1 // dwarf_reader.cc -- parse dwarf2/3 debug information
3 // Copyright 2007, 2008, 2009, 2010, 2011, 2012 Free Software Foundation, Inc.
4 // Written by Ian Lance Taylor <iant@google.com>.
6 // This file is part of gold.
8 // This program is free software; you can redistribute it and/or modify
9 // it under the terms of the GNU General Public License as published by
10 // the Free Software Foundation; either version 3 of the License, or
11 // (at your option) any later version.
13 // This program is distributed in the hope that it will be useful,
14 // but WITHOUT ANY WARRANTY; without even the implied warranty of
15 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 // GNU General Public License for more details.
18 // You should have received a copy of the GNU General Public License
19 // along with this program; if not, write to the Free Software
20 // Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston,
21 // MA 02110-1301, USA.
28 #include "elfcpp_swap.h"
32 #include "dwarf_reader.h"
33 #include "int_encoding.h"
34 #include "compressed_output.h"
38 // Class Sized_elf_reloc_mapper
40 // Initialize the relocation tracker for section RELOC_SHNDX.
42 template<int size
, bool big_endian
>
44 Sized_elf_reloc_mapper
<size
, big_endian
>::do_initialize(
45 unsigned int reloc_shndx
, unsigned int reloc_type
)
47 this->reloc_type_
= reloc_type
;
48 return this->track_relocs_
.initialize(this->object_
, reloc_shndx
,
52 // Looks in the symtab to see what section a symbol is in.
54 template<int size
, bool big_endian
>
56 Sized_elf_reloc_mapper
<size
, big_endian
>::symbol_section(
57 unsigned int symndx
, Address
* value
, bool* is_ordinary
)
59 const int symsize
= elfcpp::Elf_sizes
<size
>::sym_size
;
60 gold_assert(static_cast<off_t
>((symndx
+ 1) * symsize
) <= this->symtab_size_
);
61 elfcpp::Sym
<size
, big_endian
> elfsym(this->symtab_
+ symndx
* symsize
);
62 *value
= elfsym
.get_st_value();
63 return this->object_
->adjust_sym_shndx(symndx
, elfsym
.get_st_shndx(),
67 // Return the section index and offset within the section of
68 // the target of the relocation for RELOC_OFFSET.
70 template<int size
, bool big_endian
>
72 Sized_elf_reloc_mapper
<size
, big_endian
>::do_get_reloc_target(
73 off_t reloc_offset
, off_t
* target_offset
)
75 this->track_relocs_
.advance(reloc_offset
);
76 if (reloc_offset
!= this->track_relocs_
.next_offset())
78 unsigned int symndx
= this->track_relocs_
.next_symndx();
79 typename
elfcpp::Elf_types
<size
>::Elf_Addr value
;
81 unsigned int target_shndx
= this->symbol_section(symndx
, &value
,
85 if (this->reloc_type_
== elfcpp::SHT_RELA
)
86 value
+= this->track_relocs_
.next_addend();
87 *target_offset
= value
;
91 static inline Elf_reloc_mapper
*
92 make_elf_reloc_mapper(Relobj
* object
, const unsigned char* symtab
,
95 if (object
->elfsize() == 32)
97 if (object
->is_big_endian())
99 #ifdef HAVE_TARGET_32_BIG
100 return new Sized_elf_reloc_mapper
<32, true>(object
, symtab
,
108 #ifdef HAVE_TARGET_32_LITTLE
109 return new Sized_elf_reloc_mapper
<32, false>(object
, symtab
,
116 else if (object
->elfsize() == 64)
118 if (object
->is_big_endian())
120 #ifdef HAVE_TARGET_64_BIG
121 return new Sized_elf_reloc_mapper
<64, true>(object
, symtab
,
129 #ifdef HAVE_TARGET_64_LITTLE
130 return new Sized_elf_reloc_mapper
<64, false>(object
, symtab
,
141 // class Dwarf_abbrev_table
144 Dwarf_abbrev_table::clear_abbrev_codes()
146 for (unsigned int code
= 0; code
< this->low_abbrev_code_max_
; ++code
)
148 if (this->low_abbrev_codes_
[code
] != NULL
)
150 delete this->low_abbrev_codes_
[code
];
151 this->low_abbrev_codes_
[code
] = NULL
;
154 for (Abbrev_code_table::iterator it
= this->high_abbrev_codes_
.begin();
155 it
!= this->high_abbrev_codes_
.end();
158 if (it
->second
!= NULL
)
161 this->high_abbrev_codes_
.clear();
164 // Read the abbrev table from an object file.
167 Dwarf_abbrev_table::do_read_abbrevs(
169 unsigned int abbrev_shndx
,
172 this->clear_abbrev_codes();
174 // If we don't have relocations, abbrev_shndx will be 0, and
175 // we'll have to hunt for the .debug_abbrev section.
176 if (abbrev_shndx
== 0 && this->abbrev_shndx_
> 0)
177 abbrev_shndx
= this->abbrev_shndx_
;
178 else if (abbrev_shndx
== 0)
180 for (unsigned int i
= 1; i
< object
->shnum(); ++i
)
182 std::string name
= object
->section_name(i
);
183 if (name
== ".debug_abbrev")
186 // Correct the offset. For incremental update links, we have a
187 // relocated offset that is relative to the output section, but
188 // here we need an offset relative to the input section.
189 abbrev_offset
-= object
->output_section_offset(i
);
193 if (abbrev_shndx
== 0)
197 // Get the section contents and decompress if necessary.
198 if (abbrev_shndx
!= this->abbrev_shndx_
)
200 if (this->owns_buffer_
&& this->buffer_
!= NULL
)
202 delete[] this->buffer_
;
203 this->owns_buffer_
= false;
206 section_size_type buffer_size
;
208 object
->decompressed_section_contents(abbrev_shndx
,
210 &this->owns_buffer_
);
211 this->buffer_end_
= this->buffer_
+ buffer_size
;
212 this->abbrev_shndx_
= abbrev_shndx
;
215 this->buffer_pos_
= this->buffer_
+ abbrev_offset
;
219 // Lookup the abbrev code entry for CODE. This function is called
220 // only when the abbrev code is not in the direct lookup table.
221 // It may be in the hash table, it may not have been read yet,
222 // or it may not exist in the abbrev table.
224 const Dwarf_abbrev_table::Abbrev_code
*
225 Dwarf_abbrev_table::do_get_abbrev(unsigned int code
)
227 // See if the abbrev code is already in the hash table.
228 Abbrev_code_table::const_iterator it
= this->high_abbrev_codes_
.find(code
);
229 if (it
!= this->high_abbrev_codes_
.end())
232 // Read and store abbrev code definitions until we find the
233 // one we're looking for.
236 // Read the abbrev code. A zero here indicates the end of the
239 if (this->buffer_pos_
>= this->buffer_end_
)
241 uint64_t nextcode
= read_unsigned_LEB_128(this->buffer_pos_
, &len
);
244 this->buffer_pos_
= this->buffer_end_
;
247 this->buffer_pos_
+= len
;
250 if (this->buffer_pos_
>= this->buffer_end_
)
252 uint64_t tag
= read_unsigned_LEB_128(this->buffer_pos_
, &len
);
253 this->buffer_pos_
+= len
;
255 // Read the has_children flag.
256 if (this->buffer_pos_
>= this->buffer_end_
)
258 bool has_children
= *this->buffer_pos_
== elfcpp::DW_CHILDREN_yes
;
259 this->buffer_pos_
+= 1;
261 // Read the list of (attribute, form) pairs.
262 Abbrev_code
* entry
= new Abbrev_code(tag
, has_children
);
265 // Read the attribute.
