie_interior,
125 but they have this common head. */
126 struct trie_node
127 {
128 /* If zero, we are an interior node.
129 Otherwise, how many ranges we have room for in this leaf. */
130 unsigned int num_room_in_leaf;
131 };
132
133 struct trie_leaf
134 {
135 struct trie_node head;
136 unsigned int num_stored_in_leaf;
137 struct {
138 struct comp_unit *unit;
139 bfd_vma low_pc, high_pc;
140 } ranges[];
141 };
142
143 struct trie_interior
144 {
145 struct trie_node head;
146 struct trie_node *children[256];
147 };
148
149 static struct trie_node *alloc_trie_leaf (bfd *abfd)
150 {
151 struct trie_leaf *leaf;
152 size_t amt = sizeof (*leaf) + TRIE_LEAF_SIZE * sizeof (leaf->ranges[0]);
153 leaf = bfd_zalloc (abfd, amt);
154 if (leaf == NULL)
155 return NULL;
156 leaf->head.num_room_in_leaf = TRIE_LEAF_SIZE;
157 return &leaf->head;
158 }
159
160 struct addr_range
161 {
162 bfd_byte *start;
163 bfd_byte *end;
164 };
165
166 /* Return true if address range do intersect. */
167
168 static bool
169 addr_range_intersects (struct addr_range *r1, struct addr_range *r2)
170 {
171 return (r1->start <= r2->start && r2->start < r1->end)
172 || (r1->start <= (r2->end - 1) && (r2->end - 1) < r1->end);
173 }
174
175 /* Compare function for splay tree of addr_ranges. */
176
177 static int
178 splay_tree_compare_addr_range (splay_tree_key xa, splay_tree_key xb)
179 {
180 struct addr_range *r1 = (struct addr_range *) xa;
181 struct addr_range *r2 = (struct addr_range *) xb;
182
183 if (addr_range_intersects (r1, r2) || addr_range_intersects (r2, r1))
184 return 0;
185 else if (r1->end <= r2->start)
186 return -1;
187 else
188 return 1;
189 }
190
191 /* Splay tree release function for keys (addr_range). */
192
193 static void
194 splay_tree_free_addr_range (splay_tree_key key)
195 {
196 free ((struct addr_range *)key);
197 }
198
199 struct dwarf2_debug_file
200 {
201 /* The actual bfd from which debug info was loaded. Might be
202 different to orig_bfd because of gnu_debuglink sections. */
203 bfd *bfd_ptr;
204
205 /* Pointer to the symbol table. */
206 asymbol **syms;
207
208 /* The current info pointer for the .debug_info section being parsed. */
209 bfd_byte *info_ptr;
210
211 /* A pointer to the memory block allocated for .debug_info sections. */
212 bfd_byte *dwarf_info_buffer;
213
214 /* Length of the loaded .debug_info sections. */
215 bfd_size_type dwarf_info_size;
216
217 /* Pointer to the .debug_abbrev section loaded into memory. */
218 bfd_byte *dwarf_abbrev_buffer;
219
220 /* Length of the loaded .debug_abbrev section. */
221 bfd_size_type dwarf_abbrev_size;
222
223 /* Buffer for decode_line_info. */
224 bfd_byte *dwarf_line_buffer;
225
226 /* Length of the loaded .debug_line section. */
227 bfd_size_type dwarf_line_size;
228
229 /* Pointer to the .debug_str section loaded into memory. */
230 bfd_byte *dwarf_str_buffer;
231
232 /* Length of the loaded .debug_str section. */
233 bfd_size_type dwarf_str_size;
234
235 /* Pointer to the .debug_str_offsets section loaded into memory. */
236 bfd_byte *dwarf_str_offsets_buffer;
237
238 /* Length of the loaded .debug_str_offsets section. */
239 bfd_size_type dwarf_str_offsets_size;
240
241 /* Pointer to the .debug_addr section loaded into memory. */
242 bfd_byte *dwarf_addr_buffer;
243
244 /* Length of the loaded .debug_addr section. */
245 bfd_size_type dwarf_addr_size;
246
247 /* Pointer to the .debug_line_str section loaded into memory. */
248 bfd_byte *dwarf_line_str_buffer;
249
250 /* Length of the loaded .debug_line_str section. */
251 bfd_size_type dwarf_line_str_size;
252
253 /* Pointer to the .debug_ranges section loaded into memory. */
254 bfd_byte *dwarf_ranges_buffer;
255
256 /* Length of the loaded .debug_ranges section. */
257 bfd_size_type dwarf_ranges_size;
258
259 /* Pointer to the .debug_rnglists section loaded into memory. */
260 bfd_byte *dwarf_rnglists_buffer;
261
262 /* Length of the loaded .debug_rnglists section. */
263 bfd_size_type dwarf_rnglists_size;
264
265 /* A list of all previously read comp_units. */
266 struct comp_unit *all_comp_units;
267
268 /* A list of all previously read comp_units with no ranges (yet). */
269 struct comp_unit *all_comp_units_without_ranges;
270
271 /* Last comp unit in list above. */
272 struct comp_unit *last_comp_unit;
273
274 /* Line table at line_offset zero. */
275 struct line_info_table *line_table;
276
277 /* Hash table to map offsets to decoded abbrevs. */
278 htab_t abbrev_offsets;
279
280 /* Root of a trie to map addresses to compilation units. */
281 struct trie_node *trie_root;
282
283 /* Splay tree to map info_ptr address to compilation units. */
284 splay_tree comp_unit_tree;
285 };
286
287 struct dwarf2_debug
288 {
289 /* Names of the debug sections. */
290 const struct dwarf_debug_section *debug_sections;
291
292 /* Per-file stuff. */
293 struct dwarf2_debug_file f, alt;
294
295 /* Pointer to the original bfd for which debug was loaded. This is what
296 we use to compare and so check that the cached debug data is still
297 valid - it saves having to possibly dereference the gnu_debuglink each
298 time. */
299 bfd *orig_bfd;
300
301 /* If the most recent call to bfd_find_nearest_line was given an
302 address in an inlined function, preserve a pointer into the
303 calling chain for subsequent calls to bfd_find_inliner_info to
304 use. */
305 struct funcinfo *inliner_chain;
306
307 /* Section VMAs at the time the stash was built. */
308 bfd_vma *sec_vma;
309 /* Number of sections in the SEC_VMA table. */
310 unsigned int sec_vma_count;
311
312 /* Number of sections whose VMA we must adjust. */
313 int adjusted_section_count;
314
315 /* Array of sections with adjusted VMA. */
316 struct adjusted_section *adjusted_sections;
317
318 /* Number of times find_line is called. This is used in
319 the heuristic for enabling the info hash tables. */
320 int info_hash_count;
321
322 #define STASH_INFO_HASH_TRIGGER 100
323
324 /* Hash table mapping symbol names to function infos. */
325 struct info_hash_table *funcinfo_hash_table;
326
327 /* Hash table mapping symbol names to variable infos. */
328 struct info_hash_table *varinfo_hash_table;
329
330 /* Head of comp_unit list in the last hash table update. */
331 struct comp_unit *hash_units_head;
332
333 /* Status of info hash. */
334 int info_hash_status;
335 #define STASH_INFO_HASH_OFF 0
336 #define STASH_INFO_HASH_ON 1
337 #define STASH_INFO_HASH_DISABLED 2
338
339 /* True if we opened bfd_ptr. */
340 bool close_on_cleanup;
341 };
342
343 struct arange
344 {
345 struct arange *next;
346 bfd_vma low;
347 bfd_vma high;
348 };
349
350 /* A minimal decoding of DWARF2 compilation units. We only decode
351 what's needed to get to the line number information. */
352
353 struct comp_unit
354 {
355 /* Chain the previously read compilation units. */
356 struct comp_unit *next_unit;
357
358 /* Chain the previously read compilation units that have no ranges yet.
359 We scan these separately when we have a trie over the ranges.
360 Unused if arange.high != 0. */
361 struct comp_unit *next_unit_without_ranges;
362
363 /* Likewise, chain the compilation unit read after this one.
364 The comp units are stored in reversed reading order. */
365 struct comp_unit *prev_unit;
366
367 /* Keep the bfd convenient (for memory allocation). */
368 bfd *abfd;
369
370 /* The lowest and highest addresses contained in this compilation
371 unit as specified in the compilation unit header. */
372 struct arange arange;
373
374 /* The DW_AT_name attribute (for error messages). */
375 char *name;
376
377 /* The abbrev hash table. */
378 struct abbrev_info **abbrevs;
379
380 /* DW_AT_language. */
381 int lang;
382
383 /* Note that an error was found by comp_unit_find_nearest_line. */
384 int error;
385
386 /* The DW_AT_comp_dir attribute. */
387 char *comp_dir;
388
389 /* TRUE if there is a line number table associated with this comp. unit. */
390 int stmtlist;
391
392 /* Pointer to the current comp_unit so that we can find a given entry
393 by its reference. */
394 bfd_byte *info_ptr_unit;
395
396 /* The offset into .debug_line of the line number table. */
397 unsigned long line_offset;
398
399 /* Pointer to the first child die for the comp unit. */
400 bfd_byte *first_child_die_ptr;
401
402 /* The end of the comp unit. */
403 bfd_byte *end_ptr;
404
405 /* The decoded line number, NULL if not yet decoded. */
406 struct line_info_table *line_table;
407
408 /* A list of the functions found in this comp. unit. */
409 struct funcinfo *function_table;
410
411 /* A table of function information references searchable by address. */
412 struct lookup_funcinfo *lookup_funcinfo_table;
413
414 /* Number of functions in the function_table and sorted_function_table. */
415 bfd_size_type number_of_functions;
416
417 /* A list of the variables found in this comp. unit. */
418 struct varinfo *variable_table;
419
420 /* Pointers to dwarf2_debug structures. */
421 struct dwarf2_debug *stash;
422 struct dwarf2_debug_file *file;
423
424 /* DWARF format version for this unit - from unit header. */
425 int version;
426
427 /* Address size for this unit - from unit header. */
428 unsigned char addr_size;
429
430 /* Offset size for this unit - from unit header. */
431 unsigned char offset_size;
432
433 /* Base address for this unit - from DW_AT_low_pc attribute of
434 DW_TAG_compile_unit DIE */
435 bfd_vma base_address;
436
437 /* TRUE if symbols are cached in hash table for faster lookup by name. */
438 bool cached;
439
440 /* Used when iterating over trie leaves to know which units we have
441 already seen in this iteration. */
442 bool mark;
443
444 /* Base address of debug_addr section. */
445 size_t dwarf_addr_offset;
446
447 /* Base address of string offset table. */
448 size_t dwarf_str_offset;
449 };
450
451 /* This data structure holds the information of an abbrev. */
452 struct abbrev_info
453 {
454 unsigned int number; /* Number identifying abbrev. */
455 enum dwarf_tag tag; /* DWARF tag. */
456 bool has_children; /* TRUE if the abbrev has children. */
457 unsigned int num_attrs; /* Number of attributes. */
458 struct attr_abbrev * attrs; /* An array of attribute descriptions. */
459 struct abbrev_info * next; /* Next in chain. */
460 };
461
462 struct attr_abbrev
463 {
464 enum dwarf_attribute name;
465 enum dwarf_form form;
466 bfd_vma implicit_const;
467 };
468
469 /* Map of uncompressed DWARF debug section name to compressed one. It
470 is terminated by NULL uncompressed_name. */
471
472 const struct dwarf_debug_section dwarf_debug_sections[] =
473 {
474 { ".debug_abbrev", ".zdebug_abbrev" },
475 { ".debug_aranges", ".zdebug_aranges" },
476 { ".debug_frame", ".zdebug_frame" },
477 { ".debug_info", ".zdebug_info" },
478 { ".debug_info", ".zdebug_info" },
479 { ".debug_line", ".zdebug_line" },
480 { ".debug_loc", ".zdebug_loc" },
481 { ".debug_macinfo", ".zdebug_macinfo" },
482 { ".debug_macro", ".zdebug_macro" },
483 { ".debug_pubnames", ".zdebug_pubnames" },
484 { ".debug_pubtypes", ".zdebug_pubtypes" },
485 { ".debug_ranges", ".zdebug_ranges" },
486 { ".debug_rnglists", ".zdebug_rnglist" },
487 { ".debug_static_func", ".zdebug_static_func" },
488 { ".debug_static_vars", ".zdebug_static_vars" },
489 { ".debug_str", ".zdebug_str", },
490 { ".debug_str", ".zdebug_str", },
491 { ".debug_str_offsets", ".zdebug_str_offsets", },
492 { ".debug_addr", ".zdebug_addr", },
493 { ".debug_line_str", ".zdebug_line_str", },
494 { ".debug_types", ".zdebug_types" },
495 /* GNU DWARF 1 extensions */
496 { ".debug_sfnames", ".zdebug_sfnames" },
497 { ".debug_srcinfo", ".zebug_srcinfo" },
498 /* SGI/MIPS DWARF 2 extensions */
499 { ".debug_funcnames", ".zdebug_funcnames" },
500 { ".debug_typenames", ".zdebug_typenames" },
501 { ".debug_varnames", ".zdebug_varnames" },
502 { ".debug_weaknames", ".zdebug_weaknames" },
503 { NULL, NULL },
504 };
505
506 /* NB/ Numbers in this enum must match up with indices
507 into the dwarf_debug_sections[] array above. */
508 enum dwarf_debug_section_enum
509 {
510 debug_abbrev = 0,
511 debug_aranges,
512 debug_frame,
513 debug_info,
514 debug_info_alt,
515 debug_line,
516 debug_loc,
517 debug_macinfo,
518 debug_macro,
519 debug_pubnames,
520 debug_pubtypes,
521 debug_ranges,
522 debug_rnglists,
523 debug_static_func,
524 debug_static_vars,
525 debug_str,
526 debug_str_alt,
527 debug_str_offsets,
528 debug_addr,
529 debug_line_str,
530 debug_types,
531 debug_sfnames,
532 debug_srcinfo,
533 debug_funcnames,
534 debug_typenames,
535 debug_varnames,
536 debug_weaknames,
537 debug_max
538 };
539
540 /* A static assertion. */
541 extern int dwarf_debug_section_assert[ARRAY_SIZE (dwarf_debug_sections)
542 == debug_max + 1 ? 1 : -1];
543
544 #ifndef ABBREV_HASH_SIZE
545 #define ABBREV_HASH_SIZE 121
546 #endif
547 #ifndef ATTR_ALLOC_CHUNK
548 #define ATTR_ALLOC_CHUNK 4
549 #endif
550
551 /* Variable and function hash tables. This is used to speed up look-up
552 in lookup_symbol_in_var_table() and lookup_symbol_in_function_table().
553 In order to share code between variable and function infos, we use
554 a list of untyped pointer for all variable/function info associated with
555 a symbol. We waste a bit of memory for list with one node but that
556 simplifies the code. */
557
558 struct info_list_node
559 {
560 struct info_list_node *next;
561 void *info;
562 };
563
564 /* Info hash entry. */
565 struct info_hash_entry
566 {
567 struct bfd_hash_entry root;
568 struct info_list_node *head;
569 };
570
571 struct info_hash_table
572 {
573 struct bfd_hash_table base;
574 };
575
576 /* Function to create a new entry in info hash table. */
577
578 static struct bfd_hash_entry *
579 info_hash_table_newfunc (struct bfd_hash_entry *entry,
580 struct bfd_hash_table *table,
581 const char *string)
582 {
583 struct info_hash_entry *ret = (struct info_hash_entry *) entry;
584
585 /* Allocate the structure if it has not already been allocated by a
586 derived class. */
587 if (ret == NULL)
588 {
589 ret = (struct info_hash_entry *) bfd_hash_allocate (table,
590 sizeof (* ret));
591 if (ret == NULL)
592 return NULL;
593 }
594
595 /* Call the allocation method of the base class. */
596 ret = ((struct info_hash_entry *)
597 bfd_hash_newfunc ((struct bfd_hash_entry *) ret, table, string));
598
599 /* Initialize the local fields here. */
600 if (ret)
601 ret->head = NULL;
602
603 return (struct bfd_hash_entry *) ret;
604 }
605
606 /* Function to create a new info hash table. It returns a pointer to the
607 newly created table or NULL if there is any error. We need abfd
608 solely for memory allocation. */
609
610 static struct info_hash_table *
611 create_info_hash_table (bfd *abfd)
612 {
613 struct info_hash_table *hash_table;
614
615 hash_table = ((struct info_hash_table *)
616 bfd_alloc (abfd, sizeof (struct info_hash_table)));
617 if (!hash_table)
618 return hash_table;
619
620 if (!bfd_hash_table_init (&hash_table->base, info_hash_table_newfunc,
621 sizeof (struct info_hash_entry)))
622 {
623 bfd_release (abfd, hash_table);
624 return NULL;
625 }
626
627 return hash_table;
628 }
629
630 /* Insert an info entry into an info hash table. We do not check of
631 duplicate entries. Also, the caller need to guarantee that the
632 right type of info in inserted as info is passed as a void* pointer.
633 This function returns true if there is no error. */
634
635 static bool
636 insert_info_hash_table (struct info_hash_table *hash_table,
637 const char *key,
638 void *info,
639 bool copy_p)
640 {
641 struct info_hash_entry *entry;
642 struct info_list_node *node;
643
644 entry = (struct info_hash_entry*) bfd_hash_lookup (&hash_table->base,
645 key, true, copy_p);
646 if (!entry)
647 return false;
648
649 node = (struct info_list_node *) bfd_hash_allocate (&hash_table->base,
650 sizeof (*node));
651 if (!node)
652 return false;
653
654 node->info = info;
655 node->next = entry->head;
656 entry->head = node;
657
658 return true;
659 }
660
661 /* Look up an info entry list from an info hash table. Return NULL
662 if there is none. */
663
664 static struct info_list_node *
665 lookup_info_hash_table (struct info_hash_table *hash_table, const char *key)
666 {
667 struct info_hash_entry *entry;
668
669 entry = (struct info_hash_entry*) bfd_hash_lookup (&hash_table->base, key,
670 false, false);
671 return entry ? entry->head : NULL;
672 }
673
674 /* Read a section into its appropriate place in the dwarf2_debug
675 struct (indicated by SECTION_BUFFER and SECTION_SIZE). If SYMS is
676 not NULL, use bfd_simple_get_relocated_section_contents to read the
677 section contents, otherwise use bfd_get_section_contents. Fail if
678 the located section does not contain at least OFFSET bytes. */
679
680 static bool
681 read_section (bfd *abfd,
682 const struct dwarf_debug_section *sec,
683 asymbol **syms,
684 uint64_t offset,
685 bfd_byte **section_buffer,
686 bfd_size_type *section_size)
687 {
688 const char *section_name = sec->uncompressed_name;
689 bfd_byte *contents = *section_buffer;
690
691 /* The section may have already been read. */
692 if (contents == NULL)
693 {
694 bfd_size_type amt;
695 asection *msec;
696
697 msec = bfd_get_section_by_name (abfd, section_name);
698 if (msec == NULL)
699 {
700 section_name = sec->compressed_name;
701 msec = bfd_get_section_by_name (abfd, section_name);
702 }
703 if (msec == NULL)
704 {
705 _bfd_error_handler (_("DWARF error: can't find %s section."),
706 sec->uncompressed_name);
707 bfd_set_error (bfd_error_bad_value);
708 return false;
709 }
710
711 if ((msec->flags & SEC_HAS_CONTENTS) == 0)
712 {
713 _bfd_error_handler (_("DWARF error: section %s has no contents"),
714 section_name);
715 bfd_set_error (bfd_error_no_contents);
716 return false;
717 }
718
719 if (_bfd_section_size_insane (abfd, msec))
720 {
721 /* PR 26946 */
722 _bfd_error_handler (_("DWARF error: section %s is too big"),
723 section_name);
724 return false;
725 }
726 amt = bfd_get_section_limit_octets (abfd, msec);
727 *section_size = amt;
728 /* Paranoia - alloc one extra so that we can make sure a string
729 section is NUL terminated. */
730 amt += 1;
731 if (amt == 0)
732 {
733 /* Paranoia - this should never happen. */
734 bfd_set_error (bfd_error_no_memory);
735 return false;
736 }
737 contents = (bfd_byte *) bfd_malloc (amt);
738 if (contents == NULL)
739 return false;
740 if (syms
741 ? !bfd_simple_get_relocated_section_contents (abfd, msec, contents,
742 syms)
743 : !bfd_get_section_contents (abfd, msec, contents, 0, *section_size))
744 {
745 free (contents);
746 return false;
747 }
748 contents[*section_size] = 0;
749 *section_buffer = contents;
750 }
751
752 /* It is possible to get a bad value for the offset into the section
753 that the client wants. Validate it here to avoid trouble later. */
754 if (offset != 0 && offset >= *section_size)
755 {
756 /* xgettext: c-format */
757 _bfd_error_handler (_("DWARF error: offset (%" PRIu64 ")"
758 " greater than or equal to %s size (%" PRIu64 ")"),
759 (uint64_t) offset, section_name,
760 (uint64_t) *section_size);
761 bfd_set_error (bfd_error_bad_value);
762 return false;
763 }
764
765 return true;
766 }
767
768 /* Read dwarf information from a buffer. */
769
770 static inline uint64_t
771 read_n_bytes (bfd *abfd, bfd_byte **ptr, bfd_byte *end, int n)
772 {
773 bfd_byte *buf = *ptr;
774 if (end - buf < n)
775 {
776 *ptr = end;
777 return 0;
778 }
779 *ptr = buf + n;
780 return bfd_get (n * 8, abfd, buf);
781 }
782
783 static unsigned int
784 read_1_byte (bfd *abfd, bfd_byte **ptr, bfd_byte *end)
785 {
786 return read_n_bytes (abfd, ptr, end, 1);
787 }
788
789 static int
790 read_1_signed_byte (bfd *abfd ATTRIBUTE_UNUSED, bfd_byte **ptr, bfd_byte *end)
791 {
792 bfd_byte *buf = *ptr;
793 if (end - buf < 1)
794 {
795 *ptr = end;
796 return 0;
797 }
798 *ptr = buf + 1;
799 return bfd_get_signed_8 (abfd, buf);
800 }
801
802 static unsigned int
803 read_2_bytes (bfd *abfd, bfd_byte **ptr, bfd_byte *end)
804 {
805 return read_n_bytes (abfd, ptr, end, 2);
806 }
807
808 static unsigned int
809 read_3_bytes (bfd *abfd, bfd_byte **ptr, bfd_byte *end)
810 {
811 unsigned int val = read_1_byte (abfd, ptr, end);
812 val <<= 8;
813 val |= read_1_byte (abfd, ptr, end);
814 val <<= 8;
815 val |= read_1_byte (abfd, ptr, end);
816 if (bfd_little_endian (abfd))
817 val = (((val >> 16) & 0xff)
818 | (val & 0xff00)
819 | ((val & 0xff) << 16));
820 return val;
821 }
822
823 static unsigned int
824 read_4_bytes (bfd *abfd, bfd_byte **ptr, bfd_byte *end)
825 {
826 return read_n_bytes (abfd, ptr, end, 4);
827 }
828
829 static uint64_t
830 read_8_bytes (bfd *abfd, bfd_byte **ptr, bfd_byte *end)
831 {
832 return read_n_bytes (abfd, ptr, end, 8);
833 }
834
835 static struct dwarf_block *
836 read_blk (bfd *abfd, bfd_byte **ptr, bfd_byte *end, size_t size)
837 {
838 bfd_byte *buf = *ptr;
839 struct dwarf_block *block;
840
841 block = (struct dwarf_block *) bfd_alloc (abfd, sizeof (*block));
842 if (block == NULL)
843 return NULL;
844
845 if (size > (size_t) (end - buf))
846 {
847 *ptr = end;
848 block->data = NULL;
849 block->size = 0;
850 }
851 else
852 {
853 *ptr = buf + size;
854 block->data = buf;
855 block->size = size;
856 }
857 return block;
858 }
859
860 /* Scans a NUL terminated string starting at *PTR, returning a pointer to it.
861 Bytes at or beyond BUF_END will not be read. Returns NULL if the
862 terminator is not found or if the string is empty. *PTR is
863 incremented over the bytes scanned, including the terminator. */
864
865 static char *
866 read_string (bfd_byte **ptr,
867 bfd_byte *buf_end)
868 {
869 bfd_byte *buf = *ptr;
870 bfd_byte *str = buf;
871
872 while (buf < buf_end)
873 if (*buf++ == 0)
874 {
875 if (str == buf - 1)
876 break;
877 *ptr = buf;
878 return (char *) str;
879 }
880
881 *ptr = buf;
882 return NULL;
883 }
884
885 /* Reads an offset from *PTR and then locates the string at this offset
886 inside the debug string section. Returns a pointer to the string.
887 Increments *PTR by the number of bytes read for the offset. This
888 value is set even if the function fails. Bytes at or beyond
889 BUF_END will not be read. Returns NULL if there was a problem, or
890 if the string is empty. Does not check for NUL termination of the
891 string. */
892
893 static char *
894 read_indirect_string (struct comp_unit *unit,
895 bfd_byte **ptr,
896 bfd_byte *buf_end)
897 {
898 uint64_t offset;
899 struct dwarf2_debug *stash = unit->stash;
900 struct dwarf2_debug_file *file = unit->file;
901 char *str;
902
903 if (unit->offset_size > (size_t) (buf_end - *ptr))
904 {
905 *ptr = buf_end;
906 return NULL;
907 }
908
909 if (unit->offset_size == 4)
910 offset = read_4_bytes (unit->abfd, ptr, buf_end);
911 else
912 offset = read_8_bytes (unit->abfd, ptr, buf_end);
913
914 if (! read_section (unit->abfd, &stash->debug_sections[debug_str],
915 file->syms, offset,
916 &file->dwarf_str_buffer, &file->dwarf_str_size))
917 return NULL;
918
919 str = (char *) file->dwarf_str_buffer + offset;
920 if (*str == '\0')
921 return NULL;
922 return str;
923 }
924
925 /* Like read_indirect_string but from .debug_line_str section. */
926
927 static char *
928 read_indirect_line_string (struct comp_unit *unit,
929 bfd_byte **ptr,
930 bfd_byte *buf_end)
931 {
932 uint64_t offset;
933 struct dwarf2_debug *stash = unit->stash;
934 struct dwarf2_debug_file *file = unit->file;
935 char *str;
936
937 if (unit->offset_size > (size_t) (buf_end - *ptr))
938 {
939 *ptr = buf_end;
940 return NULL;
941 }
942
943 if (unit->offset_size == 4)
944 offset = read_4_bytes (unit->abfd, ptr, buf_end);
945 else
946 offset = read_8_bytes (unit->abfd, ptr, buf_end);
947
948 if (! read_section (unit->abfd, &stash->debug_sections[debug_line_str],
949 file->syms, offset,
950 &file->dwarf_line_str_buffer,
951 &file->dwarf_line_str_size))
952 return NULL;
953
954 str = (char *) file->dwarf_line_str_buffer + offset;
955 if (*str == '\0')
956 return NULL;
957 return str;
958 }
959
960 /* Like read_indirect_string but uses a .debug_str located in
961 an alternate file pointed to by the .gnu_debugaltlink section.
