x86: decouple broadcast type and bytes fields
[binutils-gdb.git] / bfd / dwarf2.c
blob15862dc2037fcb8a09ecda1b8f50a68bc25df6c3
1 /* DWARF 2 support.
2 Copyright (C) 1994-2023 Free Software Foundation, Inc.
4 Adapted from gdb/dwarf2read.c by Gavin Koch of Cygnus Solutions
5 (gavin@cygnus.com).
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
15 This file is part of BFD.
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.
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.
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. */
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"
42 /* The data in the .debug_line statement prologue looks like this. */
44 struct line_head
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;
58 /* Attributes have a name and a value. */
60 struct attribute
62 enum dwarf_attribute name;
63 enum dwarf_form form;
64 union
66 char *str;
67 struct dwarf_block *blk;
68 uint64_t val;
69 int64_t sval;
74 /* Blocks are a bunch of untyped bytes. */
75 struct dwarf_block
77 unsigned int size;
78 bfd_byte *data;
81 struct adjusted_section
83 asection *section;
84 bfd_vma adj_vma;
85 bfd_vma orig_vma;
88 /* A trie to map quickly from address range to compilation unit.
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.
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.
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.
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.
122 #define TRIE_LEAF_SIZE 16
124 /* All trie_node pointers will really be trie_leaf or trie_interior,
125 but they have this common head. */
126 struct trie_node
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;
133 struct trie_leaf
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[TRIE_LEAF_SIZE];
143 struct trie_interior
145 struct trie_node head;
146 struct trie_node *children[256];
149 static struct trie_node *alloc_trie_leaf (bfd *abfd)
151 struct trie_leaf *leaf = bfd_zalloc (abfd, sizeof (struct trie_leaf));
152 if (leaf == NULL)
153 return NULL;
154 leaf->head.num_room_in_leaf = TRIE_LEAF_SIZE;
155 return &leaf->head;
158 struct addr_range
160 bfd_byte *start;
161 bfd_byte *end;
164 /* Return true if address range do intersect. */
166 static bool
167 addr_range_intersects (struct addr_range *r1, struct addr_range *r2)
169 return (r1->start <= r2->start && r2->start < r1->end)
170 || (r1->start <= (r2->end - 1) && (r2->end - 1) < r1->end);
173 /* Compare function for splay tree of addr_ranges. */
175 static int
176 splay_tree_compare_addr_range (splay_tree_key xa, splay_tree_key xb)
178 struct addr_range *r1 = (struct addr_range *) xa;
179 struct addr_range *r2 = (struct addr_range *) xb;
181 if (addr_range_intersects (r1, r2) || addr_range_intersects (r2, r1))
182 return 0;
183 else if (r1->end <= r2->start)
184 return -1;
185 else
186 return 1;
189 /* Splay tree release function for keys (addr_range). */
191 static void
192 splay_tree_free_addr_range (splay_tree_key key)
194 free ((struct addr_range *)key);
197 struct dwarf2_debug_file
199 /* The actual bfd from which debug info was loaded. Might be
200 different to orig_bfd because of gnu_debuglink sections. */
201 bfd *bfd_ptr;
203 /* Pointer to the symbol table. */
204 asymbol **syms;
206 /* The current info pointer for the .debug_info section being parsed. */
207 bfd_byte *info_ptr;
209 /* A pointer to the memory block allocated for .debug_info sections. */
210 bfd_byte *dwarf_info_buffer;
212 /* Length of the loaded .debug_info sections. */
213 bfd_size_type dwarf_info_size;
215 /* Pointer to the .debug_abbrev section loaded into memory. */
216 bfd_byte *dwarf_abbrev_buffer;
218 /* Length of the loaded .debug_abbrev section. */
219 bfd_size_type dwarf_abbrev_size;
221 /* Buffer for decode_line_info. */
222 bfd_byte *dwarf_line_buffer;
224 /* Length of the loaded .debug_line section. */
225 bfd_size_type dwarf_line_size;
227 /* Pointer to the .debug_str section loaded into memory. */
228 bfd_byte *dwarf_str_buffer;
230 /* Length of the loaded .debug_str section. */
231 bfd_size_type dwarf_str_size;
233 /* Pointer to the .debug_str_offsets section loaded into memory. */
234 bfd_byte *dwarf_str_offsets_buffer;
236 /* Length of the loaded .debug_str_offsets section. */
237 bfd_size_type dwarf_str_offsets_size;
239 /* Pointer to the .debug_addr section loaded into memory. */
240 bfd_byte *dwarf_addr_buffer;
242 /* Length of the loaded .debug_addr section. */
243 bfd_size_type dwarf_addr_size;
245 /* Pointer to the .debug_line_str section loaded into memory. */
246 bfd_byte *dwarf_line_str_buffer;
248 /* Length of the loaded .debug_line_str section. */
249 bfd_size_type dwarf_line_str_size;
251 /* Pointer to the .debug_ranges section loaded into memory. */
252 bfd_byte *dwarf_ranges_buffer;
254 /* Length of the loaded .debug_ranges section. */
255 bfd_size_type dwarf_ranges_size;
257 /* Pointer to the .debug_rnglists section loaded into memory. */
258 bfd_byte *dwarf_rnglists_buffer;
260 /* Length of the loaded .debug_rnglists section. */
261 bfd_size_type dwarf_rnglists_size;
263 /* A list of all previously read comp_units. */
264 struct comp_unit *all_comp_units;
266 /* A list of all previously read comp_units with no ranges (yet). */
267 struct comp_unit *all_comp_units_without_ranges;
269 /* Last comp unit in list above. */
270 struct comp_unit *last_comp_unit;
272 /* Line table at line_offset zero. */
273 struct line_info_table *line_table;
275 /* Hash table to map offsets to decoded abbrevs. */
276 htab_t abbrev_offsets;
278 /* Root of a trie to map addresses to compilation units. */
279 struct trie_node *trie_root;
281 /* Splay tree to map info_ptr address to compilation units. */
282 splay_tree comp_unit_tree;
285 struct dwarf2_debug
287 /* Names of the debug sections. */
288 const struct dwarf_debug_section *debug_sections;
290 /* Per-file stuff. */
291 struct dwarf2_debug_file f, alt;
293 /* Pointer to the original bfd for which debug was loaded. This is what
294 we use to compare and so check that the cached debug data is still
295 valid - it saves having to possibly dereference the gnu_debuglink each
296 time. */
297 bfd *orig_bfd;
299 /* If the most recent call to bfd_find_nearest_line was given an
300 address in an inlined function, preserve a pointer into the
301 calling chain for subsequent calls to bfd_find_inliner_info to
302 use. */
303 struct funcinfo *inliner_chain;
305 /* Section VMAs at the time the stash was built. */
306 bfd_vma *sec_vma;
307 /* Number of sections in the SEC_VMA table. */
308 unsigned int sec_vma_count;
310 /* Number of sections whose VMA we must adjust. */
311 int adjusted_section_count;
313 /* Array of sections with adjusted VMA. */
314 struct adjusted_section *adjusted_sections;
316 /* Number of times find_line is called. This is used in
317 the heuristic for enabling the info hash tables. */
318 int info_hash_count;
320 #define STASH_INFO_HASH_TRIGGER 100
322 /* Hash table mapping symbol names to function infos. */
323 struct info_hash_table *funcinfo_hash_table;
325 /* Hash table mapping symbol names to variable infos. */
326 struct info_hash_table *varinfo_hash_table;
328 /* Head of comp_unit list in the last hash table update. */
329 struct comp_unit *hash_units_head;
331 /* Status of info hash. */
332 int info_hash_status;
333 #define STASH_INFO_HASH_OFF 0
334 #define STASH_INFO_HASH_ON 1
335 #define STASH_INFO_HASH_DISABLED 2
337 /* True if we opened bfd_ptr. */
338 bool close_on_cleanup;
341 struct arange
343 struct arange *next;
344 bfd_vma low;
345 bfd_vma high;
348 /* A minimal decoding of DWARF2 compilation units. We only decode
349 what's needed to get to the line number information. */
351 struct comp_unit
353 /* Chain the previously read compilation units. */
354 struct comp_unit *next_unit;
356 /* Chain the previously read compilation units that have no ranges yet.
357 We scan these separately when we have a trie over the ranges.
358 Unused if arange.high != 0. */
359 struct comp_unit *next_unit_without_ranges;
361 /* Likewise, chain the compilation unit read after this one.
362 The comp units are stored in reversed reading order. */
363 struct comp_unit *prev_unit;
365 /* Keep the bfd convenient (for memory allocation). */
366 bfd *abfd;
368 /* The lowest and highest addresses contained in this compilation
369 unit as specified in the compilation unit header. */
370 struct arange arange;
372 /* The DW_AT_name attribute (for error messages). */
373 char *name;
375 /* The abbrev hash table. */
376 struct abbrev_info **abbrevs;
378 /* DW_AT_language. */
379 int lang;
381 /* Note that an error was found by comp_unit_find_nearest_line. */
382 int error;
384 /* The DW_AT_comp_dir attribute. */
385 char *comp_dir;
387 /* TRUE if there is a line number table associated with this comp. unit. */
388 int stmtlist;
390 /* Pointer to the current comp_unit so that we can find a given entry
391 by its reference. */
392 bfd_byte *info_ptr_unit;
394 /* The offset into .debug_line of the line number table. */
395 unsigned long line_offset;
397 /* Pointer to the first child die for the comp unit. */
398 bfd_byte *first_child_die_ptr;
400 /* The end of the comp unit. */
401 bfd_byte *end_ptr;
403 /* The decoded line number, NULL if not yet decoded. */
404 struct line_info_table *line_table;
406 /* A list of the functions found in this comp. unit. */
407 struct funcinfo *function_table;
409 /* A table of function information references searchable by address. */
410 struct lookup_funcinfo *lookup_funcinfo_table;
412 /* Number of functions in the function_table and sorted_function_table. */
413 bfd_size_type number_of_functions;
415 /* A list of the variables found in this comp. unit. */
416 struct varinfo *variable_table;
418 /* Pointers to dwarf2_debug structures. */
419 struct dwarf2_debug *stash;
420 struct dwarf2_debug_file *file;
422 /* DWARF format version for this unit - from unit header. */
423 int version;
425 /* Address size for this unit - from unit header. */
426 unsigned char addr_size;
428 /* Offset size for this unit - from unit header. */
429 unsigned char offset_size;
431 /* Base address for this unit - from DW_AT_low_pc attribute of
432 DW_TAG_compile_unit DIE */
433 bfd_vma base_address;
435 /* TRUE if symbols are cached in hash table for faster lookup by name. */
436 bool cached;
438 /* Used when iterating over trie leaves to know which units we have
439 already seen in this iteration. */
440 bool mark;
442 /* Base address of debug_addr section. */
443 size_t dwarf_addr_offset;
445 /* Base address of string offset table. */
446 size_t dwarf_str_offset;
449 /* This data structure holds the information of an abbrev. */
450 struct abbrev_info
452 unsigned int number; /* Number identifying abbrev. */
453 enum dwarf_tag tag; /* DWARF tag. */
454 bool has_children; /* TRUE if the abbrev has children. */
455 unsigned int num_attrs; /* Number of attributes. */
456 struct attr_abbrev * attrs; /* An array of attribute descriptions. */
457 struct abbrev_info * next; /* Next in chain. */
460 struct attr_abbrev
462 enum dwarf_attribute name;
463 enum dwarf_form form;
464 bfd_vma implicit_const;
467 /* Map of uncompressed DWARF debug section name to compressed one. It
468 is terminated by NULL uncompressed_name. */
470 const struct dwarf_debug_section dwarf_debug_sections[] =
472 { ".debug_abbrev", ".zdebug_abbrev" },
473 { ".debug_aranges", ".zdebug_aranges" },
474 { ".debug_frame", ".zdebug_frame" },
475 { ".debug_info", ".zdebug_info" },
476 { ".debug_info", ".zdebug_info" },
477 { ".debug_line", ".zdebug_line" },
478 { ".debug_loc", ".zdebug_loc" },
479 { ".debug_macinfo", ".zdebug_macinfo" },
480 { ".debug_macro", ".zdebug_macro" },
481 { ".debug_pubnames", ".zdebug_pubnames" },
482 { ".debug_pubtypes", ".zdebug_pubtypes" },
483 { ".debug_ranges", ".zdebug_ranges" },
484 { ".debug_rnglists", ".zdebug_rnglist" },
485 { ".debug_static_func", ".zdebug_static_func" },
486 { ".debug_static_vars", ".zdebug_static_vars" },
487 { ".debug_str", ".zdebug_str", },
488 { ".debug_str", ".zdebug_str", },
489 { ".debug_str_offsets", ".zdebug_str_offsets", },
490 { ".debug_addr", ".zdebug_addr", },
491 { ".debug_line_str", ".zdebug_line_str", },
492 { ".debug_types", ".zdebug_types" },
493 /* GNU DWARF 1 extensions */
494 { ".debug_sfnames", ".zdebug_sfnames" },
495 { ".debug_srcinfo", ".zebug_srcinfo" },
496 /* SGI/MIPS DWARF 2 extensions */
497 { ".debug_funcnames", ".zdebug_funcnames" },
498 { ".debug_typenames", ".zdebug_typenames" },
499 { ".debug_varnames", ".zdebug_varnames" },
500 { ".debug_weaknames", ".zdebug_weaknames" },
501 { NULL, NULL },
504 /* NB/ Numbers in this enum must match up with indices
505 into the dwarf_debug_sections[] array above. */
506 enum dwarf_debug_section_enum
508 debug_abbrev = 0,
509 debug_aranges,
510 debug_frame,
511 debug_info,
512 debug_info_alt,
513 debug_line,
514 debug_loc,
515 debug_macinfo,
516 debug_macro,
517 debug_pubnames,
518 debug_pubtypes,
519 debug_ranges,
520 debug_rnglists,
521 debug_static_func,
522 debug_static_vars,
523 debug_str,
524 debug_str_alt,
525 debug_str_offsets,
526 debug_addr,
527 debug_line_str,
528 debug_types,
529 debug_sfnames,
530 debug_srcinfo,
531 debug_funcnames,
532 debug_typenames,
533 debug_varnames,
534 debug_weaknames,
535 debug_max
538 /* A static assertion. */
539 extern int dwarf_debug_section_assert[ARRAY_SIZE (dwarf_debug_sections)
540 == debug_max + 1 ? 1 : -1];
542 #ifndef ABBREV_HASH_SIZE
543 #define ABBREV_HASH_SIZE 121
544 #endif
545 #ifndef ATTR_ALLOC_CHUNK
546 #define ATTR_ALLOC_CHUNK 4
547 #endif
549 /* Variable and function hash tables. This is used to speed up look-up
550 in lookup_symbol_in_var_table() and lookup_symbol_in_function_table().
551 In order to share code between variable and function infos, we use
552 a list of untyped pointer for all variable/function info associated with
553 a symbol. We waste a bit of memory for list with one node but that
554 simplifies the code. */
556 struct info_list_node
558 struct info_list_node *next;
559 void *info;
562 /* Info hash entry. */
563 struct info_hash_entry
565 struct bfd_hash_entry root;
566 struct info_list_node *head;
569 struct info_hash_table
571 struct bfd_hash_table base;
574 /* Function to create a new entry in info hash table. */
576 static struct bfd_hash_entry *
577 info_hash_table_newfunc (struct bfd_hash_entry *entry,
578 struct bfd_hash_table *table,
579 const char *string)
581 struct info_hash_entry *ret = (struct info_hash_entry *) entry;
583 /* Allocate the structure if it has not already been allocated by a
584 derived class. */
585 if (ret == NULL)
587 ret = (struct info_hash_entry *) bfd_hash_allocate (table,
588 sizeof (* ret));
589 if (ret == NULL)
590 return NULL;
593 /* Call the allocation method of the base class. */
594 ret = ((struct info_hash_entry *)
595 bfd_hash_newfunc ((struct bfd_hash_entry *) ret, table, string));
597 /* Initialize the local fields here. */
598 if (ret)
599 ret->head = NULL;
601 return (struct bfd_hash_entry *) ret;
604 /* Function to create a new info hash table. It returns a pointer to the
605 newly created table or NULL if there is any error. We need abfd
606 solely for memory allocation. */
608 static struct info_hash_table *
609 create_info_hash_table (bfd *abfd)
611 struct info_hash_table *hash_table;
613 hash_table = ((struct info_hash_table *)
614 bfd_alloc (abfd, sizeof (struct info_hash_table)));
615 if (!hash_table)
616 return hash_table;
618 if (!bfd_hash_table_init (&hash_table->base, info_hash_table_newfunc,
619 sizeof (struct info_hash_entry)))
621 bfd_release (abfd, hash_table);
622 return NULL;
625 return hash_table;
628 /* Insert an info entry into an info hash table. We do not check of
629 duplicate entries. Also, the caller need to guarantee that the
630 right type of info in inserted as info is passed as a void* pointer.
631 This function returns true if there is no error. */
633 static bool
634 insert_info_hash_table (struct info_hash_table *hash_table,
635 const char *key,
636 void *info,
637 bool copy_p)
639 struct info_hash_entry *entry;
640 struct info_list_node *node;
642 entry = (struct info_hash_entry*) bfd_hash_lookup (&hash_table->base,
643 key, true, copy_p);
644 if (!entry)
645 return false;
647 node = (struct info_list_node *) bfd_hash_allocate (&hash_table->base,
648 sizeof (*node));
649 if (!node)
650 return false;
652 node->info = info;
653 node->next = entry->head;
654 entry->head = node;
656 return true;
659 /* Look up an info entry list from an info hash table. Return NULL
660 if there is none. */
662 static struct info_list_node *
663 lookup_info_hash_table (struct info_hash_table *hash_table, const char *key)
665 struct info_hash_entry *entry;
667 entry = (struct info_hash_entry*) bfd_hash_lookup (&hash_table->base, key,
668 false, false);
669 return entry ? entry->head : NULL;
672 /* Read a section into its appropriate place in the dwarf2_debug
673 struct (indicated by SECTION_BUFFER and SECTION_SIZE). If SYMS is
674 not NULL, use bfd_simple_get_relocated_section_contents to read the
675 section contents, otherwise use bfd_get_section_contents. Fail if
676 the located section does not contain at least OFFSET bytes. */
678 static bool
679 read_section (bfd *abfd,
680 const struct dwarf_debug_section *sec,
681 asymbol **syms,
682 uint64_t offset,
683 bfd_byte **section_buffer,
684 bfd_size_type *section_size)
686 const char *section_name = sec->uncompressed_name;
687 bfd_byte *contents = *section_buffer;
689 /* The section may have already been read. */
690 if (contents == NULL)
692 bfd_size_type amt;
693 asection *msec;
695 msec = bfd_get_section_by_name (abfd, section_name);
696 if (msec == NULL)
698 section_name = sec->compressed_name;
699 msec = bfd_get_section_by_name (abfd, section_name);
701 if (msec == NULL)
703 _bfd_error_handler (_("DWARF error: can't find %s section."),
704 sec->uncompressed_name);
705 bfd_set_error (bfd_error_bad_value);
706 return false;
709 if ((msec->flags & SEC_HAS_CONTENTS) == 0)
711 _bfd_error_handler (_("DWARF error: section %s has no contents"),
712 section_name);
713 bfd_set_error (bfd_error_no_contents);
714 return false;
717 if (_bfd_section_size_insane (abfd, msec))
719 /* PR 26946 */
720 _bfd_error_handler (_("DWARF error: section %s is too big"),
721 section_name);
722 return false;
724 amt = bfd_get_section_limit_octets (abfd, msec);
725 *section_size = amt;
726 /* Paranoia - alloc one extra so that we can make sure a string
727 section is NUL terminated. */
728 amt += 1;
729 if (amt == 0)
731 /* Paranoia - this should never happen. */
732 bfd_set_error (bfd_error_no_memory);
733 return false;
735 contents = (bfd_byte *) bfd_malloc (amt);
736 if (contents == NULL)
737 return false;
738 if (syms
739 ? !bfd_simple_get_relocated_section_contents (abfd, msec, contents,
740 syms)
741 : !bfd_get_section_contents (abfd, msec, contents, 0, *section_size))
743 free (contents);
744 return false;
746 contents[*section_size] = 0;
747 *section_buffer = contents;
750 /* It is possible to get a bad value for the offset into the section
751 that the client wants. Validate it here to avoid trouble later. */
752 if (offset != 0 && offset >= *section_size)
754 /* xgettext: c-format */
755 _bfd_error_handler (_("DWARF error: offset (%" PRIu64 ")"
756 " greater than or equal to %s size (%" PRIu64 ")"),
757 (uint64_t) offset, section_name,
758 (uint64_t) *section_size);
759 bfd_set_error (bfd_error_bad_value);
760 return false;
763 return true;
766 /* Read dwarf information from a buffer. */
768 static inline uint64_t
769 read_n_bytes (bfd *abfd, bfd_byte **ptr, bfd_byte *end, int n)
771 bfd_byte *buf = *ptr;
772 if (end - buf < n)
774 *ptr = end;
775 return 0;
777 *ptr = buf + n;
778 return bfd_get (n * 8, abfd, buf);
781 static unsigned int
782 read_1_byte (bfd *abfd, bfd_byte **ptr, bfd_byte *end)
784 return read_n_bytes (abfd, ptr, end, 1);
787 static int
788 read_1_signed_byte (bfd *abfd ATTRIBUTE_UNUSED, bfd_byte **ptr, bfd_byte *end)
790 bfd_byte *buf = *ptr;
791 if (end - buf < 1)
793 *ptr = end;
794 return 0;
796 *ptr = buf + 1;
797 return bfd_get_signed_8 (abfd, buf);
800 static unsigned int
801 read_2_bytes (bfd *abfd, bfd_byte **ptr, bfd_byte *end)
803 return read_n_bytes (abfd, ptr, end, 2);
806 static unsigned int
807 read_3_bytes (bfd *abfd, bfd_byte **ptr, bfd_byte *end)
809 unsigned int val = read_1_byte (abfd, ptr, end);
810 val <<= 8;
811 val |= read_1_byte (abfd, ptr, end);
812 val <<= 8;
813 val |= read_1_byte (abfd, ptr, end);
814 if (bfd_little_endian (abfd))
815 val = (((val >> 16) & 0xff)
816 | (val & 0xff00)
817 | ((val & 0xff) << 16));
818 return val;
821 static unsigned int
822 read_4_bytes (bfd *abfd, bfd_byte **ptr, bfd_byte *end)
824 return read_n_bytes (abfd, ptr, end, 4);
827 static uint64_t
828 read_8_bytes (bfd *abfd, bfd_byte **ptr, bfd_byte *end)
830 return read_n_bytes (abfd, ptr, end, 8);
833 static struct dwarf_block *
834 read_blk (bfd *abfd, bfd_byte **ptr, bfd_byte *end, size_t size)
836 bfd_byte *buf = *ptr;
837 struct dwarf_block *block;
839 block = (struct dwarf_block *) bfd_alloc (abfd, sizeof (*block));
840 if (block == NULL)
841 return NULL;
843 if (size > (size_t) (end - buf))
845 *ptr = end;
846 block->data = NULL;
847 block->size = 0;
849 else
851 *ptr = buf + size;
852 block->data = buf;
853 block->size = size;
855 return block;
858 /* Scans a NUL terminated string starting at *PTR, returning a pointer to it.
859 Bytes at or beyond BUF_END will not be read. Returns NULL if the
860 terminator is not found or if the string is empty. *PTR is
861 incremented over the bytes scanned, including the terminator. */
863 static char *
864 read_string (bfd_byte **ptr,
865 bfd_byte *buf_end)
867 bfd_byte *buf = *ptr;
868 bfd_byte *str = buf;
870 while (buf < buf_end)
871 if (*buf++ == 0)
873 if (str == buf - 1)
874 break;
875 *ptr = buf;
876 return (char *) str;
879 *ptr = buf;
880 return NULL;
883 /* Reads an offset from *PTR and then locates the string at this offset
884 inside the debug string section. Returns a pointer to the string.
885 Increments *PTR by the number of bytes read for the offset. This
886 value is set even if the function fails. Bytes at or beyond
887 BUF_END will not be read. Returns NULL if there was a problem, or
888 if the string is empty. Does not check for NUL termination of the
889 string. */
891 static char *
892 read_indirect_string (struct comp_unit *unit,
893 bfd_byte **ptr,
894 bfd_byte *buf_end)
896 uint64_t offset;
897 struct dwarf2_debug *stash = unit->stash;
898 struct dwarf2_debug_file *file = unit->file;
899 char *str;
901 if (unit->offset_size > (size_t) (buf_end - *ptr))
903 *ptr = buf_end;
904 return NULL;
907 if (unit->offset_size == 4)
908 offset = read_4_bytes (unit->abfd, ptr, buf_end);
909 else
910 offset = read_8_bytes (unit->abfd, ptr, buf_end);
912 if (! read_section (unit->abfd, &stash->debug_sections[debug_str],
913 file->syms, offset,
914 &file->dwarf_str_buffer, &file->dwarf_str_size))
915 return NULL;
917 str = (char *) file->dwarf_str_buffer + offset;
918 if (*str == '\0')
919 return NULL;
920 return str;
923 /* Like read_indirect_string but from .debug_line_str section. */
925 static char *
926 read_indirect_line_string (struct comp_unit *unit,
927 bfd_byte **ptr,
928 bfd_byte *buf_end)
930 uint64_t offset;
931 struct dwarf2_debug *stash = unit->stash;
932 struct dwarf2_debug_file *file = unit->file;
933 char *str;
935 if (unit->offset_size > (size_t) (buf_end - *ptr))
937 *ptr = buf_end;
938 return NULL;
941 if (unit->offset_size == 4)
942 offset = read_4_bytes (unit->abfd, ptr, buf_end);
943 else
944 offset = read_8_bytes (unit->abfd, ptr, buf_end);
946 if (! read_section (unit->abfd, &stash->debug_sections[debug_line_str],
947 file->syms, offset,
948 &file->dwarf_line_str_buffer,
949 &file->dwarf_line_str_size))
950 return NULL;
952 str = (char *) file->dwarf_line_str_buffer + offset;
953 if (*str == '\0')
954 return NULL;
955 return str;
958 /* Like read_indirect_string but uses a .debug_str located in
959 an alternate file pointed to by the .gnu_debugaltlink section.