266 if (this->buffer_pos_
>= this->buffer_end_
)
268 uint64_t attr
= read_unsigned_LEB_128(this->buffer_pos_
, &len
);
269 this->buffer_pos_
+= len
;
272 if (this->buffer_pos_
>= this->buffer_end_
)
274 uint64_t form
= read_unsigned_LEB_128(this->buffer_pos_
, &len
);
275 this->buffer_pos_
+= len
;
277 // A (0,0) pair terminates the list.
278 if (attr
== 0 && form
== 0)
281 if (attr
== elfcpp::DW_AT_sibling
)
282 entry
->has_sibling_attribute
= true;
284 entry
->add_attribute(attr
, form
);
287 this->store_abbrev(nextcode
, entry
);
288 if (nextcode
== code
)
295 // class Dwarf_ranges_table
297 // Read the ranges table from an object file.
300 Dwarf_ranges_table::read_ranges_table(
302 const unsigned char* symtab
,
304 unsigned int ranges_shndx
)
306 // If we've already read this abbrev table, return immediately.
307 if (this->ranges_shndx_
> 0
308 && this->ranges_shndx_
== ranges_shndx
)
311 // If we don't have relocations, ranges_shndx will be 0, and
312 // we'll have to hunt for the .debug_ranges section.
313 if (ranges_shndx
== 0 && this->ranges_shndx_
> 0)
314 ranges_shndx
= this->ranges_shndx_
;
315 else if (ranges_shndx
== 0)
317 for (unsigned int i
= 1; i
< object
->shnum(); ++i
)
319 std::string name
= object
->section_name(i
);
320 if (name
== ".debug_ranges")
323 this->output_section_offset_
= object
->output_section_offset(i
);
327 if (ranges_shndx
== 0)
331 // Get the section contents and decompress if necessary.
332 if (ranges_shndx
!= this->ranges_shndx_
)
334 if (this->owns_ranges_buffer_
&& this->ranges_buffer_
!= NULL
)
336 delete[] this->ranges_buffer_
;
337 this->owns_ranges_buffer_
= false;
340 section_size_type buffer_size
;
341 this->ranges_buffer_
=
342 object
->decompressed_section_contents(ranges_shndx
,
344 &this->owns_ranges_buffer_
);
345 this->ranges_buffer_end_
= this->ranges_buffer_
+ buffer_size
;
346 this->ranges_shndx_
= ranges_shndx
;
349 if (this->ranges_reloc_mapper_
!= NULL
)
351 delete this->ranges_reloc_mapper_
;
352 this->ranges_reloc_mapper_
= NULL
;
355 // For incremental objects, we have no relocations.
356 if (object
->is_incremental())
359 // Find the relocation section for ".debug_ranges".
360 unsigned int reloc_shndx
= 0;
361 unsigned int reloc_type
= 0;
362 for (unsigned int i
= 0; i
< object
->shnum(); ++i
)
364 reloc_type
= object
->section_type(i
);
365 if ((reloc_type
== elfcpp::SHT_REL
366 || reloc_type
== elfcpp::SHT_RELA
)
367 && object
->section_info(i
) == ranges_shndx
)
374 this->ranges_reloc_mapper_
= make_elf_reloc_mapper(object
, symtab
,
376 this->ranges_reloc_mapper_
->initialize(reloc_shndx
, reloc_type
);
381 // Read a range list from section RANGES_SHNDX at offset RANGES_OFFSET.
384 Dwarf_ranges_table::read_range_list(
386 const unsigned char* symtab
,
388 unsigned int addr_size
,
389 unsigned int ranges_shndx
,
392 Dwarf_range_list
* ranges
;
394 if (!this->read_ranges_table(object
, symtab
, symtab_size
, ranges_shndx
))
397 // Correct the offset. For incremental update links, we have a
398 // relocated offset that is relative to the output section, but
399 // here we need an offset relative to the input section.
400 offset
-= this->output_section_offset_
;
402 // Read the range list at OFFSET.
403 ranges
= new Dwarf_range_list();
406 this->ranges_buffer_
+ offset
< this->ranges_buffer_end_
;
407 offset
+= 2 * addr_size
)
412 // Read the raw contents of the section.
415 start
= this->dwinfo_
->read_from_pointer
<32>(this->ranges_buffer_
417 end
= this->dwinfo_
->read_from_pointer
<32>(this->ranges_buffer_
422 start
= this->dwinfo_
->read_from_pointer
<64>(this->ranges_buffer_
424 end
= this->dwinfo_
->read_from_pointer
<64>(this->ranges_buffer_
428 // Check for relocations and adjust the values.
429 unsigned int shndx1
= 0;
430 unsigned int shndx2
= 0;
431 if (this->ranges_reloc_mapper_
!= NULL
)
434 this->ranges_reloc_mapper_
->get_reloc_target(offset
, &start
);
436 this->ranges_reloc_mapper_
->get_reloc_target(offset
+ addr_size
,
440 // End of list is marked by a pair of zeroes.
441 if (shndx1
== 0 && start
== 0 && end
== 0)
444 // A "base address selection entry" is identified by
445 // 0xffffffff for the first value of the pair. The second
446 // value is used as a base for subsequent range list entries.
447 if (shndx1
== 0 && start
== -1)
449 else if (shndx1
== shndx2
)
451 if (shndx1
== 0 || object
->is_section_included(shndx1
))
452 ranges
->add(shndx1
, base
+ start
, base
+ end
);
455 gold_warning(_("%s: DWARF info may be corrupt; offsets in a "
456 "range list entry are in different sections"),
457 object
->name().c_str());
463 // class Dwarf_pubnames_table
465 // Read the pubnames section SHNDX from the object file.
468 Dwarf_pubnames_table::read_section(Relobj
* object
, unsigned int shndx
)
470 section_size_type buffer_size
;
472 // If we don't have relocations, shndx will be 0, and
473 // we'll have to hunt for the .debug_pubnames/pubtypes section.
476 const char* name
= (this->is_pubtypes_
478 : ".debug_pubnames");
479 for (unsigned int i
= 1; i
< object
->shnum(); ++i
)
481 if (object
->section_name(i
) == name
)
484 this->output_section_offset_
= object
->output_section_offset(i
);
492 this->buffer_
= object
->decompressed_section_contents(shndx
,
494 &this->owns_buffer_
);
495 if (this->buffer_
== NULL
)
497 this->buffer_end_
= this->buffer_
+ buffer_size
;
501 // Read the header for the set at OFFSET.
504 Dwarf_pubnames_table::read_header(off_t offset
)
506 // Correct the offset. For incremental update links, we have a
507 // relocated offset that is relative to the output section, but
508 // here we need an offset relative to the input section.
509 offset
-= this->output_section_offset_
;
511 if (offset
< 0 || offset
+ 14 >= this->buffer_end_
- this->buffer_
)
514 const unsigned char* pinfo
= this->buffer_
+ offset
;
516 // Read the unit_length field.
517 uint32_t unit_length
= this->dwinfo_
->read_from_pointer
<32>(pinfo
);
519 if (unit_length
== 0xffffffff)
521 unit_length
= this->dwinfo_
->read_from_pointer
<64>(pinfo
);
523 this->offset_size_
= 8;
526 this->offset_size_
= 4;
528 // Check the version.
529 unsigned int version
= this->dwinfo_
->read_from_pointer
<16>(pinfo
);
534 // Skip the debug_info_offset and debug_info_size fields.
535 pinfo
+= 2 * this->offset_size_
;
537 if (pinfo
>= this->buffer_end_
)
540 this->pinfo_
= pinfo
;
544 // Read the next name from the set.
547 Dwarf_pubnames_table::next_name()
549 const unsigned char* pinfo
= this->pinfo_
;
551 // Read the offset within the CU. If this is zero, we have reached
552 // the end of the list.
554 if (this->offset_size_
== 4)
555 offset
= this->dwinfo_
->read_from_pointer
<32>(&pinfo
);
557 offset
= this->dwinfo_
->read_from_pointer
<64>(&pinfo
);
561 // Return a pointer to the string at the current location,
562 // and advance the pointer to the next entry.