962 Used to impement DW_FORM_GNU_strp_alt. */
963
964 static char *
965 read_alt_indirect_string (struct comp_unit *unit,
966 bfd_byte **ptr,
967 bfd_byte *buf_end)
968 {
969 uint64_t offset;
970 struct dwarf2_debug *stash = unit->stash;
971 char *str;
972
973 if (unit->offset_size > (size_t) (buf_end - *ptr))
974 {
975 *ptr = buf_end;
976 return NULL;
977 }
978
979 if (unit->offset_size == 4)
980 offset = read_4_bytes (unit->abfd, ptr, buf_end);
981 else
982 offset = read_8_bytes (unit->abfd, ptr, buf_end);
983
984 if (stash->alt.bfd_ptr == NULL)
985 {
986 bfd *debug_bfd;
987 char *debug_filename = bfd_follow_gnu_debugaltlink (unit->abfd, DEBUGDIR);
988
989 if (debug_filename == NULL)
990 return NULL;
991
992 debug_bfd = bfd_openr (debug_filename, NULL);
993 free (debug_filename);
994 if (debug_bfd == NULL)
995 /* FIXME: Should we report our failure to follow the debuglink ? */
996 return NULL;
997
998 if (!bfd_check_format (debug_bfd, bfd_object))
999 {
1000 bfd_close (debug_bfd);
1001 return NULL;
1002 }
1003 stash->alt.bfd_ptr = debug_bfd;
1004 }
1005
1006 if (! read_section (unit->stash->alt.bfd_ptr,
1007 stash->debug_sections + debug_str_alt,
1008 stash->alt.syms, offset,
1009 &stash->alt.dwarf_str_buffer,
1010 &stash->alt.dwarf_str_size))
1011 return NULL;
1012
1013 str = (char *) stash->alt.dwarf_str_buffer + offset;
1014 if (*str == '\0')
1015 return NULL;
1016
1017 return str;
1018 }
1019
1020 /* Resolve an alternate reference from UNIT at OFFSET.
1021 Returns a pointer into the loaded alternate CU upon success
1022 or NULL upon failure. */
1023
1024 static bfd_byte *
1025 read_alt_indirect_ref (struct comp_unit *unit, uint64_t offset)
1026 {
1027 struct dwarf2_debug *stash = unit->stash;
1028
1029 if (stash->alt.bfd_ptr == NULL)
1030 {
1031 bfd *debug_bfd;
1032 char *debug_filename = bfd_follow_gnu_debugaltlink (unit->abfd, DEBUGDIR);
1033
1034 if (debug_filename == NULL)
1035 return NULL;
1036
1037 debug_bfd = bfd_openr (debug_filename, NULL);
1038 free (debug_filename);
1039 if (debug_bfd == NULL)
1040 /* FIXME: Should we report our failure to follow the debuglink ? */
1041 return NULL;
1042
1043 if (!bfd_check_format (debug_bfd, bfd_object))
1044 {
1045 bfd_close (debug_bfd);
1046 return NULL;
1047 }
1048 stash->alt.bfd_ptr = debug_bfd;
1049 }
1050
1051 if (! read_section (unit->stash->alt.bfd_ptr,
1052 stash->debug_sections + debug_info_alt,
1053 stash->alt.syms, offset,
1054 &stash->alt.dwarf_info_buffer,
1055 &stash->alt.dwarf_info_size))
1056 return NULL;
1057
1058 return stash->alt.dwarf_info_buffer + offset;
1059 }
1060
1061 static uint64_t
1062 read_address (struct comp_unit *unit, bfd_byte **ptr, bfd_byte *buf_end)
1063 {
1064 bfd_byte *buf = *ptr;
1065 int signed_vma = 0;
1066
1067 if (bfd_get_flavour (unit->abfd) == bfd_target_elf_flavour)
1068 signed_vma = get_elf_backend_data (unit->abfd)->sign_extend_vma;
1069
1070 if (unit->addr_size > (size_t) (buf_end - buf))
1071 {
1072 *ptr = buf_end;
1073 return 0;
1074 }
1075
1076 *ptr = buf + unit->addr_size;
1077 if (signed_vma)
1078 {
1079 switch (unit->addr_size)
1080 {
1081 case 8:
1082 return bfd_get_signed_64 (unit->abfd, buf);
1083 case 4:
1084 return bfd_get_signed_32 (unit->abfd, buf);
1085 case 2:
1086 return bfd_get_signed_16 (unit->abfd, buf);
1087 default:
1088 abort ();
1089 }
1090 }
1091 else
1092 {
1093 switch (unit->addr_size)
1094 {
1095 case 8:
1096 return bfd_get_64 (unit->abfd, buf);
1097 case 4:
1098 return bfd_get_32 (unit->abfd, buf);
1099 case 2:
1100 return bfd_get_16 (unit->abfd, buf);
1101 default:
1102 abort ();
1103 }
1104 }
1105 }
1106
1107 /* Lookup an abbrev_info structure in the abbrev hash table. */
1108
1109 static struct abbrev_info *
1110 lookup_abbrev (unsigned int number, struct abbrev_info **abbrevs)
1111 {
1112 unsigned int hash_number;
1113 struct abbrev_info *abbrev;
1114
1115 hash_number = number % ABBREV_HASH_SIZE;
1116 abbrev = abbrevs[hash_number];
1117
1118 while (abbrev)
1119 {
1120 if (abbrev->number == number)
1121 return abbrev;
1122 else
1123 abbrev = abbrev->next;
1124 }
1125
1126 return NULL;
1127 }
1128
1129 /* We keep a hash table to map .debug_abbrev section offsets to the
1130 array of abbrevs, so that compilation units using the same set of
1131 abbrevs do not waste memory. */
1132
1133 struct abbrev_offset_entry
1134 {
1135 size_t offset;
1136 struct abbrev_info **abbrevs;
1137 };
1138
1139 static hashval_t
1140 hash_abbrev (const void *p)
1141 {
1142 const struct abbrev_offset_entry *ent = p;
1143 return htab_hash_pointer ((void *) ent->offset);
1144 }
1145
1146 static int
1147 eq_abbrev (const void *pa, const void *pb)
1148 {
1149 const struct abbrev_offset_entry *a = pa;
1150 const struct abbrev_offset_entry *b = pb;
1151 return a->offset == b->offset;
1152 }
1153
1154 static void
1155 del_abbrev (void *p)
1156 {
1157 struct abbrev_offset_entry *ent = p;
1158 struct abbrev_info **abbrevs = ent->abbrevs;
1159 size_t i;
1160
1161 for (i = 0; i < ABBREV_HASH_SIZE; i++)
1162 {
1163 struct abbrev_info *abbrev = abbrevs[i];
1164
1165 while (abbrev)
1166 {
1167 free (abbrev->attrs);
1168 abbrev = abbrev->next;
1169 }
1170 }
1171 free (ent);
1172 }
1173
1174 /* In DWARF version 2, the description of the debugging information is
1175 stored in a separate .debug_abbrev section. Before we read any
1176 dies from a section we read in all abbreviations and install them
1177 in a hash table. */
1178
1179 static struct abbrev_info**
1180 read_abbrevs (bfd *abfd, uint64_t offset, struct dwarf2_debug *stash,
1181 struct dwarf2_debug_file *file)
1182 {
1183 struct abbrev_info **abbrevs;
1184 bfd_byte *abbrev_ptr;
1185 bfd_byte *abbrev_end;
1186 struct abbrev_info *cur_abbrev;
1187 unsigned int abbrev_number, abbrev_name;
1188 unsigned int abbrev_form, hash_number;
1189 size_t amt;
1190 void **slot;
1191 struct abbrev_offset_entry ent = { offset, NULL };
1192
1193 if (ent.offset != offset)
1194 return NULL;
1195
1196 slot = htab_find_slot (file->abbrev_offsets, &ent, INSERT);
1197 if (slot == NULL)
1198 return NULL;
1199 if (*slot != NULL)
1200 return ((struct abbrev_offset_entry *) (*slot))->abbrevs;
1201
1202 if (! read_section (abfd, &stash->debug_sections[debug_abbrev],
1203 file->syms, offset,
1204 &file->dwarf_abbrev_buffer,
1205 &file->dwarf_abbrev_size))
1206 return NULL;
1207
1208 amt = sizeof (struct abbrev_info*) * ABBREV_HASH_SIZE;
1209 abbrevs = (struct abbrev_info **) bfd_zalloc (abfd, amt);
1210 if (abbrevs == NULL)
1211 return NULL;
1212
1213 abbrev_ptr = file->dwarf_abbrev_buffer + offset;
1214 abbrev_end = file->dwarf_abbrev_buffer + file->dwarf_abbrev_size;
1215 abbrev_number = _bfd_safe_read_leb128 (abfd, &abbrev_ptr,
1216 false, abbrev_end);
1217
1218 /* Loop until we reach an abbrev number of 0. */
1219 while (abbrev_number)
1220 {
1221 amt = sizeof (struct abbrev_info);
1222 cur_abbrev = (struct abbrev_info *) bfd_zalloc (abfd, amt);
1223 if (cur_abbrev == NULL)
1224 goto fail;
1225
1226 /* Read in abbrev header. */
1227 cur_abbrev->number = abbrev_number;
1228 cur_abbrev->tag = (enum dwarf_tag)
1229 _bfd_safe_read_leb128 (abfd, &abbrev_ptr,
1230 false, abbrev_end);
1231 cur_abbrev->has_children = read_1_byte (abfd, &abbrev_ptr, abbrev_end);
1232
1233 /* Now read in declarations. */
1234 for (;;)
1235 {
1236 /* Initialize it just to avoid a GCC false warning. */
1237 bfd_vma implicit_const = -1;
1238
1239 abbrev_name = _bfd_safe_read_leb128 (abfd, &abbrev_ptr,
1240 false, abbrev_end);
1241 abbrev_form = _bfd_safe_read_leb128 (abfd, &abbrev_ptr,
1242 false, abbrev_end);
1243 if (abbrev_form == DW_FORM_implicit_const)
1244 implicit_const = _bfd_safe_read_leb128 (abfd, &abbrev_ptr,
1245 true, abbrev_end);
1246 if (abbrev_name == 0)
1247 break;
1248
1249 if ((cur_abbrev->num_attrs % ATTR_ALLOC_CHUNK) == 0)
1250 {
1251 struct attr_abbrev *tmp;
1252
1253 amt = cur_abbrev->num_attrs + ATTR_ALLOC_CHUNK;
1254 amt *= sizeof (struct attr_abbrev);
1255 tmp = (struct attr_abbrev *) bfd_realloc (cur_abbrev->attrs, amt);
1256 if (tmp == NULL)
1257 goto fail;
1258 cur_abbrev->attrs = tmp;
1259 }
1260
1261 cur_abbrev->attrs[cur_abbrev->num_attrs].name
1262 = (enum dwarf_attribute) abbrev_name;
1263 cur_abbrev->attrs[cur_abbrev->num_attrs].form
1264 = (enum dwarf_form) abbrev_form;
1265 cur_abbrev->attrs[cur_abbrev->num_attrs].implicit_const
1266 = implicit_const;
1267 ++cur_abbrev->num_attrs;
1268 }
1269
1270 hash_number = abbrev_number % ABBREV_HASH_SIZE;
1271 cur_abbrev->next = abbrevs[hash_number];
1272 abbrevs[hash_number] = cur_abbrev;
1273
1274 /* Get next abbreviation.
1275 Under Irix6 the abbreviations for a compilation unit are not
1276 always properly terminated with an abbrev number of 0.
1277 Exit loop if we encounter an abbreviation which we have
1278 already read (which means we are about to read the abbreviations
1279 for the next compile unit) or if the end of the abbreviation
1280 table is reached. */
1281 if ((size_t) (abbrev_ptr - file->dwarf_abbrev_buffer)
1282 >= file->dwarf_abbrev_size)
1283 break;
1284 abbrev_number = _bfd_safe_read_leb128 (abfd, &abbrev_ptr,
1285 false, abbrev_end);
1286 if (lookup_abbrev (abbrev_number, abbrevs) != NULL)
1287 break;
1288 }
1289
1290 *slot = bfd_malloc (sizeof ent);
1291 if (!*slot)
1292 goto fail;
1293 ent.abbrevs = abbrevs;
1294 memcpy (*slot, &ent, sizeof ent);
1295 return abbrevs;
1296
1297 fail:
1298 if (abbrevs != NULL)
1299 {
1300 size_t i;
1301
1302 for (i = 0; i < ABBREV_HASH_SIZE; i++)
1303 {
1304 struct abbrev_info *abbrev = abbrevs[i];
1305
1306 while (abbrev)
1307 {
1308 free (abbrev->attrs);
1309 abbrev = abbrev->next;
1310 }
1311 }
1312 free (abbrevs);
1313 }
1314 return NULL;
1315 }
1316
1317 /* Returns true if the form is one which has a string value. */
1318
1319 static bool
1320 is_str_form (const struct attribute *attr)
1321 {
1322 switch (attr->form)
1323 {
1324 case DW_FORM_string:
1325 case DW_FORM_strp:
1326 case DW_FORM_strx:
1327 case DW_FORM_strx1:
1328 case DW_FORM_strx2:
1329 case DW_FORM_strx3:
1330 case DW_FORM_strx4:
1331 case DW_FORM_line_strp:
1332 case DW_FORM_GNU_strp_alt:
1333 return true;
1334
1335 default:
1336 return false;
1337 }
1338 }
1339
1340 /* Returns true if the form is one which has an integer value. */
1341
1342 static bool
1343 is_int_form (const struct attribute *attr)
1344 {
1345 switch (attr->form)
1346 {
1347 case DW_FORM_addr:
1348 case DW_FORM_data2:
1349 case DW_FORM_data4:
1350 case DW_FORM_data8:
1351 case DW_FORM_data1:
1352 case DW_FORM_flag:
1353 case DW_FORM_sdata:
1354 case DW_FORM_udata:
1355 case DW_FORM_ref_addr:
1356 case DW_FORM_ref1:
1357 case DW_FORM_ref2:
1358 case DW_FORM_ref4:
1359 case DW_FORM_ref8:
1360 case DW_FORM_ref_udata:
1361 case DW_FORM_sec_offset:
1362 case DW_FORM_flag_present:
1363 case DW_FORM_ref_sig8:
1364 case DW_FORM_addrx:
1365 case DW_FORM_implicit_const:
1366 case DW_FORM_addrx1:
1367 case DW_FORM_addrx2:
1368 case DW_FORM_addrx3:
1369 case DW_FORM_addrx4:
1370 case DW_FORM_GNU_ref_alt:
1371 return true;
1372
1373 default:
1374 return false;
1375 }
1376 }
1377
1378 /* Returns true if the form is strx[1-4]. */
1379
1380 static inline bool
1381 is_strx_form (enum dwarf_form form)
1382 {
1383 return (form == DW_FORM_strx
1384 || form == DW_FORM_strx1
1385 || form == DW_FORM_strx2
1386 || form == DW_FORM_strx3
1387 || form == DW_FORM_strx4);
1388 }
1389
1390 /* Return true if the form is addrx[1-4]. */
1391
1392 static inline bool
1393 is_addrx_form (enum dwarf_form form)
1394 {
1395 return (form == DW_FORM_addrx
1396 || form == DW_FORM_addrx1
1397 || form == DW_FORM_addrx2
1398 || form == DW_FORM_addrx3
1399 || form == DW_FORM_addrx4);
1400 }
1401
1402 /* Returns the address in .debug_addr section using DW_AT_addr_base.
1403 Used to implement DW_FORM_addrx*. */
1404 static uint64_t
1405 read_indexed_address (uint64_t idx, struct comp_unit *unit)
1406 {
1407 struct dwarf2_debug *stash = unit->stash;
1408 struct dwarf2_debug_file *file = unit->file;
1409 bfd_byte *info_ptr;
1410 size_t offset;
1411
1412 if (stash == NULL)
1413 return 0;
1414
1415 if (!read_section (unit->abfd, &stash->debug_sections[debug_addr],
1416 file->syms, 0,
1417 &file->dwarf_addr_buffer, &file->dwarf_addr_size))
1418 return 0;
1419
1420 if (_bfd_mul_overflow (idx, unit->addr_size, &offset))
1421 return 0;
1422
1423 offset += unit->dwarf_addr_offset;
1424 if (offset < unit->dwarf_addr_offset
1425 || offset > file->dwarf_addr_size
1426 || file->dwarf_addr_size - offset < unit->addr_size)
1427 return 0;
1428
1429 info_ptr = file->dwarf_addr_buffer + offset;
1430
1431 if (unit->addr_size == 4)
1432 return bfd_get_32 (unit->abfd, info_ptr);
1433 else if (unit->addr_size == 8)
1434 return bfd_get_64 (unit->abfd, info_ptr);
1435 else
1436 return 0;
1437 }
1438
1439 /* Returns the string using DW_AT_str_offsets_base.
1440 Used to implement DW_FORM_strx*. */
1441 static const char *
1442 read_indexed_string (uint64_t idx, struct comp_unit *unit)
1443 {
1444 struct dwarf2_debug *stash = unit->stash;
1445 struct dwarf2_debug_file *file = unit->file;
1446 bfd_byte *info_ptr;
1447 uint64_t str_offset;
1448 size_t offset;
1449
1450 if (stash == NULL)
1451 return NULL;
1452
1453 if (!read_section (unit->abfd, &stash->debug_sections[debug_str],
1454 file->syms, 0,
1455 &file->dwarf_str_buffer, &file->dwarf_str_size))
1456 return NULL;
1457
1458 if (!read_section (unit->abfd, &stash->debug_sections[debug_str_offsets],
1459 file->syms, 0,
1460 &file->dwarf_str_offsets_buffer,
1461 &file->dwarf_str_offsets_size))
1462 return NULL;
1463
1464 if (_bfd_mul_overflow (idx, unit->offset_size, &offset))
1465 return NULL;
1466
1467 offset += unit->dwarf_str_offset;
1468 if (offset < unit->dwarf_str_offset
1469 || offset > file->dwarf_str_offsets_size
1470 || file->dwarf_str_offsets_size - offset < unit->offset_size)
1471 return NULL;
1472
1473 info_ptr = file->dwarf_str_offsets_buffer + offset;
1474
1475 if (unit->offset_size == 4)
1476 str_offset = bfd_get_32 (unit->abfd, info_ptr);
1477 else if (unit->offset_size == 8)
1478 str_offset = bfd_get_64 (unit->abfd, info_ptr);
1479 else
1480 return NULL;
1481
1482 if (str_offset >= file->dwarf_str_size)
1483 return NULL;
1484 return (const char *) file->dwarf_str_buffer + str_offset;
1485 }
1486
1487 /* Read and fill in the value of attribute ATTR as described by FORM.
1488 Read data starting from INFO_PTR, but never at or beyond INFO_PTR_END.
1489 Returns an updated INFO_PTR taking into account the amount of data read. */
1490
1491 static bfd_byte *
1492 read_attribute_value (struct attribute * attr,
1493 unsigned form,
1494 bfd_vma implicit_const,
1495 struct comp_unit * unit,
1496 bfd_byte * info_ptr,
1497 bfd_byte * info_ptr_end)
1498 {
1499 bfd *abfd = unit->abfd;
1500 size_t amt;
1501
1502 if (info_ptr >= info_ptr_end && form != DW_FORM_flag_present)
1503 {
1504 _bfd_error_handler (_("DWARF error: info pointer extends beyond end of attributes"));
1505 bfd_set_error (bfd_error_bad_value);
1506 return NULL;
1507 }
1508
1509 attr->form = (enum dwarf_form) form;
1510
1511 switch (form)
1512 {
1513 case DW_FORM_flag_present:
1514 attr->u.val = 1;
1515 break;
1516 case DW_FORM_ref_addr:
1517 /* DW_FORM_ref_addr is an address in DWARF2, and an offset in
1518 DWARF3. */
1519 if (unit->version >= 3)
1520 {
1521 if (unit->offset_size == 4)
1522 attr->u.val = read_4_bytes (unit->abfd, &info_ptr, info_ptr_end);
1523 else
1524 attr->u.val = read_8_bytes (unit->abfd, &info_ptr, info_ptr_end);
1525 break;
1526 }
1527 /* FALLTHROUGH */
1528 case DW_FORM_addr:
1529 attr->u.val = read_address (unit, &info_ptr, info_ptr_end);
1530 break;
1531 case DW_FORM_GNU_ref_alt:
1532 case DW_FORM_sec_offset:
1533 if (unit->offset_size == 4)
1534 attr->u.val = read_4_bytes (unit->abfd, &info_ptr, info_ptr_end);
1535 else
1536 attr->u.val = read_8_bytes (unit->abfd, &info_ptr, info_ptr_end);
1537 break;
1538 case DW_FORM_block2:
1539 amt = read_2_bytes (abfd, &info_ptr, info_ptr_end);
1540 attr->u.blk = read_blk (abfd, &info_ptr, info_ptr_end, amt);
1541 if (attr->u.blk == NULL)
1542 return NULL;
1543 break;
1544 case DW_FORM_block4:
1545 amt = read_4_bytes (abfd, &info_ptr, info_ptr_end);
1546 attr->u.blk = read_blk (abfd, &info_ptr, info_ptr_end, amt);
1547 if (attr->u.blk == NULL)
1548 return NULL;
1549 break;
1550 case DW_FORM_ref1:
1551 case DW_FORM_flag:
1552 case DW_FORM_data1:
1553 attr->u.val = read_1_byte (abfd, &info_ptr, info_ptr_end);
1554 break;
1555 case DW_FORM_addrx1:
1556 attr->u.val = read_1_byte (abfd, &info_ptr, info_ptr_end);
1557 /* dwarf_addr_offset value 0 indicates the attribute DW_AT_addr_base
1558 is not yet read. */
1559 if (unit->dwarf_addr_offset != 0)
1560 attr->u.val = read_indexed_address (attr->u.val, unit);
1561 break;
1562 case DW_FORM_data2:
1563 case DW_FORM_ref2:
1564 attr->u.val = read_2_bytes (abfd, &info_ptr, info_ptr_end);
1565 break;
1566 case DW_FORM_addrx2:
1567 attr->u.val = read_2_bytes (abfd, &info_ptr, info_ptr_end);
1568 if (unit->dwarf_addr_offset != 0)
1569 attr->u.val = read_indexed_address (attr->u.val, unit);
1570 break;
1571 case DW_FORM_addrx3:
1572 attr->u.val = read_3_bytes (abfd, &info_ptr, info_ptr_end);
1573 if (unit->dwarf_addr_offset != 0)
1574 attr->u.val = read_indexed_address(attr->u.val, unit);
1575 break;
1576 case DW_FORM_ref4:
1577 case DW_FORM_data4:
1578 attr->u.val = read_4_bytes (abfd, &info_ptr, info_ptr_end);
1579 break;
1580 case DW_FORM_addrx4:
1581 attr->u.val = read_4_bytes (abfd, &info_ptr, info_ptr_end);
1582 if (unit->dwarf_addr_offset != 0)
1583 attr->u.val = read_indexed_address (attr->u.val, unit);
1584 break;
1585 case DW_FORM_data8:
1586 case DW_FORM_ref8:
1587 case DW_FORM_ref_sig8:
1588 attr->u.val = read_8_bytes (abfd, &info_ptr, info_ptr_end);
1589 break;
1590 case DW_FORM_string:
1591 attr->u.str = read_string (&info_ptr, info_ptr_end);
1592 break;
1593 case DW_FORM_strp:
1594 attr->u.str = read_indirect_string (unit, &info_ptr, info_ptr_end);
1595 break;
1596 case DW_FORM_line_strp:
1597 attr->u.str = read_indirect_line_string (unit, &info_ptr, info_ptr_end);
1598 break;
1599 case DW_FORM_GNU_strp_alt:
1600 attr->u.str = read_alt_indirect_string (unit, &info_ptr, info_ptr_end);
1601 break;
1602 case DW_FORM_strx1:
1603 attr->u.val = read_1_byte (abfd, &info_ptr, info_ptr_end);
1604 /* dwarf_str_offset value 0 indicates the attribute DW_AT_str_offsets_base
1605 is not yet read. */
1606 if (unit->dwarf_str_offset != 0)
1607 attr->u.str = (char *) read_indexed_string (attr->u.val, unit);
1608 else
1609 attr->u.str = NULL;
1610 break;
1611 case DW_FORM_strx2:
1612 attr->u.val = read_2_bytes (abfd, &info_ptr, info_ptr_end);
1613 if (unit->dwarf_str_offset != 0)
1614 attr->u.str = (char *) read_indexed_string (attr->u.val, unit);
1615 else
1616 attr->u.str = NULL;
1617 break;
1618 case DW_FORM_strx3:
1619 attr->u.val = read_3_bytes (abfd, &info_ptr, info_ptr_end);
1620 if (unit->dwarf_str_offset != 0)
1621 attr->u.str = (char *) read_indexed_string (attr->u.val, unit);
1622 else
1623 attr->u.str = NULL;
1624 break;
1625 case DW_FORM_strx4:
1626 attr->u.val = read_4_bytes (abfd, &info_ptr, info_ptr_end);
1627 if (unit->dwarf_str_offset != 0)
1628 attr->u.str = (char *) read_indexed_string (attr->u.val, unit);
1629 else
1630 attr->u.str = NULL;
1631 break;
1632 case DW_FORM_strx:
1633 attr->u.val = _bfd_safe_read_leb128 (abfd, &info_ptr,
1634 false, info_ptr_end);
1635 if (unit->dwarf_str_offset != 0)
1636 attr->u.str = (char *) read_indexed_string (attr->u.val, unit);
1637 else
1638 attr->u.str = NULL;
1639 break;
1640 case DW_FORM_exprloc:
1641 case DW_FORM_block:
1642 amt = _bfd_safe_read_leb128 (abfd, &info_ptr,
1643 false, info_ptr_end);
1644 attr->u.blk = read_blk (abfd, &info_ptr, info_ptr_end, amt);
1645 if (attr->u.blk == NULL)
1646 return NULL;
1647 break;
1648 case DW_FORM_block1:
1649 amt = read_1_byte (abfd, &info_ptr, info_ptr_end);
1650 attr->u.blk = read_blk (abfd, &info_ptr, info_ptr_end, amt);
1651 if (attr->u.blk == NULL)
1652 return NULL;
1653 break;
1654 case DW_FORM_sdata:
1655 attr->u.sval = _bfd_safe_read_leb128 (abfd, &info_ptr,
1656 true, info_ptr_end);
1657 break;
1658
1659 case DW_FORM_rnglistx:
1660 case DW_FORM_loclistx:
1661 /* FIXME: Add support for these forms! */
1662 /* Fall through. */
1663 case DW_FORM_ref_udata:
1664 case DW_FORM_udata:
1665 attr->u.val = _bfd_safe_read_leb128 (abfd, &info_ptr,
1666 false, info_ptr_end);
1667 break;
1668 case DW_FORM_addrx:
1669 attr->u.val = _bfd_safe_read_leb128 (abfd, &info_ptr,
1670 false, info_ptr_end);
1671 if (unit->dwarf_addr_offset != 0)
1672 attr->u.val = read_indexed_address (attr->u.val, unit);
1673 break;
1674 case DW_FORM_indirect:
1675 form = _bfd_safe_read_leb128 (abfd, &info_ptr,
1676 false, info_ptr_end);
1677 if (form == DW_FORM_implicit_const)
1678 implicit_const = _bfd_safe_read_leb128 (abfd, &info_ptr,
1679 true, info_ptr_end);
1680 info_ptr = read_attribute_value (attr, form, implicit_const, unit,
1681 info_ptr, info_ptr_end);
1682 break;
1683 case DW_FORM_implicit_const:
1684 attr->form = DW_FORM_sdata;
1685 attr->u.sval = implicit_const;
1686 break;
1687 case DW_FORM_data16:
1688 /* This is really a "constant", but there is no way to store that
1689 so pretend it is a 16 byte block instead. */
1690 attr->u.blk = read_blk (abfd, &info_ptr, info_ptr_end, 16);
1691 if (attr->u.blk == NULL)
1692 return NULL;
1693 break;
1694
1695 default:
1696 _bfd_error_handler (_("DWARF error: invalid or unhandled FORM value: %#x"),
1697 form);
1698 bfd_set_error (bfd_error_bad_value);
1699 return NULL;
1700 }
1701 return info_ptr;
1702 }
1703
1704 /* Read an attribute described by an abbreviated attribute. */
1705
1706 static bfd_byte *
1707 read_attribute (struct attribute * attr,
1708 struct attr_abbrev * abbrev,
1709 struct comp_unit * unit,
1710 bfd_byte * info_ptr,
1711 bfd_byte * info_ptr_end)
1712 {
1713 attr->name = abbrev->name;
1714 info_ptr = read_attribute_value (attr, abbrev->form, abbrev->implicit_const,
1715 unit, info_ptr, info_ptr_end);
1716 return info_ptr;
1717 }
1718
1719 /* Return mangling style given LANG. */
1720
1721 static int
1722 mangle_style (int lang)
1723 {
1724 switch (lang)
1725 {
1726 case DW_LANG_Ada83:
1727 case DW_LANG_Ada95:
1728 return DMGL_GNAT;
1729
1730 case DW_LANG_C_plus_plus:
1731 case DW_LANG_C_plus_plus_03:
1732 case DW_LANG_C_plus_plus_11:
1733 case DW_LANG_C_plus_plus_14:
1734 return DMGL_GNU_V3;
1735
1736 case DW_LANG_Java:
1737 return DMGL_JAVA;
1738
1739 case DW_LANG_D:
1740 return DMGL_DLANG;
1741
1742 case DW_LANG_Rust:
1743 case DW_LANG_Rust_old:
1744 return DMGL_RUST;
1745
1746 default:
1747 return DMGL_AUTO;
1748
1749 case DW_LANG_C89:
1750 case DW_LANG_C:
1751 case DW_LANG_Cobol74:
1752 case DW_LANG_Cobol85:
1753 case DW_LANG_Fortran77:
1754 case DW_LANG_Pascal83:
1755 case DW_LANG_PLI:
1756 case DW_LANG_C99:
1757 case DW_LANG_UPC:
1758 case DW_LANG_C11:
1759 case DW_LANG_Mips_Assembler:
1760 case DW_LANG_Upc:
1761 case DW_LANG_HP_Basic91:
1762 case DW_LANG_HP_IMacro:
1763 case DW_LANG_HP_Assembler:
1764 return 0;
1765 }
1766 }
1767
1768 /* Source line information table routines. */
1769
1770 #define FILE_ALLOC_CHUNK 5
1771 #define DIR_ALLOC_CHUNK 5
1772
1773 struct line_info
1774 {
1775 struct line_info * prev_line;
1776 bfd_vma address;
1777 char * filename;
1778 unsigned int line;
1779 unsigned int column;
1780 unsigned int discriminator;
1781 unsigned char op_index;
1782 unsigned char end_sequence; /* End of (sequential) code sequence. */
1783 };
1784
1785 struct fileinfo
1786 {
1787 char * name;
1788 unsigned int dir;
1789 unsigned int time;
1790 unsigned int size;
1791 };
1792
1793 struct line_sequence
1794 {
1795 bfd_vma low_pc;
1796 struct line_sequence* prev_sequence;
1797 struct line_info* last_line; /* Largest VMA. */
1798 struct line_info** line_info_lookup;
1799 bfd_size_type num_lines;
1800 };
1801
1802 struct line_info_table
1803 {
1804 bfd * abfd;
1805 unsigned int num_files;
1806 unsigned int num_dirs;
1807 unsigned int num_sequences;
1808 bool use_dir_and_file_0;
1809 char * comp_dir;
1810 char ** dirs;
1811 struct fileinfo* files;
1812 struct line_sequence* sequences;
1813 struct line_info* lcl_head; /* Local head; used in 'add_line_info'. */
1814 };
1815
1816 /* Remember some information about each function. If the function is
1817 inlined (DW_TAG_inlined_subroutine) it may have two additional
1818 attributes, DW_AT_call_file and DW_AT_call_line, which specify the
1819 source code location where this function was inlined. */
1820
1821 struct funcinfo
1822 {
1823 /* Pointer to previous function in list of all functions. */
1824 struct funcinfo *prev_func;
1825 /* Pointer to function one scope higher. */
1826 struct funcinfo *caller_func;
1827 /* Source location file name where caller_func inlines this func. */
1828 char *caller_file;
1829 /* Source location file name. */
1830 char *file;
1831 /* Source location line number where caller_func inlines this func. */
1832 int caller_line;
1833 /* Source location line number. */
1834 int line;
1835 int tag;
1836 bool is_linkage;
1837 const char *name;
1838 struct arange arange;
1839 /* The offset of the funcinfo from the start of the unit. */
1840 uint64_t unit_offset;
1841 };
1842
1843 struct lookup_funcinfo
1844 {
1845 /* Function information corresponding to this lookup table entry. */
1846 struct funcinfo *funcinfo;
1847
1848 /* The lowest address for this specific function. */
1849 bfd_vma low_addr;
1850
1851 /* The highest address of this function before the lookup table is sorted.