960 Used to impement DW_FORM_GNU_strp_alt. */
962 static char *
963 read_alt_indirect_string (struct comp_unit *unit,
964 bfd_byte **ptr,
965 bfd_byte *buf_end)
967 uint64_t offset;
968 struct dwarf2_debug *stash = unit->stash;
969 char *str;
971 if (unit->offset_size > (size_t) (buf_end - *ptr))
973 *ptr = buf_end;
974 return NULL;
977 if (unit->offset_size == 4)
978 offset = read_4_bytes (unit->abfd, ptr, buf_end);
979 else
980 offset = read_8_bytes (unit->abfd, ptr, buf_end);
982 if (stash->alt.bfd_ptr == NULL)
984 bfd *debug_bfd;
985 char *debug_filename = bfd_follow_gnu_debugaltlink (unit->abfd, DEBUGDIR);
987 if (debug_filename == NULL)
988 return NULL;
990 debug_bfd = bfd_openr (debug_filename, NULL);
991 free (debug_filename);
992 if (debug_bfd == NULL)
993 /* FIXME: Should we report our failure to follow the debuglink ? */
994 return NULL;
996 if (!bfd_check_format (debug_bfd, bfd_object))
998 bfd_close (debug_bfd);
999 return NULL;
1001 stash->alt.bfd_ptr = debug_bfd;
1004 if (! read_section (unit->stash->alt.bfd_ptr,
1005 stash->debug_sections + debug_str_alt,
1006 stash->alt.syms, offset,
1007 &stash->alt.dwarf_str_buffer,
1008 &stash->alt.dwarf_str_size))
1009 return NULL;
1011 str = (char *) stash->alt.dwarf_str_buffer + offset;
1012 if (*str == '\0')
1013 return NULL;
1015 return str;
1018 /* Resolve an alternate reference from UNIT at OFFSET.
1019 Returns a pointer into the loaded alternate CU upon success
1020 or NULL upon failure. */
1022 static bfd_byte *
1023 read_alt_indirect_ref (struct comp_unit *unit, uint64_t offset)
1025 struct dwarf2_debug *stash = unit->stash;
1027 if (stash->alt.bfd_ptr == NULL)
1029 bfd *debug_bfd;
1030 char *debug_filename = bfd_follow_gnu_debugaltlink (unit->abfd, DEBUGDIR);
1032 if (debug_filename == NULL)
1033 return NULL;
1035 debug_bfd = bfd_openr (debug_filename, NULL);
1036 free (debug_filename);
1037 if (debug_bfd == NULL)
1038 /* FIXME: Should we report our failure to follow the debuglink ? */
1039 return NULL;
1041 if (!bfd_check_format (debug_bfd, bfd_object))
1043 bfd_close (debug_bfd);
1044 return NULL;
1046 stash->alt.bfd_ptr = debug_bfd;
1049 if (! read_section (unit->stash->alt.bfd_ptr,
1050 stash->debug_sections + debug_info_alt,
1051 stash->alt.syms, offset,
1052 &stash->alt.dwarf_info_buffer,
1053 &stash->alt.dwarf_info_size))
1054 return NULL;
1056 return stash->alt.dwarf_info_buffer + offset;
1059 static uint64_t
1060 read_address (struct comp_unit *unit, bfd_byte **ptr, bfd_byte *buf_end)
1062 bfd_byte *buf = *ptr;
1063 int signed_vma = 0;
1065 if (bfd_get_flavour (unit->abfd) == bfd_target_elf_flavour)
1066 signed_vma = get_elf_backend_data (unit->abfd)->sign_extend_vma;
1068 if (unit->addr_size > (size_t) (buf_end - buf))
1070 *ptr = buf_end;
1071 return 0;
1074 *ptr = buf + unit->addr_size;
1075 if (signed_vma)
1077 switch (unit->addr_size)
1079 case 8:
1080 return bfd_get_signed_64 (unit->abfd, buf);
1081 case 4:
1082 return bfd_get_signed_32 (unit->abfd, buf);
1083 case 2:
1084 return bfd_get_signed_16 (unit->abfd, buf);
1085 default:
1086 abort ();
1089 else
1091 switch (unit->addr_size)
1093 case 8:
1094 return bfd_get_64 (unit->abfd, buf);
1095 case 4:
1096 return bfd_get_32 (unit->abfd, buf);
1097 case 2:
1098 return bfd_get_16 (unit->abfd, buf);
1099 default:
1100 abort ();
1105 /* Lookup an abbrev_info structure in the abbrev hash table. */
1107 static struct abbrev_info *
1108 lookup_abbrev (unsigned int number, struct abbrev_info **abbrevs)
1110 unsigned int hash_number;
1111 struct abbrev_info *abbrev;
1113 hash_number = number % ABBREV_HASH_SIZE;
1114 abbrev = abbrevs[hash_number];
1116 while (abbrev)
1118 if (abbrev->number == number)
1119 return abbrev;
1120 else
1121 abbrev = abbrev->next;
1124 return NULL;
1127 /* We keep a hash table to map .debug_abbrev section offsets to the
1128 array of abbrevs, so that compilation units using the same set of
1129 abbrevs do not waste memory. */
1131 struct abbrev_offset_entry
1133 size_t offset;
1134 struct abbrev_info **abbrevs;
1137 static hashval_t
1138 hash_abbrev (const void *p)
1140 const struct abbrev_offset_entry *ent = p;
1141 return htab_hash_pointer ((void *) ent->offset);
1144 static int
1145 eq_abbrev (const void *pa, const void *pb)
1147 const struct abbrev_offset_entry *a = pa;
1148 const struct abbrev_offset_entry *b = pb;
1149 return a->offset == b->offset;
1152 static void
1153 del_abbrev (void *p)
1155 struct abbrev_offset_entry *ent = p;
1156 struct abbrev_info **abbrevs = ent->abbrevs;
1157 size_t i;
1159 for (i = 0; i < ABBREV_HASH_SIZE; i++)
1161 struct abbrev_info *abbrev = abbrevs[i];
1163 while (abbrev)
1165 free (abbrev->attrs);
1166 abbrev = abbrev->next;
1169 free (ent);
1172 /* In DWARF version 2, the description of the debugging information is
1173 stored in a separate .debug_abbrev section. Before we read any
1174 dies from a section we read in all abbreviations and install them
1175 in a hash table. */
1177 static struct abbrev_info**
1178 read_abbrevs (bfd *abfd, uint64_t offset, struct dwarf2_debug *stash,
1179 struct dwarf2_debug_file *file)
1181 struct abbrev_info **abbrevs;
1182 bfd_byte *abbrev_ptr;
1183 bfd_byte *abbrev_end;
1184 struct abbrev_info *cur_abbrev;
1185 unsigned int abbrev_number, abbrev_name;
1186 unsigned int abbrev_form, hash_number;
1187 size_t amt;
1188 void **slot;
1189 struct abbrev_offset_entry ent = { offset, NULL };
1191 if (ent.offset != offset)
1192 return NULL;
1194 slot = htab_find_slot (file->abbrev_offsets, &ent, INSERT);
1195 if (slot == NULL)
1196 return NULL;
1197 if (*slot != NULL)
1198 return ((struct abbrev_offset_entry *) (*slot))->abbrevs;
1200 if (! read_section (abfd, &stash->debug_sections[debug_abbrev],
1201 file->syms, offset,
1202 &file->dwarf_abbrev_buffer,
1203 &file->dwarf_abbrev_size))
1204 return NULL;
1206 amt = sizeof (struct abbrev_info*) * ABBREV_HASH_SIZE;
1207 abbrevs = (struct abbrev_info **) bfd_zalloc (abfd, amt);
1208 if (abbrevs == NULL)
1209 return NULL;
1211 abbrev_ptr = file->dwarf_abbrev_buffer + offset;
1212 abbrev_end = file->dwarf_abbrev_buffer + file->dwarf_abbrev_size;
1213 abbrev_number = _bfd_safe_read_leb128 (abfd, &abbrev_ptr,
1214 false, abbrev_end);
1216 /* Loop until we reach an abbrev number of 0. */
1217 while (abbrev_number)
1219 amt = sizeof (struct abbrev_info);
1220 cur_abbrev = (struct abbrev_info *) bfd_zalloc (abfd, amt);
1221 if (cur_abbrev == NULL)
1222 goto fail;
1224 /* Read in abbrev header. */
1225 cur_abbrev->number = abbrev_number;
1226 cur_abbrev->tag = (enum dwarf_tag)
1227 _bfd_safe_read_leb128 (abfd, &abbrev_ptr,
1228 false, abbrev_end);
1229 cur_abbrev->has_children = read_1_byte (abfd, &abbrev_ptr, abbrev_end);
1231 /* Now read in declarations. */
1232 for (;;)
1234 /* Initialize it just to avoid a GCC false warning. */
1235 bfd_vma implicit_const = -1;
1237 abbrev_name = _bfd_safe_read_leb128 (abfd, &abbrev_ptr,
1238 false, abbrev_end);
1239 abbrev_form = _bfd_safe_read_leb128 (abfd, &abbrev_ptr,
1240 false, abbrev_end);
1241 if (abbrev_form == DW_FORM_implicit_const)
1242 implicit_const = _bfd_safe_read_leb128 (abfd, &abbrev_ptr,
1243 true, abbrev_end);
1244 if (abbrev_name == 0)
1245 break;
1247 if ((cur_abbrev->num_attrs % ATTR_ALLOC_CHUNK) == 0)
1249 struct attr_abbrev *tmp;
1251 amt = cur_abbrev->num_attrs + ATTR_ALLOC_CHUNK;
1252 amt *= sizeof (struct attr_abbrev);
1253 tmp = (struct attr_abbrev *) bfd_realloc (cur_abbrev->attrs, amt);
1254 if (tmp == NULL)
1255 goto fail;
1256 cur_abbrev->attrs = tmp;
1259 cur_abbrev->attrs[cur_abbrev->num_attrs].name
1260 = (enum dwarf_attribute) abbrev_name;
1261 cur_abbrev->attrs[cur_abbrev->num_attrs].form
1262 = (enum dwarf_form) abbrev_form;
1263 cur_abbrev->attrs[cur_abbrev->num_attrs].implicit_const
1264 = implicit_const;
1265 ++cur_abbrev->num_attrs;
1268 hash_number = abbrev_number % ABBREV_HASH_SIZE;
1269 cur_abbrev->next = abbrevs[hash_number];
1270 abbrevs[hash_number] = cur_abbrev;
1272 /* Get next abbreviation.
1273 Under Irix6 the abbreviations for a compilation unit are not
1274 always properly terminated with an abbrev number of 0.
1275 Exit loop if we encounter an abbreviation which we have
1276 already read (which means we are about to read the abbreviations
1277 for the next compile unit) or if the end of the abbreviation
1278 table is reached. */
1279 if ((size_t) (abbrev_ptr - file->dwarf_abbrev_buffer)
1280 >= file->dwarf_abbrev_size)
1281 break;
1282 abbrev_number = _bfd_safe_read_leb128 (abfd, &abbrev_ptr,
1283 false, abbrev_end);
1284 if (lookup_abbrev (abbrev_number, abbrevs) != NULL)
1285 break;
1288 *slot = bfd_malloc (sizeof ent);
1289 if (!*slot)
1290 goto fail;
1291 ent.abbrevs = abbrevs;
1292 memcpy (*slot, &ent, sizeof ent);
1293 return abbrevs;
1295 fail:
1296 if (abbrevs != NULL)
1298 size_t i;
1300 for (i = 0; i < ABBREV_HASH_SIZE; i++)
1302 struct abbrev_info *abbrev = abbrevs[i];
1304 while (abbrev)
1306 free (abbrev->attrs);
1307 abbrev = abbrev->next;
1310 free (abbrevs);
1312 return NULL;
1315 /* Returns true if the form is one which has a string value. */
1317 static bool
1318 is_str_form (const struct attribute *attr)
1320 switch (attr->form)
1322 case DW_FORM_string:
1323 case DW_FORM_strp:
1324 case DW_FORM_strx:
1325 case DW_FORM_strx1:
1326 case DW_FORM_strx2:
1327 case DW_FORM_strx3:
1328 case DW_FORM_strx4:
1329 case DW_FORM_line_strp:
1330 case DW_FORM_GNU_strp_alt:
1331 return true;
1333 default:
1334 return false;
1338 /* Returns true if the form is one which has an integer value. */
1340 static bool
1341 is_int_form (const struct attribute *attr)
1343 switch (attr->form)
1345 case DW_FORM_addr:
1346 case DW_FORM_data2:
1347 case DW_FORM_data4:
1348 case DW_FORM_data8:
1349 case DW_FORM_data1:
1350 case DW_FORM_flag:
1351 case DW_FORM_sdata:
1352 case DW_FORM_udata:
1353 case DW_FORM_ref_addr:
1354 case DW_FORM_ref1:
1355 case DW_FORM_ref2:
1356 case DW_FORM_ref4:
1357 case DW_FORM_ref8:
1358 case DW_FORM_ref_udata:
1359 case DW_FORM_sec_offset:
1360 case DW_FORM_flag_present:
1361 case DW_FORM_ref_sig8:
1362 case DW_FORM_addrx:
1363 case DW_FORM_implicit_const:
1364 case DW_FORM_addrx1:
1365 case DW_FORM_addrx2:
1366 case DW_FORM_addrx3:
1367 case DW_FORM_addrx4:
1368 case DW_FORM_GNU_ref_alt:
1369 return true;
1371 default:
1372 return false;
1376 /* Returns true if the form is strx[1-4]. */
1378 static inline bool
1379 is_strx_form (enum dwarf_form form)
1381 return (form == DW_FORM_strx
1382 || form == DW_FORM_strx1
1383 || form == DW_FORM_strx2
1384 || form == DW_FORM_strx3
1385 || form == DW_FORM_strx4);
1388 /* Return true if the form is addrx[1-4]. */
1390 static inline bool
1391 is_addrx_form (enum dwarf_form form)
1393 return (form == DW_FORM_addrx
1394 || form == DW_FORM_addrx1
1395 || form == DW_FORM_addrx2
1396 || form == DW_FORM_addrx3
1397 || form == DW_FORM_addrx4);
1400 /* Returns the address in .debug_addr section using DW_AT_addr_base.
1401 Used to implement DW_FORM_addrx*. */
1402 static uint64_t
1403 read_indexed_address (uint64_t idx, struct comp_unit *unit)
1405 struct dwarf2_debug *stash = unit->stash;
1406 struct dwarf2_debug_file *file = unit->file;
1407 bfd_byte *info_ptr;
1408 size_t offset;
1410 if (stash == NULL)
1411 return 0;
1413 if (!read_section (unit->abfd, &stash->debug_sections[debug_addr],
1414 file->syms, 0,
1415 &file->dwarf_addr_buffer, &file->dwarf_addr_size))
1416 return 0;
1418 if (_bfd_mul_overflow (idx, unit->addr_size, &offset))
1419 return 0;
1421 offset += unit->dwarf_addr_offset;
1422 if (offset < unit->dwarf_addr_offset
1423 || offset > file->dwarf_addr_size
1424 || file->dwarf_addr_size - offset < unit->addr_size)
1425 return 0;
1427 info_ptr = file->dwarf_addr_buffer + offset;
1429 if (unit->addr_size == 4)
1430 return bfd_get_32 (unit->abfd, info_ptr);
1431 else if (unit->addr_size == 8)
1432 return bfd_get_64 (unit->abfd, info_ptr);
1433 else
1434 return 0;
1437 /* Returns the string using DW_AT_str_offsets_base.
1438 Used to implement DW_FORM_strx*. */
1439 static const char *
1440 read_indexed_string (uint64_t idx, struct comp_unit *unit)
1442 struct dwarf2_debug *stash = unit->stash;
1443 struct dwarf2_debug_file *file = unit->file;
1444 bfd_byte *info_ptr;
1445 uint64_t str_offset;
1446 size_t offset;
1448 if (stash == NULL)
1449 return NULL;
1451 if (!read_section (unit->abfd, &stash->debug_sections[debug_str],
1452 file->syms, 0,
1453 &file->dwarf_str_buffer, &file->dwarf_str_size))
1454 return NULL;
1456 if (!read_section (unit->abfd, &stash->debug_sections[debug_str_offsets],
1457 file->syms, 0,
1458 &file->dwarf_str_offsets_buffer,
1459 &file->dwarf_str_offsets_size))
1460 return NULL;
1462 if (_bfd_mul_overflow (idx, unit->offset_size, &offset))
1463 return NULL;
1465 offset += unit->dwarf_str_offset;
1466 if (offset < unit->dwarf_str_offset
1467 || offset > file->dwarf_str_offsets_size
1468 || file->dwarf_str_offsets_size - offset < unit->offset_size)
1469 return NULL;
1471 info_ptr = file->dwarf_str_offsets_buffer + offset;
1473 if (unit->offset_size == 4)
1474 str_offset = bfd_get_32 (unit->abfd, info_ptr);
1475 else if (unit->offset_size == 8)
1476 str_offset = bfd_get_64 (unit->abfd, info_ptr);
1477 else
1478 return NULL;
1480 if (str_offset >= file->dwarf_str_size)
1481 return NULL;
1482 return (const char *) file->dwarf_str_buffer + str_offset;
1485 /* Read and fill in the value of attribute ATTR as described by FORM.