563 const char* ret
= reinterpret_cast<const char*>(pinfo
);
564 while (pinfo
< this->buffer_end_
&& *pinfo
!= '\0')
566 if (pinfo
< this->buffer_end_
)
569 this->pinfo_
= pinfo
;
575 Dwarf_die::Dwarf_die(
576 Dwarf_info_reader
* dwinfo
,
579 : dwinfo_(dwinfo
), parent_(parent
), die_offset_(die_offset
),
580 child_offset_(0), sibling_offset_(0), abbrev_code_(NULL
), attributes_(),
581 attributes_read_(false), name_(NULL
), name_off_(-1), linkage_name_(NULL
),
582 linkage_name_off_(-1), string_shndx_(0), specification_(0),
586 const unsigned char* pdie
= dwinfo
->buffer_at_offset(die_offset
);
589 unsigned int code
= read_unsigned_LEB_128(pdie
, &len
);
593 parent
->set_sibling_offset(die_offset
+ len
);
596 this->attr_offset_
= len
;
598 // Lookup the abbrev code in the abbrev table.
599 this->abbrev_code_
= dwinfo
->get_abbrev(code
);
602 // Read all the attributes of the DIE.
605 Dwarf_die::read_attributes()
607 if (this->attributes_read_
)
610 gold_assert(this->abbrev_code_
!= NULL
);
612 const unsigned char* pdie
=
613 this->dwinfo_
->buffer_at_offset(this->die_offset_
);
616 const unsigned char* pattr
= pdie
+ this->attr_offset_
;
618 unsigned int nattr
= this->abbrev_code_
->attributes
.size();
619 this->attributes_
.reserve(nattr
);
620 for (unsigned int i
= 0; i
< nattr
; ++i
)
623 unsigned int attr
= this->abbrev_code_
->attributes
[i
].attr
;
624 unsigned int form
= this->abbrev_code_
->attributes
[i
].form
;
625 if (form
== elfcpp::DW_FORM_indirect
)
627 form
= read_unsigned_LEB_128(pattr
, &len
);
630 off_t attr_off
= this->die_offset_
+ (pattr
- pdie
);
631 bool ref_form
= false;
632 Attribute_value attr_value
;
633 attr_value
.attr
= attr
;
634 attr_value
.form
= form
;
635 attr_value
.aux
.shndx
= 0;
638 case elfcpp::DW_FORM_flag_present
:
639 attr_value
.val
.intval
= 1;
641 case elfcpp::DW_FORM_strp
:
644 if (this->dwinfo_
->offset_size() == 4)
645 str_off
= this->dwinfo_
->read_from_pointer
<32>(&pattr
);
647 str_off
= this->dwinfo_
->read_from_pointer
<64>(&pattr
);
649 this->dwinfo_
->lookup_reloc(attr_off
, &str_off
);
650 attr_value
.aux
.shndx
= shndx
;
651 attr_value
.val
.refval
= str_off
;
654 case elfcpp::DW_FORM_sec_offset
:
657 if (this->dwinfo_
->offset_size() == 4)
658 sec_off
= this->dwinfo_
->read_from_pointer
<32>(&pattr
);
660 sec_off
= this->dwinfo_
->read_from_pointer
<64>(&pattr
);
662 this->dwinfo_
->lookup_reloc(attr_off
, &sec_off
);
663 attr_value
.aux
.shndx
= shndx
;
664 attr_value
.val
.refval
= sec_off
;
668 case elfcpp::DW_FORM_addr
:
669 case elfcpp::DW_FORM_ref_addr
:
672 if (this->dwinfo_
->address_size() == 4)
673 sec_off
= this->dwinfo_
->read_from_pointer
<32>(&pattr
);
675 sec_off
= this->dwinfo_
->read_from_pointer
<64>(&pattr
);
677 this->dwinfo_
->lookup_reloc(attr_off
, &sec_off
);
678 attr_value
.aux
.shndx
= shndx
;
679 attr_value
.val
.refval
= sec_off
;
683 case elfcpp::DW_FORM_block1
:
684 attr_value
.aux
.blocklen
= *pattr
++;
685 attr_value
.val
.blockval
= pattr
;
686 pattr
+= attr_value
.aux
.blocklen
;
688 case elfcpp::DW_FORM_block2
:
689 attr_value
.aux
.blocklen
=
690 this->dwinfo_
->read_from_pointer
<16>(&pattr
);
691 attr_value
.val
.blockval
= pattr
;
692 pattr
+= attr_value
.aux
.blocklen
;
694 case elfcpp::DW_FORM_block4
:
695 attr_value
.aux
.blocklen
=
696 this->dwinfo_
->read_from_pointer
<32>(&pattr
);
697 attr_value
.val
.blockval
= pattr
;
698 pattr
+= attr_value
.aux
.blocklen
;
700 case elfcpp::DW_FORM_block
:
701 case elfcpp::DW_FORM_exprloc
:
702 attr_value
.aux
.blocklen
= read_unsigned_LEB_128(pattr
, &len
);
703 attr_value
.val
.blockval
= pattr
+ len
;
704 pattr
+= len
+ attr_value
.aux
.blocklen
;
706 case elfcpp::DW_FORM_data1
:
707 case elfcpp::DW_FORM_flag
:
708 attr_value
.val
.intval
= *pattr
++;
710 case elfcpp::DW_FORM_ref1
:
711 attr_value
.val
.refval
= *pattr
++;
714 case elfcpp::DW_FORM_data2
:
715 attr_value
.val
.intval
=
716 this->dwinfo_
->read_from_pointer
<16>(&pattr
);
718 case elfcpp::DW_FORM_ref2
:
719 attr_value
.val
.refval
=
720 this->dwinfo_
->read_from_pointer
<16>(&pattr
);
723 case elfcpp::DW_FORM_data4
:
726 sec_off
= this->dwinfo_
->read_from_pointer
<32>(&pattr
);
728 this->dwinfo_
->lookup_reloc(attr_off
, &sec_off
);
729 attr_value
.aux
.shndx
= shndx
;
730 attr_value
.val
.intval
= sec_off
;
733 case elfcpp::DW_FORM_ref4
:
736 sec_off
= this->dwinfo_
->read_from_pointer
<32>(&pattr
);
738 this->dwinfo_
->lookup_reloc(attr_off
, &sec_off
);
739 attr_value
.aux
.shndx
= shndx
;
740 attr_value
.val
.refval
= sec_off
;
744 case elfcpp::DW_FORM_data8
:
747 sec_off
= this->dwinfo_
->read_from_pointer
<64>(&pattr
);
749 this->dwinfo_
->lookup_reloc(attr_off
, &sec_off
);
750 attr_value
.aux
.shndx
= shndx
;
751 attr_value
.val
.intval
= sec_off
;
754 case elfcpp::DW_FORM_ref_sig8
:
755 attr_value
.val
.uintval
=
756 this->dwinfo_
->read_from_pointer
<64>(&pattr
);
758 case elfcpp::DW_FORM_ref8
:
761 sec_off
= this->dwinfo_
->read_from_pointer
<64>(&pattr
);
763 this->dwinfo_
->lookup_reloc(attr_off
, &sec_off
);
764 attr_value
.aux
.shndx
= shndx
;
765 attr_value
.val
.refval
= sec_off
;
769 case elfcpp::DW_FORM_ref_udata
:
770 attr_value
.val
.refval
= read_unsigned_LEB_128(pattr
, &len
);
774 case elfcpp::DW_FORM_udata
:
775 case elfcpp::DW_FORM_GNU_addr_index
:
776 case elfcpp::DW_FORM_GNU_str_index
:
777 attr_value
.val
.uintval
= read_unsigned_LEB_128(pattr
, &len
);
780 case elfcpp::DW_FORM_sdata
:
781 attr_value
.val
.intval
= read_signed_LEB_128(pattr
, &len
);
784 case elfcpp::DW_FORM_string
:
785 attr_value
.val
.stringval
= reinterpret_cast<const char*>(pattr
);
786 len
= strlen(attr_value
.val
.stringval
);
793 // Cache the most frequently-requested attributes.