1852 The highest address of all prior functions after the lookup table is
1853 sorted, which is used for binary search. */
1854 bfd_vma high_addr;
1855 /* Index of this function, used to ensure qsort is stable. */
1856 unsigned int idx;
1857 };
1858
1859 struct varinfo
1860 {
1861 /* Pointer to previous variable in list of all variables. */
1862 struct varinfo *prev_var;
1863 /* The offset of the varinfo from the start of the unit. */
1864 uint64_t unit_offset;
1865 /* Source location file name. */
1866 char *file;
1867 /* Source location line number. */
1868 int line;
1869 /* The type of this variable. */
1870 int tag;
1871 /* The name of the variable, if it has one. */
1872 const char *name;
1873 /* The address of the variable. */
1874 bfd_vma addr;
1875 /* Is this a stack variable? */
1876 bool stack;
1877 };
1878
1879 /* Return TRUE if NEW_LINE should sort after LINE. */
1880
1881 static inline bool
1882 new_line_sorts_after (struct line_info *new_line, struct line_info *line)
1883 {
1884 return (new_line->address > line->address
1885 || (new_line->address == line->address
1886 && new_line->op_index > line->op_index));
1887 }
1888
1889
1890 /* Adds a new entry to the line_info list in the line_info_table, ensuring
1891 that the list is sorted. Note that the line_info list is sorted from
1892 highest to lowest VMA (with possible duplicates); that is,
1893 line_info->prev_line always accesses an equal or smaller VMA. */
1894
1895 static bool
1896 add_line_info (struct line_info_table *table,
1897 bfd_vma address,
1898 unsigned char op_index,
1899 char *filename,
1900 unsigned int line,
1901 unsigned int column,
1902 unsigned int discriminator,
1903 int end_sequence)
1904 {
1905 size_t amt = sizeof (struct line_info);
1906 struct line_sequence* seq = table->sequences;
1907 struct line_info* info = (struct line_info *) bfd_alloc (table->abfd, amt);
1908
1909 if (info == NULL)
1910 return false;
1911
1912 /* Set member data of 'info'. */
1913 info->prev_line = NULL;
1914 info->address = address;
1915 info->op_index = op_index;
1916 info->line = line;
1917 info->column = column;
1918 info->discriminator = discriminator;
1919 info->end_sequence = end_sequence;
1920
1921 if (filename && filename[0])
1922 {
1923 info->filename = (char *) bfd_alloc (table->abfd, strlen (filename) + 1);
1924 if (info->filename == NULL)
1925 return false;
1926 strcpy (info->filename, filename);
1927 }
1928 else
1929 info->filename = NULL;
1930
1931 /* Find the correct location for 'info'. Normally we will receive
1932 new line_info data 1) in order and 2) with increasing VMAs.
1933 However some compilers break the rules (cf. decode_line_info) and
1934 so we include some heuristics for quickly finding the correct
1935 location for 'info'. In particular, these heuristics optimize for
1936 the common case in which the VMA sequence that we receive is a
1937 list of locally sorted VMAs such as
1938 p...z a...j (where a < j < p < z)
1939
1940 Note: table->lcl_head is used to head an *actual* or *possible*
1941 sub-sequence within the list (such as a...j) that is not directly
1942 headed by table->last_line
1943
1944 Note: we may receive duplicate entries from 'decode_line_info'. */
1945
1946 if (seq
1947 && seq->last_line->address == address
1948 && seq->last_line->op_index == op_index
1949 && seq->last_line->end_sequence == end_sequence)
1950 {
1951 /* We only keep the last entry with the same address and end
1952 sequence. See PR ld/4986. */
1953 if (table->lcl_head == seq->last_line)
1954 table->lcl_head = info;
1955 info->prev_line = seq->last_line->prev_line;
1956 seq->last_line = info;
1957 }
1958 else if (!seq || seq->last_line->end_sequence)
1959 {
1960 /* Start a new line sequence. */
1961 amt = sizeof (struct line_sequence);
1962 seq = (struct line_sequence *) bfd_malloc (amt);
1963 if (seq == NULL)
1964 return false;
1965 seq->low_pc = address;
1966 seq->prev_sequence = table->sequences;
1967 seq->last_line = info;
1968 table->lcl_head = info;
1969 table->sequences = seq;
1970 table->num_sequences++;
1971 }
1972 else if (info->end_sequence
1973 || new_line_sorts_after (info, seq->last_line))
1974 {
1975 /* Normal case: add 'info' to the beginning of the current sequence. */
1976 info->prev_line = seq->last_line;
1977 seq->last_line = info;
1978
1979 /* lcl_head: initialize to head a *possible* sequence at the end. */
1980 if (!table->lcl_head)
1981 table->lcl_head = info;
1982 }
1983 else if (!new_line_sorts_after (info, table->lcl_head)
1984 && (!table->lcl_head->prev_line
1985 || new_line_sorts_after (info, table->lcl_head->prev_line)))
1986 {
1987 /* Abnormal but easy: lcl_head is the head of 'info'. */
1988 info->prev_line = table->lcl_head->prev_line;
1989 table->lcl_head->prev_line = info;
1990 }
1991 else
1992 {
1993 /* Abnormal and hard: Neither 'last_line' nor 'lcl_head'
1994 are valid heads for 'info'. Reset 'lcl_head'. */
1995 struct line_info* li2 = seq->last_line; /* Always non-NULL. */
1996 struct line_info* li1 = li2->prev_line;
1997
1998 while (li1)
1999 {
2000 if (!new_line_sorts_after (info, li2)
2001 && new_line_sorts_after (info, li1))
2002 break;
2003
2004 li2 = li1; /* always non-NULL */
2005 li1 = li1->prev_line;
2006 }
2007 table->lcl_head = li2;
2008 info->prev_line = table->lcl_head->prev_line;
2009 table->lcl_head->prev_line = info;
2010 if (address < seq->low_pc)
2011 seq->low_pc = address;
2012 }
2013 return true;
2014 }
2015
2016 /* Extract a fully qualified filename from a line info table.
2017 The returned string has been malloc'ed and it is the caller's
2018 responsibility to free it. */
2019
2020 static char *
2021 concat_filename (struct line_info_table *table, unsigned int file)
2022 {
2023 char *filename;
2024
2025 /* Pre DWARF-5 entry 0 in the directory and filename tables was not used.
2026 So in order to save space in the tables used here the info for, eg
2027 directory 1 is stored in slot 0 of the directory table, directory 2
2028 in slot 1 and so on.
2029
2030 Starting with DWARF-5 the 0'th entry is used so there is a one to one
2031 mapping between DWARF slots and internal table entries. */
2032 if (! table->use_dir_and_file_0)
2033 {
2034 /* Pre DWARF-5, FILE == 0 means unknown. */
2035 if (file == 0)
2036 return strdup ("<unknown>");
2037 -- file;
2038 }
2039
2040 if (table == NULL || file >= table->num_files)
2041 {
2042 _bfd_error_handler
2043 (_("DWARF error: mangled line number section (bad file number)"));
2044 return strdup ("<unknown>");
2045 }
2046
2047 filename = table->files[file].name;
2048
2049 if (filename == NULL)
2050 return strdup ("<unknown>");
2051
2052 if (!IS_ABSOLUTE_PATH (filename))
2053 {
2054 char *dir_name = NULL;
2055 char *subdir_name = NULL;
2056 char *name;
2057 size_t len;
2058 unsigned int dir = table->files[file].dir;
2059
2060 if (!table->use_dir_and_file_0)
2061 --dir;
2062 /* Wrapping from 0 to -1u above gives the intended result with
2063 the test below of leaving subdir_name NULL for pre-DWARF5 dir
2064 of 0. */
2065 /* PR 17512: file: 0317e960, file: 7f3d2e4b. */
2066 if (dir < table->num_dirs)
2067 subdir_name = table->dirs[dir];
2068
2069 if (!subdir_name || !IS_ABSOLUTE_PATH (subdir_name))
2070 dir_name = table->comp_dir;
2071
2072 if (!dir_name)
2073 {
2074 dir_name = subdir_name;
2075 subdir_name = NULL;
2076 }
2077
2078 if (!dir_name)
2079 return strdup (filename);
2080
2081 len = strlen (dir_name) + strlen (filename) + 2;
2082
2083 if (subdir_name)
2084 {
2085 len += strlen (subdir_name) + 1;
2086 name = (char *) bfd_malloc (len);
2087 if (name)
2088 sprintf (name, "%s/%s/%s", dir_name, subdir_name, filename);
2089 }
2090 else
2091 {
2092 name = (char *) bfd_malloc (len);
2093 if (name)
2094 sprintf (name, "%s/%s", dir_name, filename);
2095 }
2096
2097 return name;
2098 }
2099
2100 return strdup (filename);
2101 }
2102
2103 /* Number of bits in a bfd_vma. */
2104 #define VMA_BITS (8 * sizeof (bfd_vma))
2105
2106 /* Check whether [low1, high1) can be combined with [low2, high2),
2107 i.e., they touch or overlap. */
2108
2109 static bool
2110 ranges_overlap (bfd_vma low1,
2111 bfd_vma high1,
2112 bfd_vma low2,
2113 bfd_vma high2)
2114 {
2115 if (low1 == low2 || high1 == high2)
2116 return true;
2117
2118 /* Sort so that low1 is below low2. */
2119 if (low1 > low2)
2120 {
2121 bfd_vma tmp;
2122
2123 tmp = low1;
2124 low1 = low2;
2125 low2 = tmp;
2126
2127 tmp = high1;
2128 high1 = high2;
2129 high2 = tmp;
2130 }
2131
2132 /* We touch iff low2 == high1.
2133 We overlap iff low2 is within [low1, high1). */
2134 return low2 <= high1;
2135 }
2136
2137 /* Insert an address range in the trie mapping addresses to compilation units.
2138 Will return the new trie node (usually the same as is being sent in, but
2139 in case of a leaf-to-interior conversion, or expansion of a leaf, it may be
2140 different), or NULL on failure. */
2141
2142 static struct trie_node *
2143 insert_arange_in_trie (bfd *abfd,
2144 struct trie_node *trie,
2145 bfd_vma trie_pc,
2146 unsigned int trie_pc_bits,
2147 struct comp_unit *unit,
2148 bfd_vma low_pc,
2149 bfd_vma high_pc)
2150 {
2151 bfd_vma clamped_low_pc, clamped_high_pc;
2152 int ch, from_ch, to_ch;
2153 bool is_full_leaf = false;
2154 bool splitting_leaf_will_help = false;
2155
2156 /* See if we can extend any of the existing ranges. This merging
2157 isn't perfect (if merging opens up the possibility of merging two existing
2158 ranges, we won't find them), but it takes the majority of the cases. */
2159 if (trie->num_room_in_leaf > 0)
2160 {
2161 struct trie_leaf *leaf = (struct trie_leaf *) trie;
2162 unsigned int i;
2163
2164 for (i = 0; i < leaf->num_stored_in_leaf; ++i)
2165 {
2166 if (leaf->ranges[i].unit == unit
2167 && ranges_overlap (low_pc, high_pc,
2168 leaf->ranges[i].low_pc,
2169 leaf->ranges[i].high_pc))
2170 {
2171 if (low_pc < leaf->ranges[i].low_pc)
2172 leaf->ranges[i].low_pc = low_pc;
2173 if (high_pc > leaf->ranges[i].high_pc)
2174 leaf->ranges[i].high_pc = high_pc;
2175 return trie;
2176 }
2177 }
2178
2179 is_full_leaf = leaf->num_stored_in_leaf == trie->num_room_in_leaf;
2180
2181 if (is_full_leaf && trie_pc_bits < VMA_BITS)
2182 {
2183 /* See if we have at least one leaf that does _not_ cover the
2184 entire bucket, so that splitting will actually reduce the number
2185 of elements in at least one of the child nodes. (For simplicity,
2186 we don't test the range we're inserting, but it will be counted
2187 on the next insertion where we're full, if any.) */
2188 bfd_vma bucket_high_pc =
2189 trie_pc + ((bfd_vma) -1 >> trie_pc_bits); /* Inclusive. */
2190 for (i = 0; i < leaf->num_stored_in_leaf; ++i)
2191 {
2192 if (leaf->ranges[i].low_pc > trie_pc
2193 || leaf->ranges[i].high_pc <= bucket_high_pc)
2194 {
2195 splitting_leaf_will_help = true;
2196 break;
2197 }
2198 }
2199 }
2200 }
2201
2202 /* If we're a leaf with no more room and we're _not_ at the bottom,
2203 convert to an interior node. */
2204 if (is_full_leaf && splitting_leaf_will_help)
2205 {
2206 const struct trie_leaf *leaf = (struct trie_leaf *) trie;
2207 unsigned int i;
2208
2209 trie = bfd_zalloc (abfd, sizeof (struct trie_interior));
2210 if (!trie)
2211 return NULL;
2212 is_full_leaf = false;
2213
2214 /* TODO: If we wanted to save a little more memory at the cost of
2215 complexity, we could have reused the old leaf node as one of the
2216 children of the new interior node, instead of throwing it away. */
2217 for (i = 0; i < leaf->num_stored_in_leaf; ++i)
2218 {
2219 if (!insert_arange_in_trie (abfd, trie, trie_pc, trie_pc_bits,
2220 leaf->ranges[i].unit, leaf->ranges[i].low_pc,
2221 leaf->ranges[i].high_pc))
2222 return NULL;
2223 }
2224 }
2225
2226 /* If we're a leaf with no more room and we _are_ at the bottom
2227 (or splitting it won't help), we have no choice but to just
2228 make it larger. */
2229 if (is_full_leaf)
2230 {
2231 const struct trie_leaf *leaf = (struct trie_leaf *) trie;
2232 unsigned int new_room_in_leaf = trie->num_room_in_leaf * 2;
2233 struct trie_leaf *new_leaf;
2234 size_t amt = sizeof (*leaf) + new_room_in_leaf * sizeof (leaf->ranges[0]);
2235 new_leaf = bfd_zalloc (abfd, amt);
2236 new_leaf->head.num_room_in_leaf = new_room_in_leaf;
2237 new_leaf->num_stored_in_leaf = leaf->num_stored_in_leaf;
2238
2239 memcpy (new_leaf->ranges,
2240 leaf->ranges,
2241 leaf->num_stored_in_leaf * sizeof (leaf->ranges[0]));
2242 trie = &new_leaf->head;
2243 is_full_leaf = false;
2244
2245 /* Now the insert below will go through. */
2246 }
2247
2248 /* If we're a leaf (now with room), we can just insert at the end. */
2249 if (trie->num_room_in_leaf > 0)
2250 {
2251 struct trie_leaf *leaf = (struct trie_leaf *) trie;
2252
2253 unsigned int i = leaf->num_stored_in_leaf++;
2254 leaf->ranges[i].unit = unit;
2255 leaf->ranges[i].low_pc = low_pc;
2256 leaf->ranges[i].high_pc = high_pc;
2257 return trie;
2258 }
2259
2260 /* Now we are definitely an interior node, so recurse into all
2261 the relevant buckets. */
2262
2263 /* Clamp the range to the current trie bucket. */
2264 clamped_low_pc = low_pc;
2265 clamped_high_pc = high_pc;
2266 if (trie_pc_bits > 0)
2267 {
2268 bfd_vma bucket_high_pc =
2269 trie_pc + ((bfd_vma) -1 >> trie_pc_bits); /* Inclusive. */
2270 if (clamped_low_pc < trie_pc)
2271 clamped_low_pc = trie_pc;
2272 if (clamped_high_pc > bucket_high_pc)
2273 clamped_high_pc = bucket_high_pc;
2274 }
2275
2276 /* Insert the ranges in all buckets that it spans. */
2277 from_ch = (clamped_low_pc >> (VMA_BITS - trie_pc_bits - 8)) & 0xff;
2278 to_ch = ((clamped_high_pc - 1) >> (VMA_BITS - trie_pc_bits - 8)) & 0xff;
2279 for (ch = from_ch; ch <= to_ch; ++ch)
2280 {
2281 struct trie_interior *interior = (struct trie_interior *) trie;
2282 struct trie_node *child = interior->children[ch];
2283
2284 if (child == NULL)
2285 {
2286 child = alloc_trie_leaf (abfd);
2287 if (!child)
2288 return NULL;
2289 }
2290 bfd_vma bucket = (bfd_vma) ch << (VMA_BITS - trie_pc_bits - 8);
2291 child = insert_arange_in_trie (abfd,
2292 child,
2293 trie_pc + bucket,
2294 trie_pc_bits + 8,
2295 unit,
2296 low_pc,
2297 high_pc);
2298 if (!child)
2299 return NULL;
2300
2301 interior->children[ch] = child;
2302 }
2303
2304 return trie;
2305 }
2306
2307 static bool
2308 arange_add (struct comp_unit *unit, struct arange *first_arange,
2309 struct trie_node **trie_root, bfd_vma low_pc, bfd_vma high_pc)
2310 {
2311 struct arange *arange;
2312
2313 /* Ignore empty ranges. */
2314 if (low_pc == high_pc)
2315 return true;
2316
2317 if (trie_root != NULL)
2318 {
2319 *trie_root = insert_arange_in_trie (unit->file->bfd_ptr,
2320 *trie_root,
2321 0,
2322 0,
2323 unit,
2324 low_pc,
2325 high_pc);
2326 if (*trie_root == NULL)
2327 return false;
2328 }
2329
2330 /* If the first arange is empty, use it. */
2331 if (first_arange->high == 0)
2332 {
2333 first_arange->low = low_pc;
2334 first_arange->high = high_pc;
2335 return true;
2336 }
2337
2338 /* Next see if we can cheaply extend an existing range. */
2339 arange = first_arange;
2340 do
2341 {
2342 if (low_pc == arange->high)
2343 {
2344 arange->high = high_pc;
2345 return true;
2346 }
2347 if (high_pc == arange->low)
2348 {
2349 arange->low = low_pc;
2350 return true;
2351 }
2352 arange = arange->next;
2353 }
2354 while (arange);
2355
2356 /* Need to allocate a new arange and insert it into the arange list.