1486 Read data starting from INFO_PTR, but never at or beyond INFO_PTR_END.
1487 Returns an updated INFO_PTR taking into account the amount of data read. */
1489 static bfd_byte *
1490 read_attribute_value (struct attribute * attr,
1491 unsigned form,
1492 bfd_vma implicit_const,
1493 struct comp_unit * unit,
1494 bfd_byte * info_ptr,
1495 bfd_byte * info_ptr_end)
1497 bfd *abfd = unit->abfd;
1498 size_t amt;
1500 if (info_ptr >= info_ptr_end && form != DW_FORM_flag_present)
1502 _bfd_error_handler (_("DWARF error: info pointer extends beyond end of attributes"));
1503 bfd_set_error (bfd_error_bad_value);
1504 return NULL;
1507 attr->form = (enum dwarf_form) form;
1509 switch (form)
1511 case DW_FORM_flag_present:
1512 attr->u.val = 1;
1513 break;
1514 case DW_FORM_ref_addr:
1515 /* DW_FORM_ref_addr is an address in DWARF2, and an offset in
1516 DWARF3. */
1517 if (unit->version >= 3)
1519 if (unit->offset_size == 4)
1520 attr->u.val = read_4_bytes (unit->abfd, &info_ptr, info_ptr_end);
1521 else
1522 attr->u.val = read_8_bytes (unit->abfd, &info_ptr, info_ptr_end);
1523 break;
1525 /* FALLTHROUGH */
1526 case DW_FORM_addr:
1527 attr->u.val = read_address (unit, &info_ptr, info_ptr_end);
1528 break;
1529 case DW_FORM_GNU_ref_alt:
1530 case DW_FORM_sec_offset:
1531 if (unit->offset_size == 4)
1532 attr->u.val = read_4_bytes (unit->abfd, &info_ptr, info_ptr_end);
1533 else
1534 attr->u.val = read_8_bytes (unit->abfd, &info_ptr, info_ptr_end);
1535 break;
1536 case DW_FORM_block2:
1537 amt = read_2_bytes (abfd, &info_ptr, info_ptr_end);
1538 attr->u.blk = read_blk (abfd, &info_ptr, info_ptr_end, amt);
1539 if (attr->u.blk == NULL)
1540 return NULL;
1541 break;
1542 case DW_FORM_block4:
1543 amt = read_4_bytes (abfd, &info_ptr, info_ptr_end);
1544 attr->u.blk = read_blk (abfd, &info_ptr, info_ptr_end, amt);
1545 if (attr->u.blk == NULL)
1546 return NULL;
1547 break;
1548 case DW_FORM_ref1:
1549 case DW_FORM_flag:
1550 case DW_FORM_data1:
1551 attr->u.val = read_1_byte (abfd, &info_ptr, info_ptr_end);
1552 break;
1553 case DW_FORM_addrx1:
1554 attr->u.val = read_1_byte (abfd, &info_ptr, info_ptr_end);
1555 /* dwarf_addr_offset value 0 indicates the attribute DW_AT_addr_base
1556 is not yet read. */
1557 if (unit->dwarf_addr_offset != 0)
1558 attr->u.val = read_indexed_address (attr->u.val, unit);
1559 break;
1560 case DW_FORM_data2:
1561 case DW_FORM_ref2:
1562 attr->u.val = read_2_bytes (abfd, &info_ptr, info_ptr_end);
1563 break;
1564 case DW_FORM_addrx2:
1565 attr->u.val = read_2_bytes (abfd, &info_ptr, info_ptr_end);
1566 if (unit->dwarf_addr_offset != 0)
1567 attr->u.val = read_indexed_address (attr->u.val, unit);
1568 break;
1569 case DW_FORM_addrx3:
1570 attr->u.val = read_3_bytes (abfd, &info_ptr, info_ptr_end);
1571 if (unit->dwarf_addr_offset != 0)
1572 attr->u.val = read_indexed_address(attr->u.val, unit);
1573 break;
1574 case DW_FORM_ref4:
1575 case DW_FORM_data4:
1576 attr->u.val = read_4_bytes (abfd, &info_ptr, info_ptr_end);
1577 break;
1578 case DW_FORM_addrx4:
1579 attr->u.val = read_4_bytes (abfd, &info_ptr, info_ptr_end);
1580 if (unit->dwarf_addr_offset != 0)
1581 attr->u.val = read_indexed_address (attr->u.val, unit);
1582 break;
1583 case DW_FORM_data8:
1584 case DW_FORM_ref8:
1585 case DW_FORM_ref_sig8:
1586 attr->u.val = read_8_bytes (abfd, &info_ptr, info_ptr_end);
1587 break;
1588 case DW_FORM_string:
1589 attr->u.str = read_string (&info_ptr, info_ptr_end);
1590 break;
1591 case DW_FORM_strp:
1592 attr->u.str = read_indirect_string (unit, &info_ptr, info_ptr_end);
1593 break;
1594 case DW_FORM_line_strp:
1595 attr->u.str = read_indirect_line_string (unit, &info_ptr, info_ptr_end);
1596 break;
1597 case DW_FORM_GNU_strp_alt:
1598 attr->u.str = read_alt_indirect_string (unit, &info_ptr, info_ptr_end);
1599 break;
1600 case DW_FORM_strx1:
1601 attr->u.val = read_1_byte (abfd, &info_ptr, info_ptr_end);
1602 /* dwarf_str_offset value 0 indicates the attribute DW_AT_str_offsets_base
1603 is not yet read. */
1604 if (unit->dwarf_str_offset != 0)
1605 attr->u.str = (char *) read_indexed_string (attr->u.val, unit);
1606 else
1607 attr->u.str = NULL;
1608 break;
1609 case DW_FORM_strx2:
1610 attr->u.val = read_2_bytes (abfd, &info_ptr, info_ptr_end);
1611 if (unit->dwarf_str_offset != 0)
1612 attr->u.str = (char *) read_indexed_string (attr->u.val, unit);
1613 else
1614 attr->u.str = NULL;
1615 break;
1616 case DW_FORM_strx3:
1617 attr->u.val = read_3_bytes (abfd, &info_ptr, info_ptr_end);
1618 if (unit->dwarf_str_offset != 0)
1619 attr->u.str = (char *) read_indexed_string (attr->u.val, unit);
1620 else
1621 attr->u.str = NULL;
1622 break;
1623 case DW_FORM_strx4:
1624 attr->u.val = read_4_bytes (abfd, &info_ptr, info_ptr_end);
1625 if (unit->dwarf_str_offset != 0)
1626 attr->u.str = (char *) read_indexed_string (attr->u.val, unit);
1627 else
1628 attr->u.str = NULL;
1629 break;
1630 case DW_FORM_strx:
1631 attr->u.val = _bfd_safe_read_leb128 (abfd, &info_ptr,
1632 false, info_ptr_end);
1633 if (unit->dwarf_str_offset != 0)
1634 attr->u.str = (char *) read_indexed_string (attr->u.val, unit);
1635 else
1636 attr->u.str = NULL;
1637 break;
1638 case DW_FORM_exprloc:
1639 case DW_FORM_block:
1640 amt = _bfd_safe_read_leb128 (abfd, &info_ptr,
1641 false, info_ptr_end);
1642 attr->u.blk = read_blk (abfd, &info_ptr, info_ptr_end, amt);
1643 if (attr->u.blk == NULL)
1644 return NULL;
1645 break;
1646 case DW_FORM_block1:
1647 amt = read_1_byte (abfd, &info_ptr, info_ptr_end);
1648 attr->u.blk = read_blk (abfd, &info_ptr, info_ptr_end, amt);
1649 if (attr->u.blk == NULL)
1650 return NULL;
1651 break;
1652 case DW_FORM_sdata:
1653 attr->u.sval = _bfd_safe_read_leb128 (abfd, &info_ptr,
1654 true, info_ptr_end);
1655 break;
1657 case DW_FORM_rnglistx:
1658 case DW_FORM_loclistx:
1659 /* FIXME: Add support for these forms! */
1660 /* Fall through. */
1661 case DW_FORM_ref_udata:
1662 case DW_FORM_udata:
1663 attr->u.val = _bfd_safe_read_leb128 (abfd, &info_ptr,
1664 false, info_ptr_end);
1665 break;
1666 case DW_FORM_addrx:
1667 attr->u.val = _bfd_safe_read_leb128 (abfd, &info_ptr,
1668 false, info_ptr_end);
1669 if (unit->dwarf_addr_offset != 0)
1670 attr->u.val = read_indexed_address (attr->u.val, unit);
1671 break;
1672 case DW_FORM_indirect:
1673 form = _bfd_safe_read_leb128 (abfd, &info_ptr,
1674 false, info_ptr_end);
1675 if (form == DW_FORM_implicit_const)
1676 implicit_const = _bfd_safe_read_leb128 (abfd, &info_ptr,
1677 true, info_ptr_end);
1678 info_ptr = read_attribute_value (attr, form, implicit_const, unit,
1679 info_ptr, info_ptr_end);
1680 break;
1681 case DW_FORM_implicit_const:
1682 attr->form = DW_FORM_sdata;
1683 attr->u.sval = implicit_const;
1684 break;
1685 case DW_FORM_data16:
1686 /* This is really a "constant", but there is no way to store that
1687 so pretend it is a 16 byte block instead. */
1688 attr->u.blk = read_blk (abfd, &info_ptr, info_ptr_end, 16);
1689 if (attr->u.blk == NULL)
1690 return NULL;
1691 break;
1693 default:
1694 _bfd_error_handler (_("DWARF error: invalid or unhandled FORM value: %#x"),
1695 form);
1696 bfd_set_error (bfd_error_bad_value);
1697 return NULL;
1699 return info_ptr;
1702 /* Read an attribute described by an abbreviated attribute. */
1704 static bfd_byte *
1705 read_attribute (struct attribute * attr,
1706 struct attr_abbrev * abbrev,
1707 struct comp_unit * unit,
1708 bfd_byte * info_ptr,
1709 bfd_byte * info_ptr_end)
1711 attr->name = abbrev->name;
1712 info_ptr = read_attribute_value (attr, abbrev->form, abbrev->implicit_const,
1713 unit, info_ptr, info_ptr_end);
1714 return info_ptr;
1717 /* Return mangling style given LANG. */
1719 static int
1720 mangle_style (int lang)
1722 switch (lang)
1724 case DW_LANG_Ada83:
1725 case DW_LANG_Ada95:
1726 return DMGL_GNAT;
1728 case DW_LANG_C_plus_plus:
1729 case DW_LANG_C_plus_plus_03:
1730 case DW_LANG_C_plus_plus_11:
1731 case DW_LANG_C_plus_plus_14:
1732 return DMGL_GNU_V3;
1734 case DW_LANG_Java:
1735 return DMGL_JAVA;
1737 case DW_LANG_D:
1738 return DMGL_DLANG;
1740 case DW_LANG_Rust:
1741 case DW_LANG_Rust_old:
1742 return DMGL_RUST;
1744 default:
1745 return DMGL_AUTO;
1747 case DW_LANG_C89:
1748 case DW_LANG_C:
1749 case DW_LANG_Cobol74:
1750 case DW_LANG_Cobol85:
1751 case DW_LANG_Fortran77:
1752 case DW_LANG_Pascal83:
1753 case DW_LANG_PLI:
1754 case DW_LANG_C99:
1755 case DW_LANG_UPC:
1756 case DW_LANG_C11:
1757 case DW_LANG_Mips_Assembler:
1758 case DW_LANG_Upc:
1759 case DW_LANG_HP_Basic91:
1760 case DW_LANG_HP_IMacro:
1761 case DW_LANG_HP_Assembler:
1762 return 0;
1766 /* Source line information table routines. */
1768 #define FILE_ALLOC_CHUNK 5
1769 #define DIR_ALLOC_CHUNK 5
1771 struct line_info
1773 struct line_info * prev_line;
1774 bfd_vma address;
1775 char * filename;
1776 unsigned int line;
1777 unsigned int column;
1778 unsigned int discriminator;
1779 unsigned char op_index;
1780 unsigned char end_sequence; /* End of (sequential) code sequence. */
1783 struct fileinfo
1785 char * name;
1786 unsigned int dir;
1787 unsigned int time;
1788 unsigned int size;
1791 struct line_sequence
1793 bfd_vma low_pc;
1794 struct line_sequence* prev_sequence;
1795 struct line_info* last_line; /* Largest VMA. */
1796 struct line_info** line_info_lookup;
1797 bfd_size_type num_lines;
1800 struct line_info_table
1802 bfd * abfd;
1803 unsigned int num_files;
1804 unsigned int num_dirs;
1805 unsigned int num_sequences;
1806 bool use_dir_and_file_0;
1807 char * comp_dir;
1808 char ** dirs;
1809 struct fileinfo* files;
1810 struct line_sequence* sequences;
1811 struct line_info* lcl_head; /* Local head; used in 'add_line_info'. */
1814 /* Remember some information about each function. If the function is
1815 inlined (DW_TAG_inlined_subroutine) it may have two additional
1816 attributes, DW_AT_call_file and DW_AT_call_line, which specify the
1817 source code location where this function was inlined. */
1819 struct funcinfo
1821 /* Pointer to previous function in list of all functions. */
1822 struct funcinfo *prev_func;
1823 /* Pointer to function one scope higher. */
1824 struct funcinfo *caller_func;
1825 /* Source location file name where caller_func inlines this func. */
1826 char *caller_file;
1827 /* Source location file name. */
1828 char *file;
1829 /* Source location line number where caller_func inlines this func. */
1830 int caller_line;
1831 /* Source location line number. */
1832 int line;
1833 int tag;
1834 bool is_linkage;
1835 const char *name;
1836 struct arange arange;
1837 /* The offset of the funcinfo from the start of the unit. */
1838 uint64_t unit_offset;
1841 struct lookup_funcinfo
1843 /* Function information corresponding to this lookup table entry. */
1844 struct funcinfo *funcinfo;
1846 /* The lowest address for this specific function. */
1847 bfd_vma low_addr;
1849 /* The highest address of this function before the lookup table is sorted.
1850 The highest address of all prior functions after the lookup table is
1851 sorted, which is used for binary search. */
1852 bfd_vma high_addr;
1853 /* Index of this function, used to ensure qsort is stable. */
1854 unsigned int idx;
1857 struct varinfo
1859 /* Pointer to previous variable in list of all variables. */
1860 struct varinfo *prev_var;
1861 /* The offset of the varinfo from the start of the unit. */
1862 uint64_t unit_offset;
1863 /* Source location file name. */
1864 char *file;
1865 /* Source location line number. */
1866 int line;
1867 /* The type of this variable. */
1868 int tag;
1869 /* The name of the variable, if it has one. */
1870 const char *name;
1871 /* The address of the variable. */
1872 bfd_vma addr;
1873 /* Is this a stack variable? */
1874 bool stack;
1877 /* Return TRUE if NEW_LINE should sort after LINE. */
1879 static inline bool
1880 new_line_sorts_after (struct line_info *new_line, struct line_info *line)
1882 return (new_line->address > line->address
1883 || (new_line->address == line->address
1884 && new_line->op_index > line->op_index));
1888 /* Adds a new entry to the line_info list in the line_info_table, ensuring
1889 that the list is sorted. Note that the line_info list is sorted from
1890 highest to lowest VMA (with possible duplicates); that is,
1891 line_info->prev_line always accesses an equal or smaller VMA. */
1893 static bool
1894 add_line_info (struct line_info_table *table,
1895 bfd_vma address,
1896 unsigned char op_index,
1897 char *filename,
1898 unsigned int line,
1899 unsigned int column,
1900 unsigned int discriminator,
1901 int end_sequence)
1903 size_t amt = sizeof (struct line_info);
1904 struct line_sequence* seq = table->sequences;
1905 struct line_info* info = (struct line_info *) bfd_alloc (table->abfd, amt);
1907 if (info == NULL)
1908 return false;
1910 /* Set member data of 'info'. */
1911 info->prev_line = NULL;
1912 info->address = address;
1913 info->op_index = op_index;
1914 info->line = line;
1915 info->column = column;
1916 info->discriminator = discriminator;
1917 info->end_sequence = end_sequence;
1919 if (filename && filename[0])
1921 info->filename = (char *) bfd_alloc (table->abfd, strlen (filename) + 1);
1922 if (info->filename == NULL)
1923 return false;
1924 strcpy (info->filename, filename);
1926 else
1927 info->filename = NULL;
1929 /* Find the correct location for 'info'. Normally we will receive
1930 new line_info data 1) in order and 2) with increasing VMAs.
1931 However some compilers break the rules (cf. decode_line_info) and
1932 so we include some heuristics for quickly finding the correct
1933 location for 'info'. In particular, these heuristics optimize for
1934 the common case in which the VMA sequence that we receive is a
1935 list of locally sorted VMAs such as
1936 p...z a...j (where a < j < p < z)
1938 Note: table->lcl_head is used to head an *actual* or *possible*
1939 sub-sequence within the list (such as a...j) that is not directly
1940 headed by table->last_line
1942 Note: we may receive duplicate entries from 'decode_line_info'. */
1944 if (seq
1945 && seq->last_line->address == address
1946 && seq->last_line->op_index == op_index
1947 && seq->last_line->end_sequence == end_sequence)
1949 /* We only keep the last entry with the same address and end
1950 sequence. See PR ld/4986. */
1951 if (table->lcl_head == seq->last_line)
1952 table->lcl_head = info;
1953 info->prev_line = seq->last_line->prev_line;
1954 seq->last_line = info;
1956 else if (!seq || seq->last_line->end_sequence)
1958 /* Start a new line sequence. */
1959 amt = sizeof (struct line_sequence);
1960 seq = (struct line_sequence *) bfd_malloc (amt);
1961 if (seq == NULL)
1962 return false;
1963 seq->low_pc = address;
1964 seq->prev_sequence = table->sequences;
1965 seq->last_line = info;
1966 table->lcl_head = info;
1967 table->sequences = seq;
1968 table->num_sequences++;
1970 else if (info->end_sequence
1971 || new_line_sorts_after (info, seq->last_line))
1973 /* Normal case: add 'info' to the beginning of the current sequence. */
1974 info->prev_line = seq->last_line;
1975 seq->last_line = info;
1977 /* lcl_head: initialize to head a *possible* sequence at the end. */
1978 if (!table->lcl_head)
1979 table->lcl_head = info;
1981 else if (!new_line_sorts_after (info, table->lcl_head)
1982 && (!table->lcl_head->prev_line
1983 || new_line_sorts_after (info, table->lcl_head->prev_line)))
1985 /* Abnormal but easy: lcl_head is the head of 'info'. */
1986 info->prev_line = table->lcl_head->prev_line;
1987 table->lcl_head->prev_line = info;
1989 else
1991 /* Abnormal and hard: Neither 'last_line' nor 'lcl_head'
1992 are valid heads for 'info'. Reset 'lcl_head'. */
1993 struct line_info* li2 = seq->last_line; /* Always non-NULL. */
1994 struct line_info* li1 = li2->prev_line;
1996 while (li1)
1998 if (!new_line_sorts_after (info, li2)
1999 && new_line_sorts_after (info, li1))
2000 break;
2002 li2 = li1; /* always non-NULL */
2003 li1 = li1->prev_line;
2005 table->lcl_head = li2;
2006 info->prev_line = table->lcl_head->prev_line;
2007 table->lcl_head->prev_line = info;
2008 if (address < seq->low_pc)
2009 seq->low_pc = address;
2011 return true;
2014 /* Extract a fully qualified filename from a line info table.
2015 The returned string has been malloc'ed and it is the caller's
2016 responsibility to free it. */
2018 static char *
2019 concat_filename (struct line_info_table *table, unsigned int file)
2021 char *filename;
2023 /* Pre DWARF-5 entry 0 in the directory and filename tables was not used.
2024 So in order to save space in the tables used here the info for, eg
2025 directory 1 is stored in slot 0 of the directory table, directory 2
2026 in slot 1 and so on.
2028 Starting with DWARF-5 the 0'th entry is used so there is a one to one
2029 mapping between DWARF slots and internal table entries. */
2030 if (! table->use_dir_and_file_0)
2032 /* Pre DWARF-5, FILE == 0 means unknown. */
2033 if (file == 0)
2034 return strdup ("<unknown>");
2035 -- file;
2038 if (table == NULL || file >= table->num_files)
2040 _bfd_error_handler
2041 (_("DWARF error: mangled line number section (bad file number)"));
2042 return strdup ("<unknown>");
2045 filename = table->files[file].name;
2047 if (filename == NULL)
2048 return strdup ("<unknown>");
2050 if (!IS_ABSOLUTE_PATH (filename))
2052 char *dir_name = NULL;
2053 char *subdir_name = NULL;
2054 char *name;
2055 size_t len;
2056 unsigned int dir = table->files[file].dir;
2058 if (!table->use_dir_and_file_0)
2059 --dir;
2060 /* Wrapping from 0 to -1u above gives the intended result with
2061 the test below of leaving subdir_name NULL for pre-DWARF5 dir
2062 of 0. */
2063 /* PR 17512: file: 0317e960, file: 7f3d2e4b. */
2064 if (dir < table->num_dirs)
2065 subdir_name = table->dirs[dir];
2067 if (!subdir_name || !IS_ABSOLUTE_PATH (subdir_name))
2068 dir_name = table->comp_dir;
2070 if (!dir_name)
2072 dir_name = subdir_name;
2073 subdir_name = NULL;
2076 if (!dir_name)
2077 return strdup (filename);
2079 len = strlen (dir_name) + strlen (filename) + 2;
2081 if (subdir_name)
2083 len += strlen (subdir_name) + 1;
2084 name = (char *) bfd_malloc (len);
2085 if (name)
2086 sprintf (name, "%s/%s/%s", dir_name, subdir_name, filename);
2088 else
2090 name = (char *) bfd_malloc (len);
2091 if (name)
2092 sprintf (name, "%s/%s", dir_name, filename);
2095 return name;
2098 return strdup (filename);
2101 /* Number of bits in a bfd_vma. */
2102 #define VMA_BITS (8 * sizeof (bfd_vma))
2104 /* Check whether [low1, high1) can be combined with [low2, high2),
2105 i.e., they touch or overlap. */
2107 static bool
2108 ranges_overlap (bfd_vma low1,
2109 bfd_vma high1,
2110 bfd_vma low2,
2111 bfd_vma high2)
2113 if (low1 == low2 || high1 == high2)
2114 return true;
2116 /* Sort so that low1 is below low2. */
2117 if (low1 > low2)
2119 bfd_vma tmp;
2121 tmp = low1;
2122 low1 = low2;
2123 low2 = tmp;
2125 tmp = high1;
2126 high1 = high2;
2127 high2 = tmp;
2130 /* We touch iff low2 == high1.
2131 We overlap iff low2 is within [low1, high1). */
2132 return low2 <= high1;
2135 /* Insert an address range in the trie mapping addresses to compilation units.
2136 Will return the new trie node (usually the same as is being sent in, but
2137 in case of a leaf-to-interior conversion, or expansion of a leaf, it may be
2138 different), or NULL on failure. */
2140 static struct trie_node *
2141 insert_arange_in_trie (bfd *abfd,
2142 struct trie_node *trie,
2143 bfd_vma trie_pc,
2144 unsigned int trie_pc_bits,
2145 struct comp_unit *unit,
2146 bfd_vma low_pc,
2147 bfd_vma high_pc)
2149 bfd_vma clamped_low_pc, clamped_high_pc;
2150 int ch, from_ch, to_ch;
2151 bool is_full_leaf = false;
2153 /* See if we can extend any of the existing ranges. This merging
2154 isn't perfect (if merging opens up the possibility of merging two existing
2155 ranges, we won't find them), but it takes the majority of the cases. */
2156 if (trie->num_room_in_leaf > 0)
2158 struct trie_leaf *leaf = (struct trie_leaf *) trie;
2159 unsigned int i;
2161 for (i = 0; i < leaf->num_stored_in_leaf; ++i)
2163 if (leaf->ranges[i].unit == unit
2164 && ranges_overlap (low_pc, high_pc,
2165 leaf->ranges[i].low_pc,
2166 leaf->ranges[i].high_pc))
2168 if (low_pc < leaf->ranges[i].low_pc)
2169 leaf->ranges[i].low_pc = low_pc;
2170 if (high_pc > leaf->ranges[i].high_pc)
2171 leaf->ranges[i].high_pc = high_pc;
2172 return trie;
2176 is_full_leaf = leaf->num_stored_in_leaf == trie->num_room_in_leaf;
2179 /* If we're a leaf with no more room and we're _not_ at the bottom,
2180 convert to an interior node. */
2181 if (is_full_leaf && trie_pc_bits < VMA_BITS)
2183 const struct trie_leaf *leaf = (struct trie_leaf *) trie;
2184 unsigned int i;
2186 trie = bfd_zalloc (abfd, sizeof (struct trie_interior));
2187 if (!trie)
2188 return NULL;
2189 is_full_leaf = false;
2191 /* TODO: If we wanted to save a little more memory at the cost of
2192 complexity, we could have reused the old leaf node as one of the
2193 children of the new interior node, instead of throwing it away. */
2194 for (i = 0; i < leaf->num_stored_in_leaf; ++i)
2196 if (!insert_arange_in_trie (abfd, trie, trie_pc, trie_pc_bits,
2197 leaf->ranges[i].unit, leaf->ranges[i].low_pc,
2198 leaf->ranges[i].high_pc))
2199 return NULL;
2203 /* If we're a leaf with no more room and we _are_ at the bottom,
2204 we have no choice but to just make it larger. */
2205 if (is_full_leaf)
2207 const struct trie_leaf *leaf = (struct trie_leaf *) trie;
2208 unsigned int new_room_in_leaf = trie->num_room_in_leaf * 2;
2209 struct trie_leaf *new_leaf;
2210 size_t amt = (sizeof (struct trie_leaf)
2211 + ((new_room_in_leaf - TRIE_LEAF_SIZE)
2212 * sizeof (leaf->ranges[0])));
2213 new_leaf = bfd_zalloc (abfd, amt);
2214 new_leaf->head.num_room_in_leaf = new_room_in_leaf;
2215 new_leaf->num_stored_in_leaf = leaf->num_stored_in_leaf;
2217 memcpy (new_leaf->ranges,
2218 leaf->ranges,
2219 leaf->num_stored_in_leaf * sizeof (leaf->ranges[0]));
2220 trie = &new_leaf->head;
2221 is_full_leaf = false;
2223 /* Now the insert below will go through. */
2226 /* If we're a leaf (now with room), we can just insert at the end. */
2227 if (trie->num_room_in_leaf > 0)
2229 struct trie_leaf *leaf = (struct trie_leaf *) trie;
2231 unsigned int i = leaf->num_stored_in_leaf++;
2232 leaf->ranges[i].unit = unit;
2233 leaf->ranges[i].low_pc = low_pc;
2234 leaf->ranges[i].high_pc = high_pc;
2235 return trie;
2238 /* Now we are definitely an interior node, so recurse into all
2239 the relevant buckets. */
2241 /* Clamp the range to the current trie bucket. */
2242 clamped_low_pc = low_pc;
2243 clamped_high_pc = high_pc;
2244 if (trie_pc_bits > 0)
2246 bfd_vma bucket_high_pc =
2247 trie_pc + ((bfd_vma) -1 >> trie_pc_bits); /* Inclusive. */
2248 if (clamped_low_pc < trie_pc)
2249 clamped_low_pc = trie_pc;
2250 if (clamped_high_pc > bucket_high_pc)
2251 clamped_high_pc = bucket_high_pc;
2254 /* Insert the ranges in all buckets that it spans. */
2255 from_ch = (clamped_low_pc >> (VMA_BITS - trie_pc_bits - 8)) & 0xff;
2256 to_ch = ((clamped_high_pc - 1) >> (VMA_BITS - trie_pc_bits - 8)) & 0xff;
2257 for (ch = from_ch; ch <= to_ch; ++ch)
2259 struct trie_interior *interior = (struct trie_interior *) trie;
2260 struct trie_node *child = interior->children[ch];
2262 if (child == NULL)
2264 child = alloc_trie_leaf (abfd);
2265 if (!child)
2266 return NULL;
2268 bfd_vma bucket = (bfd_vma) ch << (VMA_BITS - trie_pc_bits - 8);
2269 child = insert_arange_in_trie (abfd,
2270 child,
2271 trie_pc + bucket,
2272 trie_pc_bits + 8,
2273 unit,
2274 low_pc,
2275 high_pc);
2276 if (!child)
2277 return NULL;
2279 interior->children[ch] = child;
2282 return trie;
2285 static bool
2286 arange_add (struct comp_unit *unit, struct arange *first_arange,
2287 struct trie_node **trie_root, bfd_vma low_pc, bfd_vma high_pc)
2289 struct arange *arange;
2291 /* Ignore empty ranges. */
2292 if (low_pc == high_pc)
2293 return true;
2295 if (trie_root != NULL)
2297 *trie_root = insert_arange_in_trie (unit->file->bfd_ptr,
2298 *trie_root,
2301 unit,
2302 low_pc,
2303 high_pc);
2304 if (*trie_root == NULL)
2305 return false;
2308 /* If the first arange is empty, use it. */
2309 if (first_arange->high == 0)
2311 first_arange->low = low_pc;
2312 first_arange->high = high_pc;
2313 return true;
2316 /* Next see if we can cheaply extend an existing range. */
2317 arange = first_arange;
2320 if (low_pc == arange->high)
2322 arange->high = high_pc;
2323 return true;
2325 if (high_pc == arange->low)
2327 arange->low = low_pc;
2328 return true;
2330 arange = arange->next;
2332 while (arange);
2334 /* Need to allocate a new arange and insert it into the arange list.