796 case elfcpp::DW_AT_name
:
797 if (form
== elfcpp::DW_FORM_string
)
798 this->name_
= attr_value
.val
.stringval
;
799 else if (form
== elfcpp::DW_FORM_strp
)
801 // All indirect strings should refer to the same
802 // string section, so we just save the last one seen.
803 this->string_shndx_
= attr_value
.aux
.shndx
;
804 this->name_off_
= attr_value
.val
.refval
;
807 case elfcpp::DW_AT_linkage_name
:
808 case elfcpp::DW_AT_MIPS_linkage_name
:
809 if (form
== elfcpp::DW_FORM_string
)
810 this->linkage_name_
= attr_value
.val
.stringval
;
811 else if (form
== elfcpp::DW_FORM_strp
)
813 // All indirect strings should refer to the same
814 // string section, so we just save the last one seen.
815 this->string_shndx_
= attr_value
.aux
.shndx
;
816 this->linkage_name_off_
= attr_value
.val
.refval
;
819 case elfcpp::DW_AT_specification
:
821 this->specification_
= attr_value
.val
.refval
;
823 case elfcpp::DW_AT_abstract_origin
:
825 this->abstract_origin_
= attr_value
.val
.refval
;
827 case elfcpp::DW_AT_sibling
:
828 if (ref_form
&& attr_value
.aux
.shndx
== 0)
829 this->sibling_offset_
= attr_value
.val
.refval
;
834 this->attributes_
.push_back(attr_value
);
837 // Now that we know where the next DIE begins, record the offset
838 // to avoid later recalculation.
839 if (this->has_children())
840 this->child_offset_
= this->die_offset_
+ (pattr
- pdie
);
842 this->sibling_offset_
= this->die_offset_
+ (pattr
- pdie
);
844 this->attributes_read_
= true;
848 // Skip all the attributes of the DIE and return the offset of the next DIE.
851 Dwarf_die::skip_attributes()
853 gold_assert(this->abbrev_code_
!= NULL
);
855 const unsigned char* pdie
=
856 this->dwinfo_
->buffer_at_offset(this->die_offset_
);
859 const unsigned char* pattr
= pdie
+ this->attr_offset_
;
861 for (unsigned int i
= 0; i
< this->abbrev_code_
->attributes
.size(); ++i
)
864 unsigned int form
= this->abbrev_code_
->attributes
[i
].form
;
865 if (form
== elfcpp::DW_FORM_indirect
)
867 form
= read_unsigned_LEB_128(pattr
, &len
);
872 case elfcpp::DW_FORM_flag_present
:
874 case elfcpp::DW_FORM_strp
:
875 case elfcpp::DW_FORM_sec_offset
:
876 pattr
+= this->dwinfo_
->offset_size();
878 case elfcpp::DW_FORM_addr
:
879 case elfcpp::DW_FORM_ref_addr
:
880 pattr
+= this->dwinfo_
->address_size();
882 case elfcpp::DW_FORM_block1
:
885 case elfcpp::DW_FORM_block2
:
888 block_size
= this->dwinfo_
->read_from_pointer
<16>(&pattr
);
892 case elfcpp::DW_FORM_block4
:
895 block_size
= this->dwinfo_
->read_from_pointer
<32>(&pattr
);
899 case elfcpp::DW_FORM_block
:
900 case elfcpp::DW_FORM_exprloc
:
903 block_size
= read_unsigned_LEB_128(pattr
, &len
);
904 pattr
+= len
+ block_size
;
907 case elfcpp::DW_FORM_data1
:
908 case elfcpp::DW_FORM_ref1
:
909 case elfcpp::DW_FORM_flag
:
912 case elfcpp::DW_FORM_data2
:
913 case elfcpp::DW_FORM_ref2
:
916 case elfcpp::DW_FORM_data4
:
917 case elfcpp::DW_FORM_ref4
:
920 case elfcpp::DW_FORM_data8
:
921 case elfcpp::DW_FORM_ref8
:
922 case elfcpp::DW_FORM_ref_sig8
:
925 case elfcpp::DW_FORM_ref_udata
:
926 case elfcpp::DW_FORM_udata
:
927 case elfcpp::DW_FORM_GNU_addr_index
:
928 case elfcpp::DW_FORM_GNU_str_index
:
929 read_unsigned_LEB_128(pattr
, &len
);
932 case elfcpp::DW_FORM_sdata
:
933 read_signed_LEB_128(pattr
, &len
);
936 case elfcpp::DW_FORM_string
:
937 len
= strlen(reinterpret_cast<const char*>(pattr
));
945 return this->die_offset_
+ (pattr
- pdie
);
948 // Get the name of the DIE and cache it.
951 Dwarf_die::set_name()
953 if (this->name_
!= NULL
|| !this->read_attributes())
955 if (this->name_off_
!= -1)
956 this->name_
= this->dwinfo_
->get_string(this->name_off_
,
957 this->string_shndx_
);
960 // Get the linkage name of the DIE and cache it.
963 Dwarf_die::set_linkage_name()
965 if (this->linkage_name_
!= NULL
|| !this->read_attributes())
967 if (this->linkage_name_off_
!= -1)
968 this->linkage_name_
= this->dwinfo_
->get_string(this->linkage_name_off_
,
969 this->string_shndx_
);
972 // Return the value of attribute ATTR.
974 const Dwarf_die::Attribute_value
*
975 Dwarf_die::attribute(unsigned int attr
)
977 if (!this->read_attributes())
979 for (unsigned int i
= 0; i
< this->attributes_
.size(); ++i
)
981 if (this->attributes_
[i
].attr
== attr
)
982 return &this->attributes_
[i
];
988 Dwarf_die::string_attribute(unsigned int attr
)
990 const Attribute_value
* attr_val
= this->attribute(attr
);
991 if (attr_val
== NULL
)
993 switch (attr_val
->form
)
995 case elfcpp::DW_FORM_string
:
996 return attr_val
->val
.stringval
;
997 case elfcpp::DW_FORM_strp
:
998 return this->dwinfo_
->get_string(attr_val
->val
.refval
,
999 attr_val
->aux
.shndx
);
1006 Dwarf_die::int_attribute(unsigned int attr
)
1008 const Attribute_value
* attr_val
= this->attribute(attr
);
1009 if (attr_val
== NULL
)
1011 switch (attr_val
->form
)
1013 case elfcpp::DW_FORM_flag_present
:
1014 case elfcpp::DW_FORM_data1
:
1015 case elfcpp::DW_FORM_flag
:
1016 case elfcpp::DW_FORM_data2
:
1017 case elfcpp::DW_FORM_data4
:
1018 case elfcpp::DW_FORM_data8
:
1019 case elfcpp::DW_FORM_sdata
:
1020 return attr_val
->val
.intval
;
1027 Dwarf_die::uint_attribute(unsigned int attr
)
1029 const Attribute_value
* attr_val
= this->attribute(attr
);
1030 if (attr_val
== NULL
)
1032 switch (attr_val
->form
)
1034 case elfcpp::DW_FORM_flag_present
:
1035 case elfcpp::DW_FORM_data1
:
1036 case elfcpp::DW_FORM_flag
:
1037 case elfcpp::DW_FORM_data4
:
1038 case elfcpp::DW_FORM_data8
:
1039 case elfcpp::DW_FORM_ref_sig8
:
1040 case elfcpp::DW_FORM_udata
:
1041 return attr_val
->val
.uintval
;
1048 Dwarf_die::ref_attribute(unsigned int attr
, unsigned int* shndx
)
1050 const Attribute_value
* attr_val
= this->attribute(attr
);
1051 if (attr_val
== NULL
)
1053 switch (attr_val
->form
)
1055 case elfcpp::DW_FORM_sec_offset
:
1056 case elfcpp::DW_FORM_addr
:
1057 case elfcpp::DW_FORM_ref_addr
:
1058 case elfcpp::DW_FORM_ref1
:
1059 case elfcpp::DW_FORM_ref2
:
1060 case elfcpp::DW_FORM_ref4
:
1061 case elfcpp::DW_FORM_ref8
:
1062 case elfcpp::DW_FORM_ref_udata
:
1063 *shndx
= attr_val
->aux
.shndx
;
1064 return attr_val
->val
.refval
;
1065 case elfcpp::DW_FORM_ref_sig8
:
1066 *shndx
= attr_val
->aux
.shndx
;
1067 return attr_val
->val
.uintval
;
1068 case elfcpp::DW_FORM_data4
:
1069 case elfcpp::DW_FORM_data8
:
1070 *shndx
= attr_val
->aux
.shndx
;
1071 return attr_val
->val
.intval
;
1078 Dwarf_die::address_attribute(unsigned int attr
, unsigned int* shndx
)
1080 const Attribute_value
* attr_val
= this->attribute(attr
);
1081 if (attr_val
== NULL
|| attr_val
->form
!= elfcpp::DW_FORM_addr
)
1084 *shndx
= attr_val
->aux
.shndx
;
1085 return attr_val
->val
.refval
;
1088 // Return the offset of this DIE's first child.