2357 Order isn't significant, so just insert after the first arange. */
2358 arange = (struct arange *) bfd_alloc (unit->abfd, sizeof (*arange));
2359 if (arange == NULL)
2360 return false;
2361 arange->low = low_pc;
2362 arange->high = high_pc;
2363 arange->next = first_arange->next;
2364 first_arange->next = arange;
2365 return true;
2366 }
2367
2368 /* Compare function for line sequences. */
2369
2370 static int
2371 compare_sequences (const void* a, const void* b)
2372 {
2373 const struct line_sequence* seq1 = a;
2374 const struct line_sequence* seq2 = b;
2375
2376 /* Sort by low_pc as the primary key. */
2377 if (seq1->low_pc < seq2->low_pc)
2378 return -1;
2379 if (seq1->low_pc > seq2->low_pc)
2380 return 1;
2381
2382 /* If low_pc values are equal, sort in reverse order of
2383 high_pc, so that the largest region comes first. */
2384 if (seq1->last_line->address < seq2->last_line->address)
2385 return 1;
2386 if (seq1->last_line->address > seq2->last_line->address)
2387 return -1;
2388
2389 if (seq1->last_line->op_index < seq2->last_line->op_index)
2390 return 1;
2391 if (seq1->last_line->op_index > seq2->last_line->op_index)
2392 return -1;
2393
2394 /* num_lines is initially an index, to make the sort stable. */
2395 if (seq1->num_lines < seq2->num_lines)
2396 return -1;
2397 if (seq1->num_lines > seq2->num_lines)
2398 return 1;
2399 return 0;
2400 }
2401
2402 /* Construct the line information table for quick lookup. */
2403
2404 static bool
2405 build_line_info_table (struct line_info_table * table,
2406 struct line_sequence * seq)
2407 {
2408 size_t amt;
2409 struct line_info **line_info_lookup;
2410 struct line_info *each_line;
2411 unsigned int num_lines;
2412 unsigned int line_index;
2413
2414 if (seq->line_info_lookup != NULL)
2415 return true;
2416
2417 /* Count the number of line information entries. We could do this while
2418 scanning the debug information, but some entries may be added via
2419 lcl_head without having a sequence handy to increment the number of
2420 lines. */
2421 num_lines = 0;
2422 for (each_line = seq->last_line; each_line; each_line = each_line->prev_line)
2423 num_lines++;
2424
2425 seq->num_lines = num_lines;
2426 if (num_lines == 0)
2427 return true;
2428
2429 /* Allocate space for the line information lookup table. */
2430 amt = sizeof (struct line_info*) * num_lines;
2431 line_info_lookup = (struct line_info**) bfd_alloc (table->abfd, amt);
2432 seq->line_info_lookup = line_info_lookup;
2433 if (line_info_lookup == NULL)
2434 return false;
2435
2436 /* Create the line information lookup table. */
2437 line_index = num_lines;
2438 for (each_line = seq->last_line; each_line; each_line = each_line->prev_line)
2439 line_info_lookup[--line_index] = each_line;
2440
2441 BFD_ASSERT (line_index == 0);
2442 return true;
2443 }
2444
2445 /* Sort the line sequences for quick lookup. */
2446
2447 static bool
2448 sort_line_sequences (struct line_info_table* table)
2449 {
2450 size_t amt;
2451 struct line_sequence *sequences;
2452 struct line_sequence *seq;
2453 unsigned int n = 0;
2454 unsigned int num_sequences = table->num_sequences;
2455 bfd_vma last_high_pc;
2456
2457 if (num_sequences == 0)
2458 return true;
2459
2460 /* Allocate space for an array of sequences. */
2461 amt = sizeof (struct line_sequence) * num_sequences;
2462 sequences = (struct line_sequence *) bfd_alloc (table->abfd, amt);
2463 if (sequences == NULL)
2464 return false;
2465
2466 /* Copy the linked list into the array, freeing the original nodes. */
2467 seq = table->sequences;
2468 for (n = 0; n < num_sequences; n++)
2469 {
2470 struct line_sequence* last_seq = seq;
2471
2472 BFD_ASSERT (seq);
2473 sequences[n].low_pc = seq->low_pc;
2474 sequences[n].prev_sequence = NULL;
2475 sequences[n].last_line = seq->last_line;
2476 sequences[n].line_info_lookup = NULL;
2477 sequences[n].num_lines = n;
2478 seq = seq->prev_sequence;
2479 free (last_seq);
2480 }
2481 BFD_ASSERT (seq == NULL);
2482
2483 qsort (sequences, n, sizeof (struct line_sequence), compare_sequences);
2484
2485 /* Make the list binary-searchable by trimming overlapping entries
2486 and removing nested entries. */
2487 num_sequences = 1;
2488 last_high_pc = sequences[0].last_line->address;
2489 for (n = 1; n < table->num_sequences; n++)
2490 {
2491 if (sequences[n].low_pc < last_high_pc)
2492 {
2493 if (sequences[n].last_line->address <= last_high_pc)
2494 /* Skip nested entries. */
2495 continue;
2496
2497 /* Trim overlapping entries. */
2498 sequences[n].low_pc = last_high_pc;
2499 }
2500 last_high_pc = sequences[n].last_line->address;
2501 if (n > num_sequences)
2502 {
2503 /* Close up the gap. */
2504 sequences[num_sequences].low_pc = sequences[n].low_pc;
2505 sequences[num_sequences].last_line = sequences[n].last_line;
2506 }
2507 num_sequences++;
2508 }
2509
2510 table->sequences = sequences;
2511 table->num_sequences = num_sequences;
2512 return true;
2513 }
2514
2515 /* Add directory to TABLE. CUR_DIR memory ownership is taken by TABLE. */
2516
2517 static bool
2518 line_info_add_include_dir (struct line_info_table *table, char *cur_dir)
2519 {
2520 if ((table->num_dirs % DIR_ALLOC_CHUNK) == 0)
2521 {
2522 char **tmp;
2523 size_t amt;
2524
2525 amt = table->num_dirs + DIR_ALLOC_CHUNK;
2526 amt *= sizeof (char *);
2527
2528 tmp = (char **) bfd_realloc (table->dirs, amt);
2529 if (tmp == NULL)
2530 return false;
2531 table->dirs = tmp;
2532 }
2533
2534 table->dirs[table->num_dirs++] = cur_dir;
2535 return true;
2536 }
2537
2538 static bool
2539 line_info_add_include_dir_stub (struct line_info_table *table, char *cur_dir,
2540 unsigned int dir ATTRIBUTE_UNUSED,
2541 unsigned int xtime ATTRIBUTE_UNUSED,
2542 unsigned int size ATTRIBUTE_UNUSED)
2543 {
2544 return line_info_add_include_dir (table, cur_dir);
2545 }
2546
2547 /* Add file to TABLE. CUR_FILE memory ownership is taken by TABLE. */
2548
2549 static bool
2550 line_info_add_file_name (struct line_info_table *table, char *cur_file,
2551 unsigned int dir, unsigned int xtime,
2552 unsigned int size)
2553 {
2554 if ((table->num_files % FILE_ALLOC_CHUNK) == 0)
2555 {
2556 struct fileinfo *tmp;
2557 size_t amt;
2558
2559 amt = table->num_files + FILE_ALLOC_CHUNK;
2560 amt *= sizeof (struct fileinfo);
2561
2562 tmp = (struct fileinfo *) bfd_realloc (table->files, amt);
2563 if (tmp == NULL)
2564 return false;
2565 table->files = tmp;
2566 }
2567
2568 table->files[table->num_files].name = cur_file;
2569 table->files[table->num_files].dir = dir;
2570 table->files[table->num_files].time = xtime;
2571 table->files[table->num_files].size = size;
2572 table->num_files++;
2573 return true;
2574 }
2575
2576 /* Read directory or file name entry format, starting with byte of
2577 format count entries, ULEB128 pairs of entry formats, ULEB128 of
2578 entries count and the entries themselves in the described entry
2579 format. */
2580
2581 static bool
2582 read_formatted_entries (struct comp_unit *unit, bfd_byte **bufp,
2583 bfd_byte *buf_end, struct line_info_table *table,
2584 bool (*callback) (struct line_info_table *table,
2585 char *cur_file,
2586 unsigned int dir,
2587 unsigned int time,
2588 unsigned int size))
2589 {
2590 bfd *abfd = unit->abfd;
2591 bfd_byte format_count, formati;
2592 bfd_vma data_count, datai;
2593 bfd_byte *buf = *bufp;
2594 bfd_byte *format_header_data;
2595
2596 format_count = read_1_byte (abfd, &buf, buf_end);
2597 format_header_data = buf;
2598 for (formati = 0; formati < format_count; formati++)
2599 {
2600 _bfd_safe_read_leb128 (abfd, &buf, false, buf_end);
2601 _bfd_safe_read_leb128 (abfd, &buf, false, buf_end);
2602 }
2603
2604 data_count = _bfd_safe_read_leb128 (abfd, &buf, false, buf_end);
2605 if (format_count == 0 && data_count != 0)
2606 {
2607 _bfd_error_handler (_("DWARF error: zero format count"));
2608 bfd_set_error (bfd_error_bad_value);
2609 return false;
2610 }
2611
2612 /* PR 22210. Paranoia check. Don't bother running the loop
2613 if we know that we are going to run out of buffer. */
2614 if (data_count > (bfd_vma) (buf_end - buf))
2615 {
2616 _bfd_error_handler
2617 (_("DWARF error: data count (%" PRIx64 ") larger than buffer size"),
2618 (uint64_t) data_count);
2619 bfd_set_error (bfd_error_bad_value);
2620 return false;
2621 }
2622
2623 for (datai = 0; datai < data_count; datai++)
2624 {
2625 bfd_byte *format = format_header_data;
2626 struct fileinfo fe;
2627
2628 memset (&fe, 0, sizeof fe);
2629 for (formati = 0; formati < format_count; formati++)
2630 {
2631 bfd_vma content_type, form;
2632 char *string_trash;
2633 char **stringp = &string_trash;
2634 unsigned int uint_trash, *uintp = &uint_trash;
2635 struct attribute attr;
2636
2637 content_type = _bfd_safe_read_leb128 (abfd, &format, false, buf_end);
2638 switch (content_type)
2639 {
2640 case DW_LNCT_path:
2641 stringp = &fe.name;
2642 break;
2643 case DW_LNCT_directory_index:
2644 uintp = &fe.dir;
2645 break;
2646 case DW_LNCT_timestamp:
2647 uintp = &fe.time;
2648 break;
2649 case DW_LNCT_size:
2650 uintp = &fe.size;
2651 break;
2652 case DW_LNCT_MD5:
2653 break;
2654 default:
2655 _bfd_error_handler
2656 (_("DWARF error: unknown format content type %" PRIu64),
2657 (uint64_t) content_type);
2658 bfd_set_error (bfd_error_bad_value);
2659 return false;
2660 }
2661
2662 form = _bfd_safe_read_leb128 (abfd, &format, false, buf_end);
2663 buf = read_attribute_value (&attr, form, 0, unit, buf, buf_end);
2664 if (buf == NULL)
2665 return false;
2666 switch (form)
2667 {
2668 case DW_FORM_string:
2669 case DW_FORM_line_strp:
2670 case DW_FORM_strx:
2671 case DW_FORM_strx1:
2672 case DW_FORM_strx2:
2673 case DW_FORM_strx3:
2674 case DW_FORM_strx4:
2675 *stringp = attr.u.str;
2676 break;
2677
2678 case DW_FORM_data1:
2679 case DW_FORM_data2:
2680 case DW_FORM_data4:
2681 case DW_FORM_data8:
2682 case DW_FORM_udata:
2683 *uintp = attr.u.val;
2684 break;
2685
2686 case DW_FORM_data16:
2687 /* MD5 data is in the attr.blk, but we are ignoring those. */
2688 break;
2689 }
2690 }
2691
2692 if (!callback (table, fe.name, fe.dir, fe.time, fe.size))
2693 return false;
2694 }
2695
2696 *bufp = buf;
2697 return true;
2698 }
2699
2700 /* Decode the line number information for UNIT. */
2701
2702 static struct line_info_table*
2703 decode_line_info (struct comp_unit *unit)
2704 {
2705 bfd *abfd = unit->abfd;
2706 struct dwarf2_debug *stash = unit->stash;
2707 struct dwarf2_debug_file *file = unit->file;
2708 struct line_info_table* table;
2709 bfd_byte *line_ptr;
2710 bfd_byte *line_end;
2711 struct line_head lh;
2712 unsigned int i, offset_size;
2713 char *cur_file, *cur_dir;
2714 unsigned char op_code, extended_op, adj_opcode;
2715 unsigned int exop_len;
2716 size_t amt;
2717
2718 if (unit->line_offset == 0 && file->line_table)
2719 return file->line_table;
2720
2721 if (! read_section (abfd, &stash->debug_sections[debug_line],
2722 file->syms, unit->line_offset,
2723 &file->dwarf_line_buffer, &file->dwarf_line_size))
2724 return NULL;
2725
2726 if (file->dwarf_line_size < 16)
2727 {
2728 _bfd_error_handler
2729 (_("DWARF error: line info section is too small (%" PRId64 ")"),
2730 (int64_t) file->dwarf_line_size);
2731 bfd_set_error (bfd_error_bad_value);
2732 return NULL;
2733 }
2734 line_ptr = file->dwarf_line_buffer + unit->line_offset;
2735 line_end = file->dwarf_line_buffer + file->dwarf_line_size;
2736
2737 /* Read in the prologue. */
2738 lh.total_length = read_4_bytes (abfd, &line_ptr, line_end);
2739 offset_size = 4;
2740 if (lh.total_length == 0xffffffff)
2741 {
2742 lh.total_length = read_8_bytes (abfd, &line_ptr, line_end);
2743 offset_size = 8;
2744 }
2745 else if (lh.total_length == 0 && unit->addr_size == 8)
2746 {
2747 /* Handle (non-standard) 64-bit DWARF2 formats. */
2748 lh.total_length = read_4_bytes (abfd, &line_ptr, line_end);
2749 offset_size = 8;
2750 }
2751
2752 if (lh.total_length > (size_t) (line_end - line_ptr))
2753 {
2754 _bfd_error_handler
2755 /* xgettext: c-format */
2756 (_("DWARF error: line info data is bigger (%#" PRIx64 ")"
2757 " than the space remaining in the section (%#lx)"),
2758 (uint64_t) lh.total_length, (unsigned long) (line_end - line_ptr));
2759 bfd_set_error (bfd_error_bad_value);
2760 return NULL;
2761 }
2762
2763 line_end = line_ptr + lh.total_length;
2764
2765 lh.version = read_2_bytes (abfd, &line_ptr, line_end);
2766 if (lh.version < 2 || lh.version > 5)
2767 {
2768 _bfd_error_handler
2769 (_("DWARF error: unhandled .debug_line version %d"), lh.version);
2770 bfd_set_error (bfd_error_bad_value);
2771 return NULL;
2772 }
2773
2774 if (line_ptr + offset_size + (lh.version >= 5 ? 8 : (lh.version >= 4 ? 6 : 5))
2775 >= line_end)
2776 {
2777 _bfd_error_handler
2778 (_("DWARF error: ran out of room reading prologue"));
2779 bfd_set_error (bfd_error_bad_value);
2780 return NULL;
2781 }
2782
2783 if (lh.version >= 5)
2784 {
2785 unsigned int segment_selector_size;
2786
2787 /* Skip address size. */
2788 read_1_byte (abfd, &line_ptr, line_end);
2789
2790 segment_selector_size = read_1_byte (abfd, &line_ptr, line_end);
2791 if (segment_selector_size != 0)
2792 {
2793 _bfd_error_handler
2794 (_("DWARF error: line info unsupported segment selector size %u"),
2795 segment_selector_size);
2796 bfd_set_error (bfd_error_bad_value);
2797 return NULL;
2798 }
2799 }
2800
2801 if (offset_size == 4)
2802 lh.prologue_length = read_4_bytes (abfd, &line_ptr, line_end);
2803 else
2804 lh.prologue_length = read_8_bytes (abfd, &line_ptr, line_end);
2805
2806 lh.minimum_instruction_length = read_1_byte (abfd, &line_ptr, line_end);
2807
2808 if (lh.version >= 4)
2809 lh.maximum_ops_per_insn = read_1_byte (abfd, &line_ptr, line_end);
2810 else
2811 lh.maximum_ops_per_insn = 1;
2812
2813 if (lh.maximum_ops_per_insn == 0)
2814 {
2815 _bfd_error_handler
2816 (_("DWARF error: invalid maximum operations per instruction"));
2817 bfd_set_error (bfd_error_bad_value);
2818 return NULL;
2819 }
2820
2821 lh.default_is_stmt = read_1_byte (abfd, &line_ptr, line_end);
2822 lh.line_base = read_1_signed_byte (abfd, &line_ptr, line_end);
2823 lh.line_range = read_1_byte (abfd, &line_ptr, line_end);
2824 lh.opcode_base = read_1_byte (abfd, &line_ptr, line_end);
2825
2826 if (line_ptr + (lh.opcode_base - 1) >= line_end)
2827 {
2828 _bfd_error_handler (_("DWARF error: ran out of room reading opcodes"));
2829 bfd_set_error (bfd_error_bad_value);
2830 return NULL;
2831 }
2832
2833 amt = lh.opcode_base * sizeof (unsigned char);
2834 lh.standard_opcode_lengths = (unsigned char *) bfd_alloc (abfd, amt);
2835
2836 lh.standard_opcode_lengths[0] = 1;
2837
2838 for (i = 1; i < lh.opcode_base; ++i)
2839 lh.standard_opcode_lengths[i] = read_1_byte (abfd, &line_ptr, line_end);
2840
2841 amt = sizeof (struct line_info_table);
2842 table = (struct line_info_table *) bfd_alloc (abfd, amt);
2843 if (table == NULL)
2844 return NULL;
2845 table->abfd = abfd;
2846 table->comp_dir = unit->comp_dir;
2847
2848 table->num_files = 0;
2849 table->files = NULL;
2850
2851 table->num_dirs = 0;
2852 table->dirs = NULL;
2853
2854 table->num_sequences = 0;
2855 table->sequences = NULL;
2856
2857 table->lcl_head = NULL;
2858
2859 if (lh.version >= 5)
2860 {
2861 /* Read directory table. */
2862 if (!read_formatted_entries (unit, &line_ptr, line_end, table,
2863 line_info_add_include_dir_stub))
2864 goto fail;
2865
2866 /* Read file name table. */
2867 if (!read_formatted_entries (unit, &line_ptr, line_end, table,
2868 line_info_add_file_name))
2869 goto fail;
2870 table->use_dir_and_file_0 = true;
2871 }
2872 else
2873 {
2874 /* Read directory table. */
2875 while ((cur_dir = read_string (&line_ptr, line_end)) != NULL)
2876 {
2877 if (!line_info_add_include_dir (table, cur_dir))
2878 goto fail;
2879 }
2880
2881 /* Read file name table. */
2882 while ((cur_file = read_string (&line_ptr, line_end)) != NULL)
2883 {
2884 unsigned int dir, xtime, size;
2885
2886 dir = _bfd_safe_read_leb128 (abfd, &line_ptr, false, line_end);
2887 xtime = _bfd_safe_read_leb128 (abfd, &line_ptr, false, line_end);
2888 size = _bfd_safe_read_leb128 (abfd, &line_ptr, false, line_end);
2889
2890 if (!line_info_add_file_name (table, cur_file, dir, xtime, size))
2891 goto fail;
2892 }
2893 table->use_dir_and_file_0 = false;
2894 }
2895
2896 /* Read the statement sequences until there's nothing left. */
2897 while (line_ptr < line_end)
2898 {
2899 /* State machine registers. */
2900 bfd_vma address = 0;
2901 unsigned char op_index = 0;
2902 char * filename = NULL;
2903 unsigned int line = 1;
2904 unsigned int column = 0;
2905 unsigned int discriminator = 0;
2906 int is_stmt = lh.default_is_stmt;
2907 int end_sequence = 0;
2908 unsigned int dir, xtime, size;
2909 /* eraxxon@alumni.rice.edu: Against the DWARF2 specs, some
2910 compilers generate address sequences that are wildly out of
2911 order using DW_LNE_set_address (e.g. Intel C++ 6.0 compiler
2912 for ia64-Linux). Thus, to determine the low and high
2913 address, we must compare on every DW_LNS_copy, etc. */
2914 bfd_vma low_pc = (bfd_vma) -1;
2915 bfd_vma high_pc = 0;
2916
2917 if (table->num_files)
2918 {
2919 if (table->use_dir_and_file_0)
2920 filename = concat_filename (table, 0);
2921 else
2922 filename = concat_filename (table, 1);
2923 }
2924
2925 /* Decode the table. */
2926 while (!end_sequence && line_ptr < line_end)
2927 {
2928 op_code = read_1_byte (abfd, &line_ptr, line_end);
2929
2930 if (op_code >= lh.opcode_base)
2931 {
2932 /* Special operand. */
2933 adj_opcode = op_code - lh.opcode_base;
2934 if (lh.line_range == 0)
2935 goto line_fail;
2936 if (lh.maximum_ops_per_insn == 1)
2937 address += (adj_opcode / lh.line_range
2938 * lh.minimum_instruction_length);
2939 else
2940 {
2941 address += ((op_index + adj_opcode / lh.line_range)
2942 / lh.maximum_ops_per_insn
2943 * lh.minimum_instruction_length);
2944 op_index = ((op_index + adj_opcode / lh.line_range)
2945 % lh.maximum_ops_per_insn);
2946 }
2947 line += lh.line_base + (adj_opcode % lh.line_range);
2948 /* Append row to matrix using current values. */
2949 if (!add_line_info (table, address, op_index, filename,
2950 line, column, discriminator, 0))
2951 goto line_fail;
2952 discriminator = 0;
2953 if (address < low_pc)
2954 low_pc = address;
2955 if (address > high_pc)
2956 high_pc = address;
2957 }
2958 else switch (op_code)
2959 {
2960 case DW_LNS_extended_op:
2961 exop_len = _bfd_safe_read_leb128 (abfd, &line_ptr,
2962 false, line_end);
2963 extended_op = read_1_byte (abfd, &line_ptr, line_end);
2964
2965 switch (extended_op)
2966 {
2967 case DW_LNE_end_sequence:
2968 end_sequence = 1;
2969 if (!add_line_info (table, address, op_index, filename, line,
2970 column, discriminator, end_sequence))
2971 goto line_fail;
2972 discriminator = 0;
2973 if (address < low_pc)
2974 low_pc = address;
2975 if (address > high_pc)
2976 high_pc = address;
2977 if (!arange_add (unit, &unit->arange, &unit->file->trie_root,
2978 low_pc, high_pc))
2979 goto line_fail;
2980 break;
2981 case DW_LNE_set_address:
2982 address = read_address (unit, &line_ptr, line_end);
2983 op_index = 0;
2984 break;
2985 case DW_LNE_define_file:
2986 cur_file = read_string (&line_ptr, line_end);
2987 dir = _bfd_safe_read_leb128 (abfd, &line_ptr,
2988 false, line_end);
2989 xtime = _bfd_safe_read_leb128 (abfd, &line_ptr,
2990 false, line_end);
2991 size = _bfd_safe_read_leb128 (abfd, &line_ptr,
2992 false, line_end);
2993 if (!line_info_add_file_name (table, cur_file, dir,
2994 xtime, size))
2995 goto line_fail;
2996 break;
2997 case DW_LNE_set_discriminator:
2998 discriminator = _bfd_safe_read_leb128 (abfd, &line_ptr,
2999 false, line_end);
3000 break;
3001 case DW_LNE_HP_source_file_correlation:
3002 line_ptr += exop_len - 1;
3003 break;
3004 default:
3005 _bfd_error_handler
3006 (_("DWARF error: mangled line number section"));
3007 bfd_set_error (bfd_error_bad_value);
3008 line_fail:
3009 free (filename);
3010 goto fail;
3011 }
3012 break;
3013 case DW_LNS_copy:
3014 if (!add_line_info (table, address, op_index,
3015 filename, line, column, discriminator, 0))
3016 goto line_fail;
3017 discriminator = 0;
3018 if (address < low_pc)
3019 low_pc = address;
3020 if (address > high_pc)
3021 high_pc = address;
3022 break;
3023 case DW_LNS_advance_pc:
3024 if (lh.maximum_ops_per_insn == 1)
3025 address += (lh.minimum_instruction_length
3026 * _bfd_safe_read_leb128 (abfd, &line_ptr,
3027 false, line_end));
3028 else
3029 {
3030 bfd_vma adjust = _bfd_safe_read_leb128 (abfd, &line_ptr,
3031 false, line_end);
3032 address = ((op_index + adjust) / lh.maximum_ops_per_insn
3033 * lh.minimum_instruction_length);
3034 op_index = (op_index + adjust) % lh.maximum_ops_per_insn;
3035 }
3036 break;
3037 case DW_LNS_advance_line:
3038 line += _bfd_safe_read_leb128 (abfd, &line_ptr,
3039 true, line_end);
3040 break;
3041 case DW_LNS_set_file:
3042 {
3043 unsigned int filenum;
3044
3045 /* The file and directory tables are 0
3046 based, the references are 1 based. */
3047 filenum = _bfd_safe_read_leb128 (abfd, &line_ptr,
3048 false, line_end);
3049 free (filename);
3050 filename = concat_filename (table, filenum);
3051 break;
3052 }
3053 case DW_LNS_set_column:
3054 column = _bfd_safe_read_leb128 (abfd, &line_ptr,
3055 false, line_end);
3056 break;
3057 case DW_LNS_negate_stmt:
3058 is_stmt = (!is_stmt);
3059 break;
3060 case DW_LNS_set_basic_block:
3061 break;
3062 case DW_LNS_const_add_pc:
3063 if (lh.line_range == 0)
3064 goto line_fail;
3065 if (lh.maximum_ops_per_insn == 1)
3066 address += (lh.minimum_instruction_length
3067 * ((255 - lh.opcode_base) / lh.line_range));
3068 else
3069 {
3070 bfd_vma adjust = ((255 - lh.opcode_base) / lh.line_range);
3071 address += (lh.minimum_instruction_length
3072 * ((op_index + adjust)
3073 / lh.maximum_ops_per_insn));
3074 op_index = (op_index + adjust) % lh.maximum_ops_per_insn;
3075 }
3076 break;
3077 case DW_LNS_fixed_advance_pc:
3078 address += read_2_bytes (abfd, &line_ptr, line_end);
3079 op_index = 0;
3080 break;
3081 default:
3082 /* Unknown standard opcode, ignore it. */
3083 for (i = 0; i < lh.standard_opcode_lengths[op_code]; i++)
3084 (void) _bfd_safe_read_leb128 (abfd, &line_ptr,
3085 false, line_end);
3086 break;
3087 }
3088 }
3089
3090 free (filename);
3091 }
3092
3093 if (unit->line_offset == 0)
3094 file->line_table = table;
3095 if (sort_line_sequences (table))
3096 return table;
3097
3098 fail:
3099 while (table->sequences != NULL)
3100 {
3101 struct line_sequence* seq = table->sequences;
3102 table->sequences = table->sequences->prev_sequence;
3103 free (seq);
3104 }
3105 free (table->files);
3106 free (table->dirs);
3107 return NULL;
3108 }
3109
3110 /* If ADDR is within TABLE set the output parameters and return TRUE,
3111 otherwise set *FILENAME_PTR to NULL and return FALSE.