2335 Order isn't significant, so just insert after the first arange. */
2336 arange = (struct arange *) bfd_alloc (unit->abfd, sizeof (*arange));
2337 if (arange == NULL)
2338 return false;
2339 arange->low = low_pc;
2340 arange->high = high_pc;
2341 arange->next = first_arange->next;
2342 first_arange->next = arange;
2343 return true;
2346 /* Compare function for line sequences. */
2348 static int
2349 compare_sequences (const void* a, const void* b)
2351 const struct line_sequence* seq1 = a;
2352 const struct line_sequence* seq2 = b;
2354 /* Sort by low_pc as the primary key. */
2355 if (seq1->low_pc < seq2->low_pc)
2356 return -1;
2357 if (seq1->low_pc > seq2->low_pc)
2358 return 1;
2360 /* If low_pc values are equal, sort in reverse order of
2361 high_pc, so that the largest region comes first. */
2362 if (seq1->last_line->address < seq2->last_line->address)
2363 return 1;
2364 if (seq1->last_line->address > seq2->last_line->address)
2365 return -1;
2367 if (seq1->last_line->op_index < seq2->last_line->op_index)
2368 return 1;
2369 if (seq1->last_line->op_index > seq2->last_line->op_index)
2370 return -1;
2372 /* num_lines is initially an index, to make the sort stable. */
2373 if (seq1->num_lines < seq2->num_lines)
2374 return -1;
2375 if (seq1->num_lines > seq2->num_lines)
2376 return 1;
2377 return 0;
2380 /* Construct the line information table for quick lookup. */
2382 static bool
2383 build_line_info_table (struct line_info_table * table,
2384 struct line_sequence * seq)
2386 size_t amt;
2387 struct line_info **line_info_lookup;
2388 struct line_info *each_line;
2389 unsigned int num_lines;
2390 unsigned int line_index;
2392 if (seq->line_info_lookup != NULL)
2393 return true;
2395 /* Count the number of line information entries. We could do this while
2396 scanning the debug information, but some entries may be added via
2397 lcl_head without having a sequence handy to increment the number of
2398 lines. */
2399 num_lines = 0;
2400 for (each_line = seq->last_line; each_line; each_line = each_line->prev_line)
2401 num_lines++;
2403 seq->num_lines = num_lines;
2404 if (num_lines == 0)
2405 return true;
2407 /* Allocate space for the line information lookup table. */
2408 amt = sizeof (struct line_info*) * num_lines;
2409 line_info_lookup = (struct line_info**) bfd_alloc (table->abfd, amt);
2410 seq->line_info_lookup = line_info_lookup;
2411 if (line_info_lookup == NULL)
2412 return false;
2414 /* Create the line information lookup table. */
2415 line_index = num_lines;
2416 for (each_line = seq->last_line; each_line; each_line = each_line->prev_line)
2417 line_info_lookup[--line_index] = each_line;
2419 BFD_ASSERT (line_index == 0);
2420 return true;
2423 /* Sort the line sequences for quick lookup. */
2425 static bool
2426 sort_line_sequences (struct line_info_table* table)
2428 size_t amt;
2429 struct line_sequence *sequences;
2430 struct line_sequence *seq;
2431 unsigned int n = 0;
2432 unsigned int num_sequences = table->num_sequences;
2433 bfd_vma last_high_pc;
2435 if (num_sequences == 0)
2436 return true;
2438 /* Allocate space for an array of sequences. */
2439 amt = sizeof (struct line_sequence) * num_sequences;
2440 sequences = (struct line_sequence *) bfd_alloc (table->abfd, amt);
2441 if (sequences == NULL)
2442 return false;
2444 /* Copy the linked list into the array, freeing the original nodes. */
2445 seq = table->sequences;
2446 for (n = 0; n < num_sequences; n++)
2448 struct line_sequence* last_seq = seq;
2450 BFD_ASSERT (seq);
2451 sequences[n].low_pc = seq->low_pc;
2452 sequences[n].prev_sequence = NULL;
2453 sequences[n].last_line = seq->last_line;
2454 sequences[n].line_info_lookup = NULL;
2455 sequences[n].num_lines = n;
2456 seq = seq->prev_sequence;
2457 free (last_seq);
2459 BFD_ASSERT (seq == NULL);
2461 qsort (sequences, n, sizeof (struct line_sequence), compare_sequences);
2463 /* Make the list binary-searchable by trimming overlapping entries
2464 and removing nested entries. */
2465 num_sequences = 1;
2466 last_high_pc = sequences[0].last_line->address;
2467 for (n = 1; n < table->num_sequences; n++)
2469 if (sequences[n].low_pc < last_high_pc)
2471 if (sequences[n].last_line->address <= last_high_pc)
2472 /* Skip nested entries. */
2473 continue;
2475 /* Trim overlapping entries. */
2476 sequences[n].low_pc = last_high_pc;
2478 last_high_pc = sequences[n].last_line->address;
2479 if (n > num_sequences)
2481 /* Close up the gap. */
2482 sequences[num_sequences].low_pc = sequences[n].low_pc;
2483 sequences[num_sequences].last_line = sequences[n].last_line;
2485 num_sequences++;
2488 table->sequences = sequences;
2489 table->num_sequences = num_sequences;
2490 return true;
2493 /* Add directory to TABLE. CUR_DIR memory ownership is taken by TABLE. */
2495 static bool
2496 line_info_add_include_dir (struct line_info_table *table, char *cur_dir)
2498 if ((table->num_dirs % DIR_ALLOC_CHUNK) == 0)
2500 char **tmp;
2501 size_t amt;
2503 amt = table->num_dirs + DIR_ALLOC_CHUNK;
2504 amt *= sizeof (char *);
2506 tmp = (char **) bfd_realloc (table->dirs, amt);
2507 if (tmp == NULL)
2508 return false;
2509 table->dirs = tmp;
2512 table->dirs[table->num_dirs++] = cur_dir;
2513 return true;
2516 static bool
2517 line_info_add_include_dir_stub (struct line_info_table *table, char *cur_dir,
2518 unsigned int dir ATTRIBUTE_UNUSED,
2519 unsigned int xtime ATTRIBUTE_UNUSED,
2520 unsigned int size ATTRIBUTE_UNUSED)
2522 return line_info_add_include_dir (table, cur_dir);
2525 /* Add file to TABLE. CUR_FILE memory ownership is taken by TABLE. */
2527 static bool
2528 line_info_add_file_name (struct line_info_table *table, char *cur_file,
2529 unsigned int dir, unsigned int xtime,
2530 unsigned int size)
2532 if ((table->num_files % FILE_ALLOC_CHUNK) == 0)
2534 struct fileinfo *tmp;
2535 size_t amt;
2537 amt = table->num_files + FILE_ALLOC_CHUNK;
2538 amt *= sizeof (struct fileinfo);
2540 tmp = (struct fileinfo *) bfd_realloc (table->files, amt);
2541 if (tmp == NULL)
2542 return false;
2543 table->files = tmp;
2546 table->files[table->num_files].name = cur_file;
2547 table->files[table->num_files].dir = dir;
2548 table->files[table->num_files].time = xtime;
2549 table->files[table->num_files].size = size;
2550 table->num_files++;
2551 return true;
2554 /* Read directory or file name entry format, starting with byte of
2555 format count entries, ULEB128 pairs of entry formats, ULEB128 of
2556 entries count and the entries themselves in the described entry
2557 format. */
2559 static bool
2560 read_formatted_entries (struct comp_unit *unit, bfd_byte **bufp,
2561 bfd_byte *buf_end, struct line_info_table *table,
2562 bool (*callback) (struct line_info_table *table,
2563 char *cur_file,
2564 unsigned int dir,
2565 unsigned int time,
2566 unsigned int size))
2568 bfd *abfd = unit->abfd;
2569 bfd_byte format_count, formati;
2570 bfd_vma data_count, datai;
2571 bfd_byte *buf = *bufp;
2572 bfd_byte *format_header_data;
2574 format_count = read_1_byte (abfd, &buf, buf_end);
2575 format_header_data = buf;
2576 for (formati = 0; formati < format_count; formati++)
2578 _bfd_safe_read_leb128 (abfd, &buf, false, buf_end);
2579 _bfd_safe_read_leb128 (abfd, &buf, false, buf_end);
2582 data_count = _bfd_safe_read_leb128 (abfd, &buf, false, buf_end);
2583 if (format_count == 0 && data_count != 0)
2585 _bfd_error_handler (_("DWARF error: zero format count"));
2586 bfd_set_error (bfd_error_bad_value);
2587 return false;
2590 /* PR 22210. Paranoia check. Don't bother running the loop
2591 if we know that we are going to run out of buffer. */
2592 if (data_count > (bfd_vma) (buf_end - buf))
2594 _bfd_error_handler
2595 (_("DWARF error: data count (%" PRIx64 ") larger than buffer size"),
2596 (uint64_t) data_count);
2597 bfd_set_error (bfd_error_bad_value);
2598 return false;
2601 for (datai = 0; datai < data_count; datai++)
2603 bfd_byte *format = format_header_data;
2604 struct fileinfo fe;
2606 memset (&fe, 0, sizeof fe);
2607 for (formati = 0; formati < format_count; formati++)
2609 bfd_vma content_type, form;
2610 char *string_trash;
2611 char **stringp = &string_trash;
2612 unsigned int uint_trash, *uintp = &uint_trash;
2613 struct attribute attr;
2615 content_type = _bfd_safe_read_leb128 (abfd, &format, false, buf_end);
2616 switch (content_type)
2618 case DW_LNCT_path:
2619 stringp = &fe.name;
2620 break;
2621 case DW_LNCT_directory_index:
2622 uintp = &fe.dir;
2623 break;
2624 case DW_LNCT_timestamp:
2625 uintp = &fe.time;
2626 break;
2627 case DW_LNCT_size:
2628 uintp = &fe.size;
2629 break;
2630 case DW_LNCT_MD5:
2631 break;
2632 default:
2633 _bfd_error_handler
2634 (_("DWARF error: unknown format content type %" PRIu64),
2635 (uint64_t) content_type);
2636 bfd_set_error (bfd_error_bad_value);
2637 return false;
2640 form = _bfd_safe_read_leb128 (abfd, &format, false, buf_end);
2641 buf = read_attribute_value (&attr, form, 0, unit, buf, buf_end);
2642 if (buf == NULL)
2643 return false;
2644 switch (form)
2646 case DW_FORM_string:
2647 case DW_FORM_line_strp:
2648 case DW_FORM_strx:
2649 case DW_FORM_strx1:
2650 case DW_FORM_strx2:
2651 case DW_FORM_strx3:
2652 case DW_FORM_strx4:
2653 *stringp = attr.u.str;
2654 break;
2656 case DW_FORM_data1:
2657 case DW_FORM_data2:
2658 case DW_FORM_data4:
2659 case DW_FORM_data8:
2660 case DW_FORM_udata:
2661 *uintp = attr.u.val;
2662 break;
2664 case DW_FORM_data16:
2665 /* MD5 data is in the attr.blk, but we are ignoring those. */
2666 break;
2670 if (!callback (table, fe.name, fe.dir, fe.time, fe.size))
2671 return false;
2674 *bufp = buf;
2675 return true;
2678 /* Decode the line number information for UNIT. */
2680 static struct line_info_table*
2681 decode_line_info (struct comp_unit *unit)
2683 bfd *abfd = unit->abfd;
2684 struct dwarf2_debug *stash = unit->stash;
2685 struct dwarf2_debug_file *file = unit->file;
2686 struct line_info_table* table;
2687 bfd_byte *line_ptr;
2688 bfd_byte *line_end;
2689 struct line_head lh;
2690 unsigned int i, offset_size;
2691 char *cur_file, *cur_dir;
2692 unsigned char op_code, extended_op, adj_opcode;
2693 unsigned int exop_len;
2694 size_t amt;
2696 if (unit->line_offset == 0 && file->line_table)
2697 return file->line_table;
2699 if (! read_section (abfd, &stash->debug_sections[debug_line],
2700 file->syms, unit->line_offset,
2701 &file->dwarf_line_buffer, &file->dwarf_line_size))
2702 return NULL;
2704 if (file->dwarf_line_size < 16)
2706 _bfd_error_handler
2707 (_("DWARF error: line info section is too small (%" PRId64 ")"),
2708 (int64_t) file->dwarf_line_size);
2709 bfd_set_error (bfd_error_bad_value);
2710 return NULL;
2712 line_ptr = file->dwarf_line_buffer + unit->line_offset;
2713 line_end = file->dwarf_line_buffer + file->dwarf_line_size;
2715 /* Read in the prologue. */
2716 lh.total_length = read_4_bytes (abfd, &line_ptr, line_end);
2717 offset_size = 4;
2718 if (lh.total_length == 0xffffffff)
2720 lh.total_length = read_8_bytes (abfd, &line_ptr, line_end);
2721 offset_size = 8;
2723 else if (lh.total_length == 0 && unit->addr_size == 8)
2725 /* Handle (non-standard) 64-bit DWARF2 formats. */
2726 lh.total_length = read_4_bytes (abfd, &line_ptr, line_end);
2727 offset_size = 8;
2730 if (lh.total_length > (size_t) (line_end - line_ptr))
2732 _bfd_error_handler
2733 /* xgettext: c-format */
2734 (_("DWARF error: line info data is bigger (%#" PRIx64 ")"
2735 " than the space remaining in the section (%#lx)"),
2736 (uint64_t) lh.total_length, (unsigned long) (line_end - line_ptr));
2737 bfd_set_error (bfd_error_bad_value);
2738 return NULL;
2741 line_end = line_ptr + lh.total_length;
2743 lh.version = read_2_bytes (abfd, &line_ptr, line_end);
2744 if (lh.version < 2 || lh.version > 5)
2746 _bfd_error_handler
2747 (_("DWARF error: unhandled .debug_line version %d"), lh.version);
2748 bfd_set_error (bfd_error_bad_value);
2749 return NULL;
2752 if (line_ptr + offset_size + (lh.version >= 5 ? 8 : (lh.version >= 4 ? 6 : 5))
2753 >= line_end)
2755 _bfd_error_handler
2756 (_("DWARF error: ran out of room reading prologue"));
2757 bfd_set_error (bfd_error_bad_value);
2758 return NULL;
2761 if (lh.version >= 5)
2763 unsigned int segment_selector_size;
2765 /* Skip address size. */
2766 read_1_byte (abfd, &line_ptr, line_end);
2768 segment_selector_size = read_1_byte (abfd, &line_ptr, line_end);
2769 if (segment_selector_size != 0)
2771 _bfd_error_handler
2772 (_("DWARF error: line info unsupported segment selector size %u"),
2773 segment_selector_size);
2774 bfd_set_error (bfd_error_bad_value);
2775 return NULL;
2779 if (offset_size == 4)
2780 lh.prologue_length = read_4_bytes (abfd, &line_ptr, line_end);
2781 else
2782 lh.prologue_length = read_8_bytes (abfd, &line_ptr, line_end);
2784 lh.minimum_instruction_length = read_1_byte (abfd, &line_ptr, line_end);
2786 if (lh.version >= 4)
2787 lh.maximum_ops_per_insn = read_1_byte (abfd, &line_ptr, line_end);
2788 else
2789 lh.maximum_ops_per_insn = 1;
2791 if (lh.maximum_ops_per_insn == 0)
2793 _bfd_error_handler
2794 (_("DWARF error: invalid maximum operations per instruction"));
2795 bfd_set_error (bfd_error_bad_value);
2796 return NULL;
2799 lh.default_is_stmt = read_1_byte (abfd, &line_ptr, line_end);
2800 lh.line_base = read_1_signed_byte (abfd, &line_ptr, line_end);
2801 lh.line_range = read_1_byte (abfd, &line_ptr, line_end);
2802 lh.opcode_base = read_1_byte (abfd, &line_ptr, line_end);
2804 if (line_ptr + (lh.opcode_base - 1) >= line_end)
2806 _bfd_error_handler (_("DWARF error: ran out of room reading opcodes"));
2807 bfd_set_error (bfd_error_bad_value);
2808 return NULL;
2811 amt = lh.opcode_base * sizeof (unsigned char);
2812 lh.standard_opcode_lengths = (unsigned char *) bfd_alloc (abfd, amt);
2814 lh.standard_opcode_lengths[0] = 1;
2816 for (i = 1; i < lh.opcode_base; ++i)
2817 lh.standard_opcode_lengths[i] = read_1_byte (abfd, &line_ptr, line_end);
2819 amt = sizeof (struct line_info_table);
2820 table = (struct line_info_table *) bfd_alloc (abfd, amt);
2821 if (table == NULL)
2822 return NULL;
2823 table->abfd = abfd;
2824 table->comp_dir = unit->comp_dir;
2826 table->num_files = 0;
2827 table->files = NULL;
2829 table->num_dirs = 0;
2830 table->dirs = NULL;
2832 table->num_sequences = 0;
2833 table->sequences = NULL;
2835 table->lcl_head = NULL;
2837 if (lh.version >= 5)
2839 /* Read directory table. */
2840 if (!read_formatted_entries (unit, &line_ptr, line_end, table,
2841 line_info_add_include_dir_stub))
2842 goto fail;
2844 /* Read file name table. */
2845 if (!read_formatted_entries (unit, &line_ptr, line_end, table,
2846 line_info_add_file_name))
2847 goto fail;
2848 table->use_dir_and_file_0 = true;
2850 else
2852 /* Read directory table. */
2853 while ((cur_dir = read_string (&line_ptr, line_end)) != NULL)
2855 if (!line_info_add_include_dir (table, cur_dir))
2856 goto fail;
2859 /* Read file name table. */
2860 while ((cur_file = read_string (&line_ptr, line_end)) != NULL)
2862 unsigned int dir, xtime, size;
2864 dir = _bfd_safe_read_leb128 (abfd, &line_ptr, false, line_end);
2865 xtime = _bfd_safe_read_leb128 (abfd, &line_ptr, false, line_end);
2866 size = _bfd_safe_read_leb128 (abfd, &line_ptr, false, line_end);
2868 if (!line_info_add_file_name (table, cur_file, dir, xtime, size))
2869 goto fail;
2871 table->use_dir_and_file_0 = false;
2874 /* Read the statement sequences until there's nothing left. */
2875 while (line_ptr < line_end)
2877 /* State machine registers. */
2878 bfd_vma address = 0;
2879 unsigned char op_index = 0;
2880 char * filename = NULL;
2881 unsigned int line = 1;
2882 unsigned int column = 0;
2883 unsigned int discriminator = 0;
2884 int is_stmt = lh.default_is_stmt;
2885 int end_sequence = 0;
2886 unsigned int dir, xtime, size;
2887 /* eraxxon@alumni.rice.edu: Against the DWARF2 specs, some
2888 compilers generate address sequences that are wildly out of
2889 order using DW_LNE_set_address (e.g. Intel C++ 6.0 compiler
2890 for ia64-Linux). Thus, to determine the low and high
2891 address, we must compare on every DW_LNS_copy, etc. */
2892 bfd_vma low_pc = (bfd_vma) -1;
2893 bfd_vma high_pc = 0;
2895 if (table->num_files)
2897 if (table->use_dir_and_file_0)
2898 filename = concat_filename (table, 0);
2899 else
2900 filename = concat_filename (table, 1);
2903 /* Decode the table. */
2904 while (!end_sequence && line_ptr < line_end)
2906 op_code = read_1_byte (abfd, &line_ptr, line_end);
2908 if (op_code >= lh.opcode_base)
2910 /* Special operand. */
2911 adj_opcode = op_code - lh.opcode_base;
2912 if (lh.line_range == 0)
2913 goto line_fail;
2914 if (lh.maximum_ops_per_insn == 1)
2915 address += (adj_opcode / lh.line_range
2916 * lh.minimum_instruction_length);
2917 else
2919 address += ((op_index + adj_opcode / lh.line_range)
2920 / lh.maximum_ops_per_insn
2921 * lh.minimum_instruction_length);
2922 op_index = ((op_index + adj_opcode / lh.line_range)
2923 % lh.maximum_ops_per_insn);
2925 line += lh.line_base + (adj_opcode % lh.line_range);
2926 /* Append row to matrix using current values. */
2927 if (!add_line_info (table, address, op_index, filename,
2928 line, column, discriminator, 0))
2929 goto line_fail;
2930 discriminator = 0;
2931 if (address < low_pc)
2932 low_pc = address;
2933 if (address > high_pc)
2934 high_pc = address;
2936 else switch (op_code)
2938 case DW_LNS_extended_op:
2939 exop_len = _bfd_safe_read_leb128 (abfd, &line_ptr,
2940 false, line_end);
2941 extended_op = read_1_byte (abfd, &line_ptr, line_end);
2943 switch (extended_op)
2945 case DW_LNE_end_sequence:
2946 end_sequence = 1;
2947 if (!add_line_info (table, address, op_index, filename, line,
2948 column, discriminator, end_sequence))
2949 goto line_fail;
2950 discriminator = 0;
2951 if (address < low_pc)
2952 low_pc = address;
2953 if (address > high_pc)
2954 high_pc = address;
2955 if (!arange_add (unit, &unit->arange, &unit->file->trie_root,
2956 low_pc, high_pc))
2957 goto line_fail;
2958 break;
2959 case DW_LNE_set_address:
2960 address = read_address (unit, &line_ptr, line_end);
2961 op_index = 0;
2962 break;
2963 case DW_LNE_define_file:
2964 cur_file = read_string (&line_ptr, line_end);
2965 dir = _bfd_safe_read_leb128 (abfd, &line_ptr,
2966 false, line_end);
2967 xtime = _bfd_safe_read_leb128 (abfd, &line_ptr,
2968 false, line_end);
2969 size = _bfd_safe_read_leb128 (abfd, &line_ptr,
2970 false, line_end);
2971 if (!line_info_add_file_name (table, cur_file, dir,
2972 xtime, size))
2973 goto line_fail;
2974 break;
2975 case DW_LNE_set_discriminator:
2976 discriminator = _bfd_safe_read_leb128 (abfd, &line_ptr,
2977 false, line_end);
2978 break;
2979 case DW_LNE_HP_source_file_correlation:
2980 line_ptr += exop_len - 1;
2981 break;
2982 default:
2983 _bfd_error_handler
2984 (_("DWARF error: mangled line number section"));
2985 bfd_set_error (bfd_error_bad_value);
2986 line_fail:
2987 free (filename);
2988 goto fail;
2990 break;
2991 case DW_LNS_copy:
2992 if (!add_line_info (table, address, op_index,
2993 filename, line, column, discriminator, 0))
2994 goto line_fail;
2995 discriminator = 0;
2996 if (address < low_pc)
2997 low_pc = address;
2998 if (address > high_pc)
2999 high_pc = address;
3000 break;
3001 case DW_LNS_advance_pc:
3002 if (lh.maximum_ops_per_insn == 1)
3003 address += (lh.minimum_instruction_length
3004 * _bfd_safe_read_leb128 (abfd, &line_ptr,
3005 false, line_end));
3006 else
3008 bfd_vma adjust = _bfd_safe_read_leb128 (abfd, &line_ptr,
3009 false, line_end);
3010 address = ((op_index + adjust) / lh.maximum_ops_per_insn
3011 * lh.minimum_instruction_length);
3012 op_index = (op_index + adjust) % lh.maximum_ops_per_insn;
3014 break;
3015 case DW_LNS_advance_line:
3016 line += _bfd_safe_read_leb128 (abfd, &line_ptr,
3017 true, line_end);
3018 break;
3019 case DW_LNS_set_file:
3021 unsigned int filenum;
3023 /* The file and directory tables are 0
3024 based, the references are 1 based. */
3025 filenum = _bfd_safe_read_leb128 (abfd, &line_ptr,
3026 false, line_end);
3027 free (filename);
3028 filename = concat_filename (table, filenum);
3029 break;
3031 case DW_LNS_set_column:
3032 column = _bfd_safe_read_leb128 (abfd, &line_ptr,
3033 false, line_end);
3034 break;
3035 case DW_LNS_negate_stmt:
3036 is_stmt = (!is_stmt);
3037 break;
3038 case DW_LNS_set_basic_block:
3039 break;
3040 case DW_LNS_const_add_pc:
3041 if (lh.line_range == 0)
3042 goto line_fail;
3043 if (lh.maximum_ops_per_insn == 1)
3044 address += (lh.minimum_instruction_length
3045 * ((255 - lh.opcode_base) / lh.line_range));
3046 else
3048 bfd_vma adjust = ((255 - lh.opcode_base) / lh.line_range);
3049 address += (lh.minimum_instruction_length
3050 * ((op_index + adjust)
3051 / lh.maximum_ops_per_insn));
3052 op_index = (op_index + adjust) % lh.maximum_ops_per_insn;
3054 break;
3055 case DW_LNS_fixed_advance_pc:
3056 address += read_2_bytes (abfd, &line_ptr, line_end);
3057 op_index = 0;
3058 break;
3059 default:
3060 /* Unknown standard opcode, ignore it. */
3061 for (i = 0; i < lh.standard_opcode_lengths[op_code]; i++)
3062 (void) _bfd_safe_read_leb128 (abfd, &line_ptr,
3063 false, line_end);
3064 break;
3068 free (filename);
3071 if (unit->line_offset == 0)
3072 file->line_table = table;
3073 if (sort_line_sequences (table))
3074 return table;
3076 fail:
3077 while (table->sequences != NULL)
3079 struct line_sequence* seq = table->sequences;
3080 table->sequences = table->sequences->prev_sequence;
3081 free (seq);
3083 free (table->files);
3084 free (table->dirs);
3085 return NULL;
3088 /* If ADDR is within TABLE set the output parameters and return TRUE,
3089 otherwise set *FILENAME_PTR to NULL and return FALSE.