1091 Dwarf_die::child_offset()
1093 gold_assert(this->abbrev_code_
!= NULL
);
1094 if (!this->has_children())
1096 if (this->child_offset_
== 0)
1097 this->child_offset_
= this->skip_attributes();
1098 return this->child_offset_
;
1101 // Return the offset of this DIE's next sibling.
1104 Dwarf_die::sibling_offset()
1106 gold_assert(this->abbrev_code_
!= NULL
);
1108 if (this->sibling_offset_
!= 0)
1109 return this->sibling_offset_
;
1111 if (!this->has_children())
1113 this->sibling_offset_
= this->skip_attributes();
1114 return this->sibling_offset_
;
1117 if (this->has_sibling_attribute())
1119 if (!this->read_attributes())
1121 if (this->sibling_offset_
!= 0)
1122 return this->sibling_offset_
;
1125 // Skip over the children.
1126 off_t child_offset
= this->child_offset();
1127 while (child_offset
> 0)
1129 Dwarf_die
die(this->dwinfo_
, child_offset
, this);
1130 // The Dwarf_die ctor will set this DIE's sibling offset
1131 // when it reads a zero abbrev code.
1134 child_offset
= die
.sibling_offset();
1137 // This should be set by now. If not, there was a problem reading
1138 // the DWARF info, and we return 0.
1139 return this->sibling_offset_
;
1142 // class Dwarf_info_reader
1144 // Check that the pointer P is within the current compilation unit.
1147 Dwarf_info_reader::check_buffer(const unsigned char* p
) const
1149 if (p
> this->buffer_
+ this->cu_offset_
+ this->cu_length_
)
1151 gold_warning(_("%s: corrupt debug info in %s"),
1152 this->object_
->name().c_str(),
1153 this->object_
->section_name(this->shndx_
).c_str());
1159 // Begin parsing the debug info. This calls visit_compilation_unit()
1160 // or visit_type_unit() for each compilation or type unit found in the
1161 // section, and visit_die() for each top-level DIE.
1164 Dwarf_info_reader::parse()
1166 if (this->object_
->is_big_endian())
1168 #if defined(HAVE_TARGET_32_BIG) || defined(HAVE_TARGET_64_BIG)
1169 this->do_parse
<true>();
1176 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_64_LITTLE)
1177 this->do_parse
<false>();
1184 template<bool big_endian
>
1186 Dwarf_info_reader::do_parse()
1188 // Get the section contents and decompress if necessary.
1189 section_size_type buffer_size
;
1191 this->buffer_
= this->object_
->decompressed_section_contents(this->shndx_
,
1194 if (this->buffer_
== NULL
|| buffer_size
== 0)
1196 this->buffer_end_
= this->buffer_
+ buffer_size
;
1198 // The offset of this input section in the output section.
1199 off_t section_offset
= this->object_
->output_section_offset(this->shndx_
);
1201 // Start tracking relocations for this section.
1202 this->reloc_mapper_
= make_elf_reloc_mapper(this->object_
, this->symtab_
,
1203 this->symtab_size_
);
1204 this->reloc_mapper_
->initialize(this->reloc_shndx_
, this->reloc_type_
);
1206 // Loop over compilation units (or type units).
1207 unsigned int abbrev_shndx
= this->abbrev_shndx_
;
1208 off_t abbrev_offset
= 0;
1209 const unsigned char* pinfo
= this->buffer_
;
1210 while (pinfo
< this->buffer_end_
)
1212 // Read the compilation (or type) unit header.
1213 const unsigned char* cu_start
= pinfo
;
1214 this->cu_offset_
= cu_start
- this->buffer_
;
1215 this->cu_length_
= this->buffer_end_
- cu_start
;
1217 // Read unit_length (4 or 12 bytes).
1218 if (!this->check_buffer(pinfo
+ 4))
1220 uint32_t unit_length
=
1221 elfcpp::Swap_unaligned
<32, big_endian
>::readval(pinfo
);
1223 if (unit_length
== 0xffffffff)
1225 if (!this->check_buffer(pinfo
+ 8))
1227 unit_length
= elfcpp::Swap_unaligned
<64, big_endian
>::readval(pinfo
);
1229 this->offset_size_
= 8;
1232 this->offset_size_
= 4;
1233 if (!this->check_buffer(pinfo
+ unit_length
))
1235 const unsigned char* cu_end
= pinfo
+ unit_length
;
1236 this->cu_length_
= cu_end
- cu_start
;
1237 if (!this->check_buffer(pinfo
+ 2 + this->offset_size_
+ 1))
1240 // Read version (2 bytes).
1242 elfcpp::Swap_unaligned
<16, big_endian
>::readval(pinfo
);
1245 // Read debug_abbrev_offset (4 or 8 bytes).
1246 if (this->offset_size_
== 4)
1247 abbrev_offset
= elfcpp::Swap_unaligned
<32, big_endian
>::readval(pinfo
);
1249 abbrev_offset
= elfcpp::Swap_unaligned
<64, big_endian
>::readval(pinfo
);
1250 if (this->reloc_shndx_
> 0)
1252 off_t reloc_offset
= pinfo
- this->buffer_
;
1255 this->reloc_mapper_
->get_reloc_target(reloc_offset
, &value
);
1256 if (abbrev_shndx
== 0)
1258 if (this->reloc_type_
== elfcpp::SHT_REL
)
1259 abbrev_offset
+= value
;
1261 abbrev_offset
= value
;
1263 pinfo
+= this->offset_size_
;
1265 // Read address_size (1 byte).
1266 this->address_size_
= *pinfo
++;
1268 // For type units, read the two extra fields.
1269 uint64_t signature
= 0;
1270 off_t type_offset
= 0;
1271 if (this->is_type_unit_
)
1273 if (!this->check_buffer(pinfo
+ 8 + this->offset_size_
))
1276 // Read type_signature (8 bytes).
1277 signature
= elfcpp::Swap_unaligned
<64, big_endian
>::readval(pinfo
);
1280 // Read type_offset (4 or 8 bytes).
1281 if (this->offset_size_
== 4)
1283 elfcpp::Swap_unaligned
<32, big_endian
>::readval(pinfo
);
1286 elfcpp::Swap_unaligned
<64, big_endian
>::readval(pinfo
);
1287 pinfo
+= this->offset_size_
;
1290 // Read the .debug_abbrev table.
1291 this->abbrev_table_
.read_abbrevs(this->object_
, abbrev_shndx
,
1294 // Visit the root DIE.