3112 The parameters FILENAME_PTR, LINENUMBER_PTR and DISCRIMINATOR_PTR
3113 are pointers to the objects to be filled in. */
3114
3115 static bool
3116 lookup_address_in_line_info_table (struct line_info_table *table,
3117 bfd_vma addr,
3118 const char **filename_ptr,
3119 unsigned int *linenumber_ptr,
3120 unsigned int *discriminator_ptr)
3121 {
3122 struct line_sequence *seq = NULL;
3123 struct line_info *info;
3124 int low, high, mid;
3125
3126 /* Binary search the array of sequences. */
3127 low = 0;
3128 high = table->num_sequences;
3129 while (low < high)
3130 {
3131 mid = (low + high) / 2;
3132 seq = &table->sequences[mid];
3133 if (addr < seq->low_pc)
3134 high = mid;
3135 else if (addr >= seq->last_line->address)
3136 low = mid + 1;
3137 else
3138 break;
3139 }
3140
3141 /* Check for a valid sequence. */
3142 if (!seq || addr < seq->low_pc || addr >= seq->last_line->address)
3143 goto fail;
3144
3145 if (!build_line_info_table (table, seq))
3146 goto fail;
3147
3148 /* Binary search the array of line information. */
3149 low = 0;
3150 high = seq->num_lines;
3151 info = NULL;
3152 while (low < high)
3153 {
3154 mid = (low + high) / 2;
3155 info = seq->line_info_lookup[mid];
3156 if (addr < info->address)
3157 high = mid;
3158 else if (addr >= seq->line_info_lookup[mid + 1]->address)
3159 low = mid + 1;
3160 else
3161 break;
3162 }
3163
3164 /* Check for a valid line information entry. */
3165 if (info
3166 && addr >= info->address
3167 && addr < seq->line_info_lookup[mid + 1]->address
3168 && !(info->end_sequence || info == seq->last_line))
3169 {
3170 *filename_ptr = info->filename;
3171 *linenumber_ptr = info->line;
3172 if (discriminator_ptr)
3173 *discriminator_ptr = info->discriminator;
3174 return true;
3175 }
3176
3177 fail:
3178 *filename_ptr = NULL;
3179 return false;
3180 }
3181
3182 /* Read in the .debug_ranges section for future reference. */
3183
3184 static bool
3185 read_debug_ranges (struct comp_unit * unit)
3186 {
3187 struct dwarf2_debug *stash = unit->stash;
3188 struct dwarf2_debug_file *file = unit->file;
3189
3190 return read_section (unit->abfd, &stash->debug_sections[debug_ranges],
3191 file->syms, 0,
3192 &file->dwarf_ranges_buffer, &file->dwarf_ranges_size);
3193 }
3194
3195 /* Read in the .debug_rnglists section for future reference. */
3196
3197 static bool
3198 read_debug_rnglists (struct comp_unit * unit)
3199 {
3200 struct dwarf2_debug *stash = unit->stash;
3201 struct dwarf2_debug_file *file = unit->file;
3202
3203 return read_section (unit->abfd, &stash->debug_sections[debug_rnglists],
3204 file->syms, 0,
3205 &file->dwarf_rnglists_buffer, &file->dwarf_rnglists_size);
3206 }
3207
3208 /* Function table functions. */
3209
3210 static int
3211 compare_lookup_funcinfos (const void * a, const void * b)
3212 {
3213 const struct lookup_funcinfo * lookup1 = a;
3214 const struct lookup_funcinfo * lookup2 = b;
3215
3216 if (lookup1->low_addr < lookup2->low_addr)
3217 return -1;
3218 if (lookup1->low_addr > lookup2->low_addr)
3219 return 1;
3220 if (lookup1->high_addr < lookup2->high_addr)
3221 return -1;
3222 if (lookup1->high_addr > lookup2->high_addr)
3223 return 1;
3224
3225 if (lookup1->idx < lookup2->idx)
3226 return -1;
3227 if (lookup1->idx > lookup2->idx)
3228 return 1;
3229 return 0;
3230 }
3231
3232 static bool
3233 build_lookup_funcinfo_table (struct comp_unit * unit)
3234 {
3235 struct lookup_funcinfo *lookup_funcinfo_table = unit->lookup_funcinfo_table;
3236 unsigned int number_of_functions = unit->number_of_functions;
3237 struct funcinfo *each;
3238 struct lookup_funcinfo *entry;
3239 size_t func_index;
3240 struct arange *range;
3241 bfd_vma low_addr, high_addr;
3242
3243 if (lookup_funcinfo_table || number_of_functions == 0)
3244 return true;
3245
3246 /* Create the function info lookup table. */
3247 lookup_funcinfo_table = (struct lookup_funcinfo *)
3248 bfd_malloc (number_of_functions * sizeof (struct lookup_funcinfo));
3249 if (lookup_funcinfo_table == NULL)
3250 return false;
3251
3252 /* Populate the function info lookup table. */
3253 func_index = number_of_functions;
3254 for (each = unit->function_table; each; each = each->prev_func)
3255 {
3256 entry = &lookup_funcinfo_table[--func_index];
3257 entry->funcinfo = each;
3258 entry->idx = func_index;
3259
3260 /* Calculate the lowest and highest address for this function entry. */
3261 low_addr = entry->funcinfo->arange.low;
3262 high_addr = entry->funcinfo->arange.high;
3263
3264 for (range = entry->funcinfo->arange.next; range; range = range->next)
3265 {
3266 if (range->low < low_addr)
3267 low_addr = range->low;
3268 if (range->high > high_addr)
3269 high_addr = range->high;
3270 }
3271
3272 entry->low_addr = low_addr;
3273 entry->high_addr = high_addr;
3274 }
3275
3276 BFD_ASSERT (func_index == 0);
3277
3278 /* Sort the function by address. */
3279 qsort (lookup_funcinfo_table,
3280 number_of_functions,
3281 sizeof (struct lookup_funcinfo),
3282 compare_lookup_funcinfos);
3283
3284 /* Calculate the high watermark for each function in the lookup table. */
3285 high_addr = lookup_funcinfo_table[0].high_addr;
3286 for (func_index = 1; func_index < number_of_functions; func_index++)
3287 {
3288 entry = &lookup_funcinfo_table[func_index];
3289 if (entry->high_addr > high_addr)
3290 high_addr = entry->high_addr;
3291 else
3292 entry->high_addr = high_addr;
3293 }
3294
3295 unit->lookup_funcinfo_table = lookup_funcinfo_table;
3296 return true;
3297 }
3298
3299 /* If ADDR is within UNIT's function tables, set FUNCTION_PTR, and return
3300 TRUE. Note that we need to find the function that has the smallest range
3301 that contains ADDR, to handle inlined functions without depending upon
3302 them being ordered in TABLE by increasing range. */
3303
3304 static bool
3305 lookup_address_in_function_table (struct comp_unit *unit,
3306 bfd_vma addr,
3307 struct funcinfo **function_ptr)
3308 {
3309 unsigned int number_of_functions = unit->number_of_functions;
3310 struct lookup_funcinfo* lookup_funcinfo = NULL;
3311 struct funcinfo* funcinfo = NULL;
3312 struct funcinfo* best_fit = NULL;
3313 bfd_vma best_fit_len = (bfd_vma) -1;
3314 bfd_size_type low, high, mid, first;
3315 struct arange *arange;
3316
3317 if (number_of_functions == 0)
3318 return false;
3319
3320 if (!build_lookup_funcinfo_table (unit))
3321 return false;
3322
3323 if (unit->lookup_funcinfo_table[number_of_functions - 1].high_addr < addr)
3324 return false;
3325
3326 /* Find the first function in the lookup table which may contain the
3327 specified address. */
3328 low = 0;
3329 high = number_of_functions;
3330 first = high;
3331 while (low < high)
3332 {
3333 mid = (low + high) / 2;
3334 lookup_funcinfo = &unit->lookup_funcinfo_table[mid];
3335 if (addr < lookup_funcinfo->low_addr)
3336 high = mid;
3337 else if (addr >= lookup_funcinfo->high_addr)
3338 low = mid + 1;
3339 else
3340 high = first = mid;
3341 }
3342
3343 /* Find the 'best' match for the address. The prior algorithm defined the
3344 best match as the function with the smallest address range containing
3345 the specified address. This definition should probably be changed to the
3346 innermost inline routine containing the address, but right now we want
3347 to get the same results we did before. */
3348 while (first < number_of_functions)
3349 {
3350 if (addr < unit->lookup_funcinfo_table[first].low_addr)
3351 break;
3352 funcinfo = unit->lookup_funcinfo_table[first].funcinfo;
3353
3354 for (arange = &funcinfo->arange; arange; arange = arange->next)
3355 {
3356 if (addr < arange->low || addr >= arange->high)
3357 continue;
3358
3359 if (arange->high - arange->low < best_fit_len
3360 /* The following comparison is designed to return the same
3361 match as the previous algorithm for routines which have the
3362 same best fit length. */
3363 || (arange->high - arange->low == best_fit_len
3364 && funcinfo > best_fit))
3365 {
3366 best_fit = funcinfo;
3367 best_fit_len = arange->high - arange->low;
3368 }
3369 }
3370
3371 first++;
3372 }
3373
3374 if (!best_fit)
3375 return false;
3376
3377 *function_ptr = best_fit;
3378 return true;
3379 }
3380
3381 /* If SYM at ADDR is within function table of UNIT, set FILENAME_PTR
3382 and LINENUMBER_PTR, and return TRUE. */
3383
3384 static bool
3385 lookup_symbol_in_function_table (struct comp_unit *unit,
3386 asymbol *sym,
3387 bfd_vma addr,
3388 const char **filename_ptr,
3389 unsigned int *linenumber_ptr)
3390 {
3391 struct funcinfo* each;
3392 struct funcinfo* best_fit = NULL;
3393 bfd_vma best_fit_len = (bfd_vma) -1;
3394 struct arange *arange;
3395 const char *name = bfd_asymbol_name (sym);
3396
3397 for (each = unit->function_table; each; each = each->prev_func)
3398 for (arange = &each->arange; arange; arange = arange->next)
3399 if (addr >= arange->low
3400 && addr < arange->high
3401 && arange->high - arange->low < best_fit_len
3402 && each->file
3403 && each->name
3404 && strstr (name, each->name) != NULL)
3405 {
3406 best_fit = each;
3407 best_fit_len = arange->high - arange->low;
3408 }
3409
3410 if (best_fit)
3411 {
3412 *filename_ptr = best_fit->file;
3413 *linenumber_ptr = best_fit->line;
3414 return true;
3415 }
3416
3417 return false;
3418 }
3419
3420 /* Variable table functions. */
3421
3422 /* If SYM is within variable table of UNIT, set FILENAME_PTR and
3423 LINENUMBER_PTR, and return TRUE. */
3424
3425 static bool
3426 lookup_symbol_in_variable_table (struct comp_unit *unit,
3427 asymbol *sym,
3428 bfd_vma addr,
3429 const char **filename_ptr,
3430 unsigned int *linenumber_ptr)
3431 {
3432 struct varinfo* each;
3433 const char *name = bfd_asymbol_name (sym);
3434
3435 for (each = unit->variable_table; each; each = each->prev_var)
3436 if (each->addr == addr
3437 && !each->stack
3438 && each->file != NULL
3439 && each->name != NULL
3440 && strstr (name, each->name) != NULL)
3441 break;
3442
3443 if (each)
3444 {
3445 *filename_ptr = each->file;
3446 *linenumber_ptr = each->line;
3447 return true;
3448 }
3449
3450 return false;
3451 }
3452
3453 static struct comp_unit *stash_comp_unit (struct dwarf2_debug *,
3454 struct dwarf2_debug_file *);
3455 static bool comp_unit_maybe_decode_line_info (struct comp_unit *);
3456
3457 static bool
3458 find_abstract_instance (struct comp_unit *unit,
3459 struct attribute *attr_ptr,
3460 unsigned int recur_count,
3461 const char **pname,
3462 bool *is_linkage,
3463 char **filename_ptr,
3464 int *linenumber_ptr)
3465 {
3466 bfd *abfd = unit->abfd;
3467 bfd_byte *info_ptr = NULL;
3468 bfd_byte *info_ptr_end;
3469 unsigned int abbrev_number, i;
3470 struct abbrev_info *abbrev;
3471 uint64_t die_ref = attr_ptr->u.val;
3472 struct attribute attr;
3473
3474 if (recur_count == 100)
3475 {
3476 _bfd_error_handler
3477 (_("DWARF error: abstract instance recursion detected"));
3478 bfd_set_error (bfd_error_bad_value);
3479 return false;
3480 }
3481
3482 /* DW_FORM_ref_addr can reference an entry in a different CU. It
3483 is an offset from the .debug_info section, not the current CU. */
3484 if (attr_ptr->form == DW_FORM_ref_addr)
3485 {
3486 /* We only support DW_FORM_ref_addr within the same file, so
3487 any relocations should be resolved already. Check this by
3488 testing for a zero die_ref; There can't be a valid reference
3489 to the header of a .debug_info section.
3490 DW_FORM_ref_addr is an offset relative to .debug_info.
3491 Normally when using the GNU linker this is accomplished by
3492 emitting a symbolic reference to a label, because .debug_info
3493 sections are linked at zero. When there are multiple section
3494 groups containing .debug_info, as there might be in a
3495 relocatable object file, it would be reasonable to assume that
3496 a symbolic reference to a label in any .debug_info section
3497 might be used. Since we lay out multiple .debug_info
3498 sections at non-zero VMAs (see place_sections), and read
3499 them contiguously into dwarf_info_buffer, that means the
3500 reference is relative to dwarf_info_buffer. */
3501 size_t total;
3502
3503 info_ptr = unit->file->dwarf_info_buffer;
3504 info_ptr_end = info_ptr + unit->file->dwarf_info_size;
3505 total = info_ptr_end - info_ptr;
3506 if (!die_ref)
3507 return true;
3508 else if (die_ref >= total)
3509 {
3510 _bfd_error_handler
3511 (_("DWARF error: invalid abstract instance DIE ref"));
3512 bfd_set_error (bfd_error_bad_value);
3513 return false;
3514 }
3515 info_ptr += die_ref;
3516 }
3517 else if (attr_ptr->form == DW_FORM_GNU_ref_alt)
3518 {
3519 bool first_time = unit->stash->alt.dwarf_info_buffer == NULL;
3520
3521 info_ptr = read_alt_indirect_ref (unit, die_ref);
3522 if (first_time)
3523 unit->stash->alt.info_ptr = unit->stash->alt.dwarf_info_buffer;
3524 if (info_ptr == NULL)
3525 {
3526 _bfd_error_handler
3527 (_("DWARF error: unable to read alt ref %" PRIu64),
3528 (uint64_t) die_ref);
3529 bfd_set_error (bfd_error_bad_value);
3530 return false;
3531 }
3532 info_ptr_end = (unit->stash->alt.dwarf_info_buffer
3533 + unit->stash->alt.dwarf_info_size);
3534 if (unit->stash->alt.all_comp_units)
3535 unit = unit->stash->alt.all_comp_units;
3536 }
3537
3538 if (attr_ptr->form == DW_FORM_ref_addr
3539 || attr_ptr->form == DW_FORM_GNU_ref_alt)
3540 {
3541 /* Now find the CU containing this pointer. */
3542 if (info_ptr >= unit->info_ptr_unit && info_ptr < unit->end_ptr)
3543 info_ptr_end = unit->end_ptr;
3544 else
3545 {
3546 /* Check other CUs to see if they contain the abbrev. */
3547 struct comp_unit *u = NULL;
3548 struct addr_range range = { info_ptr, info_ptr };
3549 splay_tree_node v = splay_tree_lookup (unit->file->comp_unit_tree,
3550 (splay_tree_key)&range);
3551 if (v != NULL)
3552 u = (struct comp_unit *)v->value;
3553
3554 if (attr_ptr->form == DW_FORM_ref_addr)
3555 while (u == NULL)
3556 {
3557 u = stash_comp_unit (unit->stash, &unit->stash->f);
3558 if (u == NULL)
3559 break;
3560 if (info_ptr >= u->info_ptr_unit && info_ptr < u->end_ptr)
3561 break;
3562 u = NULL;
3563 }
3564
3565 if (attr_ptr->form == DW_FORM_GNU_ref_alt)
3566 while (u == NULL)
3567 {
3568 u = stash_comp_unit (unit->stash, &unit->stash->alt);
3569 if (u == NULL)
3570 break;
3571 if (info_ptr >= u->info_ptr_unit && info_ptr < u->end_ptr)
3572 break;
3573 u = NULL;
3574 }
3575
3576 if (u == NULL)
3577 {
3578 _bfd_error_handler
3579 (_("DWARF error: unable to locate abstract instance DIE ref %"
3580 PRIu64), (uint64_t) die_ref);
3581 bfd_set_error (bfd_error_bad_value);
3582 return false;
3583 }
3584 unit = u;
3585 info_ptr_end = unit->end_ptr;
3586 }
3587 }
3588 else
3589 {
3590 /* DW_FORM_ref1, DW_FORM_ref2, DW_FORM_ref4, DW_FORM_ref8 or
3591 DW_FORM_ref_udata. These are all references relative to the
3592 start of the current CU. */
3593 size_t total;
3594
3595 info_ptr = unit->info_ptr_unit;
3596 info_ptr_end = unit->end_ptr;
3597 total = info_ptr_end - info_ptr;
3598 if (!die_ref || die_ref >= total)
3599 {
3600 _bfd_error_handler
3601 (_("DWARF error: invalid abstract instance DIE ref"));
3602 bfd_set_error (bfd_error_bad_value);
3603 return false;
3604 }
3605 info_ptr += die_ref;
3606 }
3607
3608 abbrev_number = _bfd_safe_read_leb128 (abfd, &info_ptr,
3609 false, info_ptr_end);
3610 if (abbrev_number)
3611 {
3612 abbrev = lookup_abbrev (abbrev_number, unit->abbrevs);
3613 if (! abbrev)
3614 {
3615 _bfd_error_handler
3616 (_("DWARF error: could not find abbrev number %u"), abbrev_number);
3617 bfd_set_error (bfd_error_bad_value);
3618 return false;
3619 }
3620 else
3621 {
3622 for (i = 0; i < abbrev->num_attrs; ++i)
3623 {
3624 info_ptr = read_attribute (&attr, &abbrev->attrs[i], unit,
3625 info_ptr, info_ptr_end);
3626 if (info_ptr == NULL)
3627 break;
3628 switch (attr.name)
3629 {
3630 case DW_AT_name:
3631 /* Prefer DW_AT_MIPS_linkage_name or DW_AT_linkage_name
3632 over DW_AT_name. */
3633 if (*pname == NULL && is_str_form (&attr))
3634 {
3635 *pname = attr.u.str;
3636 if (mangle_style (unit->lang) == 0)
3637 *is_linkage = true;
3638 }
3639 break;
3640 case DW_AT_specification:
3641 if (is_int_form (&attr)
3642 && !find_abstract_instance (unit, &attr, recur_count + 1,
3643 pname, is_linkage,
3644 filename_ptr, linenumber_ptr))
3645 return false;
3646 break;
3647 case DW_AT_linkage_name:
3648 case DW_AT_MIPS_linkage_name:
3649 /* PR 16949: Corrupt debug info can place
3650 non-string forms into these attributes. */
3651 if (is_str_form (&attr))
3652 {
3653 *pname = attr.u.str;
3654 *is_linkage = true;
3655 }
3656 break;
3657 case DW_AT_decl_file:
3658 if (!comp_unit_maybe_decode_line_info (unit))
3659 return false;
3660 if (is_int_form (&attr))
3661 {
3662 free (*filename_ptr);
3663 *filename_ptr = concat_filename (unit->line_table,
3664 attr.u.val);
3665 }
3666 break;
3667 case DW_AT_decl_line:
3668 if (is_int_form (&attr))
3669 *linenumber_ptr = attr.u.val;
3670 break;
3671 default:
3672 break;
3673 }
3674 }
3675 }
3676 }
3677 return true;
3678 }
3679
3680 static bool
3681 read_ranges (struct comp_unit *unit, struct arange *arange,
3682 struct trie_node **trie_root, uint64_t offset)
3683 {
3684 bfd_byte *ranges_ptr;
3685 bfd_byte *ranges_end;
3686 bfd_vma base_address = unit->base_address;
3687
3688 if (! unit->file->dwarf_ranges_buffer)
3689 {
3690 if (! read_debug_ranges (unit))
3691 return false;
3692 }
3693
3694 if (offset > unit->file->dwarf_ranges_size)
3695 return false;
3696 ranges_ptr = unit->file->dwarf_ranges_buffer + offset;
3697 ranges_end = unit->file->dwarf_ranges_buffer + unit->file->dwarf_ranges_size;
3698
3699 for (;;)
3700 {
3701 bfd_vma low_pc;
3702 bfd_vma high_pc;
3703
3704 /* PR 17512: file: 62cada7d. */
3705 if (2u * unit->addr_size > (size_t) (ranges_end - ranges_ptr))
3706 return false;
3707
3708 low_pc = read_address (unit, &ranges_ptr, ranges_end);
3709 high_pc = read_address (unit, &ranges_ptr, ranges_end);
3710
3711 if (low_pc == 0 && high_pc == 0)
3712 break;
3713 if (low_pc == (bfd_vma) -1 && high_pc != (bfd_vma) -1)
3714 base_address = high_pc;
3715 else
3716 {
3717 if (!arange_add (unit, arange, trie_root,
3718 base_address + low_pc, base_address + high_pc))
3719 return false;
3720 }
3721 }
3722 return true;
3723 }
3724
3725 static bool
3726 read_rnglists (struct comp_unit *unit, struct arange *arange,
3727 struct trie_node **trie_root, uint64_t offset)
3728 {
3729 bfd_byte *rngs_ptr;
3730 bfd_byte *rngs_end;
3731 bfd_vma base_address = unit->base_address;
3732 bfd_vma low_pc;
3733 bfd_vma high_pc;
3734 bfd *abfd = unit->abfd;
3735
3736 if (! unit->file->dwarf_rnglists_buffer)
3737 {
3738 if (! read_debug_rnglists (unit))
3739 return false;
3740 }
3741
3742 rngs_ptr = unit->file->dwarf_rnglists_buffer + offset;
3743 if (rngs_ptr < unit->file->dwarf_rnglists_buffer)
3744 return false;
3745 rngs_end = unit->file->dwarf_rnglists_buffer;
3746 rngs_end += unit->file->dwarf_rnglists_size;
3747
3748 for (;;)
3749 {
3750 enum dwarf_range_list_entry rlet;
3751
3752 if (rngs_ptr >= rngs_end)
3753 return false;
3754
3755 rlet = read_1_byte (abfd, &rngs_ptr, rngs_end);
3756
3757 switch (rlet)
3758 {
3759 case DW_RLE_end_of_list:
3760 return true;
3761
3762 case DW_RLE_base_address:
3763 if (unit->addr_size > (size_t) (rngs_end - rngs_ptr))
3764 return false;
3765 base_address = read_address (unit, &rngs_ptr, rngs_end);
3766 continue;
3767
3768 case DW_RLE_start_length:
3769 if (unit->addr_size > (size_t) (rngs_end - rngs_ptr))
3770 return false;
3771 low_pc = read_address (unit, &rngs_ptr, rngs_end);
3772 high_pc = low_pc;
3773 high_pc += _bfd_safe_read_leb128 (abfd, &rngs_ptr,
3774 false, rngs_end);
3775 break;
3776
3777 case DW_RLE_offset_pair:
3778 low_pc = base_address;
3779 low_pc += _bfd_safe_read_leb128 (abfd, &rngs_ptr,
3780 false, rngs_end);
3781 high_pc = base_address;
3782 high_pc += _bfd_safe_read_leb128 (abfd, &rngs_ptr,
3783 false, rngs_end);
3784 break;
3785
3786 case DW_RLE_start_end:
3787 if (2u * unit->addr_size > (size_t) (rngs_end - rngs_ptr))
3788 return false;
3789 low_pc = read_address (unit, &rngs_ptr, rngs_end);
3790 high_pc = read_address (unit, &rngs_ptr, rngs_end);
3791 break;
3792
3793 /* TODO x-variants need .debug_addr support used for split-dwarf. */
3794 case DW_RLE_base_addressx:
3795 case DW_RLE_startx_endx:
3796 case DW_RLE_startx_length:
3797 default:
3798 return false;
3799 }
3800
3801 if (!arange_add (unit, arange, trie_root, low_pc, high_pc))
3802 return false;
3803 }
3804 }
3805
3806 static bool
3807 read_rangelist (struct comp_unit *unit, struct arange *arange,
3808 struct trie_node **trie_root, uint64_t offset)
3809 {
3810 if (unit->version <= 4)
3811 return read_ranges (unit, arange, trie_root, offset);
3812 else
3813 return read_rnglists (unit, arange, trie_root, offset);
3814 }
3815
3816 static struct funcinfo *
3817 lookup_func_by_offset (uint64_t offset, struct funcinfo * table)
3818 {
3819 for (; table != NULL; table = table->prev_func)
3820 if (table->unit_offset == offset)
3821 return table;
3822 return NULL;
3823 }
3824
3825 static struct varinfo *
3826 lookup_var_by_offset (uint64_t offset, struct varinfo * table)
3827 {
3828 while (table)
3829 {
3830 if (table->unit_offset == offset)
3831 return table;
3832 table = table->prev_var;
3833 }
3834
3835 return NULL;
3836 }
3837
3838
3839 /* DWARF2 Compilation unit functions. */
3840
3841 static struct funcinfo *
3842 reverse_funcinfo_list (struct funcinfo *head)
3843 {
3844 struct funcinfo *rhead;
3845 struct funcinfo *temp;
3846
3847 for (rhead = NULL; head; head = temp)
3848 {
3849 temp = head->prev_func;
3850 head->prev_func = rhead;
3851 rhead = head;
3852 }
3853 return rhead;
3854 }
3855
3856 static struct varinfo *
3857 reverse_varinfo_list (struct varinfo *head)
3858 {
3859 struct varinfo *rhead;
3860 struct varinfo *temp;
3861
3862 for (rhead = NULL; head; head = temp)
3863 {
3864 temp = head->prev_var;
3865 head->prev_var = rhead;
3866 rhead = head;
3867 }
3868 return rhead;
3869 }
3870
3871 /* Scan over each die in a comp. unit looking for functions to add
3872 to the function table and variables to the variable table. */
3873
3874 static bool
3875 scan_unit_for_symbols (struct comp_unit *unit)
3876 {
3877 bfd *abfd = unit->abfd;
3878 bfd_byte *info_ptr = unit->first_child_die_ptr;
3879 bfd_byte *info_ptr_end = unit->end_ptr;
3880 int nesting_level = 0;
3881 struct nest_funcinfo
3882 {
3883 struct funcinfo *func;
3884 } *nested_funcs;
3885 int nested_funcs_size;
3886 struct funcinfo *last_func;
3887 struct varinfo *last_var;
3888
3889 /* Maintain a stack of in-scope functions and inlined functions, which we
3890 can use to set the caller_func field. */
3891 nested_funcs_size = 32;
3892 nested_funcs = (struct nest_funcinfo *)
3893 bfd_malloc (nested_funcs_size * sizeof (*nested_funcs));
3894 if (nested_funcs == NULL)
3895 return false;
3896 nested_funcs[nesting_level].func = 0;
3897
3898 /* PR 27484: We must scan the DIEs twice. The first time we look for
3899 function and variable tags and accumulate them into their respective
3900 tables. The second time through we process the attributes of the
3901 functions/variables and augment the table entries. */
3902 while (nesting_level >= 0)
3903 {
3904 unsigned int abbrev_number, i;
3905 struct abbrev_info *abbrev;
3906 struct funcinfo *func;
3907 struct varinfo *var;
3908 uint64_t current_offset;
3909
3910 /* PR 17512: file: 9f405d9d. */
3911 if (info_ptr >= info_ptr_end)
3912 goto fail;
3913
3914 current_offset = info_ptr - unit->info_ptr_unit;
3915 abbrev_number = _bfd_safe_read_leb128 (abfd, &info_ptr,
3916 false, info_ptr_end);
3917 if (abbrev_number == 0)
3918 {
3919 nesting_level--;
3920 continue;
3921 }
3922
3923 abbrev = lookup_abbrev (abbrev_number, unit->abbrevs);
3924 if (! abbrev)
3925 {
3926 static unsigned int previous_failed_abbrev = -1U;
3927
3928 /* Avoid multiple reports of the same missing abbrev. */
3929 if (abbrev_number != previous_failed_abbrev)
3930 {
3931 _bfd_error_handler
3932 (_("DWARF error: could not find abbrev number %u"),
3933 abbrev_number);
3934 previous_failed_abbrev = abbrev_number;
3935 }
3936 bfd_set_error (bfd_error_bad_value);
3937 goto fail;
3938 }
3939
3940 if (abbrev->tag == DW_TAG_subprogram
3941 || abbrev->tag == DW_TAG_entry_point
3942 || abbrev->tag == DW_TAG_inlined_subroutine)
3943 {
3944 size_t amt = sizeof (struct funcinfo);
3945
3946 var = NULL;
3947 func = (struct funcinfo *) bfd_zalloc (abfd, amt);
3948 if (func == NULL)
3949 goto fail;
3950 func->tag = abbrev->tag;
3951 func->prev_func = unit->function_table;
3952 func->unit_offset = current_offset;