3090 The parameters FILENAME_PTR, LINENUMBER_PTR and DISCRIMINATOR_PTR
3091 are pointers to the objects to be filled in. */
3093 static bool
3094 lookup_address_in_line_info_table (struct line_info_table *table,
3095 bfd_vma addr,
3096 const char **filename_ptr,
3097 unsigned int *linenumber_ptr,
3098 unsigned int *discriminator_ptr)
3100 struct line_sequence *seq = NULL;
3101 struct line_info *info;
3102 int low, high, mid;
3104 /* Binary search the array of sequences. */
3105 low = 0;
3106 high = table->num_sequences;
3107 while (low < high)
3109 mid = (low + high) / 2;
3110 seq = &table->sequences[mid];
3111 if (addr < seq->low_pc)
3112 high = mid;
3113 else if (addr >= seq->last_line->address)
3114 low = mid + 1;
3115 else
3116 break;
3119 /* Check for a valid sequence. */
3120 if (!seq || addr < seq->low_pc || addr >= seq->last_line->address)
3121 goto fail;
3123 if (!build_line_info_table (table, seq))
3124 goto fail;
3126 /* Binary search the array of line information. */
3127 low = 0;
3128 high = seq->num_lines;
3129 info = NULL;
3130 while (low < high)
3132 mid = (low + high) / 2;
3133 info = seq->line_info_lookup[mid];
3134 if (addr < info->address)
3135 high = mid;
3136 else if (addr >= seq->line_info_lookup[mid + 1]->address)
3137 low = mid + 1;
3138 else
3139 break;
3142 /* Check for a valid line information entry. */
3143 if (info
3144 && addr >= info->address
3145 && addr < seq->line_info_lookup[mid + 1]->address
3146 && !(info->end_sequence || info == seq->last_line))
3148 *filename_ptr = info->filename;
3149 *linenumber_ptr = info->line;
3150 if (discriminator_ptr)
3151 *discriminator_ptr = info->discriminator;
3152 return true;
3155 fail:
3156 *filename_ptr = NULL;
3157 return false;
3160 /* Read in the .debug_ranges section for future reference. */
3162 static bool
3163 read_debug_ranges (struct comp_unit * unit)
3165 struct dwarf2_debug *stash = unit->stash;
3166 struct dwarf2_debug_file *file = unit->file;
3168 return read_section (unit->abfd, &stash->debug_sections[debug_ranges],
3169 file->syms, 0,
3170 &file->dwarf_ranges_buffer, &file->dwarf_ranges_size);
3173 /* Read in the .debug_rnglists section for future reference. */
3175 static bool
3176 read_debug_rnglists (struct comp_unit * unit)
3178 struct dwarf2_debug *stash = unit->stash;
3179 struct dwarf2_debug_file *file = unit->file;
3181 return read_section (unit->abfd, &stash->debug_sections[debug_rnglists],
3182 file->syms, 0,
3183 &file->dwarf_rnglists_buffer, &file->dwarf_rnglists_size);
3186 /* Function table functions. */
3188 static int
3189 compare_lookup_funcinfos (const void * a, const void * b)
3191 const struct lookup_funcinfo * lookup1 = a;
3192 const struct lookup_funcinfo * lookup2 = b;
3194 if (lookup1->low_addr < lookup2->low_addr)
3195 return -1;
3196 if (lookup1->low_addr > lookup2->low_addr)
3197 return 1;
3198 if (lookup1->high_addr < lookup2->high_addr)
3199 return -1;
3200 if (lookup1->high_addr > lookup2->high_addr)
3201 return 1;
3203 if (lookup1->idx < lookup2->idx)
3204 return -1;
3205 if (lookup1->idx > lookup2->idx)
3206 return 1;
3207 return 0;
3210 static bool
3211 build_lookup_funcinfo_table (struct comp_unit * unit)
3213 struct lookup_funcinfo *lookup_funcinfo_table = unit->lookup_funcinfo_table;
3214 unsigned int number_of_functions = unit->number_of_functions;
3215 struct funcinfo *each;
3216 struct lookup_funcinfo *entry;
3217 size_t func_index;
3218 struct arange *range;
3219 bfd_vma low_addr, high_addr;
3221 if (lookup_funcinfo_table || number_of_functions == 0)
3222 return true;
3224 /* Create the function info lookup table. */
3225 lookup_funcinfo_table = (struct lookup_funcinfo *)
3226 bfd_malloc (number_of_functions * sizeof (struct lookup_funcinfo));
3227 if (lookup_funcinfo_table == NULL)
3228 return false;
3230 /* Populate the function info lookup table. */
3231 func_index = number_of_functions;
3232 for (each = unit->function_table; each; each = each->prev_func)
3234 entry = &lookup_funcinfo_table[--func_index];
3235 entry->funcinfo = each;
3236 entry->idx = func_index;
3238 /* Calculate the lowest and highest address for this function entry. */
3239 low_addr = entry->funcinfo->arange.low;
3240 high_addr = entry->funcinfo->arange.high;
3242 for (range = entry->funcinfo->arange.next; range; range = range->next)
3244 if (range->low < low_addr)
3245 low_addr = range->low;
3246 if (range->high > high_addr)
3247 high_addr = range->high;
3250 entry->low_addr = low_addr;
3251 entry->high_addr = high_addr;
3254 BFD_ASSERT (func_index == 0);
3256 /* Sort the function by address. */
3257 qsort (lookup_funcinfo_table,
3258 number_of_functions,
3259 sizeof (struct lookup_funcinfo),
3260 compare_lookup_funcinfos);
3262 /* Calculate the high watermark for each function in the lookup table. */
3263 high_addr = lookup_funcinfo_table[0].high_addr;
3264 for (func_index = 1; func_index < number_of_functions; func_index++)
3266 entry = &lookup_funcinfo_table[func_index];
3267 if (entry->high_addr > high_addr)
3268 high_addr = entry->high_addr;
3269 else
3270 entry->high_addr = high_addr;
3273 unit->lookup_funcinfo_table = lookup_funcinfo_table;
3274 return true;
3277 /* If ADDR is within UNIT's function tables, set FUNCTION_PTR, and return
3278 TRUE. Note that we need to find the function that has the smallest range
3279 that contains ADDR, to handle inlined functions without depending upon
3280 them being ordered in TABLE by increasing range. */
3282 static bool
3283 lookup_address_in_function_table (struct comp_unit *unit,
3284 bfd_vma addr,
3285 struct funcinfo **function_ptr)
3287 unsigned int number_of_functions = unit->number_of_functions;
3288 struct lookup_funcinfo* lookup_funcinfo = NULL;
3289 struct funcinfo* funcinfo = NULL;
3290 struct funcinfo* best_fit = NULL;
3291 bfd_vma best_fit_len = (bfd_vma) -1;
3292 bfd_size_type low, high, mid, first;
3293 struct arange *arange;
3295 if (number_of_functions == 0)
3296 return false;
3298 if (!build_lookup_funcinfo_table (unit))
3299 return false;
3301 if (unit->lookup_funcinfo_table[number_of_functions - 1].high_addr < addr)
3302 return false;
3304 /* Find the first function in the lookup table which may contain the
3305 specified address. */
3306 low = 0;
3307 high = number_of_functions;
3308 first = high;
3309 while (low < high)
3311 mid = (low + high) / 2;
3312 lookup_funcinfo = &unit->lookup_funcinfo_table[mid];
3313 if (addr < lookup_funcinfo->low_addr)
3314 high = mid;
3315 else if (addr >= lookup_funcinfo->high_addr)
3316 low = mid + 1;
3317 else
3318 high = first = mid;
3321 /* Find the 'best' match for the address. The prior algorithm defined the
3322 best match as the function with the smallest address range containing
3323 the specified address. This definition should probably be changed to the
3324 innermost inline routine containing the address, but right now we want
3325 to get the same results we did before. */
3326 while (first < number_of_functions)
3328 if (addr < unit->lookup_funcinfo_table[first].low_addr)
3329 break;
3330 funcinfo = unit->lookup_funcinfo_table[first].funcinfo;
3332 for (arange = &funcinfo->arange; arange; arange = arange->next)
3334 if (addr < arange->low || addr >= arange->high)
3335 continue;
3337 if (arange->high - arange->low < best_fit_len
3338 /* The following comparison is designed to return the same
3339 match as the previous algorithm for routines which have the
3340 same best fit length. */
3341 || (arange->high - arange->low == best_fit_len
3342 && funcinfo > best_fit))
3344 best_fit = funcinfo;
3345 best_fit_len = arange->high - arange->low;
3349 first++;
3352 if (!best_fit)
3353 return false;
3355 *function_ptr = best_fit;
3356 return true;
3359 /* If SYM at ADDR is within function table of UNIT, set FILENAME_PTR
3360 and LINENUMBER_PTR, and return TRUE. */
3362 static bool
3363 lookup_symbol_in_function_table (struct comp_unit *unit,
3364 asymbol *sym,
3365 bfd_vma addr,
3366 const char **filename_ptr,
3367 unsigned int *linenumber_ptr)
3369 struct funcinfo* each;
3370 struct funcinfo* best_fit = NULL;
3371 bfd_vma best_fit_len = (bfd_vma) -1;
3372 struct arange *arange;
3373 const char *name = bfd_asymbol_name (sym);
3375 for (each = unit->function_table; each; each = each->prev_func)
3376 for (arange = &each->arange; arange; arange = arange->next)
3377 if (addr >= arange->low
3378 && addr < arange->high
3379 && arange->high - arange->low < best_fit_len
3380 && each->file
3381 && each->name
3382 && strstr (name, each->name) != NULL)
3384 best_fit = each;
3385 best_fit_len = arange->high - arange->low;
3388 if (best_fit)
3390 *filename_ptr = best_fit->file;
3391 *linenumber_ptr = best_fit->line;
3392 return true;
3395 return false;
3398 /* Variable table functions. */
3400 /* If SYM is within variable table of UNIT, set FILENAME_PTR and
3401 LINENUMBER_PTR, and return TRUE. */
3403 static bool
3404 lookup_symbol_in_variable_table (struct comp_unit *unit,
3405 asymbol *sym,
3406 bfd_vma addr,
3407 const char **filename_ptr,
3408 unsigned int *linenumber_ptr)
3410 struct varinfo* each;
3411 const char *name = bfd_asymbol_name (sym);
3413 for (each = unit->variable_table; each; each = each->prev_var)
3414 if (each->addr == addr
3415 && !each->stack
3416 && each->file != NULL
3417 && each->name != NULL
3418 && strstr (name, each->name) != NULL)
3419 break;
3421 if (each)
3423 *filename_ptr = each->file;
3424 *linenumber_ptr = each->line;
3425 return true;
3428 return false;
3431 static struct comp_unit *stash_comp_unit (struct dwarf2_debug *,
3432 struct dwarf2_debug_file *);
3433 static bool comp_unit_maybe_decode_line_info (struct comp_unit *);
3435 static bool
3436 find_abstract_instance (struct comp_unit *unit,
3437 struct attribute *attr_ptr,
3438 unsigned int recur_count,
3439 const char **pname,
3440 bool *is_linkage,
3441 char **filename_ptr,
3442 int *linenumber_ptr)
3444 bfd *abfd = unit->abfd;
3445 bfd_byte *info_ptr = NULL;
3446 bfd_byte *info_ptr_end;
3447 unsigned int abbrev_number, i;
3448 struct abbrev_info *abbrev;
3449 uint64_t die_ref = attr_ptr->u.val;
3450 struct attribute attr;
3452 if (recur_count == 100)
3454 _bfd_error_handler
3455 (_("DWARF error: abstract instance recursion detected"));
3456 bfd_set_error (bfd_error_bad_value);
3457 return false;
3460 /* DW_FORM_ref_addr can reference an entry in a different CU. It
3461 is an offset from the .debug_info section, not the current CU. */
3462 if (attr_ptr->form == DW_FORM_ref_addr)
3464 /* We only support DW_FORM_ref_addr within the same file, so
3465 any relocations should be resolved already. Check this by
3466 testing for a zero die_ref; There can't be a valid reference
3467 to the header of a .debug_info section.
3468 DW_FORM_ref_addr is an offset relative to .debug_info.
3469 Normally when using the GNU linker this is accomplished by
3470 emitting a symbolic reference to a label, because .debug_info
3471 sections are linked at zero. When there are multiple section
3472 groups containing .debug_info, as there might be in a
3473 relocatable object file, it would be reasonable to assume that
3474 a symbolic reference to a label in any .debug_info section
3475 might be used. Since we lay out multiple .debug_info
3476 sections at non-zero VMAs (see place_sections), and read
3477 them contiguously into dwarf_info_buffer, that means the
3478 reference is relative to dwarf_info_buffer. */
3479 size_t total;
3481 info_ptr = unit->file->dwarf_info_buffer;
3482 info_ptr_end = info_ptr + unit->file->dwarf_info_size;
3483 total = info_ptr_end - info_ptr;
3484 if (!die_ref)
3485 return true;
3486 else if (die_ref >= total)
3488 _bfd_error_handler
3489 (_("DWARF error: invalid abstract instance DIE ref"));
3490 bfd_set_error (bfd_error_bad_value);
3491 return false;
3493 info_ptr += die_ref;
3495 else if (attr_ptr->form == DW_FORM_GNU_ref_alt)
3497 bool first_time = unit->stash->alt.dwarf_info_buffer == NULL;
3499 info_ptr = read_alt_indirect_ref (unit, die_ref);
3500 if (first_time)
3501 unit->stash->alt.info_ptr = unit->stash->alt.dwarf_info_buffer;
3502 if (info_ptr == NULL)
3504 _bfd_error_handler
3505 (_("DWARF error: unable to read alt ref %" PRIu64),
3506 (uint64_t) die_ref);
3507 bfd_set_error (bfd_error_bad_value);
3508 return false;
3510 info_ptr_end = (unit->stash->alt.dwarf_info_buffer
3511 + unit->stash->alt.dwarf_info_size);
3512 if (unit->stash->alt.all_comp_units)
3513 unit = unit->stash->alt.all_comp_units;
3516 if (attr_ptr->form == DW_FORM_ref_addr
3517 || attr_ptr->form == DW_FORM_GNU_ref_alt)
3519 /* Now find the CU containing this pointer. */
3520 if (info_ptr >= unit->info_ptr_unit && info_ptr < unit->end_ptr)
3521 info_ptr_end = unit->end_ptr;
3522 else
3524 /* Check other CUs to see if they contain the abbrev. */
3525 struct comp_unit *u = NULL;
3526 struct addr_range range = { info_ptr, info_ptr };
3527 splay_tree_node v = splay_tree_lookup (unit->file->comp_unit_tree,
3528 (splay_tree_key)&range);
3529 if (v != NULL)
3530 u = (struct comp_unit *)v->value;
3532 if (attr_ptr->form == DW_FORM_ref_addr)
3533 while (u == NULL)
3535 u = stash_comp_unit (unit->stash, &unit->stash->f);
3536 if (u == NULL)
3537 break;
3538 if (info_ptr >= u->info_ptr_unit && info_ptr < u->end_ptr)
3539 break;
3540 u = NULL;
3543 if (attr_ptr->form == DW_FORM_GNU_ref_alt)
3544 while (u == NULL)
3546 u = stash_comp_unit (unit->stash, &unit->stash->alt);
3547 if (u == NULL)
3548 break;
3549 if (info_ptr >= u->info_ptr_unit && info_ptr < u->end_ptr)
3550 break;
3551 u = NULL;
3554 if (u == NULL)
3556 _bfd_error_handler
3557 (_("DWARF error: unable to locate abstract instance DIE ref %"
3558 PRIu64), (uint64_t) die_ref);
3559 bfd_set_error (bfd_error_bad_value);
3560 return false;
3562 unit = u;
3563 info_ptr_end = unit->end_ptr;
3566 else
3568 /* DW_FORM_ref1, DW_FORM_ref2, DW_FORM_ref4, DW_FORM_ref8 or
3569 DW_FORM_ref_udata. These are all references relative to the
3570 start of the current CU. */
3571 size_t total;
3573 info_ptr = unit->info_ptr_unit;
3574 info_ptr_end = unit->end_ptr;
3575 total = info_ptr_end - info_ptr;
3576 if (!die_ref || die_ref >= total)
3578 _bfd_error_handler
3579 (_("DWARF error: invalid abstract instance DIE ref"));
3580 bfd_set_error (bfd_error_bad_value);
3581 return false;
3583 info_ptr += die_ref;
3586 abbrev_number = _bfd_safe_read_leb128 (abfd, &info_ptr,
3587 false, info_ptr_end);
3588 if (abbrev_number)
3590 abbrev = lookup_abbrev (abbrev_number, unit->abbrevs);
3591 if (! abbrev)
3593 _bfd_error_handler
3594 (_("DWARF error: could not find abbrev number %u"), abbrev_number);
3595 bfd_set_error (bfd_error_bad_value);
3596 return false;
3598 else
3600 for (i = 0; i < abbrev->num_attrs; ++i)
3602 info_ptr = read_attribute (&attr, &abbrev->attrs[i], unit,
3603 info_ptr, info_ptr_end);
3604 if (info_ptr == NULL)
3605 break;
3606 switch (attr.name)
3608 case DW_AT_name:
3609 /* Prefer DW_AT_MIPS_linkage_name or DW_AT_linkage_name
3610 over DW_AT_name. */
3611 if (*pname == NULL && is_str_form (&attr))
3613 *pname = attr.u.str;
3614 if (mangle_style (unit->lang) == 0)
3615 *is_linkage = true;
3617 break;
3618 case DW_AT_specification:
3619 if (is_int_form (&attr)
3620 && !find_abstract_instance (unit, &attr, recur_count + 1,
3621 pname, is_linkage,
3622 filename_ptr, linenumber_ptr))
3623 return false;
3624 break;
3625 case DW_AT_linkage_name:
3626 case DW_AT_MIPS_linkage_name:
3627 /* PR 16949: Corrupt debug info can place
3628 non-string forms into these attributes. */
3629 if (is_str_form (&attr))
3631 *pname = attr.u.str;
3632 *is_linkage = true;
3634 break;
3635 case DW_AT_decl_file:
3636 if (!comp_unit_maybe_decode_line_info (unit))
3637 return false;
3638 if (is_int_form (&attr))
3640 free (*filename_ptr);
3641 *filename_ptr = concat_filename (unit->line_table,
3642 attr.u.val);
3644 break;
3645 case DW_AT_decl_line:
3646 if (is_int_form (&attr))
3647 *linenumber_ptr = attr.u.val;
3648 break;
3649 default:
3650 break;
3655 return true;
3658 static bool
3659 read_ranges (struct comp_unit *unit, struct arange *arange,
3660 struct trie_node **trie_root, uint64_t offset)
3662 bfd_byte *ranges_ptr;
3663 bfd_byte *ranges_end;
3664 bfd_vma base_address = unit->base_address;
3666 if (! unit->file->dwarf_ranges_buffer)
3668 if (! read_debug_ranges (unit))
3669 return false;
3672 if (offset > unit->file->dwarf_ranges_size)
3673 return false;
3674 ranges_ptr = unit->file->dwarf_ranges_buffer + offset;
3675 ranges_end = unit->file->dwarf_ranges_buffer + unit->file->dwarf_ranges_size;
3677 for (;;)
3679 bfd_vma low_pc;
3680 bfd_vma high_pc;
3682 /* PR 17512: file: 62cada7d. */
3683 if (2u * unit->addr_size > (size_t) (ranges_end - ranges_ptr))
3684 return false;
3686 low_pc = read_address (unit, &ranges_ptr, ranges_end);
3687 high_pc = read_address (unit, &ranges_ptr, ranges_end);
3689 if (low_pc == 0 && high_pc == 0)
3690 break;
3691 if (low_pc == -1UL && high_pc != -1UL)
3692 base_address = high_pc;
3693 else
3695 if (!arange_add (unit, arange, trie_root,
3696 base_address + low_pc, base_address + high_pc))
3697 return false;
3700 return true;
3703 static bool
3704 read_rnglists (struct comp_unit *unit, struct arange *arange,
3705 struct trie_node **trie_root, uint64_t offset)
3707 bfd_byte *rngs_ptr;
3708 bfd_byte *rngs_end;
3709 bfd_vma base_address = unit->base_address;
3710 bfd_vma low_pc;
3711 bfd_vma high_pc;
3712 bfd *abfd = unit->abfd;
3714 if (! unit->file->dwarf_rnglists_buffer)
3716 if (! read_debug_rnglists (unit))
3717 return false;
3720 rngs_ptr = unit->file->dwarf_rnglists_buffer + offset;
3721 if (rngs_ptr < unit->file->dwarf_rnglists_buffer)
3722 return false;
3723 rngs_end = unit->file->dwarf_rnglists_buffer;
3724 rngs_end += unit->file->dwarf_rnglists_size;
3726 for (;;)
3728 enum dwarf_range_list_entry rlet;
3730 if (rngs_ptr >= rngs_end)
3731 return false;
3733 rlet = read_1_byte (abfd, &rngs_ptr, rngs_end);
3735 switch (rlet)
3737 case DW_RLE_end_of_list:
3738 return true;
3740 case DW_RLE_base_address:
3741 if (unit->addr_size > (size_t) (rngs_end - rngs_ptr))
3742 return false;
3743 base_address = read_address (unit, &rngs_ptr, rngs_end);
3744 continue;
3746 case DW_RLE_start_length:
3747 if (unit->addr_size > (size_t) (rngs_end - rngs_ptr))
3748 return false;
3749 low_pc = read_address (unit, &rngs_ptr, rngs_end);
3750 high_pc = low_pc;
3751 high_pc += _bfd_safe_read_leb128 (abfd, &rngs_ptr,
3752 false, rngs_end);
3753 break;
3755 case DW_RLE_offset_pair:
3756 low_pc = base_address;
3757 low_pc += _bfd_safe_read_leb128 (abfd, &rngs_ptr,
3758 false, rngs_end);
3759 high_pc = base_address;
3760 high_pc += _bfd_safe_read_leb128 (abfd, &rngs_ptr,
3761 false, rngs_end);
3762 break;
3764 case DW_RLE_start_end:
3765 if (2u * unit->addr_size > (size_t) (rngs_end - rngs_ptr))
3766 return false;
3767 low_pc = read_address (unit, &rngs_ptr, rngs_end);
3768 high_pc = read_address (unit, &rngs_ptr, rngs_end);
3769 break;
3771 /* TODO x-variants need .debug_addr support used for split-dwarf. */
3772 case DW_RLE_base_addressx:
3773 case DW_RLE_startx_endx:
3774 case DW_RLE_startx_length:
3775 default:
3776 return false;
3779 if (!arange_add (unit, arange, trie_root, low_pc, high_pc))
3780 return false;
3784 static bool
3785 read_rangelist (struct comp_unit *unit, struct arange *arange,
3786 struct trie_node **trie_root, uint64_t offset)
3788 if (unit->version <= 4)
3789 return read_ranges (unit, arange, trie_root, offset);
3790 else
3791 return read_rnglists (unit, arange, trie_root, offset);
3794 static struct funcinfo *
3795 lookup_func_by_offset (uint64_t offset, struct funcinfo * table)
3797 for (; table != NULL; table = table->prev_func)
3798 if (table->unit_offset == offset)
3799 return table;
3800 return NULL;
3803 static struct varinfo *
3804 lookup_var_by_offset (uint64_t offset, struct varinfo * table)
3806 while (table)
3808 if (table->unit_offset == offset)
3809 return table;
3810 table = table->prev_var;
3813 return NULL;
3817 /* DWARF2 Compilation unit functions. */
3819 static struct funcinfo *
3820 reverse_funcinfo_list (struct funcinfo *head)
3822 struct funcinfo *rhead;
3823 struct funcinfo *temp;
3825 for (rhead = NULL; head; head = temp)
3827 temp = head->prev_func;
3828 head->prev_func = rhead;
3829 rhead = head;
3831 return rhead;
3834 static struct varinfo *
3835 reverse_varinfo_list (struct varinfo *head)
3837 struct varinfo *rhead;
3838 struct varinfo *temp;
3840 for (rhead = NULL; head; head = temp)
3842 temp = head->prev_var;
3843 head->prev_var = rhead;
3844 rhead = head;
3846 return rhead;
3849 /* Scan over each die in a comp. unit looking for functions to add
3850 to the function table and variables to the variable table. */
3852 static bool
3853 scan_unit_for_symbols (struct comp_unit *unit)
3855 bfd *abfd = unit->abfd;
3856 bfd_byte *info_ptr = unit->first_child_die_ptr;
3857 bfd_byte *info_ptr_end = unit->end_ptr;
3858 int nesting_level = 0;
3859 struct nest_funcinfo
3861 struct funcinfo *func;
3862 } *nested_funcs;
3863 int nested_funcs_size;
3864 struct funcinfo *last_func;
3865 struct varinfo *last_var;
3867 /* Maintain a stack of in-scope functions and inlined functions, which we
3868 can use to set the caller_func field. */
3869 nested_funcs_size = 32;
3870 nested_funcs = (struct nest_funcinfo *)