1295 Dwarf_die
root_die(this,
1296 pinfo
- (this->buffer_
+ this->cu_offset_
),
1298 if (root_die
.tag() != 0)
1300 // Visit the CU or TU.
1301 if (this->is_type_unit_
)
1302 this->visit_type_unit(section_offset
+ this->cu_offset_
,
1303 type_offset
, signature
, &root_die
);
1305 this->visit_compilation_unit(section_offset
+ this->cu_offset_
,
1306 cu_end
- cu_start
, &root_die
);
1309 // Advance to the next CU.
1315 delete[] this->buffer_
;
1316 this->buffer_
= NULL
;
1320 // Read the DWARF string table.
1323 Dwarf_info_reader::do_read_string_table(unsigned int string_shndx
)
1325 Relobj
* object
= this->object_
;
1327 // If we don't have relocations, string_shndx will be 0, and
1328 // we'll have to hunt for the .debug_str section.
1329 if (string_shndx
== 0)
1331 for (unsigned int i
= 1; i
< this->object_
->shnum(); ++i
)
1333 std::string name
= object
->section_name(i
);
1334 if (name
== ".debug_str")
1337 this->string_output_section_offset_
=
1338 object
->output_section_offset(i
);
1342 if (string_shndx
== 0)
1346 if (this->owns_string_buffer_
&& this->string_buffer_
!= NULL
)
1348 delete[] this->string_buffer_
;
1349 this->owns_string_buffer_
= false;
1352 // Get the secton contents and decompress if necessary.
1353 section_size_type buffer_size
;
1354 const unsigned char* buffer
=
1355 object
->decompressed_section_contents(string_shndx
,
1357 &this->owns_string_buffer_
);
1358 this->string_buffer_
= reinterpret_cast<const char*>(buffer
);
1359 this->string_buffer_end_
= this->string_buffer_
+ buffer_size
;
1360 this->string_shndx_
= string_shndx
;
1364 // Read a possibly unaligned integer of SIZE.
1365 template <int valsize
>
1366 inline typename
elfcpp::Valtype_base
<valsize
>::Valtype
1367 Dwarf_info_reader::read_from_pointer(const unsigned char* source
)
1369 typename
elfcpp::Valtype_base
<valsize
>::Valtype return_value
;
1370 if (this->object_
->is_big_endian())
1371 return_value
= elfcpp::Swap_unaligned
<valsize
, true>::readval(source
);
1373 return_value
= elfcpp::Swap_unaligned
<valsize
, false>::readval(source
);
1374 return return_value
;
1377 // Read a possibly unaligned integer of SIZE. Update SOURCE after read.
1378 template <int valsize
>
1379 inline typename
elfcpp::Valtype_base
<valsize
>::Valtype
1380 Dwarf_info_reader::read_from_pointer(const unsigned char** source
)
1382 typename
elfcpp::Valtype_base
<valsize
>::Valtype return_value
;
1383 if (this->object_
->is_big_endian())
1384 return_value
= elfcpp::Swap_unaligned
<valsize
, true>::readval(*source
);
1386 return_value
= elfcpp::Swap_unaligned
<valsize
, false>::readval(*source
);
1387 *source
+= valsize
/ 8;
1388 return return_value
;
1391 // Look for a relocation at offset ATTR_OFF in the dwarf info,
1392 // and return the section index and offset of the target.
1395 Dwarf_info_reader::lookup_reloc(off_t attr_off
, off_t
* target_off
)
1398 attr_off
+= this->cu_offset_
;
1399 unsigned int shndx
= this->reloc_mapper_
->get_reloc_target(attr_off
, &value
);
1402 if (this->reloc_type_
== elfcpp::SHT_REL
)
1403 *target_off
+= value
;
1405 *target_off
= value
;
1409 // Return a string from the DWARF string table.
1412 Dwarf_info_reader::get_string(off_t str_off
, unsigned int string_shndx
)
1414 if (!this->read_string_table(string_shndx
))
1417 // Correct the offset. For incremental update links, we have a
1418 // relocated offset that is relative to the output section, but
1419 // here we need an offset relative to the input section.
1420 str_off
-= this->string_output_section_offset_
;
1422 const char* p
= this->string_buffer_
+ str_off
;
1424 if (p
< this->string_buffer_
|| p
>= this->string_buffer_end_
)
1430 // The following are default, do-nothing, implementations of the
1431 // hook methods normally provided by a derived class. We provide
1432 // default implementations rather than no implementation so that
1433 // a derived class needs to implement only the hooks that it needs
1436 // Process a compilation unit and parse its child DIE.
1439 Dwarf_info_reader::visit_compilation_unit(off_t
, off_t
, Dwarf_die
*)
1443 // Process a type unit and parse its child DIE.
1446 Dwarf_info_reader::visit_type_unit(off_t
, off_t
, uint64_t, Dwarf_die
*)
1450 // class Sized_dwarf_line_info
1452 struct LineStateMachine
1458 unsigned int shndx
; // the section address refers to
1459 bool is_stmt
; // stmt means statement.
1465 ResetLineStateMachine(struct LineStateMachine
* lsm
, bool default_is_stmt
)
1470 lsm
->column_num
= 0;
1472 lsm
->is_stmt
= default_is_stmt
;
1473 lsm
->basic_block
= false;
1474 lsm
->end_sequence
= false;
1477 template<int size
, bool big_endian
>
1478 Sized_dwarf_line_info
<size
, big_endian
>::Sized_dwarf_line_info(
1480 unsigned int read_shndx
)
1481 : data_valid_(false), buffer_(NULL
), buffer_start_(NULL
),
1482 reloc_mapper_(NULL
), symtab_buffer_(NULL
), directories_(), files_(),
1483 current_header_index_(-1)
1485 unsigned int debug_shndx
;
1487 for (debug_shndx
= 1; debug_shndx
< object
->shnum(); ++debug_shndx
)
1489 // FIXME: do this more efficiently: section_name() isn't super-fast
1490 std::string name
= object
->section_name(debug_shndx
);
1491 if (name
== ".debug_line" || name
== ".zdebug_line")
1493 section_size_type buffer_size
;
1494 bool is_new
= false;
1495 this->buffer_
= object
->decompressed_section_contents(debug_shndx
,
1499 this->buffer_start_
= this->buffer_
;
1500 this->buffer_end_
= this->buffer_
+ buffer_size
;
1504 if (this->buffer_
== NULL
)
1507 // Find the relocation section for ".debug_line".
1508 // We expect these for relobjs (.o's) but not dynobjs (.so's).
1509 unsigned int reloc_shndx
= 0;
1510 for (unsigned int i
= 0; i
< object
->shnum(); ++i
)
1512 unsigned int reloc_sh_type
= object
->section_type(i
);
1513 if ((reloc_sh_type
== elfcpp::SHT_REL
1514 || reloc_sh_type
== elfcpp::SHT_RELA
)
1515 && object
->section_info(i
) == debug_shndx
)
1518 this->track_relocs_type_
= reloc_sh_type
;
1523 // Finally, we need the symtab section to interpret the relocs.
1524 if (reloc_shndx
!= 0)
1526 unsigned int symtab_shndx
;
1527 for (symtab_shndx
= 0; symtab_shndx
< object
->shnum(); ++symtab_shndx
)
1528 if (object
->section_type(symtab_shndx
) == elfcpp::SHT_SYMTAB
)
1530 this->symtab_buffer_
= object
->section_contents(
1531 symtab_shndx
, &this->symtab_buffer_size_
, false);
1534 if (this->symtab_buffer_
== NULL
)
1538 this->reloc_mapper_
=
1539 new Sized_elf_reloc_mapper
<size
, big_endian
>(object
,
1540 this->symtab_buffer_
,
1541 this->symtab_buffer_size_
);
1542 if (!this->reloc_mapper_
->initialize(reloc_shndx
, this->track_relocs_type_
))
1545 // Now that we have successfully read all the data, parse the debug
1547 this->data_valid_
= true;
1548 this->read_line_mappings(read_shndx
);
1551 // Read the DWARF header.