3953 unit->function_table = func;
3954 unit->number_of_functions++;
3955 BFD_ASSERT (!unit->cached);
3956
3957 if (func->tag == DW_TAG_inlined_subroutine)
3958 for (i = nesting_level; i-- != 0; )
3959 if (nested_funcs[i].func)
3960 {
3961 func->caller_func = nested_funcs[i].func;
3962 break;
3963 }
3964 nested_funcs[nesting_level].func = func;
3965 }
3966 else
3967 {
3968 func = NULL;
3969 if (abbrev->tag == DW_TAG_variable
3970 || abbrev->tag == DW_TAG_member)
3971 {
3972 size_t amt = sizeof (struct varinfo);
3973
3974 var = (struct varinfo *) bfd_zalloc (abfd, amt);
3975 if (var == NULL)
3976 goto fail;
3977 var->tag = abbrev->tag;
3978 var->stack = true;
3979 var->prev_var = unit->variable_table;
3980 unit->variable_table = var;
3981 var->unit_offset = current_offset;
3982 /* PR 18205: Missing debug information can cause this
3983 var to be attached to an already cached unit. */
3984 }
3985 else
3986 var = NULL;
3987
3988 /* No inline function in scope at this nesting level. */
3989 nested_funcs[nesting_level].func = 0;
3990 }
3991
3992 for (i = 0; i < abbrev->num_attrs; ++i)
3993 {
3994 struct attribute attr;
3995
3996 info_ptr = read_attribute (&attr, &abbrev->attrs[i],
3997 unit, info_ptr, info_ptr_end);
3998 if (info_ptr == NULL)
3999 goto fail;
4000 }
4001
4002 if (abbrev->has_children)
4003 {
4004 nesting_level++;
4005
4006 if (nesting_level >= nested_funcs_size)
4007 {
4008 struct nest_funcinfo *tmp;
4009
4010 nested_funcs_size *= 2;
4011 tmp = (struct nest_funcinfo *)
4012 bfd_realloc (nested_funcs,
4013 nested_funcs_size * sizeof (*nested_funcs));
4014 if (tmp == NULL)
4015 goto fail;
4016 nested_funcs = tmp;
4017 }
4018 nested_funcs[nesting_level].func = 0;
4019 }
4020 }
4021
4022 unit->function_table = reverse_funcinfo_list (unit->function_table);
4023 unit->variable_table = reverse_varinfo_list (unit->variable_table);
4024
4025 /* This is the second pass over the abbrevs. */
4026 info_ptr = unit->first_child_die_ptr;
4027 nesting_level = 0;
4028
4029 last_func = NULL;
4030 last_var = NULL;
4031
4032 while (nesting_level >= 0)
4033 {
4034 unsigned int abbrev_number, i;
4035 struct abbrev_info *abbrev;
4036 struct attribute attr;
4037 struct funcinfo *func;
4038 struct varinfo *var;
4039 bfd_vma low_pc = 0;
4040 bfd_vma high_pc = 0;
4041 bool high_pc_relative = false;
4042 uint64_t current_offset;
4043
4044 /* PR 17512: file: 9f405d9d. */
4045 if (info_ptr >= info_ptr_end)
4046 goto fail;
4047
4048 current_offset = info_ptr - unit->info_ptr_unit;
4049 abbrev_number = _bfd_safe_read_leb128 (abfd, &info_ptr,
4050 false, info_ptr_end);
4051 if (! abbrev_number)
4052 {
4053 nesting_level--;
4054 continue;
4055 }
4056
4057 abbrev = lookup_abbrev (abbrev_number, unit->abbrevs);
4058 /* This should have been handled above. */
4059 BFD_ASSERT (abbrev != NULL);
4060
4061 func = NULL;
4062 var = NULL;
4063 if (abbrev->tag == DW_TAG_subprogram
4064 || abbrev->tag == DW_TAG_entry_point
4065 || abbrev->tag == DW_TAG_inlined_subroutine)
4066 {
4067 if (last_func
4068 && last_func->prev_func
4069 && last_func->prev_func->unit_offset == current_offset)
4070 func = last_func->prev_func;
4071 else
4072 func = lookup_func_by_offset (current_offset, unit->function_table);
4073
4074 if (func == NULL)
4075 goto fail;
4076
4077 last_func = func;
4078 }
4079 else if (abbrev->tag == DW_TAG_variable
4080 || abbrev->tag == DW_TAG_member)
4081 {
4082 if (last_var
4083 && last_var->prev_var
4084 && last_var->prev_var->unit_offset == current_offset)
4085 var = last_var->prev_var;
4086 else
4087 var = lookup_var_by_offset (current_offset, unit->variable_table);
4088
4089 if (var == NULL)
4090 goto fail;
4091
4092 last_var = var;
4093 }
4094
4095 for (i = 0; i < abbrev->num_attrs; ++i)
4096 {
4097 info_ptr = read_attribute (&attr, &abbrev->attrs[i],
4098 unit, info_ptr, info_ptr_end);
4099 if (info_ptr == NULL)
4100 goto fail;
4101
4102 if (func)
4103 {
4104 switch (attr.name)
4105 {
4106 case DW_AT_call_file:
4107 if (is_int_form (&attr))
4108 {
4109 free (func->caller_file);
4110 func->caller_file = concat_filename (unit->line_table,
4111 attr.u.val);
4112 }
4113 break;
4114
4115 case DW_AT_call_line:
4116 if (is_int_form (&attr))
4117 func->caller_line = attr.u.val;
4118 break;
4119
4120 case DW_AT_abstract_origin:
4121 case DW_AT_specification:
4122 if (is_int_form (&attr)
4123 && !find_abstract_instance (unit, &attr, 0,
4124 &func->name,
4125 &func->is_linkage,
4126 &func->file,
4127 &func->line))
4128 goto fail;
4129 break;
4130
4131 case DW_AT_name:
4132 /* Prefer DW_AT_MIPS_linkage_name or DW_AT_linkage_name
4133 over DW_AT_name. */
4134 if (func->name == NULL && is_str_form (&attr))
4135 {
4136 func->name = attr.u.str;
4137 if (mangle_style (unit->lang) == 0)
4138 func->is_linkage = true;
4139 }
4140 break;
4141
4142 case DW_AT_linkage_name:
4143 case DW_AT_MIPS_linkage_name:
4144 /* PR 16949: Corrupt debug info can place
4145 non-string forms into these attributes. */
4146 if (is_str_form (&attr))
4147 {
4148 func->name = attr.u.str;
4149 func->is_linkage = true;
4150 }
4151 break;
4152
4153 case DW_AT_low_pc:
4154 if (is_int_form (&attr))
4155 low_pc = attr.u.val;
4156 break;
4157
4158 case DW_AT_high_pc:
4159 if (is_int_form (&attr))
4160 {
4161 high_pc = attr.u.val;
4162 high_pc_relative = attr.form != DW_FORM_addr;
4163 }
4164 break;
4165
4166 case DW_AT_ranges:
4167 if (is_int_form (&attr)
4168 && !read_rangelist (unit, &func->arange,
4169 &unit->file->trie_root, attr.u.val))
4170 goto fail;
4171 break;
4172
4173 case DW_AT_decl_file:
4174 if (is_int_form (&attr))
4175 {
4176 free (func->file);
4177 func->file = concat_filename (unit->line_table,
4178 attr.u.val);
4179 }
4180 break;
4181
4182 case DW_AT_decl_line:
4183 if (is_int_form (&attr))
4184 func->line = attr.u.val;
4185 break;
4186
4187 default:
4188 break;
4189 }
4190 }
4191 else if (var)
4192 {
4193 switch (attr.name)
4194 {
4195 case DW_AT_specification:
4196 if (is_int_form (&attr) && attr.u.val)
4197 {
4198 bool is_linkage;
4199 if (!find_abstract_instance (unit, &attr, 0,
4200 &var->name,
4201 &is_linkage,
4202 &var->file,
4203 &var->line))
4204 {
4205 _bfd_error_handler (_("DWARF error: could not find "
4206 "variable specification "
4207 "at offset 0x%lx"),
4208 (unsigned long) attr.u.val);
4209 break;
4210 }
4211 }
4212 break;
4213
4214 case DW_AT_name:
4215 if (is_str_form (&attr))
4216 var->name = attr.u.str;
4217 break;
4218
4219 case DW_AT_decl_file:
4220 if (is_int_form (&attr))
4221 {
4222 free (var->file);
4223 var->file = concat_filename (unit->line_table,
4224 attr.u.val);
4225 }
4226 break;
4227
4228 case DW_AT_decl_line:
4229 if (is_int_form (&attr))
4230 var->line = attr.u.val;
4231 break;
4232
4233 case DW_AT_external:
4234 if (is_int_form (&attr) && attr.u.val != 0)
4235 var->stack = false;
4236 break;
4237
4238 case DW_AT_location:
4239 switch (attr.form)
4240 {
4241 case DW_FORM_block:
4242 case DW_FORM_block1:
4243 case DW_FORM_block2:
4244 case DW_FORM_block4:
4245 case DW_FORM_exprloc:
4246 if (attr.u.blk->data != NULL
4247 && *attr.u.blk->data == DW_OP_addr)
4248 {
4249 var->stack = false;
4250
4251 /* Verify that DW_OP_addr is the only opcode in the
4252 location, in which case the block size will be 1
4253 plus the address size. */
4254 /* ??? For TLS variables, gcc can emit
4255 DW_OP_addr <addr> DW_OP_GNU_push_tls_address
4256 which we don't handle here yet. */
4257 if (attr.u.blk->size == unit->addr_size + 1U)
4258 var->addr = bfd_get (unit->addr_size * 8,
4259 unit->abfd,
4260 attr.u.blk->data + 1);
4261 }
4262 break;
4263
4264 default:
4265 break;
4266 }
4267 break;
4268
4269 default:
4270 break;
4271 }
4272 }
4273 }
4274
4275 if (abbrev->has_children)
4276 nesting_level++;
4277
4278 if (high_pc_relative)
4279 high_pc += low_pc;
4280
4281 if (func && high_pc != 0)
4282 {
4283 if (!arange_add (unit, &func->arange, &unit->file->trie_root,
4284 low_pc, high_pc))
4285 goto fail;
4286 }
4287 }
4288
4289 unit->function_table = reverse_funcinfo_list (unit->function_table);
4290 unit->variable_table = reverse_varinfo_list (unit->variable_table);
4291
4292 free (nested_funcs);
4293 return true;
4294
4295 fail:
4296 free (nested_funcs);
4297 return false;
4298 }
4299
4300 /* Read the attributes of the form strx and addrx. */
4301
4302 static void
4303 reread_attribute (struct comp_unit *unit,
4304 struct attribute *attr,
4305 bfd_vma *low_pc,
4306 bfd_vma *high_pc,
4307 bool *high_pc_relative,
4308 bool compunit)
4309 {
4310 if (is_strx_form (attr->form))
4311 attr->u.str = (char *) read_indexed_string (attr->u.val, unit);
4312 if (is_addrx_form (attr->form))
4313 attr->u.val = read_indexed_address (attr->u.val, unit);
4314
4315 switch (attr->name)
4316 {
4317 case DW_AT_stmt_list:
4318 unit->stmtlist = 1;
4319 unit->line_offset = attr->u.val;
4320 break;
4321
4322 case DW_AT_name:
4323 if (is_str_form (attr))
4324 unit->name = attr->u.str;
4325 break;
4326
4327 case DW_AT_low_pc:
4328 *low_pc = attr->u.val;
4329 if (compunit)
4330 unit->base_address = *low_pc;
4331 break;
4332
4333 case DW_AT_high_pc:
4334 *high_pc = attr->u.val;
4335 *high_pc_relative = attr->form != DW_FORM_addr;
4336 break;
4337
4338 case DW_AT_ranges:
4339 if (!read_rangelist (unit, &unit->arange,
4340 &unit->file->trie_root, attr->u.val))
4341 return;
4342 break;
4343
4344 case DW_AT_comp_dir:
4345 {
4346 char *comp_dir = attr->u.str;
4347
4348 if (!is_str_form (attr))
4349 {
4350 _bfd_error_handler
4351 (_("DWARF error: DW_AT_comp_dir attribute encountered "
4352 "with a non-string form"));
4353 comp_dir = NULL;
4354 }
4355
4356 if (comp_dir)
4357 {
4358 char *cp = strchr (comp_dir, ':');
4359
4360 if (cp && cp != comp_dir && cp[-1] == '.' && cp[1] == '/')
4361 comp_dir = cp + 1;
4362 }
4363 unit->comp_dir = comp_dir;
4364 break;
4365 }
4366
4367 case DW_AT_language:
4368 unit->lang = attr->u.val;
4369 default:
4370 break;
4371 }
4372 }
4373
4374 /* Parse a DWARF2 compilation unit starting at INFO_PTR. UNIT_LENGTH
4375 includes the compilation unit header that proceeds the DIE's, but
4376 does not include the length field that precedes each compilation
4377 unit header. END_PTR points one past the end of this comp unit.
4378 OFFSET_SIZE is the size of DWARF2 offsets (either 4 or 8 bytes).
4379
4380 This routine does not read the whole compilation unit; only enough
4381 to get to the line number information for the compilation unit. */
4382
4383 static struct comp_unit *
4384 parse_comp_unit (struct dwarf2_debug *stash,
4385 struct dwarf2_debug_file *file,
4386 bfd_byte *info_ptr,
4387 bfd_vma unit_length,
4388 bfd_byte *info_ptr_unit,
4389 unsigned int offset_size)
4390 {
4391 struct comp_unit* unit;
4392 unsigned int version;
4393 uint64_t abbrev_offset = 0;
4394 /* Initialize it just to avoid a GCC false warning. */
4395 unsigned int addr_size = -1;
4396 struct abbrev_info** abbrevs;
4397 unsigned int abbrev_number, i;
4398 struct abbrev_info *abbrev;
4399 struct attribute attr;
4400 bfd_byte *end_ptr = info_ptr + unit_length;
4401 size_t amt;
4402 bfd_vma low_pc = 0;
4403 bfd_vma high_pc = 0;
4404 bfd *abfd = file->bfd_ptr;
4405 bool high_pc_relative = false;
4406 enum dwarf_unit_type unit_type;
4407 struct attribute *str_addrp = NULL;
4408 size_t str_count = 0;
4409 size_t str_alloc = 0;
4410 bool compunit_flag = false;
4411
4412 version = read_2_bytes (abfd, &info_ptr, end_ptr);
4413 if (version < 2 || version > 5)
4414 {
4415 /* PR 19872: A version number of 0 probably means that there is padding
4416 at the end of the .debug_info section. Gold puts it there when
4417 performing an incremental link, for example. So do not generate
4418 an error, just return a NULL. */
4419 if (version)
4420 {
4421 _bfd_error_handler
4422 (_("DWARF error: found dwarf version '%u', this reader"
4423 " only handles version 2, 3, 4 and 5 information"), version);
4424 bfd_set_error (bfd_error_bad_value);
4425 }
4426 return NULL;
4427 }
4428
4429 if (version < 5)
4430 unit_type = DW_UT_compile;
4431 else
4432 {
4433 unit_type = read_1_byte (abfd, &info_ptr, end_ptr);
4434 addr_size = read_1_byte (abfd, &info_ptr, end_ptr);
4435 }
4436
4437 BFD_ASSERT (offset_size == 4 || offset_size == 8);
4438 if (offset_size == 4)
4439 abbrev_offset = read_4_bytes (abfd, &info_ptr, end_ptr);
4440 else
4441 abbrev_offset = read_8_bytes (abfd, &info_ptr, end_ptr);
4442
4443 if (version < 5)
4444 addr_size = read_1_byte (abfd, &info_ptr, end_ptr);
4445
4446 switch (unit_type)
4447 {
4448 case DW_UT_type:
4449 /* Skip type signature. */
4450 info_ptr += 8;
4451
4452 /* Skip type offset. */
4453 info_ptr += offset_size;
4454 break;
4455
4456 case DW_UT_skeleton:
4457 /* Skip DWO_id field. */
4458 info_ptr += 8;
4459 break;
4460
4461 default:
4462 break;
4463 }
4464
4465 if (addr_size > sizeof (bfd_vma))
4466 {
4467 _bfd_error_handler
4468 /* xgettext: c-format */
4469 (_("DWARF error: found address size '%u', this reader"
4470 " can not handle sizes greater than '%u'"),
4471 addr_size,
4472 (unsigned int) sizeof (bfd_vma));
4473 bfd_set_error (bfd_error_bad_value);
4474 return NULL;
4475 }
4476
4477 if (addr_size != 2 && addr_size != 4 && addr_size != 8)
4478 {
4479 _bfd_error_handler
4480 ("DWARF error: found address size '%u', this reader"
4481 " can only handle address sizes '2', '4' and '8'", addr_size);
4482 bfd_set_error (bfd_error_bad_value);
4483 return NULL;
4484 }
4485
4486 /* Read the abbrevs for this compilation unit into a table. */
4487 abbrevs = read_abbrevs (abfd, abbrev_offset, stash, file);
4488 if (! abbrevs)
4489 return NULL;
4490
4491 abbrev_number = _bfd_safe_read_leb128 (abfd, &info_ptr,
4492 false, end_ptr);
4493 if (! abbrev_number)
4494 {
4495 /* PR 19872: An abbrev number of 0 probably means that there is padding
4496 at the end of the .debug_abbrev section. Gold puts it there when
4497 performing an incremental link, for example. So do not generate
4498 an error, just return a NULL. */
4499 return NULL;
4500 }
4501
4502 abbrev = lookup_abbrev (abbrev_number, abbrevs);
4503 if (! abbrev)
4504 {
4505 _bfd_error_handler (_("DWARF error: could not find abbrev number %u"),
4506 abbrev_number);
4507 bfd_set_error (bfd_error_bad_value);
4508 return NULL;
4509 }
4510
4511 amt = sizeof (struct comp_unit);
4512 unit = (struct comp_unit *) bfd_zalloc (abfd, amt);
4513 if (unit == NULL)
4514 return NULL;
4515 unit->abfd = abfd;
4516 unit->version = version;
4517 unit->addr_size = addr_size;
4518 unit->offset_size = offset_size;
4519 unit->abbrevs = abbrevs;
4520 unit->end_ptr = end_ptr;
4521 unit->stash = stash;
4522 unit->file = file;
4523 unit->info_ptr_unit = info_ptr_unit;
4524
4525 if (abbrev->tag == DW_TAG_compile_unit)
4526 compunit_flag = true;
4527
4528 for (i = 0; i < abbrev->num_attrs; ++i)
4529 {
4530 info_ptr = read_attribute (&attr, &abbrev->attrs[i], unit, info_ptr, end_ptr);
4531 if (info_ptr == NULL)
4532 goto err_exit;
4533
4534 /* Identify attributes of the form strx* and addrx* which come before
4535 DW_AT_str_offsets_base and DW_AT_addr_base respectively in the CU.
4536 Store the attributes in an array and process them later. */
4537 if ((unit->dwarf_str_offset == 0 && is_strx_form (attr.form))
4538 || (unit->dwarf_addr_offset == 0 && is_addrx_form (attr.form)))
4539 {
4540 if (str_count <= str_alloc)
4541 {
4542 str_alloc = 2 * str_alloc + 200;
4543 str_addrp = bfd_realloc (str_addrp,
4544 str_alloc * sizeof (*str_addrp));
4545 if (str_addrp == NULL)
4546 goto err_exit;
4547 }
4548 str_addrp[str_count] = attr;
4549 str_count++;
4550 continue;
4551 }
4552
4553 /* Store the data if it is of an attribute we want to keep in a
4554 partial symbol table. */
4555 switch (attr.name)
4556 {
4557 case DW_AT_stmt_list:
4558 if (is_int_form (&attr))
4559 {
4560 unit->stmtlist = 1;
4561 unit->line_offset = attr.u.val;
4562 }
4563 break;
4564
4565 case DW_AT_name:
4566 if (is_str_form (&attr))
4567 unit->name = attr.u.str;
4568 break;
4569
4570 case DW_AT_low_pc:
4571 if (is_int_form (&attr))
4572 {
4573 low_pc = attr.u.val;
4574 /* If the compilation unit DIE has a DW_AT_low_pc attribute,
4575 this is the base address to use when reading location
4576 lists or range lists. */
4577 if (compunit_flag)
4578 unit->base_address = low_pc;
4579 }
4580 break;
4581
4582 case DW_AT_high_pc:
4583 if (is_int_form (&attr))
4584 {
4585 high_pc = attr.u.val;
4586 high_pc_relative = attr.form != DW_FORM_addr;
4587 }
4588 break;
4589
4590 case DW_AT_ranges:
4591 if (is_int_form (&attr)
4592 && !read_rangelist (unit, &unit->arange,
4593 &unit->file->trie_root, attr.u.val))
4594 goto err_exit;
4595 break;
4596
4597 case DW_AT_comp_dir:
4598 {
4599 char *comp_dir = attr.u.str;
4600
4601 /* PR 17512: file: 1fe726be. */
4602 if (!is_str_form (&attr))
4603 {
4604 _bfd_error_handler
4605 (_("DWARF error: DW_AT_comp_dir attribute encountered with a non-string form"));
4606 comp_dir = NULL;
4607 }
4608
4609 if (comp_dir)
4610 {
4611 /* Irix 6.2 native cc prepends <machine>.: to the compilation
4612 directory, get rid of it. */
4613 char *cp = strchr (comp_dir, ':');
4614
4615 if (cp && cp != comp_dir && cp[-1] == '.' && cp[1] == '/')
4616 comp_dir = cp + 1;
4617 }
4618 unit->comp_dir = comp_dir;
4619 break;
4620 }
4621
4622 case DW_AT_language:
4623 if (is_int_form (&attr))
4624 unit->lang = attr.u.val;
4625 break;
4626
4627 case DW_AT_addr_base:
4628 unit->dwarf_addr_offset = attr.u.val;
4629 break;
4630
4631 case DW_AT_str_offsets_base:
4632 unit->dwarf_str_offset = attr.u.val;
4633 break;
4634
4635 default:
4636 break;
4637 }
4638 }
4639
4640 for (i = 0; i < str_count; ++i)
4641 reread_attribute (unit, &str_addrp[i], &low_pc, &high_pc,
4642 &high_pc_relative, compunit_flag);
4643
4644 if (high_pc_relative)
4645 high_pc += low_pc;
4646 if (high_pc != 0)
4647 {
4648 if (!arange_add (unit, &unit->arange, &unit->file->trie_root,
4649 low_pc, high_pc))
4650 goto err_exit;
4651 }
4652
4653 unit->first_child_die_ptr = info_ptr;
4654
4655 free (str_addrp);
4656 return unit;
4657
4658 err_exit:
4659 unit->error = 1;
4660 free (str_addrp);
4661 return NULL;
4662 }
4663
4664 /* Return TRUE if UNIT may contain the address given by ADDR. When
4665 there are functions written entirely with inline asm statements, the
4666 range info in the compilation unit header may not be correct. We
4667 need to consult the line info table to see if a compilation unit
4668 really contains the given address. */
4669
4670 static bool
4671 comp_unit_may_contain_address (struct comp_unit *unit, bfd_vma addr)
4672 {
4673 struct arange *arange;
4674
4675 if (unit->error)
4676 return false;
4677
4678 if (unit->arange.high == 0 /* No ranges have been computed yet. */
4679 || unit->line_table == NULL) /* The line info table has not been loaded. */
4680 return true;
4681
4682 for (arange = &unit->arange; arange != NULL; arange = arange->next)
4683 if (addr >= arange->low && addr < arange->high)
4684 return true;
4685
4686 return false;
4687 }
4688
4689 /* If UNIT contains ADDR, set the output parameters to the values for
4690 the line containing ADDR and return TRUE. Otherwise return FALSE.
4691 The output parameters, FILENAME_PTR, FUNCTION_PTR, and
4692 LINENUMBER_PTR, are pointers to the objects to be filled in. */
4693
4694 static bool
4695 comp_unit_find_nearest_line (struct comp_unit *unit,
4696 bfd_vma addr,
4697 const char **filename_ptr,
4698 struct funcinfo **function_ptr,
4699 unsigned int *linenumber_ptr,
4700 unsigned int *discriminator_ptr)
4701 {
4702 bool line_p, func_p;
4703
4704 if (!comp_unit_maybe_decode_line_info (unit))
4705 return false;
4706
4707 *function_ptr = NULL;
4708 func_p = lookup_address_in_function_table (unit, addr, function_ptr);
4709
4710 if (func_p && (*function_ptr)->tag == DW_TAG_inlined_subroutine)
4711 unit->stash->inliner_chain = *function_ptr;
4712
4713 line_p = lookup_address_in_line_info_table (unit->line_table, addr,
4714 filename_ptr,
4715 linenumber_ptr,
4716 discriminator_ptr);
4717 return line_p || func_p;
4718 }
4719
4720 /* Check to see if line info is already decoded in a comp_unit.
4721 If not, decode it. Returns TRUE if no errors were encountered;
4722 FALSE otherwise. */
4723
4724 static bool
4725 comp_unit_maybe_decode_line_info (struct comp_unit *unit)
4726 {
4727 if (unit->error)
4728 return false;
4729
4730 if (! unit->line_table)
4731 {
4732 if (! unit->stmtlist)
4733 {
4734 unit->error = 1;
4735 return false;
4736 }
4737
4738 unit->line_table = decode_line_info (unit);
4739
4740 if (! unit->line_table)
4741 {
4742 unit->error = 1;
4743 return false;
4744 }
4745
4746 if (unit->first_child_die_ptr < unit->end_ptr
4747 && ! scan_unit_for_symbols (unit))
4748 {
4749 unit->error = 1;
4750 return false;
4751 }
4752 }
4753
4754 return true;
4755 }
4756
4757 /* If UNIT contains SYM at ADDR, set the output parameters to the
4758 values for the line containing SYM. The output parameters,
4759 FILENAME_PTR, and LINENUMBER_PTR, are pointers to the objects to be
4760 filled in.
4761
4762 Return TRUE if UNIT contains SYM, and no errors were encountered;
4763 FALSE otherwise. */
4764
4765 static bool
4766 comp_unit_find_line (struct comp_unit *unit,
4767 asymbol *sym,
4768 bfd_vma addr,
4769 const char **filename_ptr,
4770 unsigned int *linenumber_ptr)
4771 {
4772 if (!comp_unit_maybe_decode_line_info (unit))
4773 return false;
4774
4775 if (sym->flags & BSF_FUNCTION)
4776 return lookup_symbol_in_function_table (unit, sym, addr,
4777 filename_ptr,
4778 linenumber_ptr);
4779
4780 return lookup_symbol_in_variable_table (unit, sym, addr,
4781 filename_ptr,
4782 linenumber_ptr);
4783 }
4784
4785 /* Extract all interesting funcinfos and varinfos of a compilation
4786 unit into hash tables for faster lookup. Returns TRUE if no
4787 errors were enountered; FALSE otherwise. */
4788
4789 static bool
4790 comp_unit_hash_info (struct dwarf2_debug *stash,
4791 struct comp_unit *unit,
4792 struct info_hash_table *funcinfo_hash_table,
4793 struct info_hash_table *varinfo_hash_table)
4794 {
4795 struct funcinfo* each_func;
4796 struct varinfo* each_var;
4797 bool okay = true;
4798
4799 BFD_ASSERT (stash->info_hash_status != STASH_INFO_HASH_DISABLED);
4800
4801 if (!comp_unit_maybe_decode_line_info (unit))
4802 return false;
4803
4804 BFD_ASSERT (!unit->cached);
4805
4806 /* To preserve the original search order, we went to visit the function
4807 infos in the reversed order of the list. However, making the list
4808 bi-directional use quite a bit of extra memory. So we reverse
4809 the list first, traverse the list in the now reversed order and
4810 finally reverse the list again to get back the original order. */
4811 unit->function_table = reverse_funcinfo_list (unit->function_table);
4812 for (each_func = unit->function_table;
4813 each_func && okay;
4814 each_func = each_func->prev_func)
4815 {
4816 /* Skip nameless functions. */
4817 if (each_func->name)
4818 /* There is no need to copy name string into hash table as
4819 name string is either in the dwarf string buffer or
4820 info in the stash. */
4821 okay = insert_info_hash_table (funcinfo_hash_table, each_func->name,
4822 (void*) each_func, false);
4823 }
4824 unit->function_table = reverse_funcinfo_list (unit->function_table);
4825 if (!okay)
4826 return false;
4827
4828 /* We do the same for variable infos. */
4829 unit->variable_table = reverse_varinfo_list (unit->variable_table);
4830 for (each_var = unit->variable_table;
4831 each_var && okay;
4832 each_var = each_var->prev_var)
4833 {
4834 /* Skip stack vars and vars with no files or names. */
4835 if (! each_var->stack
4836 && each_var->file != NULL
4837 && each_var->name != NULL)
4838 /* There is no need to copy name string into hash table as
4839 name string is either in the dwarf string buffer or
4840 info in the stash. */
4841 okay = insert_info_hash_table (varinfo_hash_table, each_var->name,
4842 (void*) each_var, false);
4843 }
4844
4845 unit->variable_table = reverse_varinfo_list (unit->variable_table);
4846 unit->cached = true;
4847 return okay;
4848 }
4849
4850 /* Locate a section in a BFD containing debugging info. The search starts
4851 from the section after AFTER_SEC, or from the first section in the BFD if
4852 AFTER_SEC is NULL. The search works by examining the names of the
4853 sections. There are three permissiable names. The first two are given
4854 by DEBUG_SECTIONS[debug_info] (whose standard DWARF2 names are .debug_info
4855 and .zdebug_info). The third is a prefix .gnu.linkonce.wi.