3871 bfd_malloc (nested_funcs_size * sizeof (*nested_funcs));
3872 if (nested_funcs == NULL)
3873 return false;
3874 nested_funcs[nesting_level].func = 0;
3876 /* PR 27484: We must scan the DIEs twice. The first time we look for
3877 function and variable tags and accumulate them into their respective
3878 tables. The second time through we process the attributes of the
3879 functions/variables and augment the table entries. */
3880 while (nesting_level >= 0)
3882 unsigned int abbrev_number, i;
3883 struct abbrev_info *abbrev;
3884 struct funcinfo *func;
3885 struct varinfo *var;
3886 uint64_t current_offset;
3888 /* PR 17512: file: 9f405d9d. */
3889 if (info_ptr >= info_ptr_end)
3890 goto fail;
3892 current_offset = info_ptr - unit->info_ptr_unit;
3893 abbrev_number = _bfd_safe_read_leb128 (abfd, &info_ptr,
3894 false, info_ptr_end);
3895 if (abbrev_number == 0)
3897 nesting_level--;
3898 continue;
3901 abbrev = lookup_abbrev (abbrev_number, unit->abbrevs);
3902 if (! abbrev)
3904 static unsigned int previous_failed_abbrev = -1U;
3906 /* Avoid multiple reports of the same missing abbrev. */
3907 if (abbrev_number != previous_failed_abbrev)
3909 _bfd_error_handler
3910 (_("DWARF error: could not find abbrev number %u"),
3911 abbrev_number);
3912 previous_failed_abbrev = abbrev_number;
3914 bfd_set_error (bfd_error_bad_value);
3915 goto fail;
3918 if (abbrev->tag == DW_TAG_subprogram
3919 || abbrev->tag == DW_TAG_entry_point
3920 || abbrev->tag == DW_TAG_inlined_subroutine)
3922 size_t amt = sizeof (struct funcinfo);
3924 var = NULL;
3925 func = (struct funcinfo *) bfd_zalloc (abfd, amt);
3926 if (func == NULL)
3927 goto fail;
3928 func->tag = abbrev->tag;
3929 func->prev_func = unit->function_table;
3930 func->unit_offset = current_offset;
3931 unit->function_table = func;
3932 unit->number_of_functions++;
3933 BFD_ASSERT (!unit->cached);
3935 if (func->tag == DW_TAG_inlined_subroutine)
3936 for (i = nesting_level; i-- != 0; )
3937 if (nested_funcs[i].func)
3939 func->caller_func = nested_funcs[i].func;
3940 break;
3942 nested_funcs[nesting_level].func = func;
3944 else
3946 func = NULL;
3947 if (abbrev->tag == DW_TAG_variable
3948 || abbrev->tag == DW_TAG_member)
3950 size_t amt = sizeof (struct varinfo);
3952 var = (struct varinfo *) bfd_zalloc (abfd, amt);
3953 if (var == NULL)
3954 goto fail;
3955 var->tag = abbrev->tag;
3956 var->stack = true;
3957 var->prev_var = unit->variable_table;
3958 unit->variable_table = var;
3959 var->unit_offset = current_offset;
3960 /* PR 18205: Missing debug information can cause this
3961 var to be attached to an already cached unit. */
3963 else
3964 var = NULL;
3966 /* No inline function in scope at this nesting level. */
3967 nested_funcs[nesting_level].func = 0;
3970 for (i = 0; i < abbrev->num_attrs; ++i)
3972 struct attribute attr;
3974 info_ptr = read_attribute (&attr, &abbrev->attrs[i],
3975 unit, info_ptr, info_ptr_end);
3976 if (info_ptr == NULL)
3977 goto fail;
3980 if (abbrev->has_children)
3982 nesting_level++;
3984 if (nesting_level >= nested_funcs_size)
3986 struct nest_funcinfo *tmp;
3988 nested_funcs_size *= 2;
3989 tmp = (struct nest_funcinfo *)
3990 bfd_realloc (nested_funcs,
3991 nested_funcs_size * sizeof (*nested_funcs));
3992 if (tmp == NULL)
3993 goto fail;
3994 nested_funcs = tmp;
3996 nested_funcs[nesting_level].func = 0;
4000 unit->function_table = reverse_funcinfo_list (unit->function_table);
4001 unit->variable_table = reverse_varinfo_list (unit->variable_table);
4003 /* This is the second pass over the abbrevs. */
4004 info_ptr = unit->first_child_die_ptr;
4005 nesting_level = 0;
4007 last_func = NULL;
4008 last_var = NULL;
4010 while (nesting_level >= 0)
4012 unsigned int abbrev_number, i;
4013 struct abbrev_info *abbrev;
4014 struct attribute attr;
4015 struct funcinfo *func;
4016 struct varinfo *var;
4017 bfd_vma low_pc = 0;
4018 bfd_vma high_pc = 0;
4019 bool high_pc_relative = false;
4020 uint64_t current_offset;
4022 /* PR 17512: file: 9f405d9d. */
4023 if (info_ptr >= info_ptr_end)
4024 goto fail;
4026 current_offset = info_ptr - unit->info_ptr_unit;
4027 abbrev_number = _bfd_safe_read_leb128 (abfd, &info_ptr,
4028 false, info_ptr_end);
4029 if (! abbrev_number)
4031 nesting_level--;
4032 continue;
4035 abbrev = lookup_abbrev (abbrev_number, unit->abbrevs);
4036 /* This should have been handled above. */
4037 BFD_ASSERT (abbrev != NULL);
4039 func = NULL;
4040 var = NULL;
4041 if (abbrev->tag == DW_TAG_subprogram
4042 || abbrev->tag == DW_TAG_entry_point
4043 || abbrev->tag == DW_TAG_inlined_subroutine)
4045 if (last_func
4046 && last_func->prev_func
4047 && last_func->prev_func->unit_offset == current_offset)
4048 func = last_func->prev_func;
4049 else
4050 func = lookup_func_by_offset (current_offset, unit->function_table);
4052 if (func == NULL)
4053 goto fail;
4055 last_func = func;
4057 else if (abbrev->tag == DW_TAG_variable
4058 || abbrev->tag == DW_TAG_member)
4060 if (last_var
4061 && last_var->prev_var
4062 && last_var->prev_var->unit_offset == current_offset)
4063 var = last_var->prev_var;
4064 else
4065 var = lookup_var_by_offset (current_offset, unit->variable_table);
4067 if (var == NULL)
4068 goto fail;
4070 last_var = var;
4073 for (i = 0; i < abbrev->num_attrs; ++i)
4075 info_ptr = read_attribute (&attr, &abbrev->attrs[i],
4076 unit, info_ptr, info_ptr_end);
4077 if (info_ptr == NULL)
4078 goto fail;
4080 if (func)
4082 switch (attr.name)
4084 case DW_AT_call_file:
4085 if (is_int_form (&attr))
4086 func->caller_file = concat_filename (unit->line_table,
4087 attr.u.val);
4088 break;
4090 case DW_AT_call_line:
4091 if (is_int_form (&attr))
4092 func->caller_line = attr.u.val;
4093 break;
4095 case DW_AT_abstract_origin:
4096 case DW_AT_specification:
4097 if (is_int_form (&attr)
4098 && !find_abstract_instance (unit, &attr, 0,
4099 &func->name,
4100 &func->is_linkage,
4101 &func->file,
4102 &func->line))
4103 goto fail;
4104 break;
4106 case DW_AT_name:
4107 /* Prefer DW_AT_MIPS_linkage_name or DW_AT_linkage_name
4108 over DW_AT_name. */
4109 if (func->name == NULL && is_str_form (&attr))
4111 func->name = attr.u.str;
4112 if (mangle_style (unit->lang) == 0)
4113 func->is_linkage = true;
4115 break;
4117 case DW_AT_linkage_name:
4118 case DW_AT_MIPS_linkage_name:
4119 /* PR 16949: Corrupt debug info can place
4120 non-string forms into these attributes. */
4121 if (is_str_form (&attr))
4123 func->name = attr.u.str;
4124 func->is_linkage = true;
4126 break;
4128 case DW_AT_low_pc:
4129 if (is_int_form (&attr))
4130 low_pc = attr.u.val;
4131 break;
4133 case DW_AT_high_pc:
4134 if (is_int_form (&attr))
4136 high_pc = attr.u.val;
4137 high_pc_relative = attr.form != DW_FORM_addr;
4139 break;
4141 case DW_AT_ranges:
4142 if (is_int_form (&attr)
4143 && !read_rangelist (unit, &func->arange,
4144 &unit->file->trie_root, attr.u.val))
4145 goto fail;
4146 break;
4148 case DW_AT_decl_file:
4149 if (is_int_form (&attr))
4151 free (func->file);
4152 func->file = concat_filename (unit->line_table,
4153 attr.u.val);
4155 break;
4157 case DW_AT_decl_line:
4158 if (is_int_form (&attr))
4159 func->line = attr.u.val;
4160 break;
4162 default:
4163 break;
4166 else if (var)
4168 switch (attr.name)
4170 case DW_AT_specification:
4171 if (is_int_form (&attr) && attr.u.val)
4173 bool is_linkage;
4174 if (!find_abstract_instance (unit, &attr, 0,
4175 &var->name,
4176 &is_linkage,
4177 &var->file,
4178 &var->line))
4180 _bfd_error_handler (_("DWARF error: could not find "
4181 "variable specification "
4182 "at offset 0x%lx"),
4183 (unsigned long) attr.u.val);
4184 break;
4187 break;
4189 case DW_AT_name:
4190 if (is_str_form (&attr))
4191 var->name = attr.u.str;
4192 break;
4194 case DW_AT_decl_file:
4195 if (is_int_form (&attr))
4197 free (var->file);
4198 var->file = concat_filename (unit->line_table,
4199 attr.u.val);
4201 break;
4203 case DW_AT_decl_line:
4204 if (is_int_form (&attr))
4205 var->line = attr.u.val;
4206 break;
4208 case DW_AT_external:
4209 if (is_int_form (&attr) && attr.u.val != 0)
4210 var->stack = false;
4211 break;
4213 case DW_AT_location:
4214 switch (attr.form)
4216 case DW_FORM_block:
4217 case DW_FORM_block1:
4218 case DW_FORM_block2:
4219 case DW_FORM_block4:
4220 case DW_FORM_exprloc:
4221 if (attr.u.blk->data != NULL
4222 && *attr.u.blk->data == DW_OP_addr)
4224 var->stack = false;
4226 /* Verify that DW_OP_addr is the only opcode in the
4227 location, in which case the block size will be 1
4228 plus the address size. */
4229 /* ??? For TLS variables, gcc can emit
4230 DW_OP_addr <addr> DW_OP_GNU_push_tls_address
4231 which we don't handle here yet. */
4232 if (attr.u.blk->size == unit->addr_size + 1U)
4233 var->addr = bfd_get (unit->addr_size * 8,
4234 unit->abfd,
4235 attr.u.blk->data + 1);
4237 break;
4239 default:
4240 break;
4242 break;
4244 default:
4245 break;
4250 if (abbrev->has_children)
4251 nesting_level++;
4253 if (high_pc_relative)
4254 high_pc += low_pc;
4256 if (func && high_pc != 0)
4258 if (!arange_add (unit, &func->arange, &unit->file->trie_root,
4259 low_pc, high_pc))
4260 goto fail;
4264 unit->function_table = reverse_funcinfo_list (unit->function_table);
4265 unit->variable_table = reverse_varinfo_list (unit->variable_table);
4267 free (nested_funcs);
4268 return true;
4270 fail:
4271 free (nested_funcs);
4272 return false;
4275 /* Read the attributes of the form strx and addrx. */
4277 static void
4278 reread_attribute (struct comp_unit *unit,
4279 struct attribute *attr,
4280 bfd_vma *low_pc,
4281 bfd_vma *high_pc,
4282 bool *high_pc_relative,
4283 bool compunit)
4285 if (is_strx_form (attr->form))
4286 attr->u.str = (char *) read_indexed_string (attr->u.val, unit);
4287 if (is_addrx_form (attr->form))
4288 attr->u.val = read_indexed_address (attr->u.val, unit);
4290 switch (attr->name)
4292 case DW_AT_stmt_list:
4293 unit->stmtlist = 1;
4294 unit->line_offset = attr->u.val;
4295 break;
4297 case DW_AT_name:
4298 if (is_str_form (attr))
4299 unit->name = attr->u.str;
4300 break;
4302 case DW_AT_low_pc:
4303 *low_pc = attr->u.val;
4304 if (compunit)
4305 unit->base_address = *low_pc;
4306 break;
4308 case DW_AT_high_pc:
4309 *high_pc = attr->u.val;
4310 *high_pc_relative = attr->form != DW_FORM_addr;
4311 break;
4313 case DW_AT_ranges:
4314 if (!read_rangelist (unit, &unit->arange,
4315 &unit->file->trie_root, attr->u.val))
4316 return;
4317 break;
4319 case DW_AT_comp_dir:
4321 char *comp_dir = attr->u.str;
4323 if (!is_str_form (attr))
4325 _bfd_error_handler
4326 (_("DWARF error: DW_AT_comp_dir attribute encountered "
4327 "with a non-string form"));
4328 comp_dir = NULL;
4331 if (comp_dir)
4333 char *cp = strchr (comp_dir, ':');
4335 if (cp && cp != comp_dir && cp[-1] == '.' && cp[1] == '/')
4336 comp_dir = cp + 1;
4338 unit->comp_dir = comp_dir;
4339 break;
4342 case DW_AT_language:
4343 unit->lang = attr->u.val;
4344 default:
4345 break;
4349 /* Parse a DWARF2 compilation unit starting at INFO_PTR. UNIT_LENGTH
4350 includes the compilation unit header that proceeds the DIE's, but
4351 does not include the length field that precedes each compilation
4352 unit header. END_PTR points one past the end of this comp unit.
4353 OFFSET_SIZE is the size of DWARF2 offsets (either 4 or 8 bytes).
4355 This routine does not read the whole compilation unit; only enough
4356 to get to the line number information for the compilation unit. */
4358 static struct comp_unit *
4359 parse_comp_unit (struct dwarf2_debug *stash,
4360 struct dwarf2_debug_file *file,
4361 bfd_byte *info_ptr,
4362 bfd_vma unit_length,
4363 bfd_byte *info_ptr_unit,
4364 unsigned int offset_size)
4366 struct comp_unit* unit;
4367 unsigned int version;
4368 uint64_t abbrev_offset = 0;
4369 /* Initialize it just to avoid a GCC false warning. */
4370 unsigned int addr_size = -1;
4371 struct abbrev_info** abbrevs;
4372 unsigned int abbrev_number, i;
4373 struct abbrev_info *abbrev;
4374 struct attribute attr;
4375 bfd_byte *end_ptr = info_ptr + unit_length;
4376 size_t amt;
4377 bfd_vma low_pc = 0;
4378 bfd_vma high_pc = 0;
4379 bfd *abfd = file->bfd_ptr;
4380 bool high_pc_relative = false;
4381 enum dwarf_unit_type unit_type;
4382 struct attribute *str_addrp = NULL;
4383 size_t str_count = 0;
4384 size_t str_alloc = 0;
4385 bool compunit_flag = false;
4387 version = read_2_bytes (abfd, &info_ptr, end_ptr);
4388 if (version < 2 || version > 5)
4390 /* PR 19872: A version number of 0 probably means that there is padding
4391 at the end of the .debug_info section. Gold puts it there when
4392 performing an incremental link, for example. So do not generate
4393 an error, just return a NULL. */
4394 if (version)
4396 _bfd_error_handler
4397 (_("DWARF error: found dwarf version '%u', this reader"
4398 " only handles version 2, 3, 4 and 5 information"), version);
4399 bfd_set_error (bfd_error_bad_value);
4401 return NULL;
4404 if (version < 5)
4405 unit_type = DW_UT_compile;
4406 else
4408 unit_type = read_1_byte (abfd, &info_ptr, end_ptr);
4409 addr_size = read_1_byte (abfd, &info_ptr, end_ptr);
4412 BFD_ASSERT (offset_size == 4 || offset_size == 8);
4413 if (offset_size == 4)
4414 abbrev_offset = read_4_bytes (abfd, &info_ptr, end_ptr);
4415 else
4416 abbrev_offset = read_8_bytes (abfd, &info_ptr, end_ptr);
4418 if (version < 5)
4419 addr_size = read_1_byte (abfd, &info_ptr, end_ptr);
4421 switch (unit_type)
4423 case DW_UT_type:
4424 /* Skip type signature. */
4425 info_ptr += 8;
4427 /* Skip type offset. */
4428 info_ptr += offset_size;
4429 break;
4431 case DW_UT_skeleton:
4432 /* Skip DWO_id field. */
4433 info_ptr += 8;
4434 break;
4436 default:
4437 break;
4440 if (addr_size > sizeof (bfd_vma))
4442 _bfd_error_handler
4443 /* xgettext: c-format */
4444 (_("DWARF error: found address size '%u', this reader"
4445 " can not handle sizes greater than '%u'"),
4446 addr_size,
4447 (unsigned int) sizeof (bfd_vma));
4448 bfd_set_error (bfd_error_bad_value);
4449 return NULL;
4452 if (addr_size != 2 && addr_size != 4 && addr_size != 8)
4454 _bfd_error_handler
4455 ("DWARF error: found address size '%u', this reader"
4456 " can only handle address sizes '2', '4' and '8'", addr_size);
4457 bfd_set_error (bfd_error_bad_value);
4458 return NULL;
4461 /* Read the abbrevs for this compilation unit into a table. */
4462 abbrevs = read_abbrevs (abfd, abbrev_offset, stash, file);
4463 if (! abbrevs)
4464 return NULL;
4466 abbrev_number = _bfd_safe_read_leb128 (abfd, &info_ptr,
4467 false, end_ptr);
4468 if (! abbrev_number)
4470 /* PR 19872: An abbrev number of 0 probably means that there is padding
4471 at the end of the .debug_abbrev section. Gold puts it there when
4472 performing an incremental link, for example. So do not generate
4473 an error, just return a NULL. */
4474 return NULL;
4477 abbrev = lookup_abbrev (abbrev_number, abbrevs);
4478 if (! abbrev)
4480 _bfd_error_handler (_("DWARF error: could not find abbrev number %u"),
4481 abbrev_number);
4482 bfd_set_error (bfd_error_bad_value);
4483 return NULL;
4486 amt = sizeof (struct comp_unit);
4487 unit = (struct comp_unit *) bfd_zalloc (abfd, amt);
4488 if (unit == NULL)
4489 return NULL;
4490 unit->abfd = abfd;
4491 unit->version = version;
4492 unit->addr_size = addr_size;
4493 unit->offset_size = offset_size;
4494 unit->abbrevs = abbrevs;
4495 unit->end_ptr = end_ptr;
4496 unit->stash = stash;
4497 unit->file = file;
4498 unit->info_ptr_unit = info_ptr_unit;
4500 if (abbrev->tag == DW_TAG_compile_unit)
4501 compunit_flag = true;
4503 for (i = 0; i < abbrev->num_attrs; ++i)
4505 info_ptr = read_attribute (&attr, &abbrev->attrs[i], unit, info_ptr, end_ptr);
4506 if (info_ptr == NULL)
4507 goto err_exit;
4509 /* Identify attributes of the form strx* and addrx* which come before
4510 DW_AT_str_offsets_base and DW_AT_addr_base respectively in the CU.
4511 Store the attributes in an array and process them later. */
4512 if ((unit->dwarf_str_offset == 0 && is_strx_form (attr.form))
4513 || (unit->dwarf_addr_offset == 0 && is_addrx_form (attr.form)))
4515 if (str_count <= str_alloc)
4517 str_alloc = 2 * str_alloc + 200;
4518 str_addrp = bfd_realloc (str_addrp,
4519 str_alloc * sizeof (*str_addrp));
4520 if (str_addrp == NULL)
4521 goto err_exit;
4523 str_addrp[str_count] = attr;
4524 str_count++;
4525 continue;
4528 /* Store the data if it is of an attribute we want to keep in a
4529 partial symbol table. */
4530 switch (attr.name)
4532 case DW_AT_stmt_list:
4533 if (is_int_form (&attr))
4535 unit->stmtlist = 1;
4536 unit->line_offset = attr.u.val;
4538 break;
4540 case DW_AT_name:
4541 if (is_str_form (&attr))
4542 unit->name = attr.u.str;
4543 break;
4545 case DW_AT_low_pc:
4546 if (is_int_form (&attr))
4548 low_pc = attr.u.val;
4549 /* If the compilation unit DIE has a DW_AT_low_pc attribute,
4550 this is the base address to use when reading location
4551 lists or range lists. */
4552 if (compunit_flag)
4553 unit->base_address = low_pc;
4555 break;
4557 case DW_AT_high_pc:
4558 if (is_int_form (&attr))
4560 high_pc = attr.u.val;
4561 high_pc_relative = attr.form != DW_FORM_addr;
4563 break;
4565 case DW_AT_ranges:
4566 if (is_int_form (&attr)
4567 && !read_rangelist (unit, &unit->arange,
4568 &unit->file->trie_root, attr.u.val))
4569 goto err_exit;
4570 break;
4572 case DW_AT_comp_dir:
4574 char *comp_dir = attr.u.str;
4576 /* PR 17512: file: 1fe726be. */
4577 if (!is_str_form (&attr))
4579 _bfd_error_handler
4580 (_("DWARF error: DW_AT_comp_dir attribute encountered with a non-string form"));
4581 comp_dir = NULL;
4584 if (comp_dir)
4586 /* Irix 6.2 native cc prepends <machine>.: to the compilation
4587 directory, get rid of it. */
4588 char *cp = strchr (comp_dir, ':');
4590 if (cp && cp != comp_dir && cp[-1] == '.' && cp[1] == '/')
4591 comp_dir = cp + 1;
4593 unit->comp_dir = comp_dir;
4594 break;
4597 case DW_AT_language:
4598 if (is_int_form (&attr))
4599 unit->lang = attr.u.val;
4600 break;
4602 case DW_AT_addr_base:
4603 unit->dwarf_addr_offset = attr.u.val;
4604 break;
4606 case DW_AT_str_offsets_base:
4607 unit->dwarf_str_offset = attr.u.val;
4608 break;
4610 default:
4611 break;
4615 for (i = 0; i < str_count; ++i)
4616 reread_attribute (unit, &str_addrp[i], &low_pc, &high_pc,
4617 &high_pc_relative, compunit_flag);
4619 if (high_pc_relative)
4620 high_pc += low_pc;
4621 if (high_pc != 0)
4623 if (!arange_add (unit, &unit->arange, &unit->file->trie_root,
4624 low_pc, high_pc))
4625 goto err_exit;
4628 unit->first_child_die_ptr = info_ptr;
4630 free (str_addrp);
4631 return unit;
4633 err_exit:
4634 unit->error = 1;
4635 free (str_addrp);
4636 return NULL;
4639 /* Return TRUE if UNIT may contain the address given by ADDR. When
4640 there are functions written entirely with inline asm statements, the
4641 range info in the compilation unit header may not be correct. We
4642 need to consult the line info table to see if a compilation unit
4643 really contains the given address. */
4645 static bool
4646 comp_unit_may_contain_address (struct comp_unit *unit, bfd_vma addr)
4648 struct arange *arange;
4650 if (unit->error)
4651 return false;
4653 if (unit->arange.high == 0 /* No ranges have been computed yet. */
4654 || unit->line_table == NULL) /* The line info table has not been loaded. */
4655 return true;
4657 for (arange = &unit->arange; arange != NULL; arange = arange->next)
4658 if (addr >= arange->low && addr < arange->high)
4659 return true;
4661 return false;
4664 /* If UNIT contains ADDR, set the output parameters to the values for
4665 the line containing ADDR and return TRUE. Otherwise return FALSE.
4666 The output parameters, FILENAME_PTR, FUNCTION_PTR, and
4667 LINENUMBER_PTR, are pointers to the objects to be filled in. */
4669 static bool
4670 comp_unit_find_nearest_line (struct comp_unit *unit,
4671 bfd_vma addr,
4672 const char **filename_ptr,
4673 struct funcinfo **function_ptr,
4674 unsigned int *linenumber_ptr,
4675 unsigned int *discriminator_ptr)
4677 bool line_p, func_p;
4679 if (!comp_unit_maybe_decode_line_info (unit))
4680 return false;
4682 *function_ptr = NULL;
4683 func_p = lookup_address_in_function_table (unit, addr, function_ptr);
4685 if (func_p && (*function_ptr)->tag == DW_TAG_inlined_subroutine)
4686 unit->stash->inliner_chain = *function_ptr;
4688 line_p = lookup_address_in_line_info_table (unit->line_table, addr,
4689 filename_ptr,
4690 linenumber_ptr,
4691 discriminator_ptr);
4692 return line_p || func_p;
4695 /* Check to see if line info is already decoded in a comp_unit.
4696 If not, decode it. Returns TRUE if no errors were encountered;
4697 FALSE otherwise. */
4699 static bool
4700 comp_unit_maybe_decode_line_info (struct comp_unit *unit)
4702 if (unit->error)
4703 return false;
4705 if (! unit->line_table)
4707 if (! unit->stmtlist)
4709 unit->error = 1;
4710 return false;
4713 unit->line_table = decode_line_info (unit);
4715 if (! unit->line_table)
4717 unit->error = 1;
4718 return false;
4721 if (unit->first_child_die_ptr < unit->end_ptr
4722 && ! scan_unit_for_symbols (unit))
4724 unit->error = 1;
4725 return false;
4729 return true;
4732 /* If UNIT contains SYM at ADDR, set the output parameters to the
4733 values for the line containing SYM. The output parameters,
4734 FILENAME_PTR, and LINENUMBER_PTR, are pointers to the objects to be
4735 filled in.