1553 template<int size
, bool big_endian
>
1554 const unsigned char*
1555 Sized_dwarf_line_info
<size
, big_endian
>::read_header_prolog(
1556 const unsigned char* lineptr
)
1558 uint32_t initial_length
= elfcpp::Swap_unaligned
<32, big_endian
>::readval(lineptr
);
1561 // In DWARF2/3, if the initial length is all 1 bits, then the offset
1562 // size is 8 and we need to read the next 8 bytes for the real length.
1563 if (initial_length
== 0xffffffff)
1565 header_
.offset_size
= 8;
1566 initial_length
= elfcpp::Swap_unaligned
<64, big_endian
>::readval(lineptr
);
1570 header_
.offset_size
= 4;
1572 header_
.total_length
= initial_length
;
1574 gold_assert(lineptr
+ header_
.total_length
<= buffer_end_
);
1576 header_
.version
= elfcpp::Swap_unaligned
<16, big_endian
>::readval(lineptr
);
1579 if (header_
.offset_size
== 4)
1580 header_
.prologue_length
= elfcpp::Swap_unaligned
<32, big_endian
>::readval(lineptr
);
1582 header_
.prologue_length
= elfcpp::Swap_unaligned
<64, big_endian
>::readval(lineptr
);
1583 lineptr
+= header_
.offset_size
;
1585 header_
.min_insn_length
= *lineptr
;
1588 header_
.default_is_stmt
= *lineptr
;
1591 header_
.line_base
= *reinterpret_cast<const signed char*>(lineptr
);
1594 header_
.line_range
= *lineptr
;
1597 header_
.opcode_base
= *lineptr
;
1600 header_
.std_opcode_lengths
.resize(header_
.opcode_base
+ 1);
1601 header_
.std_opcode_lengths
[0] = 0;
1602 for (int i
= 1; i
< header_
.opcode_base
; i
++)
1604 header_
.std_opcode_lengths
[i
] = *lineptr
;
1611 // The header for a debug_line section is mildly complicated, because
1612 // the line info is very tightly encoded.
1614 template<int size
, bool big_endian
>
1615 const unsigned char*
1616 Sized_dwarf_line_info
<size
, big_endian
>::read_header_tables(
1617 const unsigned char* lineptr
)
1619 ++this->current_header_index_
;
1621 // Create a new directories_ entry and a new files_ entry for our new
1622 // header. We initialize each with a single empty element, because
1623 // dwarf indexes directory and filenames starting at 1.
1624 gold_assert(static_cast<int>(this->directories_
.size())
1625 == this->current_header_index_
);
1626 gold_assert(static_cast<int>(this->files_
.size())
1627 == this->current_header_index_
);
1628 this->directories_
.push_back(std::vector
<std::string
>(1));
1629 this->files_
.push_back(std::vector
<std::pair
<int, std::string
> >(1));
1631 // It is legal for the directory entry table to be empty.
1637 const char* dirname
= reinterpret_cast<const char*>(lineptr
);
1638 gold_assert(dirindex
1639 == static_cast<int>(this->directories_
.back().size()));
1640 this->directories_
.back().push_back(dirname
);
1641 lineptr
+= this->directories_
.back().back().size() + 1;
1647 // It is also legal for the file entry table to be empty.
1654 const char* filename
= reinterpret_cast<const char*>(lineptr
);
1655 lineptr
+= strlen(filename
) + 1;
1657 uint64_t dirindex
= read_unsigned_LEB_128(lineptr
, &len
);
1660 if (dirindex
>= this->directories_
.back().size())
1662 int dirindexi
= static_cast<int>(dirindex
);
1664 read_unsigned_LEB_128(lineptr
, &len
); // mod_time
1667 read_unsigned_LEB_128(lineptr
, &len
); // filelength
1670 gold_assert(fileindex
1671 == static_cast<int>(this->files_
.back().size()));
1672 this->files_
.back().push_back(std::make_pair(dirindexi
, filename
));
1681 // Process a single opcode in the .debug.line structure.
1683 template<int size
, bool big_endian
>
1685 Sized_dwarf_line_info
<size
, big_endian
>::process_one_opcode(
1686 const unsigned char* start
, struct LineStateMachine
* lsm
, size_t* len
)
1690 unsigned char opcode
= *start
;
1694 // If the opcode is great than the opcode_base, it is a special
1695 // opcode. Most line programs consist mainly of special opcodes.
1696 if (opcode
>= header_
.opcode_base
)
1698 opcode
-= header_
.opcode_base
;
1699 const int advance_address
= ((opcode
/ header_
.line_range
)
1700 * header_
.min_insn_length
);
1701 lsm
->address
+= advance_address
;
1703 const int advance_line
= ((opcode
% header_
.line_range
)
1704 + header_
.line_base
);
1705 lsm
->line_num
+= advance_line
;
1706 lsm
->basic_block
= true;
1711 // Otherwise, we have the regular opcodes
1714 case elfcpp::DW_LNS_copy
:
1715 lsm
->basic_block
= false;
1719 case elfcpp::DW_LNS_advance_pc
:
1721 const uint64_t advance_address
1722 = read_unsigned_LEB_128(start
, &templen
);
1724 lsm
->address
+= header_
.min_insn_length
* advance_address
;
1728 case elfcpp::DW_LNS_advance_line
:
1730 const uint64_t advance_line
= read_signed_LEB_128(start
, &templen
);
1732 lsm
->line_num
+= advance_line
;
1736 case elfcpp::DW_LNS_set_file
:
1738 const uint64_t fileno
= read_unsigned_LEB_128(start
, &templen
);
1740 lsm
->file_num
= fileno
;
1744 case elfcpp::DW_LNS_set_column
:
1746 const uint64_t colno
= read_unsigned_LEB_128(start
, &templen
);
1748 lsm
->column_num
= colno
;
1752 case elfcpp::DW_LNS_negate_stmt
:
1753 lsm
->is_stmt
= !lsm
->is_stmt
;
1756 case elfcpp::DW_LNS_set_basic_block
:
1757 lsm
->basic_block
= true;
1760 case elfcpp::DW_LNS_fixed_advance_pc
:
1762 int advance_address
;
1763 advance_address
= elfcpp::Swap_unaligned
<16, big_endian
>::readval(start
);
1765 lsm
->address
+= advance_address
;
1769 case elfcpp::DW_LNS_const_add_pc
:
1771 const int advance_address
= (header_
.min_insn_length
1772 * ((255 - header_
.opcode_base
)
1773 / header_
.line_range
));
1774 lsm
->address
+= advance_address
;
1778 case elfcpp::DW_LNS_extended_op
:
1780 const uint64_t extended_op_len
1781 = read_unsigned_LEB_128(start
, &templen
);
1783 oplen
+= templen
+ extended_op_len
;
1785 const unsigned char extended_op
= *start
;
1788 switch (extended_op
)
1790 case elfcpp::DW_LNE_end_sequence
:
1791 // This means that the current byte is the one immediately
1792 // after a set of instructions. Record the current line
1793 // for up to one less than the current address.
1795 lsm
->end_sequence
= true;
1799 case elfcpp::DW_LNE_set_address
:
1802 elfcpp::Swap_unaligned
<size
, big_endian
>::readval(start
);
1803 typename
Reloc_map::const_iterator it
1804 = this->reloc_map_
.find(start
- this->buffer_
);
1805 if (it
!= reloc_map_
.end())
1807 // If this is a SHT_RELA section, then ignore the
1808 // section contents. This assumes that this is a
1809 // straight reloc which just uses the reloc addend.
1810 // The reloc addend has already been included in the
1812 if (this->track_relocs_type_
== elfcpp::SHT_RELA
)
1814 // Add in the symbol value.