4856 This is a variation on the .debug_info section which has a checksum
4857 describing the contents appended onto the name. This allows the linker to
4858 identify and discard duplicate debugging sections for different
4859 compilation units. */
4860 #define GNU_LINKONCE_INFO ".gnu.linkonce.wi."
4861
4862 static asection *
4863 find_debug_info (bfd *abfd, const struct dwarf_debug_section *debug_sections,
4864 asection *after_sec)
4865 {
4866 asection *msec;
4867 const char *look;
4868
4869 if (after_sec == NULL)
4870 {
4871 look = debug_sections[debug_info].uncompressed_name;
4872 msec = bfd_get_section_by_name (abfd, look);
4873 /* Testing SEC_HAS_CONTENTS is an anti-fuzzer measure. Of
4874 course debug sections always have contents. */
4875 if (msec != NULL && (msec->flags & SEC_HAS_CONTENTS) != 0)
4876 return msec;
4877
4878 look = debug_sections[debug_info].compressed_name;
4879 msec = bfd_get_section_by_name (abfd, look);
4880 if (msec != NULL && (msec->flags & SEC_HAS_CONTENTS) != 0)
4881 return msec;
4882
4883 for (msec = abfd->sections; msec != NULL; msec = msec->next)
4884 if ((msec->flags & SEC_HAS_CONTENTS) != 0
4885 && startswith (msec->name, GNU_LINKONCE_INFO))
4886 return msec;
4887
4888 return NULL;
4889 }
4890
4891 for (msec = after_sec->next; msec != NULL; msec = msec->next)
4892 {
4893 if ((msec->flags & SEC_HAS_CONTENTS) == 0)
4894 continue;
4895
4896 look = debug_sections[debug_info].uncompressed_name;
4897 if (strcmp (msec->name, look) == 0)
4898 return msec;
4899
4900 look = debug_sections[debug_info].compressed_name;
4901 if (look != NULL && strcmp (msec->name, look) == 0)
4902 return msec;
4903
4904 if (startswith (msec->name, GNU_LINKONCE_INFO))
4905 return msec;
4906 }
4907
4908 return NULL;
4909 }
4910
4911 /* Transfer VMAs from object file to separate debug file. */
4912
4913 static void
4914 set_debug_vma (bfd *orig_bfd, bfd *debug_bfd)
4915 {
4916 asection *s, *d;
4917
4918 for (s = orig_bfd->sections, d = debug_bfd->sections;
4919 s != NULL && d != NULL;
4920 s = s->next, d = d->next)
4921 {
4922 if ((d->flags & SEC_DEBUGGING) != 0)
4923 break;
4924 /* ??? Assumes 1-1 correspondence between sections in the
4925 two files. */
4926 if (strcmp (s->name, d->name) == 0)
4927 {
4928 d->output_section = s->output_section;
4929 d->output_offset = s->output_offset;
4930 d->vma = s->vma;
4931 }
4932 }
4933 }
4934
4935 /* If the dwarf2 info was found in a separate debug file, return the
4936 debug file section corresponding to the section in the original file
4937 and the debug file symbols. */
4938
4939 static void
4940 _bfd_dwarf2_stash_syms (struct dwarf2_debug *stash, bfd *abfd,
4941 asection **sec, asymbol ***syms)
4942 {
4943 if (stash->f.bfd_ptr != abfd)
4944 {
4945 asection *s, *d;
4946
4947 if (*sec == NULL)
4948 {
4949 *syms = stash->f.syms;
4950 return;
4951 }
4952
4953 for (s = abfd->sections, d = stash->f.bfd_ptr->sections;
4954 s != NULL && d != NULL;
4955 s = s->next, d = d->next)
4956 {
4957 if ((d->flags & SEC_DEBUGGING) != 0)
4958 break;
4959 if (s == *sec
4960 && strcmp (s->name, d->name) == 0)
4961 {
4962 *sec = d;
4963 *syms = stash->f.syms;
4964 break;
4965 }
4966 }
4967 }
4968 }
4969
4970 /* Unset vmas for adjusted sections in STASH. */
4971
4972 static void
4973 unset_sections (struct dwarf2_debug *stash)
4974 {
4975 int i;
4976 struct adjusted_section *p;
4977
4978 i = stash->adjusted_section_count;
4979 p = stash->adjusted_sections;
4980 for (; i > 0; i--, p++)
4981 p->section->vma = p->orig_vma;
4982 }
4983
4984 /* Set VMAs for allocated and .debug_info sections in ORIG_BFD, a
4985 relocatable object file. VMAs are normally all zero in relocatable
4986 object files, so if we want to distinguish locations in sections by
4987 address we need to set VMAs so the sections do not overlap. We
4988 also set VMA on .debug_info so that when we have multiple
4989 .debug_info sections (or the linkonce variant) they also do not
4990 overlap. The multiple .debug_info sections make up a single
4991 logical section. ??? We should probably do the same for other
4992 debug sections. */
4993
4994 static bool
4995 place_sections (bfd *orig_bfd, struct dwarf2_debug *stash)
4996 {
4997 bfd *abfd;
4998 struct adjusted_section *p;
4999 int i;
5000 const char *debug_info_name;
5001
5002 if (stash->adjusted_section_count != 0)
5003 {
5004 i = stash->adjusted_section_count;
5005 p = stash->adjusted_sections;
5006 for (; i > 0; i--, p++)
5007 p->section->vma = p->adj_vma;
5008 return true;
5009 }
5010
5011 debug_info_name = stash->debug_sections[debug_info].uncompressed_name;
5012 i = 0;
5013 abfd = orig_bfd;
5014 while (1)
5015 {
5016 asection *sect;
5017
5018 for (sect = abfd->sections; sect != NULL; sect = sect->next)
5019 {
5020 int is_debug_info;
5021
5022 if (sect->output_section != NULL
5023 && sect->output_section != sect
5024 && (sect->flags & SEC_DEBUGGING) == 0)
5025 continue;
5026
5027 is_debug_info = (strcmp (sect->name, debug_info_name) == 0
5028 || startswith (sect->name, GNU_LINKONCE_INFO));
5029
5030 if (!((sect->flags & SEC_ALLOC) != 0 && abfd == orig_bfd)
5031 && !is_debug_info)
5032 continue;
5033
5034 i++;
5035 }
5036 if (abfd == stash->f.bfd_ptr)
5037 break;
5038 abfd = stash->f.bfd_ptr;
5039 }
5040
5041 if (i <= 1)
5042 stash->adjusted_section_count = -1;
5043 else
5044 {
5045 bfd_vma last_vma = 0, last_dwarf = 0;
5046 size_t amt = i * sizeof (struct adjusted_section);
5047
5048 p = (struct adjusted_section *) bfd_malloc (amt);
5049 if (p == NULL)
5050 return false;
5051
5052 stash->adjusted_sections = p;
5053 stash->adjusted_section_count = i;
5054
5055 abfd = orig_bfd;
5056 while (1)
5057 {
5058 asection *sect;
5059
5060 for (sect = abfd->sections; sect != NULL; sect = sect->next)
5061 {
5062 bfd_size_type sz;
5063 int is_debug_info;
5064
5065 if (sect->output_section != NULL
5066 && sect->output_section != sect
5067 && (sect->flags & SEC_DEBUGGING) == 0)
5068 continue;
5069
5070 is_debug_info = (strcmp (sect->name, debug_info_name) == 0
5071 || startswith (sect->name, GNU_LINKONCE_INFO));
5072
5073 if (!((sect->flags & SEC_ALLOC) != 0 && abfd == orig_bfd)
5074 && !is_debug_info)
5075 continue;
5076
5077 sz = sect->rawsize ? sect->rawsize : sect->size;
5078
5079 p->section = sect;
5080 p->orig_vma = sect->vma;
5081
5082 bfd_vma *v = is_debug_info ? &last_dwarf : &last_vma;
5083 /* Align the new address to the current section
5084 alignment. */
5085 bfd_vma mask = -(bfd_vma) 1 << sect->alignment_power;
5086 *v = (*v + ~mask) & mask;
5087 sect->vma = *v;
5088 *v += sz;
5089
5090 p->adj_vma = sect->vma;
5091 p++;
5092 }
5093 if (abfd == stash->f.bfd_ptr)
5094 break;
5095 abfd = stash->f.bfd_ptr;
5096 }
5097 }
5098
5099 if (orig_bfd != stash->f.bfd_ptr)
5100 set_debug_vma (orig_bfd, stash->f.bfd_ptr);
5101
5102 return true;
5103 }
5104
5105 /* Look up a funcinfo by name using the given info hash table. If found,
5106 also update the locations pointed to by filename_ptr and linenumber_ptr.
5107
5108 This function returns TRUE if a funcinfo that matches the given symbol
5109 and address is found with any error; otherwise it returns FALSE. */
5110
5111 static bool
5112 info_hash_lookup_funcinfo (struct info_hash_table *hash_table,
5113 asymbol *sym,
5114 bfd_vma addr,
5115 const char **filename_ptr,
5116 unsigned int *linenumber_ptr)
5117 {
5118 struct funcinfo* each_func;
5119 struct funcinfo* best_fit = NULL;
5120 bfd_vma best_fit_len = (bfd_vma) -1;
5121 struct info_list_node *node;
5122 struct arange *arange;
5123 const char *name = bfd_asymbol_name (sym);
5124
5125 for (node = lookup_info_hash_table (hash_table, name);
5126 node;
5127 node = node->next)
5128 {
5129 each_func = (struct funcinfo *) node->info;
5130 for (arange = &each_func->arange;
5131 arange;
5132 arange = arange->next)
5133 {
5134 if (addr >= arange->low
5135 && addr < arange->high
5136 && arange->high - arange->low < best_fit_len)
5137 {
5138 best_fit = each_func;
5139 best_fit_len = arange->high - arange->low;
5140 }
5141 }
5142 }
5143
5144 if (best_fit)
5145 {
5146 *filename_ptr = best_fit->file;
5147 *linenumber_ptr = best_fit->line;
5148 return true;
5149 }
5150
5151 return false;
5152 }
5153
5154 /* Look up a varinfo by name using the given info hash table. If found,
5155 also update the locations pointed to by filename_ptr and linenumber_ptr.
5156
5157 This function returns TRUE if a varinfo that matches the given symbol
5158 and address is found with any error; otherwise it returns FALSE. */
5159
5160 static bool
5161 info_hash_lookup_varinfo (struct info_hash_table *hash_table,
5162 asymbol *sym,
5163 bfd_vma addr,
5164 const char **filename_ptr,
5165 unsigned int *linenumber_ptr)
5166 {
5167 struct varinfo* each;
5168 struct info_list_node *node;
5169 const char *name = bfd_asymbol_name (sym);
5170
5171 for (node = lookup_info_hash_table (hash_table, name);
5172 node;
5173 node = node->next)
5174 {
5175 each = (struct varinfo *) node->info;
5176 if (each->addr == addr)
5177 {
5178 *filename_ptr = each->file;
5179 *linenumber_ptr = each->line;
5180 return true;
5181 }
5182 }
5183
5184 return false;
5185 }
5186
5187 /* Update the funcinfo and varinfo info hash tables if they are
5188 not up to date. Returns TRUE if there is no error; otherwise
5189 returns FALSE and disable the info hash tables. */
5190
5191 static bool
5192 stash_maybe_update_info_hash_tables (struct dwarf2_debug *stash)
5193 {
5194 struct comp_unit *each;
5195
5196 /* Exit if hash tables are up-to-date. */
5197 if (stash->f.all_comp_units == stash->hash_units_head)
5198 return true;
5199
5200 if (stash->hash_units_head)
5201 each = stash->hash_units_head->prev_unit;
5202 else
5203 each = stash->f.last_comp_unit;
5204
5205 while (each)
5206 {
5207 if (!comp_unit_hash_info (stash, each, stash->funcinfo_hash_table,
5208 stash->varinfo_hash_table))
5209 {
5210 stash->info_hash_status = STASH_INFO_HASH_DISABLED;
5211 return false;
5212 }
5213 each = each->prev_unit;
5214 }
5215
5216 stash->hash_units_head = stash->f.all_comp_units;
5217 return true;
5218 }
5219
5220 /* Check consistency of info hash tables. This is for debugging only. */
5221
5222 static void ATTRIBUTE_UNUSED
5223 stash_verify_info_hash_table (struct dwarf2_debug *stash)
5224 {
5225 struct comp_unit *each_unit;
5226 struct funcinfo *each_func;
5227 struct varinfo *each_var;
5228 struct info_list_node *node;
5229 bool found;
5230
5231 for (each_unit = stash->f.all_comp_units;
5232 each_unit;
5233 each_unit = each_unit->next_unit)
5234 {
5235 for (each_func = each_unit->function_table;
5236 each_func;
5237 each_func = each_func->prev_func)
5238 {
5239 if (!each_func->name)
5240 continue;
5241 node = lookup_info_hash_table (stash->funcinfo_hash_table,
5242 each_func->name);
5243 BFD_ASSERT (node);
5244 found = false;
5245 while (node && !found)
5246 {
5247 found = node->info == each_func;
5248 node = node->next;
5249 }
5250 BFD_ASSERT (found);
5251 }
5252
5253 for (each_var = each_unit->variable_table;
5254 each_var;
5255 each_var = each_var->prev_var)
5256 {
5257 if (!each_var->name || !each_var->file || each_var->stack)
5258 continue;
5259 node = lookup_info_hash_table (stash->varinfo_hash_table,
5260 each_var->name);
5261 BFD_ASSERT (node);
5262 found = false;
5263 while (node && !found)
5264 {
5265 found = node->info == each_var;
5266 node = node->next;
5267 }
5268 BFD_ASSERT (found);
5269 }
5270 }
5271 }
5272
5273 /* Check to see if we want to enable the info hash tables, which consume
5274 quite a bit of memory. Currently we only check the number times
5275 bfd_dwarf2_find_line is called. In the future, we may also want to
5276 take the number of symbols into account. */
5277
5278 static void
5279 stash_maybe_enable_info_hash_tables (bfd *abfd, struct dwarf2_debug *stash)
5280 {
5281 BFD_ASSERT (stash->info_hash_status == STASH_INFO_HASH_OFF);
5282
5283 if (stash->info_hash_count++ < STASH_INFO_HASH_TRIGGER)
5284 return;
5285
5286 /* FIXME: Maybe we should check the reduce_memory_overheads
5287 and optimize fields in the bfd_link_info structure ? */
5288
5289 /* Create hash tables. */
5290 stash->funcinfo_hash_table = create_info_hash_table (abfd);
5291 stash->varinfo_hash_table = create_info_hash_table (abfd);
5292 if (!stash->funcinfo_hash_table || !stash->varinfo_hash_table)
5293 {
5294 /* Turn off info hashes if any allocation above fails. */
5295 stash->info_hash_status = STASH_INFO_HASH_DISABLED;
5296 return;
5297 }
5298 /* We need a forced update so that the info hash tables will
5299 be created even though there is no compilation unit. That
5300 happens if STASH_INFO_HASH_TRIGGER is 0. */
5301 if (stash_maybe_update_info_hash_tables (stash))
5302 stash->info_hash_status = STASH_INFO_HASH_ON;
5303 }
5304
5305 /* Find the file and line associated with a symbol and address using the
5306 info hash tables of a stash. If there is a match, the function returns
5307 TRUE and update the locations pointed to by filename_ptr and linenumber_ptr;
5308 otherwise it returns FALSE. */
5309
5310 static bool
5311 stash_find_line_fast (struct dwarf2_debug *stash,
5312 asymbol *sym,
5313 bfd_vma addr,
5314 const char **filename_ptr,
5315 unsigned int *linenumber_ptr)
5316 {
5317 BFD_ASSERT (stash->info_hash_status == STASH_INFO_HASH_ON);
5318
5319 if (sym->flags & BSF_FUNCTION)
5320 return info_hash_lookup_funcinfo (stash->funcinfo_hash_table, sym, addr,
5321 filename_ptr, linenumber_ptr);
5322 return info_hash_lookup_varinfo (stash->varinfo_hash_table, sym, addr,
5323 filename_ptr, linenumber_ptr);
5324 }
5325
5326 /* Save current section VMAs. */
5327
5328 static bool
5329 save_section_vma (const bfd *abfd, struct dwarf2_debug *stash)
5330 {
5331 asection *s;
5332 unsigned int i;
5333
5334 if (abfd->section_count == 0)
5335 return true;
5336 stash->sec_vma = bfd_malloc (sizeof (*stash->sec_vma) * abfd->section_count);
5337 if (stash->sec_vma == NULL)
5338 return false;
5339 stash->sec_vma_count = abfd->section_count;
5340 for (i = 0, s = abfd->sections;
5341 s != NULL && i < abfd->section_count;
5342 i++, s = s->next)
5343 {
5344 if (s->output_section != NULL)
5345 stash->sec_vma[i] = s->output_section->vma + s->output_offset;
5346 else
5347 stash->sec_vma[i] = s->vma;
5348 }
5349 return true;
5350 }
5351
5352 /* Compare current section VMAs against those at the time the stash
5353 was created. If find_nearest_line is used in linker warnings or
5354 errors early in the link process, the debug info stash will be
5355 invalid for later calls. This is because we relocate debug info
5356 sections, so the stashed section contents depend on symbol values,
5357 which in turn depend on section VMAs. */
5358
5359 static bool
5360 section_vma_same (const bfd *abfd, const struct dwarf2_debug *stash)
5361 {
5362 asection *s;
5363 unsigned int i;
5364
5365 /* PR 24334: If the number of sections in ABFD has changed between
5366 when the stash was created and now, then we cannot trust the
5367 stashed vma information. */
5368 if (abfd->section_count != stash->sec_vma_count)
5369 return false;
5370
5371 for (i = 0, s = abfd->sections;
5372 s != NULL && i < abfd->section_count;
5373 i++, s = s->next)
5374 {
5375 bfd_vma vma;
5376
5377 if (s->output_section != NULL)
5378 vma = s->output_section->vma + s->output_offset;
5379 else
5380 vma = s->vma;
5381 if (vma != stash->sec_vma[i])
5382 return false;
5383 }
5384 return true;
5385 }
5386
5387 /* Read debug information from DEBUG_BFD when DEBUG_BFD is specified.
5388 If DEBUG_BFD is not specified, we read debug information from ABFD
5389 or its gnu_debuglink. The results will be stored in PINFO.
5390 The function returns TRUE iff debug information is ready. */
5391
5392 bool
5393 _bfd_dwarf2_slurp_debug_info (bfd *abfd, bfd *debug_bfd,
5394 const struct dwarf_debug_section *debug_sections,
5395 asymbol **symbols,
5396 void **pinfo,
5397 bool do_place)
5398 {
5399 bfd_size_type total_size;
5400 asection *msec;
5401 struct dwarf2_debug *stash = (struct dwarf2_debug *) *pinfo;
5402
5403 if (stash != NULL)
5404 {
5405 if (stash->orig_bfd == abfd
5406 && section_vma_same (abfd, stash))
5407 {
5408 /* Check that we did previously find some debug information
5409 before attempting to make use of it. */
5410 if (stash->f.dwarf_info_size != 0)
5411 {
5412 if (do_place && !place_sections (abfd, stash))
5413 return false;
5414 return true;
5415 }
5416
5417 return false;
5418 }
5419 _bfd_dwarf2_cleanup_debug_info (abfd, pinfo);
5420 memset (stash, 0, sizeof (*stash));
5421 }
5422 else
5423 {
5424 stash = (struct dwarf2_debug *) bfd_zalloc (abfd, sizeof (*stash));
5425 if (! stash)
5426 return false;
5427 *pinfo = stash;
5428 }
5429 stash->orig_bfd = abfd;
5430 stash->debug_sections = debug_sections;
5431 stash->f.syms = symbols;
5432 if (!save_section_vma (abfd, stash))
5433 return false;
5434
5435 stash->f.abbrev_offsets = htab_create_alloc (10, hash_abbrev, eq_abbrev,
5436 del_abbrev, calloc, free);
5437 if (!stash->f.abbrev_offsets)
5438 return false;
5439
5440 stash->alt.abbrev_offsets = htab_create_alloc (10, hash_abbrev, eq_abbrev,
5441 del_abbrev, calloc, free);
5442 if (!stash->alt.abbrev_offsets)
5443 return false;
5444
5445 stash->f.trie_root = alloc_trie_leaf (abfd);
5446 if (!stash->f.trie_root)
5447 return false;
5448
5449 stash->alt.trie_root = alloc_trie_leaf (abfd);
5450 if (!stash->alt.trie_root)
5451 return false;
5452
5453 if (debug_bfd == NULL)
5454 debug_bfd = abfd;
5455
5456 msec = find_debug_info (debug_bfd, debug_sections, NULL);
5457 if (msec == NULL && abfd == debug_bfd)
5458 {
5459 char * debug_filename;
5460
5461 debug_filename = bfd_follow_build_id_debuglink (abfd, DEBUGDIR);
5462 if (debug_filename == NULL)
5463 debug_filename = bfd_follow_gnu_debuglink (abfd, DEBUGDIR);
5464
5465 if (debug_filename == NULL)
5466 /* No dwarf2 info, and no gnu_debuglink to follow.