4737 Return TRUE if UNIT contains SYM, and no errors were encountered;
4738 FALSE otherwise. */
4740 static bool
4741 comp_unit_find_line (struct comp_unit *unit,
4742 asymbol *sym,
4743 bfd_vma addr,
4744 const char **filename_ptr,
4745 unsigned int *linenumber_ptr)
4747 if (!comp_unit_maybe_decode_line_info (unit))
4748 return false;
4750 if (sym->flags & BSF_FUNCTION)
4751 return lookup_symbol_in_function_table (unit, sym, addr,
4752 filename_ptr,
4753 linenumber_ptr);
4755 return lookup_symbol_in_variable_table (unit, sym, addr,
4756 filename_ptr,
4757 linenumber_ptr);
4760 /* Extract all interesting funcinfos and varinfos of a compilation
4761 unit into hash tables for faster lookup. Returns TRUE if no
4762 errors were enountered; FALSE otherwise. */
4764 static bool
4765 comp_unit_hash_info (struct dwarf2_debug *stash,
4766 struct comp_unit *unit,
4767 struct info_hash_table *funcinfo_hash_table,
4768 struct info_hash_table *varinfo_hash_table)
4770 struct funcinfo* each_func;
4771 struct varinfo* each_var;
4772 bool okay = true;
4774 BFD_ASSERT (stash->info_hash_status != STASH_INFO_HASH_DISABLED);
4776 if (!comp_unit_maybe_decode_line_info (unit))
4777 return false;
4779 BFD_ASSERT (!unit->cached);
4781 /* To preserve the original search order, we went to visit the function
4782 infos in the reversed order of the list. However, making the list
4783 bi-directional use quite a bit of extra memory. So we reverse
4784 the list first, traverse the list in the now reversed order and
4785 finally reverse the list again to get back the original order. */
4786 unit->function_table = reverse_funcinfo_list (unit->function_table);
4787 for (each_func = unit->function_table;
4788 each_func && okay;
4789 each_func = each_func->prev_func)
4791 /* Skip nameless functions. */
4792 if (each_func->name)
4793 /* There is no need to copy name string into hash table as
4794 name string is either in the dwarf string buffer or
4795 info in the stash. */
4796 okay = insert_info_hash_table (funcinfo_hash_table, each_func->name,
4797 (void*) each_func, false);
4799 unit->function_table = reverse_funcinfo_list (unit->function_table);
4800 if (!okay)
4801 return false;
4803 /* We do the same for variable infos. */
4804 unit->variable_table = reverse_varinfo_list (unit->variable_table);
4805 for (each_var = unit->variable_table;
4806 each_var && okay;
4807 each_var = each_var->prev_var)
4809 /* Skip stack vars and vars with no files or names. */
4810 if (! each_var->stack
4811 && each_var->file != NULL
4812 && each_var->name != NULL)
4813 /* There is no need to copy name string into hash table as
4814 name string is either in the dwarf string buffer or
4815 info in the stash. */
4816 okay = insert_info_hash_table (varinfo_hash_table, each_var->name,
4817 (void*) each_var, false);
4820 unit->variable_table = reverse_varinfo_list (unit->variable_table);
4821 unit->cached = true;
4822 return okay;
4825 /* Locate a section in a BFD containing debugging info. The search starts
4826 from the section after AFTER_SEC, or from the first section in the BFD if
4827 AFTER_SEC is NULL. The search works by examining the names of the
4828 sections. There are three permissiable names. The first two are given
4829 by DEBUG_SECTIONS[debug_info] (whose standard DWARF2 names are .debug_info
4830 and .zdebug_info). The third is a prefix .gnu.linkonce.wi.
4831 This is a variation on the .debug_info section which has a checksum
4832 describing the contents appended onto the name. This allows the linker to
4833 identify and discard duplicate debugging sections for different
4834 compilation units. */
4835 #define GNU_LINKONCE_INFO ".gnu.linkonce.wi."
4837 static asection *
4838 find_debug_info (bfd *abfd, const struct dwarf_debug_section *debug_sections,
4839 asection *after_sec)
4841 asection *msec;
4842 const char *look;
4844 if (after_sec == NULL)
4846 look = debug_sections[debug_info].uncompressed_name;
4847 msec = bfd_get_section_by_name (abfd, look);
4848 /* Testing SEC_HAS_CONTENTS is an anti-fuzzer measure. Of
4849 course debug sections always have contents. */
4850 if (msec != NULL && (msec->flags & SEC_HAS_CONTENTS) != 0)
4851 return msec;
4853 look = debug_sections[debug_info].compressed_name;
4854 msec = bfd_get_section_by_name (abfd, look);
4855 if (msec != NULL && (msec->flags & SEC_HAS_CONTENTS) != 0)
4856 return msec;
4858 for (msec = abfd->sections; msec != NULL; msec = msec->next)
4859 if ((msec->flags & SEC_HAS_CONTENTS) != 0
4860 && startswith (msec->name, GNU_LINKONCE_INFO))
4861 return msec;
4863 return NULL;
4866 for (msec = after_sec->next; msec != NULL; msec = msec->next)
4868 if ((msec->flags & SEC_HAS_CONTENTS) == 0)
4869 continue;
4871 look = debug_sections[debug_info].uncompressed_name;
4872 if (strcmp (msec->name, look) == 0)
4873 return msec;
4875 look = debug_sections[debug_info].compressed_name;
4876 if (look != NULL && strcmp (msec->name, look) == 0)
4877 return msec;
4879 if (startswith (msec->name, GNU_LINKONCE_INFO))
4880 return msec;
4883 return NULL;
4886 /* Transfer VMAs from object file to separate debug file. */
4888 static void
4889 set_debug_vma (bfd *orig_bfd, bfd *debug_bfd)
4891 asection *s, *d;
4893 for (s = orig_bfd->sections, d = debug_bfd->sections;
4894 s != NULL && d != NULL;
4895 s = s->next, d = d->next)
4897 if ((d->flags & SEC_DEBUGGING) != 0)
4898 break;
4899 /* ??? Assumes 1-1 correspondence between sections in the
4900 two files. */
4901 if (strcmp (s->name, d->name) == 0)
4903 d->output_section = s->output_section;
4904 d->output_offset = s->output_offset;
4905 d->vma = s->vma;
4910 /* If the dwarf2 info was found in a separate debug file, return the
4911 debug file section corresponding to the section in the original file
4912 and the debug file symbols. */
4914 static void
4915 _bfd_dwarf2_stash_syms (struct dwarf2_debug *stash, bfd *abfd,
4916 asection **sec, asymbol ***syms)
4918 if (stash->f.bfd_ptr != abfd)
4920 asection *s, *d;
4922 if (*sec == NULL)
4924 *syms = stash->f.syms;
4925 return;
4928 for (s = abfd->sections, d = stash->f.bfd_ptr->sections;
4929 s != NULL && d != NULL;
4930 s = s->next, d = d->next)
4932 if ((d->flags & SEC_DEBUGGING) != 0)
4933 break;
4934 if (s == *sec
4935 && strcmp (s->name, d->name) == 0)
4937 *sec = d;
4938 *syms = stash->f.syms;
4939 break;
4945 /* Unset vmas for adjusted sections in STASH. */
4947 static void
4948 unset_sections (struct dwarf2_debug *stash)
4950 int i;
4951 struct adjusted_section *p;
4953 i = stash->adjusted_section_count;
4954 p = stash->adjusted_sections;
4955 for (; i > 0; i--, p++)
4956 p->section->vma = p->orig_vma;
4959 /* Set VMAs for allocated and .debug_info sections in ORIG_BFD, a
4960 relocatable object file. VMAs are normally all zero in relocatable
4961 object files, so if we want to distinguish locations in sections by
4962 address we need to set VMAs so the sections do not overlap. We
4963 also set VMA on .debug_info so that when we have multiple
4964 .debug_info sections (or the linkonce variant) they also do not
4965 overlap. The multiple .debug_info sections make up a single
4966 logical section. ??? We should probably do the same for other
4967 debug sections. */
4969 static bool
4970 place_sections (bfd *orig_bfd, struct dwarf2_debug *stash)
4972 bfd *abfd;
4973 struct adjusted_section *p;
4974 int i;
4975 const char *debug_info_name;
4977 if (stash->adjusted_section_count != 0)
4979 i = stash->adjusted_section_count;
4980 p = stash->adjusted_sections;
4981 for (; i > 0; i--, p++)
4982 p->section->vma = p->adj_vma;
4983 return true;
4986 debug_info_name = stash->debug_sections[debug_info].uncompressed_name;
4987 i = 0;
4988 abfd = orig_bfd;
4989 while (1)
4991 asection *sect;
4993 for (sect = abfd->sections; sect != NULL; sect = sect->next)
4995 int is_debug_info;
4997 if (sect->output_section != NULL
4998 && sect->output_section != sect
4999 && (sect->flags & SEC_DEBUGGING) == 0)
5000 continue;
5002 is_debug_info = (strcmp (sect->name, debug_info_name) == 0
5003 || startswith (sect->name, GNU_LINKONCE_INFO));
5005 if (!((sect->flags & SEC_ALLOC) != 0 && abfd == orig_bfd)
5006 && !is_debug_info)
5007 continue;
5009 i++;
5011 if (abfd == stash->f.bfd_ptr)
5012 break;
5013 abfd = stash->f.bfd_ptr;
5016 if (i <= 1)
5017 stash->adjusted_section_count = -1;
5018 else
5020 bfd_vma last_vma = 0, last_dwarf = 0;
5021 size_t amt = i * sizeof (struct adjusted_section);
5023 p = (struct adjusted_section *) bfd_malloc (amt);
5024 if (p == NULL)
5025 return false;
5027 stash->adjusted_sections = p;
5028 stash->adjusted_section_count = i;
5030 abfd = orig_bfd;
5031 while (1)
5033 asection *sect;
5035 for (sect = abfd->sections; sect != NULL; sect = sect->next)
5037 bfd_size_type sz;
5038 int is_debug_info;
5040 if (sect->output_section != NULL
5041 && sect->output_section != sect
5042 && (sect->flags & SEC_DEBUGGING) == 0)
5043 continue;
5045 is_debug_info = (strcmp (sect->name, debug_info_name) == 0
5046 || startswith (sect->name, GNU_LINKONCE_INFO));
5048 if (!((sect->flags & SEC_ALLOC) != 0 && abfd == orig_bfd)
5049 && !is_debug_info)
5050 continue;
5052 sz = sect->rawsize ? sect->rawsize : sect->size;
5054 p->section = sect;
5055 p->orig_vma = sect->vma;
5057 bfd_vma *v = is_debug_info ? &last_dwarf : &last_vma;
5058 /* Align the new address to the current section
5059 alignment. */
5060 bfd_vma mask = -(bfd_vma) 1 << sect->alignment_power;
5061 *v = (*v + ~mask) & mask;
5062 sect->vma = *v;
5063 *v += sz;
5065 p->adj_vma = sect->vma;
5066 p++;
5068 if (abfd == stash->f.bfd_ptr)
5069 break;
5070 abfd = stash->f.bfd_ptr;
5074 if (orig_bfd != stash->f.bfd_ptr)
5075 set_debug_vma (orig_bfd, stash->f.bfd_ptr);
5077 return true;
5080 /* Look up a funcinfo by name using the given info hash table. If found,
5081 also update the locations pointed to by filename_ptr and linenumber_ptr.
5083 This function returns TRUE if a funcinfo that matches the given symbol
5084 and address is found with any error; otherwise it returns FALSE. */
5086 static bool
5087 info_hash_lookup_funcinfo (struct info_hash_table *hash_table,
5088 asymbol *sym,
5089 bfd_vma addr,
5090 const char **filename_ptr,
5091 unsigned int *linenumber_ptr)
5093 struct funcinfo* each_func;
5094 struct funcinfo* best_fit = NULL;
5095 bfd_vma best_fit_len = (bfd_vma) -1;
5096 struct info_list_node *node;
5097 struct arange *arange;
5098 const char *name = bfd_asymbol_name (sym);
5100 for (node = lookup_info_hash_table (hash_table, name);
5101 node;
5102 node = node->next)
5104 each_func = (struct funcinfo *) node->info;
5105 for (arange = &each_func->arange;
5106 arange;
5107 arange = arange->next)
5109 if (addr >= arange->low
5110 && addr < arange->high
5111 && arange->high - arange->low < best_fit_len)
5113 best_fit = each_func;
5114 best_fit_len = arange->high - arange->low;
5119 if (best_fit)
5121 *filename_ptr = best_fit->file;
5122 *linenumber_ptr = best_fit->line;
5123 return true;
5126 return false;
5129 /* Look up a varinfo by name using the given info hash table. If found,
5130 also update the locations pointed to by filename_ptr and linenumber_ptr.
5132 This function returns TRUE if a varinfo that matches the given symbol
5133 and address is found with any error; otherwise it returns FALSE. */
5135 static bool
5136 info_hash_lookup_varinfo (struct info_hash_table *hash_table,
5137 asymbol *sym,
5138 bfd_vma addr,
5139 const char **filename_ptr,
5140 unsigned int *linenumber_ptr)
5142 struct varinfo* each;
5143 struct info_list_node *node;
5144 const char *name = bfd_asymbol_name (sym);
5146 for (node = lookup_info_hash_table (hash_table, name);
5147 node;
5148 node = node->next)
5150 each = (struct varinfo *) node->info;
5151 if (each->addr == addr)
5153 *filename_ptr = each->file;
5154 *linenumber_ptr = each->line;
5155 return true;
5159 return false;
5162 /* Update the funcinfo and varinfo info hash tables if they are
5163 not up to date. Returns TRUE if there is no error; otherwise
5164 returns FALSE and disable the info hash tables. */
5166 static bool
5167 stash_maybe_update_info_hash_tables (struct dwarf2_debug *stash)
5169 struct comp_unit *each;
5171 /* Exit if hash tables are up-to-date. */
5172 if (stash->f.all_comp_units == stash->hash_units_head)
5173 return true;
5175 if (stash->hash_units_head)
5176 each = stash->hash_units_head->prev_unit;
5177 else
5178 each = stash->f.last_comp_unit;
5180 while (each)
5182 if (!comp_unit_hash_info (stash, each, stash->funcinfo_hash_table,
5183 stash->varinfo_hash_table))
5185 stash->info_hash_status = STASH_INFO_HASH_DISABLED;
5186 return false;
5188 each = each->prev_unit;
5191 stash->hash_units_head = stash->f.all_comp_units;
5192 return true;
5195 /* Check consistency of info hash tables. This is for debugging only. */
5197 static void ATTRIBUTE_UNUSED
5198 stash_verify_info_hash_table (struct dwarf2_debug *stash)
5200 struct comp_unit *each_unit;
5201 struct funcinfo *each_func;
5202 struct varinfo *each_var;
5203 struct info_list_node *node;
5204 bool found;
5206 for (each_unit = stash->f.all_comp_units;
5207 each_unit;
5208 each_unit = each_unit->next_unit)
5210 for (each_func = each_unit->function_table;
5211 each_func;
5212 each_func = each_func->prev_func)
5214 if (!each_func->name)
5215 continue;
5216 node = lookup_info_hash_table (stash->funcinfo_hash_table,
5217 each_func->name);
5218 BFD_ASSERT (node);
5219 found = false;
5220 while (node && !found)
5222 found = node->info == each_func;
5223 node = node->next;
5225 BFD_ASSERT (found);
5228 for (each_var = each_unit->variable_table;
5229 each_var;
5230 each_var = each_var->prev_var)
5232 if (!each_var->name || !each_var->file || each_var->stack)
5233 continue;
5234 node = lookup_info_hash_table (stash->varinfo_hash_table,
5235 each_var->name);
5236 BFD_ASSERT (node);
5237 found = false;
5238 while (node && !found)
5240 found = node->info == each_var;
5241 node = node->next;
5243 BFD_ASSERT (found);
5248 /* Check to see if we want to enable the info hash tables, which consume
5249 quite a bit of memory. Currently we only check the number times
5250 bfd_dwarf2_find_line is called. In the future, we may also want to
5251 take the number of symbols into account. */
5253 static void
5254 stash_maybe_enable_info_hash_tables (bfd *abfd, struct dwarf2_debug *stash)
5256 BFD_ASSERT (stash->info_hash_status == STASH_INFO_HASH_OFF);
5258 if (stash->info_hash_count++ < STASH_INFO_HASH_TRIGGER)
5259 return;
5261 /* FIXME: Maybe we should check the reduce_memory_overheads
5262 and optimize fields in the bfd_link_info structure ? */
5264 /* Create hash tables. */
5265 stash->funcinfo_hash_table = create_info_hash_table (abfd);
5266 stash->varinfo_hash_table = create_info_hash_table (abfd);
5267 if (!stash->funcinfo_hash_table || !stash->varinfo_hash_table)
5269 /* Turn off info hashes if any allocation above fails. */
5270 stash->info_hash_status = STASH_INFO_HASH_DISABLED;
5271 return;
5273 /* We need a forced update so that the info hash tables will
5274 be created even though there is no compilation unit. That
5275 happens if STASH_INFO_HASH_TRIGGER is 0. */
5276 if (stash_maybe_update_info_hash_tables (stash))
5277 stash->info_hash_status = STASH_INFO_HASH_ON;
5280 /* Find the file and line associated with a symbol and address using the
5281 info hash tables of a stash. If there is a match, the function returns
5282 TRUE and update the locations pointed to by filename_ptr and linenumber_ptr;
5283 otherwise it returns FALSE. */
5285 static bool
5286 stash_find_line_fast (struct dwarf2_debug *stash,
5287 asymbol *sym,
5288 bfd_vma addr,
5289 const char **filename_ptr,
5290 unsigned int *linenumber_ptr)
5292 BFD_ASSERT (stash->info_hash_status == STASH_INFO_HASH_ON);
5294 if (sym->flags & BSF_FUNCTION)
5295 return info_hash_lookup_funcinfo (stash->funcinfo_hash_table, sym, addr,
5296 filename_ptr, linenumber_ptr);
5297 return info_hash_lookup_varinfo (stash->varinfo_hash_table, sym, addr,
5298 filename_ptr, linenumber_ptr);
5301 /* Save current section VMAs. */
5303 static bool
5304 save_section_vma (const bfd *abfd, struct dwarf2_debug *stash)
5306 asection *s;
5307 unsigned int i;
5309 if (abfd->section_count == 0)
5310 return true;
5311 stash->sec_vma = bfd_malloc (sizeof (*stash->sec_vma) * abfd->section_count);
5312 if (stash->sec_vma == NULL)
5313 return false;
5314 stash->sec_vma_count = abfd->section_count;
5315 for (i = 0, s = abfd->sections;
5316 s != NULL && i < abfd->section_count;
5317 i++, s = s->next)
5319 if (s->output_section != NULL)
5320 stash->sec_vma[i] = s->output_section->vma + s->output_offset;
5321 else
5322 stash->sec_vma[i] = s->vma;
5324 return true;
5327 /* Compare current section VMAs against those at the time the stash
5328 was created. If find_nearest_line is used in linker warnings or
5329 errors early in the link process, the debug info stash will be
5330 invalid for later calls. This is because we relocate debug info
5331 sections, so the stashed section contents depend on symbol values,
5332 which in turn depend on section VMAs. */
5334 static bool
5335 section_vma_same (const bfd *abfd, const struct dwarf2_debug *stash)
5337 asection *s;
5338 unsigned int i;
5340 /* PR 24334: If the number of sections in ABFD has changed between
5341 when the stash was created and now, then we cannot trust the
5342 stashed vma information. */
5343 if (abfd->section_count != stash->sec_vma_count)
5344 return false;
5346 for (i = 0, s = abfd->sections;
5347 s != NULL && i < abfd->section_count;
5348 i++, s = s->next)
5350 bfd_vma vma;
5352 if (s->output_section != NULL)
5353 vma = s->output_section->vma + s->output_offset;
5354 else
5355 vma = s->vma;
5356 if (vma != stash->sec_vma[i])
5357 return false;
5359 return true;
5362 /* Read debug information from DEBUG_BFD when DEBUG_BFD is specified.
5363 If DEBUG_BFD is not specified, we read debug information from ABFD
5364 or its gnu_debuglink. The results will be stored in PINFO.
5365 The function returns TRUE iff debug information is ready. */
5367 bool
5368 _bfd_dwarf2_slurp_debug_info (bfd *abfd, bfd *debug_bfd,
5369 const struct dwarf_debug_section *debug_sections,
5370 asymbol **symbols,
5371 void **pinfo,
5372 bool do_place)
5374 bfd_size_type total_size;
5375 asection *msec;
5376 struct dwarf2_debug *stash = (struct dwarf2_debug *) *pinfo;
5378 if (stash != NULL)
5380 if (stash->orig_bfd == abfd
5381 && section_vma_same (abfd, stash))
5383 /* Check that we did previously find some debug information
5384 before attempting to make use of it. */
5385 if (stash->f.dwarf_info_size != 0)
5387 if (do_place && !place_sections (abfd, stash))
5388 return false;
5389 return true;
5392 return false;
5394 _bfd_dwarf2_cleanup_debug_info (abfd, pinfo);
5395 memset (stash, 0, sizeof (*stash));
5397 else
5399 stash = (struct dwarf2_debug *) bfd_zalloc (abfd, sizeof (*stash));
5400 if (! stash)
5401 return false;
5402 *pinfo = stash;
5404 stash->orig_bfd = abfd;
5405 stash->debug_sections = debug_sections;
5406 stash->f.syms = symbols;
5407 if (!save_section_vma (abfd, stash))
5408 return false;
5410 stash->f.abbrev_offsets = htab_create_alloc (10, hash_abbrev, eq_abbrev,
5411 del_abbrev, calloc, free);
5412 if (!stash->f.abbrev_offsets)
5413 return false;
5415 stash->alt.abbrev_offsets = htab_create_alloc (10, hash_abbrev, eq_abbrev,
5416 del_abbrev, calloc, free);
5417 if (!stash->alt.abbrev_offsets)
5418 return false;
5420 stash->f.trie_root = alloc_trie_leaf (abfd);
5421 if (!stash->f.trie_root)
5422 return false;
5424 stash->alt.trie_root = alloc_trie_leaf (abfd);
5425 if (!stash->alt.trie_root)
5426 return false;
5428 if (debug_bfd == NULL)
5429 debug_bfd = abfd;
5431 msec = find_debug_info (debug_bfd, debug_sections, NULL);
5432 if (msec == NULL && abfd == debug_bfd)
5434 char * debug_filename;
5436 debug_filename = bfd_follow_build_id_debuglink (abfd, DEBUGDIR);
5437 if (debug_filename == NULL)
5438 debug_filename = bfd_follow_gnu_debuglink (abfd, DEBUGDIR);
5440 if (debug_filename == NULL)
5441 /* No dwarf2 info, and no gnu_debuglink to follow.