5467 Note that at this point the stash has been allocated, but
5468 contains zeros. This lets future calls to this function
5469 fail more quickly. */
5470 return false;
5471
5472 debug_bfd = bfd_openr (debug_filename, NULL);
5473 free (debug_filename);
5474 if (debug_bfd == NULL)
5475 /* FIXME: Should we report our failure to follow the debuglink ? */
5476 return false;
5477
5478 /* Set BFD_DECOMPRESS to decompress debug sections. */
5479 debug_bfd->flags |= BFD_DECOMPRESS;
5480 if (!bfd_check_format (debug_bfd, bfd_object)
5481 || (msec = find_debug_info (debug_bfd,
5482 debug_sections, NULL)) == NULL
5483 || !bfd_generic_link_read_symbols (debug_bfd))
5484 {
5485 bfd_close (debug_bfd);
5486 return false;
5487 }
5488
5489 symbols = bfd_get_outsymbols (debug_bfd);
5490 stash->f.syms = symbols;
5491 stash->close_on_cleanup = true;
5492 }
5493 stash->f.bfd_ptr = debug_bfd;
5494
5495 if (do_place
5496 && !place_sections (abfd, stash))
5497 return false;
5498
5499 /* There can be more than one DWARF2 info section in a BFD these
5500 days. First handle the easy case when there's only one. If
5501 there's more than one, try case two: read them all in and produce
5502 one large stash. We do this in two passes - in the first pass we
5503 just accumulate the section sizes, and in the second pass we
5504 read in the section's contents. (The allows us to avoid
5505 reallocing the data as we add sections to the stash.) */
5506
5507 if (! find_debug_info (debug_bfd, debug_sections, msec))
5508 {
5509 /* Case 1: only one info section. */
5510 total_size = msec->size;
5511 if (! read_section (debug_bfd, &stash->debug_sections[debug_info],
5512 symbols, 0,
5513 &stash->f.dwarf_info_buffer, &total_size))
5514 goto restore_vma;
5515 }
5516 else
5517 {
5518 /* Case 2: multiple sections. */
5519 for (total_size = 0;
5520 msec;
5521 msec = find_debug_info (debug_bfd, debug_sections, msec))
5522 {
5523 if (_bfd_section_size_insane (debug_bfd, msec))
5524 goto restore_vma;
5525 /* Catch PR25070 testcase overflowing size calculation here. */
5526 if (total_size + msec->size < total_size)
5527 {
5528 bfd_set_error (bfd_error_no_memory);
5529 goto restore_vma;
5530 }
5531 total_size += msec->size;
5532 }
5533
5534 stash->f.dwarf_info_buffer = (bfd_byte *) bfd_malloc (total_size);
5535 if (stash->f.dwarf_info_buffer == NULL)
5536 goto restore_vma;
5537
5538 total_size = 0;
5539 for (msec = find_debug_info (debug_bfd, debug_sections, NULL);
5540 msec;
5541 msec = find_debug_info (debug_bfd, debug_sections, msec))
5542 {
5543 bfd_size_type size;
5544
5545 size = msec->size;
5546 if (size == 0)
5547 continue;
5548
5549 if (!(bfd_simple_get_relocated_section_contents
5550 (debug_bfd, msec, stash->f.dwarf_info_buffer + total_size,
5551 symbols)))
5552 goto restore_vma;
5553
5554 total_size += size;
5555 }
5556 }
5557
5558 stash->f.info_ptr = stash->f.dwarf_info_buffer;
5559 stash->f.dwarf_info_size = total_size;
5560 return true;
5561
5562 restore_vma:
5563 unset_sections (stash);
5564 return false;
5565 }
5566
5567 /* Parse the next DWARF2 compilation unit at FILE->INFO_PTR. */
5568
5569 static struct comp_unit *
5570 stash_comp_unit (struct dwarf2_debug *stash, struct dwarf2_debug_file *file)
5571 {
5572 bfd_size_type length;
5573 unsigned int offset_size;
5574 bfd_byte *info_ptr_unit = file->info_ptr;
5575 bfd_byte *info_ptr_end = file->dwarf_info_buffer + file->dwarf_info_size;
5576
5577 if (file->info_ptr >= info_ptr_end)
5578 return NULL;
5579
5580 length = read_4_bytes (file->bfd_ptr, &file->info_ptr, info_ptr_end);
5581 /* A 0xffffff length is the DWARF3 way of indicating
5582 we use 64-bit offsets, instead of 32-bit offsets. */
5583 if (length == 0xffffffff)
5584 {
5585 offset_size = 8;
5586 length = read_8_bytes (file->bfd_ptr, &file->info_ptr, info_ptr_end);
5587 }
5588 /* A zero length is the IRIX way of indicating 64-bit offsets,
5589 mostly because the 64-bit length will generally fit in 32
5590 bits, and the endianness helps. */
5591 else if (length == 0)
5592 {
5593 offset_size = 8;
5594 length = read_4_bytes (file->bfd_ptr, &file->info_ptr, info_ptr_end);
5595 }
5596 /* In the absence of the hints above, we assume 32-bit DWARF2
5597 offsets even for targets with 64-bit addresses, because:
5598 a) most of the time these targets will not have generated
5599 more than 2Gb of debug info and so will not need 64-bit
5600 offsets,
5601 and
5602 b) if they do use 64-bit offsets but they are not using
5603 the size hints that are tested for above then they are
5604 not conforming to the DWARF3 standard anyway. */
5605 else
5606 offset_size = 4;
5607
5608 if (length != 0
5609 && length <= (size_t) (info_ptr_end - file->info_ptr))
5610 {
5611 struct comp_unit *each = parse_comp_unit (stash, file,
5612 file->info_ptr, length,
5613 info_ptr_unit, offset_size);
5614 if (each)
5615 {
5616 if (file->comp_unit_tree == NULL)
5617 file->comp_unit_tree
5618 = splay_tree_new (splay_tree_compare_addr_range,
5619 splay_tree_free_addr_range, NULL);
5620
5621 struct addr_range *r
5622 = (struct addr_range *)bfd_malloc (sizeof (struct addr_range));
5623 r->start = each->info_ptr_unit;
5624 r->end = each->end_ptr;
5625 splay_tree_node v = splay_tree_lookup (file->comp_unit_tree,
5626 (splay_tree_key)r);
5627 if (v != NULL || r->end <= r->start)
5628 abort ();
5629 splay_tree_insert (file->comp_unit_tree, (splay_tree_key)r,
5630 (splay_tree_value)each);
5631
5632 if (file->all_comp_units)
5633 file->all_comp_units->prev_unit = each;
5634 else
5635 file->last_comp_unit = each;
5636
5637 each->next_unit = file->all_comp_units;
5638 file->all_comp_units = each;
5639
5640 if (each->arange.high == 0)
5641 {
5642 each->next_unit_without_ranges = file->all_comp_units_without_ranges;
5643 file->all_comp_units_without_ranges = each->next_unit_without_ranges;
5644 }
5645
5646 file->info_ptr += length;
5647 return each;
5648 }
5649 }
5650
5651 /* Don't trust any of the DWARF info after a corrupted length or
5652 parse error. */
5653 file->info_ptr = info_ptr_end;
5654 return NULL;
5655 }
5656
5657 /* Hash function for an asymbol. */
5658
5659 static hashval_t
5660 hash_asymbol (const void *sym)
5661 {
5662 const asymbol *asym = sym;
5663 return htab_hash_string (asym->name);
5664 }
5665
5666 /* Equality function for asymbols. */
5667
5668 static int
5669 eq_asymbol (const void *a, const void *b)
5670 {
5671 const asymbol *sa = a;
5672 const asymbol *sb = b;
5673 return strcmp (sa->name, sb->name) == 0;
5674 }
5675
5676 /* Scan the debug information in PINFO looking for a DW_TAG_subprogram
5677 abbrev with a DW_AT_low_pc attached to it. Then lookup that same
5678 symbol in SYMBOLS and return the difference between the low_pc and
5679 the symbol's address. Returns 0 if no suitable symbol could be found. */
5680
5681 bfd_signed_vma
5682 _bfd_dwarf2_find_symbol_bias (asymbol ** symbols, void ** pinfo)
5683 {
5684 struct dwarf2_debug *stash;
5685 struct comp_unit * unit;
5686 htab_t sym_hash;
5687 bfd_signed_vma result = 0;
5688 asymbol ** psym;
5689
5690 stash = (struct dwarf2_debug *) *pinfo;
5691
5692 if (stash == NULL || symbols == NULL)
5693 return 0;
5694
5695 sym_hash = htab_create_alloc (10, hash_asymbol, eq_asymbol,
5696 NULL, xcalloc, free);
5697 for (psym = symbols; * psym != NULL; psym++)
5698 {
5699 asymbol * sym = * psym;
5700
5701 if (sym->flags & BSF_FUNCTION && sym->section != NULL)
5702 {
5703 void **slot = htab_find_slot (sym_hash, sym, INSERT);
5704 *slot = sym;
5705 }
5706 }
5707
5708 for (unit = stash->f.all_comp_units; unit; unit = unit->next_unit)
5709 {
5710 struct funcinfo * func;
5711
5712 comp_unit_maybe_decode_line_info (unit);
5713
5714 for (func = unit->function_table; func != NULL; func = func->prev_func)
5715 if (func->name && func->arange.low)
5716 {
5717 asymbol search, *sym;
5718
5719 /* FIXME: Do we need to scan the aranges looking for the
5720 lowest pc value? */
5721
5722 search.name = func->name;
5723 sym = htab_find (sym_hash, &search);
5724 if (sym != NULL)
5725 {
5726 result = func->arange.low - (sym->value + sym->section->vma);
5727 goto done;
5728 }
5729 }
5730 }
5731
5732 done:
5733 htab_delete (sym_hash);
5734 return result;
5735 }
5736
5737 /* See _bfd_dwarf2_find_nearest_line_with_alt. */
5738
5739 int
5740 _bfd_dwarf2_find_nearest_line (bfd *abfd,
5741 asymbol **symbols,
5742 asymbol *symbol,
5743 asection *section,
5744 bfd_vma offset,
5745 const char **filename_ptr,
5746 const char **functionname_ptr,
5747 unsigned int *linenumber_ptr,
5748 unsigned int *discriminator_ptr,
5749 const struct dwarf_debug_section *debug_sections,
5750 void **pinfo)
5751 {
5752 return _bfd_dwarf2_find_nearest_line_with_alt
5753 (abfd, NULL, symbols, symbol, section, offset, filename_ptr,
5754 functionname_ptr, linenumber_ptr, discriminator_ptr, debug_sections,
5755 pinfo);
5756 }
5757
5758 /* Find the source code location of SYMBOL. If SYMBOL is NULL
5759 then find the nearest source code location corresponding to
5760 the address SECTION + OFFSET.
5761 Returns 1 if the line is found without error and fills in
5762 FILENAME_PTR and LINENUMBER_PTR. In the case where SYMBOL was
5763 NULL the FUNCTIONNAME_PTR is also filled in.
5764 Returns 2 if partial information from _bfd_elf_find_function is
5765 returned (function and maybe file) by looking at symbols. DWARF2
5766 info is present but not regarding the requested code location.
5767 Returns 0 otherwise.
5768 SYMBOLS contains the symbol table for ABFD.
5769 DEBUG_SECTIONS contains the name of the dwarf debug sections.
5770 If ALT_FILENAME is given, attempt to open the file and use it
5771 as the .gnu_debugaltlink file. Otherwise this file will be
5772 searched for when needed. */
5773
5774 int
5775 _bfd_dwarf2_find_nearest_line_with_alt
5776 (bfd *abfd,
5777 const char *alt_filename,
5778 asymbol **symbols,
5779 asymbol *symbol,
5780 asection *section,
5781 bfd_vma offset,
5782 const char **filename_ptr,
5783 const char **functionname_ptr,
5784 unsigned int *linenumber_ptr,
5785 unsigned int *discriminator_ptr,
5786 const struct dwarf_debug_section *debug_sections,
5787 void **pinfo)
5788 {
5789 /* Read each compilation unit from the section .debug_info, and check
5790 to see if it contains the address we are searching for. If yes,
5791 lookup the address, and return the line number info. If no, go
5792 on to the next compilation unit.
5793
5794 We keep a list of all the previously read compilation units, and
5795 a pointer to the next un-read compilation unit. Check the
5796 previously read units before reading more. */
5797 struct dwarf2_debug *stash;
5798 /* What address are we looking for? */
5799 bfd_vma addr;
5800 struct comp_unit* each;
5801 struct funcinfo *function = NULL;
5802 int found = false;
5803 bool do_line;
5804
5805 *filename_ptr = NULL;
5806 if (functionname_ptr != NULL)
5807 *functionname_ptr = NULL;
5808 *linenumber_ptr = 0;
5809 if (discriminator_ptr)
5810 *discriminator_ptr = 0;
5811
5812 if (! _bfd_dwarf2_slurp_debug_info (abfd, NULL, debug_sections,
5813 symbols, pinfo,
5814 (abfd->flags & (EXEC_P | DYNAMIC)) == 0))
5815 return false;
5816
5817 stash = (struct dwarf2_debug *) *pinfo;
5818
5819 if (stash->alt.bfd_ptr == NULL && alt_filename != NULL)
5820 {
5821 bfd *alt_bfd = bfd_openr (alt_filename, NULL);
5822
5823 if (alt_bfd == NULL)
5824 /* bfd_openr will have set the bfd_error. */
5825 return false;
5826 if (!bfd_check_format (alt_bfd, bfd_object))
5827 {
5828 bfd_set_error (bfd_error_wrong_format);
5829 bfd_close (alt_bfd);
5830 return false;
5831 }
5832
5833 stash->alt.bfd_ptr = alt_bfd;
5834 }
5835
5836 do_line = symbol != NULL;
5837 if (do_line)
5838 {
5839 BFD_ASSERT (section == NULL && offset == 0 && functionname_ptr == NULL);
5840 section = bfd_asymbol_section (symbol);
5841 addr = symbol->value;
5842 }
5843 else
5844 {
5845 BFD_ASSERT (section != NULL && functionname_ptr != NULL);
5846 addr = offset;
5847
5848 /* If we have no SYMBOL but the section we're looking at is not a
5849 code section, then take a look through the list of symbols to see
5850 if we have a symbol at the address we're looking for. If we do
5851 then use this to look up line information. This will allow us to
5852 give file and line results for data symbols. We exclude code
5853 symbols here, if we look up a function symbol and then look up the
5854 line information we'll actually return the line number for the
5855 opening '{' rather than the function definition line. This is
5856 because looking up by symbol uses the line table, in which the
5857 first line for a function is usually the opening '{', while
5858 looking up the function by section + offset uses the
5859 DW_AT_decl_line from the function DW_TAG_subprogram for the line,
5860 which will be the line of the function name. */
5861 if (symbols != NULL && (section->flags & SEC_CODE) == 0)
5862 {
5863 asymbol **tmp;
5864
5865 for (tmp = symbols; (*tmp) != NULL; ++tmp)
5866 if ((*tmp)->the_bfd == abfd
5867 && (*tmp)->section == section
5868 && (*tmp)->value == offset
5869 && ((*tmp)->flags & BSF_SECTION_SYM) == 0)
5870 {
5871 symbol = *tmp;
5872 do_line = true;
5873 /* For local symbols, keep going in the hope we find a
5874 global. */
5875 if ((symbol->flags & BSF_GLOBAL) != 0)
5876 break;
5877 }
5878 }
5879 }
5880
5881 if (section->output_section)
5882 addr += section->output_section->vma + section->output_offset;
5883 else
5884 addr += section->vma;
5885
5886 /* A null info_ptr indicates that there is no dwarf2 info
5887 (or that an error occured while setting up the stash). */
5888 if (! stash->f.info_ptr)
5889 return false;
5890
5891 stash->inliner_chain = NULL;
5892
5893 /* Check the previously read comp. units first. */
5894 if (do_line)
5895 {
5896 /* The info hash tables use quite a bit of memory. We may not want to
5897 always use them. We use some heuristics to decide if and when to
5898 turn it on. */
5899 if (stash->info_hash_status == STASH_INFO_HASH_OFF)
5900 stash_maybe_enable_info_hash_tables (abfd, stash);
5901
5902 /* Keep info hash table up to date if they are available. Note that we
5903 may disable the hash tables if there is any error duing update. */
5904 if (stash->info_hash_status == STASH_INFO_HASH_ON)
5905 stash_maybe_update_info_hash_tables (stash);
5906
5907 if (stash->info_hash_status == STASH_INFO_HASH_ON)
5908 {
5909 found = stash_find_line_fast (stash, symbol, addr,
5910 filename_ptr, linenumber_ptr);
5911 if (found)
5912 goto done;
5913 }
5914
5915 /* Check the previously read comp. units first. */
5916 for (each = stash->f.all_comp_units; each; each = each->next_unit)
5917 if ((symbol->flags & BSF_FUNCTION) == 0
5918 || comp_unit_may_contain_address (each, addr))
5919 {
5920 found = comp_unit_find_line (each, symbol, addr, filename_ptr,
5921 linenumber_ptr);
5922 if (found)
5923 goto done;
5924 }
5925 }
5926 else
5927 {
5928 struct trie_node *trie = stash->f.trie_root;
5929 unsigned int bits = VMA_BITS - 8;
5930 struct comp_unit **prev_each;
5931
5932 /* Traverse interior nodes until we get to a leaf. */
5933 while (trie && trie->num_room_in_leaf == 0)
5934 {
5935 int ch = (addr >> bits) & 0xff;
5936 trie = ((struct trie_interior *) trie)->children[ch];
5937 bits -= 8;
5938 }
5939
5940 if (trie)
5941 {
5942 const struct trie_leaf *leaf = (struct trie_leaf *) trie;
5943 unsigned int i;
5944
5945 for (i = 0; i < leaf->num_stored_in_leaf; ++i)
5946 leaf->ranges[i].unit->mark = false;
5947
5948 for (i = 0; i < leaf->num_stored_in_leaf; ++i)
5949 {
5950 struct comp_unit *unit = leaf->ranges[i].unit;
5951 if (unit->mark
5952 || addr < leaf->ranges[i].low_pc
5953 || addr >= leaf->ranges[i].high_pc)
5954 continue;
5955 unit->mark = true;
5956
5957 found = comp_unit_find_nearest_line (unit, addr,
5958 filename_ptr,
5959 &function,
5960 linenumber_ptr,
5961 discriminator_ptr);
5962 if (found)
5963 goto done;
5964 }
5965 }
5966
5967 /* Also scan through all compilation units without any ranges,
5968 taking them out of the list if they have acquired any since
5969 last time. */
5970 prev_each = &stash->f.all_comp_units_without_ranges;
5971 for (each = *prev_each; each; each = each->next_unit_without_ranges)
5972 {
5973 if (each->arange.high != 0)
5974 {
5975 *prev_each = each->next_unit_without_ranges;
5976 continue;
5977 }
5978
5979 found = comp_unit_find_nearest_line (each, addr,
5980 filename_ptr,
5981 &function,
5982 linenumber_ptr,
5983 discriminator_ptr);
5984 if (found)
5985 goto done;
5986 prev_each = &each->next_unit_without_ranges;
5987 }
5988 }
5989
5990 /* Read each remaining comp. units checking each as they are read. */
5991 while ((each = stash_comp_unit (stash, &stash->f)) != NULL)
5992 {
5993 /* DW_AT_low_pc and DW_AT_high_pc are optional for
5994 compilation units. If we don't have them (i.e.,
5995 unit->high == 0), we need to consult the line info table
5996 to see if a compilation unit contains the given
5997 address. */
5998 if (do_line)
5999 found = (((symbol->flags & BSF_FUNCTION) == 0
6000 || comp_unit_may_contain_address (each, addr))
6001 && comp_unit_find_line (each, symbol, addr,
6002 filename_ptr, linenumber_ptr));
6003 else
6004 found = (comp_unit_may_contain_address (each, addr)
6005 && comp_unit_find_nearest_line (each, addr,
6006 filename_ptr,
6007 &function,
6008 linenumber_ptr,
6009 discriminator_ptr));
6010
6011 if (found)
6012 break;
6013 }
6014
6015 done:
6016 if (functionname_ptr && function && function->is_linkage)
6017 {
6018 *functionname_ptr = function->name;
6019 if (!found)
6020 found = 2;
6021 }
6022 else if (functionname_ptr
6023 && (!*functionname_ptr
6024 || (function && !function->is_linkage)))
6025 {
6026 asymbol *fun;
6027 asymbol **syms = symbols;
6028 asection *sec = section;
6029
6030 _bfd_dwarf2_stash_syms (stash, abfd, &sec, &syms);
6031 fun = _bfd_elf_find_function (abfd, syms, sec, offset,
6032 *filename_ptr ? NULL : filename_ptr,
6033 functionname_ptr);
6034
6035 if (!found && fun != NULL)
6036 found = 2;
6037
6038 if (function && !function->is_linkage)
6039 {
6040 bfd_vma sec_vma;
6041
6042 sec_vma = section->vma;
6043 if (section->output_section != NULL)
6044 sec_vma = section->output_section->vma + section->output_offset;
6045 if (fun == NULL)
6046 *functionname_ptr = function->name;
6047 else if (fun->value + sec_vma == function->arange.low)
6048 function->name = *functionname_ptr;
6049 /* Even if we didn't find a linkage name, say that we have
6050 to stop a repeated search of symbols. */
6051 function->is_linkage = true;
6052 }
6053 }
6054
6055 unset_sections (stash);
6056
6057 return found;
6058 }
6059
6060 bool
6061 _bfd_dwarf2_find_inliner_info (bfd *abfd ATTRIBUTE_UNUSED,
6062 const char **filename_ptr,
6063 const char **functionname_ptr,
6064 unsigned int *linenumber_ptr,
6065 void **pinfo)
6066 {
6067 struct dwarf2_debug *stash;
6068
6069 stash = (struct dwarf2_debug *) *pinfo;
6070 if (stash)
6071 {
6072 struct funcinfo *func = stash->inliner_chain;
6073
6074 if (func && func->caller_func)
6075 {
6076 *filename_ptr = func->caller_file;
6077 *functionname_ptr = func->caller_func->name;
6078 *linenumber_ptr = func->caller_line;
6079 stash->inliner_chain = func->caller_func;
6080 return true;
6081 }
6082 }
6083
6084 return false;
6085 }
6086
6087 void
6088 _bfd_dwarf2_cleanup_debug_info (bfd *abfd, void **pinfo)
6089 {
6090 struct dwarf2_debug *stash = (struct dwarf2_debug *) *pinfo;
6091 struct comp_unit *each;
6092 struct dwarf2_debug_file *file;
6093
6094 if (abfd == NULL || stash == NULL)
6095 return;
6096
6097 if (stash->varinfo_hash_table)
6098 bfd_hash_table_free (&stash->varinfo_hash_table->base);
6099 if (stash->funcinfo_hash_table)
6100 bfd_hash_table_free (&stash->funcinfo_hash_table->base);
6101
6102 file = &stash->f;
6103 while (1)
6104 {
6105 for (each = file->all_comp_units; each; each = each->next_unit)
6106 {
6107 struct funcinfo *function_table = each->function_table;
6108 struct varinfo *variable_table = each->variable_table;
6109
6110 if (each->line_table && each->line_table != file->line_table)
6111 {
6112 free (each->line_table->files);
6113 free (each->line_table->dirs);
6114 }
6115
6116 free (each->lookup_funcinfo_table);
6117 each->lookup_funcinfo_table = NULL;
6118
6119 while (function_table)
6120 {
6121 free (function_table->file);
6122 function_table->file = NULL;
6123 free (function_table->caller_file);
6124 function_table->caller_file = NULL;
6125 function_table = function_table->prev_func;
6126 }
6127
6128 while (variable_table)
6129 {
6130 free (variable_table->file);
6131 variable_table->file = NULL;
6132 variable_table = variable_table->prev_var;
6133 }
6134 }
6135
6136 if (file->line_table)
6137 {
6138 free (file->line_table->files);
6139 free (file->line_table->dirs);
6140 }
6141 htab_delete (file->abbrev_offsets);
6142 if (file->comp_unit_tree != NULL)
6143 splay_tree_delete (file->comp_unit_tree);
6144
6145 free (file->dwarf_line_str_buffer);
6146 free (file->dwarf_str_buffer);
6147 free (file->dwarf_ranges_buffer);
6148 free (file->dwarf_rnglists_buffer);
6149 free (file->dwarf_line_buffer);
6150 free (file->dwarf_abbrev_buffer);
6151 free (file->dwarf_info_buffer);
6152 if (file == &stash->alt)
6153 break;
6154 file = &stash->alt;
6155 }
6156 free (stash->sec_vma);
6157 free (stash->adjusted_sections);
6158 if (stash->close_on_cleanup)
6159 bfd_close (stash->f.bfd_ptr);
6160 if (stash->alt.bfd_ptr)
6161 bfd_close (stash->alt.bfd_ptr);
6162 }
6163
6164 typedef struct elf_find_function_cache
6165 {
6166 asection * last_section;
6167 asymbol * func;
6168 const char * filename;
6169 bfd_size_type code_size;
6170 bfd_vma code_off;
6171
6172 } elf_find_function_cache;
6173
6174
6175 /* Returns TRUE if symbol SYM with address CODE_OFF and size CODE_SIZE
6176 is a better fit to match OFFSET than whatever is currenly stored in
6177 CACHE. */
6178
6179 static inline bool
6180 better_fit (elf_find_function_cache * cache,
6181 asymbol * sym,
6182 bfd_vma code_off,
6183 bfd_size_type code_size,
6184 bfd_vma offset)
6185 {
6186 /* If the symbol is beyond the desired offset, ignore it. */
6187 if (code_off > offset)
6188 return false;
6189
6190 /* If the symbol is further away from the desired
6191 offset than our current best, then ignore it. */
6192 if (code_off < cache->code_off)
6193 return false;
6194
6195 /* On the other hand, if it is closer, then use it. */
6196 if (code_off > cache->code_off)
6197 return true;
6198
6199 /* assert (code_off == cache->code_off); */
6200
6201 /* If our current best fit does not actually reach the desired
6202 offset... */
6203 if (cache->code_off + cache->code_size <= offset)
6204 /* ... then return whichever candidate covers
6205 more area and hence gets closer to OFFSET. */
6206 return code_size > cache->code_size;
6207
6208 /* The current cache'd symbol covers OFFSET. */
6209
6210 /* If the new symbol does not cover the desired offset then skip it. */
6211 if (code_off + code_size <= offset)
6212 return false;
6213
6214 /* Both symbols cover OFFSET. */
6215
6216 /* Prefer functions over non-functions. */
6217 flagword cache_flags = cache->func->flags;
6218 flagword sym_flags = sym->flags;
6219
6220 if ((cache_flags & BSF_FUNCTION) && ((sym_flags & BSF_FUNCTION) == 0))
6221 return false;
6222 if ((sym_flags & BSF_FUNCTION) && ((cache_flags & BSF_FUNCTION) == 0))
6223 return true;
6224
6225 /* FIXME: Should we choose LOCAL over GLOBAL ? */
6226
6227 /* Prefer typed symbols over notyped. */
6228 int cache_type = ELF_ST_TYPE (((elf_symbol_type *) cache->func)->internal_elf_sym.st_info);
6229 int sym_type = ELF_ST_TYPE (((elf_symbol_type *) sym)->internal_elf_sym.st_info);
6230
6231 if (cache_type == STT_NOTYPE && sym_type != STT_NOTYPE)
6232 return true;
6233 if (cache_type != STT_NOTYPE && sym_type == STT_NOTYPE)
6234 return false;
6235
6236 /* Otherwise choose whichever symbol covers a smaller area. */
6237 return code_size < cache->code_size;
6238 }
6239
6240 /* Find the function to a particular section and offset,
6241 for error reporting. */
6242
6243 asymbol *
6244 _bfd_elf_find_function (bfd *abfd,
6245 asymbol **symbols,
6246 asection *section,
6247 bfd_vma offset,
6248 const char **filename_ptr,
6249 const char **functionname_ptr)
6250 {
6251 if (symbols == NULL)
6252 return NULL;
6253
6254 if (bfd_get_flavour (abfd) != bfd_target_elf_flavour)
6255 return NULL;
6256
6257 elf_find_function_cache * cache = elf_tdata (abfd)->elf_find_function_cache;
6258
6259 if (cache == NULL)
6260 {
6261 cache = bfd_zalloc (abfd, sizeof (*cache));
6262 elf_tdata (abfd)->elf_find_function_cache = cache;
6263 if (cache == NULL)
6264 return NULL;
6265 }
6266
6267 if (cache->last_section != section
6268 || cache->func == NULL
6269 || offset < cache->func->value
6270 || offset >= cache->func->value + cache->code_size)
6271 {
6272 asymbol *file;
6273 asymbol **p;
6274 /* ??? Given multiple file symbols, it is impossible to reliably
6275 choose the right file name for global symbols. File symbols are
6276 local symbols, and thus all file symbols must sort before any
6277 global symbols. The ELF spec may be interpreted to say that a
6278 file symbol must sort before other local symbols, but currently
6279 ld -r doesn't do this. So, for ld -r output, it is possible to
6280 make a better choice of file name for local symbols by ignoring
6281 file symbols appearing after a given local symbol. */
6282 enum { nothing_seen, symbol_seen, file_after_symbol_seen } state;
6283 const struct elf_backend_data *bed = get_elf_backend_data (abfd);
6284
6285 file = NULL;
6286 state = nothing_seen;
6287 cache->filename = NULL;
6288 cache->func = NULL;
6289 cache->code_size = 0;
6290 cache->code_off = 0;
6291 cache->last_section = section;
6292
6293 for (p = symbols; *p != NULL; p++)
6294 {
6295 asymbol *sym = *p;
6296 bfd_vma code_off;
6297 bfd_size_type size;
6298
6299 if ((sym->flags & BSF_FILE) != 0)
6300 {
6301 file = sym;
6302 if (state == symbol_seen)
6303 state = file_after_symbol_seen;
6304 continue;
6305 }
6306
6307 if (state == nothing_seen)
6308 state = symbol_seen;
6309
6310 size = bed->maybe_function_sym (sym, section, &code_off);
6311
6312 if (size == 0)
6313 continue;
6314
6315 if (better_fit (cache, sym, code_off, size, offset))
6316 {
6317 cache->func = sym;
6318 cache->code_size = size;
6319 cache->code_off = code_off;
6320 cache->filename = NULL;
6321
6322 if (file != NULL
6323 && ((sym->flags & BSF_LOCAL) != 0
6324 || state != file_after_symbol_seen))
6325 cache->filename = bfd_asymbol_name (file);
6326 }
6327 /* Otherwise, if the symbol is beyond the desired offset but it
6328 lies within the bounds of the current best match then reduce
6329 the size of the current best match so that future searches
6330 will not not used the cached symbol by mistake. */
6331 else if (code_off > offset
6332 && code_off > cache->code_off
6333 && code_off < cache->code_off + cache->code_size)
6334 {
6335 cache->code_size = code_off - cache->code_off;
6336 }
6337 }
6338 }
6339
6340 if (cache->func == NULL)
6341 return NULL;
6342
6343 if (filename_ptr)
6344 *filename_ptr = cache->filename;
6345 if (functionname_ptr)
6346 *functionname_ptr = bfd_asymbol_name (cache->func);
6347
6348 return cache->func;
6349 }