5442 Note that at this point the stash has been allocated, but
5443 contains zeros. This lets future calls to this function
5444 fail more quickly. */
5445 return false;
5447 debug_bfd = bfd_openr (debug_filename, NULL);
5448 free (debug_filename);
5449 if (debug_bfd == NULL)
5450 /* FIXME: Should we report our failure to follow the debuglink ? */
5451 return false;
5453 /* Set BFD_DECOMPRESS to decompress debug sections. */
5454 debug_bfd->flags |= BFD_DECOMPRESS;
5455 if (!bfd_check_format (debug_bfd, bfd_object)
5456 || (msec = find_debug_info (debug_bfd,
5457 debug_sections, NULL)) == NULL
5458 || !bfd_generic_link_read_symbols (debug_bfd))
5460 bfd_close (debug_bfd);
5461 return false;
5464 symbols = bfd_get_outsymbols (debug_bfd);
5465 stash->f.syms = symbols;
5466 stash->close_on_cleanup = true;
5468 stash->f.bfd_ptr = debug_bfd;
5470 if (do_place
5471 && !place_sections (abfd, stash))
5472 return false;
5474 /* There can be more than one DWARF2 info section in a BFD these
5475 days. First handle the easy case when there's only one. If
5476 there's more than one, try case two: read them all in and produce
5477 one large stash. We do this in two passes - in the first pass we
5478 just accumulate the section sizes, and in the second pass we
5479 read in the section's contents. (The allows us to avoid
5480 reallocing the data as we add sections to the stash.) */
5482 if (! find_debug_info (debug_bfd, debug_sections, msec))
5484 /* Case 1: only one info section. */
5485 total_size = msec->size;
5486 if (! read_section (debug_bfd, &stash->debug_sections[debug_info],
5487 symbols, 0,
5488 &stash->f.dwarf_info_buffer, &total_size))
5489 goto restore_vma;
5491 else
5493 /* Case 2: multiple sections. */
5494 for (total_size = 0;
5495 msec;
5496 msec = find_debug_info (debug_bfd, debug_sections, msec))
5498 if (_bfd_section_size_insane (debug_bfd, msec))
5499 goto restore_vma;
5500 /* Catch PR25070 testcase overflowing size calculation here. */
5501 if (total_size + msec->size < total_size)
5503 bfd_set_error (bfd_error_no_memory);
5504 goto restore_vma;
5506 total_size += msec->size;
5509 stash->f.dwarf_info_buffer = (bfd_byte *) bfd_malloc (total_size);
5510 if (stash->f.dwarf_info_buffer == NULL)
5511 goto restore_vma;
5513 total_size = 0;
5514 for (msec = find_debug_info (debug_bfd, debug_sections, NULL);
5515 msec;
5516 msec = find_debug_info (debug_bfd, debug_sections, msec))
5518 bfd_size_type size;
5520 size = msec->size;
5521 if (size == 0)
5522 continue;
5524 if (!(bfd_simple_get_relocated_section_contents
5525 (debug_bfd, msec, stash->f.dwarf_info_buffer + total_size,
5526 symbols)))
5527 goto restore_vma;
5529 total_size += size;
5533 stash->f.info_ptr = stash->f.dwarf_info_buffer;
5534 stash->f.dwarf_info_size = total_size;
5535 return true;
5537 restore_vma:
5538 unset_sections (stash);
5539 return false;
5542 /* Parse the next DWARF2 compilation unit at FILE->INFO_PTR. */
5544 static struct comp_unit *
5545 stash_comp_unit (struct dwarf2_debug *stash, struct dwarf2_debug_file *file)
5547 bfd_size_type length;
5548 unsigned int offset_size;
5549 bfd_byte *info_ptr_unit = file->info_ptr;
5550 bfd_byte *info_ptr_end = file->dwarf_info_buffer + file->dwarf_info_size;
5552 if (file->info_ptr >= info_ptr_end)
5553 return NULL;
5555 length = read_4_bytes (file->bfd_ptr, &file->info_ptr, info_ptr_end);
5556 /* A 0xffffff length is the DWARF3 way of indicating
5557 we use 64-bit offsets, instead of 32-bit offsets. */
5558 if (length == 0xffffffff)
5560 offset_size = 8;
5561 length = read_8_bytes (file->bfd_ptr, &file->info_ptr, info_ptr_end);
5563 /* A zero length is the IRIX way of indicating 64-bit offsets,
5564 mostly because the 64-bit length will generally fit in 32
5565 bits, and the endianness helps. */
5566 else if (length == 0)
5568 offset_size = 8;
5569 length = read_4_bytes (file->bfd_ptr, &file->info_ptr, info_ptr_end);
5571 /* In the absence of the hints above, we assume 32-bit DWARF2
5572 offsets even for targets with 64-bit addresses, because:
5573 a) most of the time these targets will not have generated
5574 more than 2Gb of debug info and so will not need 64-bit
5575 offsets,
5577 b) if they do use 64-bit offsets but they are not using
5578 the size hints that are tested for above then they are
5579 not conforming to the DWARF3 standard anyway. */
5580 else
5581 offset_size = 4;
5583 if (length != 0
5584 && length <= (size_t) (info_ptr_end - file->info_ptr))
5586 struct comp_unit *each = parse_comp_unit (stash, file,
5587 file->info_ptr, length,
5588 info_ptr_unit, offset_size);
5589 if (each)
5591 if (file->comp_unit_tree == NULL)
5592 file->comp_unit_tree
5593 = splay_tree_new (splay_tree_compare_addr_range,
5594 splay_tree_free_addr_range, NULL);
5596 struct addr_range *r
5597 = (struct addr_range *)bfd_malloc (sizeof (struct addr_range));
5598 r->start = each->info_ptr_unit;
5599 r->end = each->end_ptr;
5600 splay_tree_node v = splay_tree_lookup (file->comp_unit_tree,
5601 (splay_tree_key)r);
5602 if (v != NULL || r->end <= r->start)
5603 abort ();
5604 splay_tree_insert (file->comp_unit_tree, (splay_tree_key)r,
5605 (splay_tree_value)each);
5607 if (file->all_comp_units)
5608 file->all_comp_units->prev_unit = each;
5609 else
5610 file->last_comp_unit = each;
5612 each->next_unit = file->all_comp_units;
5613 file->all_comp_units = each;
5615 if (each->arange.high == 0)
5617 each->next_unit_without_ranges = file->all_comp_units_without_ranges;
5618 file->all_comp_units_without_ranges = each->next_unit_without_ranges;
5621 file->info_ptr += length;
5622 return each;
5626 /* Don't trust any of the DWARF info after a corrupted length or
5627 parse error. */
5628 file->info_ptr = info_ptr_end;
5629 return NULL;
5632 /* Hash function for an asymbol. */
5634 static hashval_t
5635 hash_asymbol (const void *sym)
5637 const asymbol *asym = sym;
5638 return htab_hash_string (asym->name);
5641 /* Equality function for asymbols. */
5643 static int
5644 eq_asymbol (const void *a, const void *b)
5646 const asymbol *sa = a;
5647 const asymbol *sb = b;
5648 return strcmp (sa->name, sb->name) == 0;
5651 /* Scan the debug information in PINFO looking for a DW_TAG_subprogram
5652 abbrev with a DW_AT_low_pc attached to it. Then lookup that same
5653 symbol in SYMBOLS and return the difference between the low_pc and
5654 the symbol's address. Returns 0 if no suitable symbol could be found. */
5656 bfd_signed_vma
5657 _bfd_dwarf2_find_symbol_bias (asymbol ** symbols, void ** pinfo)
5659 struct dwarf2_debug *stash;
5660 struct comp_unit * unit;
5661 htab_t sym_hash;
5662 bfd_signed_vma result = 0;
5663 asymbol ** psym;
5665 stash = (struct dwarf2_debug *) *pinfo;
5667 if (stash == NULL || symbols == NULL)
5668 return 0;
5670 sym_hash = htab_create_alloc (10, hash_asymbol, eq_asymbol,
5671 NULL, xcalloc, free);
5672 for (psym = symbols; * psym != NULL; psym++)
5674 asymbol * sym = * psym;
5676 if (sym->flags & BSF_FUNCTION && sym->section != NULL)
5678 void **slot = htab_find_slot (sym_hash, sym, INSERT);
5679 *slot = sym;
5683 for (unit = stash->f.all_comp_units; unit; unit = unit->next_unit)
5685 struct funcinfo * func;
5687 comp_unit_maybe_decode_line_info (unit);
5689 for (func = unit->function_table; func != NULL; func = func->prev_func)
5690 if (func->name && func->arange.low)
5692 asymbol search, *sym;
5694 /* FIXME: Do we need to scan the aranges looking for the
5695 lowest pc value? */
5697 search.name = func->name;
5698 sym = htab_find (sym_hash, &search);
5699 if (sym != NULL)
5701 result = func->arange.low - (sym->value + sym->section->vma);
5702 goto done;
5707 done:
5708 htab_delete (sym_hash);
5709 return result;
5712 /* See _bfd_dwarf2_find_nearest_line_with_alt. */
5715 _bfd_dwarf2_find_nearest_line (bfd *abfd,
5716 asymbol **symbols,
5717 asymbol *symbol,
5718 asection *section,
5719 bfd_vma offset,
5720 const char **filename_ptr,
5721 const char **functionname_ptr,
5722 unsigned int *linenumber_ptr,
5723 unsigned int *discriminator_ptr,
5724 const struct dwarf_debug_section *debug_sections,
5725 void **pinfo)
5727 return _bfd_dwarf2_find_nearest_line_with_alt
5728 (abfd, NULL, symbols, symbol, section, offset, filename_ptr,
5729 functionname_ptr, linenumber_ptr, discriminator_ptr, debug_sections,
5730 pinfo);
5733 /* Find the source code location of SYMBOL. If SYMBOL is NULL
5734 then find the nearest source code location corresponding to
5735 the address SECTION + OFFSET.
5736 Returns 1 if the line is found without error and fills in
5737 FILENAME_PTR and LINENUMBER_PTR. In the case where SYMBOL was
5738 NULL the FUNCTIONNAME_PTR is also filled in.
5739 Returns 2 if partial information from _bfd_elf_find_function is
5740 returned (function and maybe file) by looking at symbols. DWARF2
5741 info is present but not regarding the requested code location.
5742 Returns 0 otherwise.
5743 SYMBOLS contains the symbol table for ABFD.
5744 DEBUG_SECTIONS contains the name of the dwarf debug sections.
5745 If ALT_FILENAME is given, attempt to open the file and use it
5746 as the .gnu_debugaltlink file. Otherwise this file will be
5747 searched for when needed. */
5750 _bfd_dwarf2_find_nearest_line_with_alt
5751 (bfd *abfd,
5752 const char *alt_filename,
5753 asymbol **symbols,
5754 asymbol *symbol,
5755 asection *section,
5756 bfd_vma offset,
5757 const char **filename_ptr,
5758 const char **functionname_ptr,
5759 unsigned int *linenumber_ptr,
5760 unsigned int *discriminator_ptr,
5761 const struct dwarf_debug_section *debug_sections,
5762 void **pinfo)
5764 /* Read each compilation unit from the section .debug_info, and check
5765 to see if it contains the address we are searching for. If yes,
5766 lookup the address, and return the line number info. If no, go
5767 on to the next compilation unit.
5769 We keep a list of all the previously read compilation units, and
5770 a pointer to the next un-read compilation unit. Check the
5771 previously read units before reading more. */
5772 struct dwarf2_debug *stash;
5773 /* What address are we looking for? */
5774 bfd_vma addr;
5775 struct comp_unit* each;
5776 struct funcinfo *function = NULL;
5777 int found = false;
5778 bool do_line;
5780 *filename_ptr = NULL;
5781 if (functionname_ptr != NULL)
5782 *functionname_ptr = NULL;
5783 *linenumber_ptr = 0;
5784 if (discriminator_ptr)
5785 *discriminator_ptr = 0;
5787 if (! _bfd_dwarf2_slurp_debug_info (abfd, NULL, debug_sections,
5788 symbols, pinfo,
5789 (abfd->flags & (EXEC_P | DYNAMIC)) == 0))
5790 return false;
5792 stash = (struct dwarf2_debug *) *pinfo;
5794 if (stash->alt.bfd_ptr == NULL && alt_filename != NULL)
5796 bfd *alt_bfd = bfd_openr (alt_filename, NULL);
5798 if (alt_bfd == NULL)
5799 /* bfd_openr will have set the bfd_error. */
5800 return false;
5801 if (!bfd_check_format (alt_bfd, bfd_object))
5803 bfd_set_error (bfd_error_wrong_format);
5804 bfd_close (alt_bfd);
5805 return false;
5808 stash->alt.bfd_ptr = alt_bfd;
5811 do_line = symbol != NULL;
5812 if (do_line)
5814 BFD_ASSERT (section == NULL && offset == 0 && functionname_ptr == NULL);
5815 section = bfd_asymbol_section (symbol);
5816 addr = symbol->value;
5818 else
5820 BFD_ASSERT (section != NULL && functionname_ptr != NULL);
5821 addr = offset;
5823 /* If we have no SYMBOL but the section we're looking at is not a
5824 code section, then take a look through the list of symbols to see
5825 if we have a symbol at the address we're looking for. If we do
5826 then use this to look up line information. This will allow us to
5827 give file and line results for data symbols. We exclude code
5828 symbols here, if we look up a function symbol and then look up the
5829 line information we'll actually return the line number for the
5830 opening '{' rather than the function definition line. This is
5831 because looking up by symbol uses the line table, in which the
5832 first line for a function is usually the opening '{', while
5833 looking up the function by section + offset uses the
5834 DW_AT_decl_line from the function DW_TAG_subprogram for the line,
5835 which will be the line of the function name. */
5836 if (symbols != NULL && (section->flags & SEC_CODE) == 0)
5838 asymbol **tmp;
5840 for (tmp = symbols; (*tmp) != NULL; ++tmp)
5841 if ((*tmp)->the_bfd == abfd
5842 && (*tmp)->section == section
5843 && (*tmp)->value == offset
5844 && ((*tmp)->flags & BSF_SECTION_SYM) == 0)
5846 symbol = *tmp;
5847 do_line = true;
5848 /* For local symbols, keep going in the hope we find a
5849 global. */
5850 if ((symbol->flags & BSF_GLOBAL) != 0)
5851 break;
5856 if (section->output_section)
5857 addr += section->output_section->vma + section->output_offset;
5858 else
5859 addr += section->vma;
5861 /* A null info_ptr indicates that there is no dwarf2 info
5862 (or that an error occured while setting up the stash). */
5863 if (! stash->f.info_ptr)
5864 return false;
5866 stash->inliner_chain = NULL;
5868 /* Check the previously read comp. units first. */
5869 if (do_line)
5871 /* The info hash tables use quite a bit of memory. We may not want to
5872 always use them. We use some heuristics to decide if and when to
5873 turn it on. */
5874 if (stash->info_hash_status == STASH_INFO_HASH_OFF)
5875 stash_maybe_enable_info_hash_tables (abfd, stash);
5877 /* Keep info hash table up to date if they are available. Note that we
5878 may disable the hash tables if there is any error duing update. */
5879 if (stash->info_hash_status == STASH_INFO_HASH_ON)
5880 stash_maybe_update_info_hash_tables (stash);
5882 if (stash->info_hash_status == STASH_INFO_HASH_ON)
5884 found = stash_find_line_fast (stash, symbol, addr,
5885 filename_ptr, linenumber_ptr);
5886 if (found)
5887 goto done;
5890 /* Check the previously read comp. units first. */
5891 for (each = stash->f.all_comp_units; each; each = each->next_unit)
5892 if ((symbol->flags & BSF_FUNCTION) == 0
5893 || comp_unit_may_contain_address (each, addr))
5895 found = comp_unit_find_line (each, symbol, addr, filename_ptr,
5896 linenumber_ptr);
5897 if (found)
5898 goto done;
5901 else
5903 struct trie_node *trie = stash->f.trie_root;
5904 unsigned int bits = VMA_BITS - 8;
5905 struct comp_unit **prev_each;
5907 /* Traverse interior nodes until we get to a leaf. */
5908 while (trie && trie->num_room_in_leaf == 0)
5910 int ch = (addr >> bits) & 0xff;
5911 trie = ((struct trie_interior *) trie)->children[ch];
5912 bits -= 8;
5915 if (trie)
5917 const struct trie_leaf *leaf = (struct trie_leaf *) trie;
5918 unsigned int i;
5920 for (i = 0; i < leaf->num_stored_in_leaf; ++i)
5921 leaf->ranges[i].unit->mark = false;
5923 for (i = 0; i < leaf->num_stored_in_leaf; ++i)
5925 struct comp_unit *unit = leaf->ranges[i].unit;
5926 if (unit->mark
5927 || addr < leaf->ranges[i].low_pc
5928 || addr >= leaf->ranges[i].high_pc)
5929 continue;
5930 unit->mark = true;
5932 found = comp_unit_find_nearest_line (unit, addr,
5933 filename_ptr,
5934 &function,
5935 linenumber_ptr,
5936 discriminator_ptr);
5937 if (found)
5938 goto done;
5942 /* Also scan through all compilation units without any ranges,
5943 taking them out of the list if they have acquired any since
5944 last time. */
5945 prev_each = &stash->f.all_comp_units_without_ranges;
5946 for (each = *prev_each; each; each = each->next_unit_without_ranges)
5948 if (each->arange.high != 0)
5950 *prev_each = each->next_unit_without_ranges;
5951 continue;
5954 found = comp_unit_find_nearest_line (each, addr,
5955 filename_ptr,
5956 &function,
5957 linenumber_ptr,
5958 discriminator_ptr);
5959 if (found)
5960 goto done;
5961 prev_each = &each->next_unit_without_ranges;
5965 /* Read each remaining comp. units checking each as they are read. */
5966 while ((each = stash_comp_unit (stash, &stash->f)) != NULL)
5968 /* DW_AT_low_pc and DW_AT_high_pc are optional for
5969 compilation units. If we don't have them (i.e.,
5970 unit->high == 0), we need to consult the line info table
5971 to see if a compilation unit contains the given
5972 address. */
5973 if (do_line)
5974 found = (((symbol->flags & BSF_FUNCTION) == 0
5975 || comp_unit_may_contain_address (each, addr))
5976 && comp_unit_find_line (each, symbol, addr,
5977 filename_ptr, linenumber_ptr));
5978 else
5979 found = (comp_unit_may_contain_address (each, addr)
5980 && comp_unit_find_nearest_line (each, addr,
5981 filename_ptr,
5982 &function,
5983 linenumber_ptr,
5984 discriminator_ptr));
5986 if (found)
5987 break;
5990 done:
5991 if (functionname_ptr && function && function->is_linkage)
5993 *functionname_ptr = function->name;
5994 if (!found)
5995 found = 2;
5997 else if (functionname_ptr
5998 && (!*functionname_ptr
5999 || (function && !function->is_linkage)))
6001 asymbol *fun;
6002 asymbol **syms = symbols;
6003 asection *sec = section;
6005 _bfd_dwarf2_stash_syms (stash, abfd, &sec, &syms);
6006 fun = _bfd_elf_find_function (abfd, syms, sec, offset,
6007 *filename_ptr ? NULL : filename_ptr,
6008 functionname_ptr);
6010 if (!found && fun != NULL)
6011 found = 2;
6013 if (function && !function->is_linkage)
6015 bfd_vma sec_vma;
6017 sec_vma = section->vma;
6018 if (section->output_section != NULL)
6019 sec_vma = section->output_section->vma + section->output_offset;
6020 if (fun == NULL)
6021 *functionname_ptr = function->name;
6022 else if (fun->value + sec_vma == function->arange.low)
6023 function->name = *functionname_ptr;
6024 /* Even if we didn't find a linkage name, say that we have
6025 to stop a repeated search of symbols. */
6026 function->is_linkage = true;
6030 unset_sections (stash);
6032 return found;
6035 bool
6036 _bfd_dwarf2_find_inliner_info (bfd *abfd ATTRIBUTE_UNUSED,
6037 const char **filename_ptr,
6038 const char **functionname_ptr,
6039 unsigned int *linenumber_ptr,
6040 void **pinfo)
6042 struct dwarf2_debug *stash;
6044 stash = (struct dwarf2_debug *) *pinfo;
6045 if (stash)
6047 struct funcinfo *func = stash->inliner_chain;
6049 if (func && func->caller_func)
6051 *filename_ptr = func->caller_file;
6052 *functionname_ptr = func->caller_func->name;
6053 *linenumber_ptr = func->caller_line;
6054 stash->inliner_chain = func->caller_func;
6055 return true;
6059 return false;
6062 void
6063 _bfd_dwarf2_cleanup_debug_info (bfd *abfd, void **pinfo)
6065 struct dwarf2_debug *stash = (struct dwarf2_debug *) *pinfo;
6066 struct comp_unit *each;
6067 struct dwarf2_debug_file *file;
6069 if (abfd == NULL || stash == NULL)
6070 return;
6072 if (stash->varinfo_hash_table)
6073 bfd_hash_table_free (&stash->varinfo_hash_table->base);
6074 if (stash->funcinfo_hash_table)
6075 bfd_hash_table_free (&stash->funcinfo_hash_table->base);
6077 file = &stash->f;
6078 while (1)
6080 for (each = file->all_comp_units; each; each = each->next_unit)
6082 struct funcinfo *function_table = each->function_table;
6083 struct varinfo *variable_table = each->variable_table;
6085 if (each->line_table && each->line_table != file->line_table)
6087 free (each->line_table->files);
6088 free (each->line_table->dirs);
6091 free (each->lookup_funcinfo_table);
6092 each->lookup_funcinfo_table = NULL;
6094 while (function_table)
6096 free (function_table->file);
6097 function_table->file = NULL;
6098 free (function_table->caller_file);
6099 function_table->caller_file = NULL;
6100 function_table = function_table->prev_func;
6103 while (variable_table)
6105 free (variable_table->file);
6106 variable_table->file = NULL;
6107 variable_table = variable_table->prev_var;
6111 if (file->line_table)
6113 free (file->line_table->files);
6114 free (file->line_table->dirs);
6116 htab_delete (file->abbrev_offsets);
6117 if (file->comp_unit_tree != NULL)
6118 splay_tree_delete (file->comp_unit_tree);
6120 free (file->dwarf_line_str_buffer);
6121 free (file->dwarf_str_buffer);
6122 free (file->dwarf_ranges_buffer);
6123 free (file->dwarf_line_buffer);
6124 free (file->dwarf_abbrev_buffer);
6125 free (file->dwarf_info_buffer);
6126 if (file == &stash->alt)
6127 break;
6128 file = &stash->alt;
6130 free (stash->sec_vma);
6131 free (stash->adjusted_sections);
6132 if (stash->close_on_cleanup)
6133 bfd_close (stash->f.bfd_ptr);
6134 if (stash->alt.bfd_ptr)
6135 bfd_close (stash->alt.bfd_ptr);
6138 typedef struct elf_find_function_cache
6140 asection * last_section;
6141 asymbol * func;
6142 const char * filename;
6143 bfd_size_type code_size;
6144 bfd_vma code_off;
6146 } elf_find_function_cache;
6149 /* Returns TRUE if symbol SYM with address CODE_OFF and size CODE_SIZE
6150 is a better fit to match OFFSET than whatever is currenly stored in
6151 CACHE. */
6153 static inline bool
6154 better_fit (elf_find_function_cache * cache,
6155 asymbol * sym,
6156 bfd_vma code_off,
6157 bfd_size_type code_size,
6158 bfd_vma offset)
6160 /* If the symbol is beyond the desired offset, ignore it. */
6161 if (code_off > offset)
6162 return false;
6164 /* If the symbol is further away from the desired
6165 offset than our current best, then ignore it. */
6166 if (code_off < cache->code_off)
6167 return false;
6169 /* On the other hand, if it is closer, then use it. */
6170 if (code_off > cache->code_off)
6171 return true;
6173 /* assert (code_off == cache->code_off); */
6175 /* If our current best fit does not actually reach the desired
6176 offset... */
6177 if (cache->code_off + cache->code_size <= offset)
6178 /* ... then return whichever candidate covers
6179 more area and hence gets closer to OFFSET. */
6180 return code_size > cache->code_size;
6182 /* The current cache'd symbol covers OFFSET. */
6184 /* If the new symbol does not cover the desired offset then skip it. */
6185 if (code_off + code_size <= offset)
6186 return false;
6188 /* Both symbols cover OFFSET. */
6190 /* Prefer functions over non-functions. */
6191 flagword cache_flags = cache->func->flags;
6192 flagword sym_flags = sym->flags;
6194 if ((cache_flags & BSF_FUNCTION) && ((sym_flags & BSF_FUNCTION) == 0))
6195 return false;
6196 if ((sym_flags & BSF_FUNCTION) && ((cache_flags & BSF_FUNCTION) == 0))
6197 return true;
6199 /* FIXME: Should we choose LOCAL over GLOBAL ? */
6201 /* Prefer typed symbols over notyped. */
6202 int cache_type = ELF_ST_TYPE (((elf_symbol_type *) cache->func)->internal_elf_sym.st_info);
6203 int sym_type = ELF_ST_TYPE (((elf_symbol_type *) sym)->internal_elf_sym.st_info);
6205 if (cache_type == STT_NOTYPE && sym_type != STT_NOTYPE)
6206 return true;
6207 if (cache_type != STT_NOTYPE && sym_type == STT_NOTYPE)
6208 return false;
6210 /* Otherwise choose whichever symbol covers a smaller area. */
6211 return code_size < cache->code_size;
6214 /* Find the function to a particular section and offset,
6215 for error reporting. */
6217 asymbol *
6218 _bfd_elf_find_function (bfd *abfd,
6219 asymbol **symbols,
6220 asection *section,
6221 bfd_vma offset,
6222 const char **filename_ptr,
6223 const char **functionname_ptr)
6225 if (symbols == NULL)
6226 return NULL;
6228 if (bfd_get_flavour (abfd) != bfd_target_elf_flavour)
6229 return NULL;
6231 elf_find_function_cache * cache = elf_tdata (abfd)->elf_find_function_cache;
6233 if (cache == NULL)
6235 cache = bfd_zalloc (abfd, sizeof (*cache));
6236 elf_tdata (abfd)->elf_find_function_cache = cache;
6237 if (cache == NULL)
6238 return NULL;
6241 if (cache->last_section != section
6242 || cache->func == NULL
6243 || offset < cache->func->value
6244 || offset >= cache->func->value + cache->code_size)
6246 asymbol *file;
6247 asymbol **p;
6248 /* ??? Given multiple file symbols, it is impossible to reliably
6249 choose the right file name for global symbols. File symbols are
6250 local symbols, and thus all file symbols must sort before any
6251 global symbols. The ELF spec may be interpreted to say that a
6252 file symbol must sort before other local symbols, but currently
6253 ld -r doesn't do this. So, for ld -r output, it is possible to
6254 make a better choice of file name for local symbols by ignoring
6255 file symbols appearing after a given local symbol. */
6256 enum { nothing_seen, symbol_seen, file_after_symbol_seen } state;
6257 const struct elf_backend_data *bed = get_elf_backend_data (abfd);
6259 file = NULL;
6260 state = nothing_seen;
6261 cache->filename = NULL;
6262 cache->func = NULL;
6263 cache->code_size = 0;
6264 cache->code_off = 0;
6265 cache->last_section = section;
6267 for (p = symbols; *p != NULL; p++)
6269 asymbol *sym = *p;
6270 bfd_vma code_off;
6271 bfd_size_type size;
6273 if ((sym->flags & BSF_FILE) != 0)
6275 file = sym;
6276 if (state == symbol_seen)
6277 state = file_after_symbol_seen;
6278 continue;
6281 if (state == nothing_seen)
6282 state = symbol_seen;
6284 size = bed->maybe_function_sym (sym, section, &code_off);
6286 if (size == 0)
6287 continue;
6289 if (better_fit (cache, sym, code_off, size, offset))
6291 cache->func = sym;
6292 cache->code_size = size;
6293 cache->code_off = code_off;
6294 cache->filename = NULL;
6296 if (file != NULL
6297 && ((sym->flags & BSF_LOCAL) != 0
6298 || state != file_after_symbol_seen))
6299 cache->filename = bfd_asymbol_name (file);
6301 /* Otherwise, if the symbol is beyond the desired offset but it
6302 lies within the bounds of the current best match then reduce
6303 the size of the current best match so that future searches
6304 will not not used the cached symbol by mistake. */
6305 else if (code_off > offset
6306 && code_off > cache->code_off
6307 && code_off < cache->code_off + cache->code_size)
6309 cache->code_size = code_off - cache->code_off;
6314 if (cache->func == NULL)
6315 return NULL;
6317 if (filename_ptr)
6318 *filename_ptr = cache->filename;
6319 if (functionname_ptr)
6320 *functionname_ptr = bfd_asymbol_name (cache->func);
6322 return cache->func;