1815 lsm
->address
+= it
->second
.second
;
1816 lsm
->shndx
= it
->second
.first
;
1820 // If we're a normal .o file, with relocs, every
1821 // set_address should have an associated relocation.
1822 if (this->input_is_relobj())
1823 this->data_valid_
= false;
1827 case elfcpp::DW_LNE_define_file
:
1829 const char* filename
= reinterpret_cast<const char*>(start
);
1830 templen
= strlen(filename
) + 1;
1833 uint64_t dirindex
= read_unsigned_LEB_128(start
, &templen
);
1835 if (dirindex
>= this->directories_
.back().size())
1837 int dirindexi
= static_cast<int>(dirindex
);
1839 // This opcode takes two additional ULEB128 parameters
1840 // (mod_time and filelength), but we don't use those
1841 // values. Because OPLEN already tells us how far to
1842 // skip to the next opcode, we don't need to read
1845 this->files_
.back().push_back(std::make_pair(dirindexi
,
1855 // Ignore unknown opcode silently
1856 for (int i
= 0; i
< header_
.std_opcode_lengths
[opcode
]; i
++)
1859 read_unsigned_LEB_128(start
, &templen
);
1870 // Read the debug information at LINEPTR and store it in the line
1873 template<int size
, bool big_endian
>
1874 unsigned const char*
1875 Sized_dwarf_line_info
<size
, big_endian
>::read_lines(unsigned const char* lineptr
,
1878 struct LineStateMachine lsm
;
1880 // LENGTHSTART is the place the length field is based on. It is the
1881 // point in the header after the initial length field.
1882 const unsigned char* lengthstart
= buffer_
;
1884 // In 64 bit dwarf, the initial length is 12 bytes, because of the
1885 // 0xffffffff at the start.
1886 if (header_
.offset_size
== 8)
1891 while (lineptr
< lengthstart
+ header_
.total_length
)
1893 ResetLineStateMachine(&lsm
, header_
.default_is_stmt
);
1894 while (!lsm
.end_sequence
)
1897 bool add_line
= this->process_one_opcode(lineptr
, &lsm
, &oplength
);
1899 && (shndx
== -1U || lsm
.shndx
== -1U || shndx
== lsm
.shndx
))
1901 Offset_to_lineno_entry entry
1902 = { static_cast<off_t
>(lsm
.address
),
1903 this->current_header_index_
,
1904 static_cast<unsigned int>(lsm
.file_num
),
1905 true, lsm
.line_num
};
1906 std::vector
<Offset_to_lineno_entry
>&
1907 map(this->line_number_map_
[lsm
.shndx
]);
1908 // If we see two consecutive entries with the same
1909 // offset and a real line number, then mark the first
1910 // one as non-canonical.
1912 && (map
.back().offset
== static_cast<off_t
>(lsm
.address
))
1913 && lsm
.line_num
!= -1
1914 && map
.back().line_num
!= -1)
1915 map
.back().last_line_for_offset
= false;
1916 map
.push_back(entry
);
1918 lineptr
+= oplength
;
1922 return lengthstart
+ header_
.total_length
;
1925 // Read the relocations into a Reloc_map.
1927 template<int size
, bool big_endian
>
1929 Sized_dwarf_line_info
<size
, big_endian
>::read_relocs()
1931 if (this->symtab_buffer_
== NULL
)
1936 while ((reloc_offset
= this->reloc_mapper_
->next_offset()) != -1)
1938 const unsigned int shndx
=
1939 this->reloc_mapper_
->get_reloc_target(reloc_offset
, &value
);
1941 // There is no reason to record non-ordinary section indexes, or
1942 // SHN_UNDEF, because they will never match the real section.
1944 this->reloc_map_
[reloc_offset
] = std::make_pair(shndx
, value
);
1946 this->reloc_mapper_
->advance(reloc_offset
+ 1);
1950 // Read the line number info.
1952 template<int size
, bool big_endian
>
1954 Sized_dwarf_line_info
<size
, big_endian
>::read_line_mappings(unsigned int shndx
)
1956 gold_assert(this->data_valid_
== true);
1958 this->read_relocs();
1959 while (this->buffer_
< this->buffer_end_
)
1961 const unsigned char* lineptr
= this->buffer_
;
1962 lineptr
= this->read_header_prolog(lineptr
);
1963 lineptr
= this->read_header_tables(lineptr
);
1964 lineptr
= this->read_lines(lineptr
, shndx
);
1965 this->buffer_
= lineptr
;
1968 // Sort the lines numbers, so addr2line can use binary search.
1969 for (typename
Lineno_map::iterator it
= line_number_map_
.begin();
1970 it
!= line_number_map_
.end();
1972 // Each vector needs to be sorted by offset.
1973 std::sort(it
->second
.begin(), it
->second
.end());
1976 // Some processing depends on whether the input is a .o file or not.
1977 // For instance, .o files have relocs, and have .debug_lines
1978 // information on a per section basis. .so files, on the other hand,
1979 // lack relocs, and offsets are unique, so we can ignore the section
1982 template<int size
, bool big_endian
>
1984 Sized_dwarf_line_info
<size
, big_endian
>::input_is_relobj()
1986 // Only .o files have relocs and the symtab buffer that goes with them.
1987 return this->symtab_buffer_
!= NULL
;
1990 // Given an Offset_to_lineno_entry vector, and an offset, figure out
1991 // if the offset points into a function according to the vector (see
1992 // comments below for the algorithm). If it does, return an iterator
1993 // into the vector that points to the line-number that contains that
1994 // offset. If not, it returns vector::end().
1996 static std::vector
<Offset_to_lineno_entry
>::const_iterator
1997 offset_to_iterator(const std::vector
<Offset_to_lineno_entry
>* offsets
,
2000 const Offset_to_lineno_entry lookup_key
= { offset
, 0, 0, true, 0 };
2002 // lower_bound() returns the smallest offset which is >= lookup_key.
2003 // If no offset in offsets is >= lookup_key, returns end().
2004 std::vector
<Offset_to_lineno_entry
>::const_iterator it
2005 = std::lower_bound(offsets
->begin(), offsets
->end(), lookup_key
);
2007 // This code is easiest to understand with a concrete example.
2008 // Here's a possible offsets array:
2009 // {{offset = 3211, header_num = 0, file_num = 1, last, line_num = 16}, // 0
2010 // {offset = 3224, header_num = 0, file_num = 1, last, line_num = 20}, // 1
2011 // {offset = 3226, header_num = 0, file_num = 1, last, line_num = 22}, // 2
2012 // {offset = 3231, header_num = 0, file_num = 1, last, line_num = 25}, // 3
2013 // {offset = 3232, header_num = 0, file_num = 1, last, line_num = -1}, // 4
2014 // {offset = 3232, header_num = 0, file_num = 1, last, line_num = 65}, // 5
2015 // {offset = 3235, header_num = 0, file_num = 1, last, line_num = 66}, // 6
2016 // {offset = 3236, header_num = 0, file_num = 1, last, line_num = -1}, // 7
2017 // {offset = 5764, header_num = 0, file_num = 1, last, line_num = 48}, // 8
2018 // {offset = 5764, header_num = 0, file_num = 1,!last, line_num = 47}, // 9
2019 // {offset = 5765, header_num = 0, file_num = 1, last, line_num = 49}, // 10
2020 // {offset = 5767, header_num = 0, file_num = 1, last, line_num = 50}, // 11
2021 // {offset = 5768, header_num = 0, file_num = 1, last, line_num = 51}, // 12
2022 // {offset = 5773, header_num = 0, file_num = 1, last, line_num = -1}, // 13
2023 // {offset = 5787, header_num = 1, file_num = 1, last, line_num = 19}, // 14
2024 // {offset = 5790, header_num = 1, file_num = 1, last, line_num = 20}, // 15
2025 // {offset = 5793, header_num = 1, file_num = 1, last, line_num = 67}, // 16
2026 // {offset = 5793, header_num = 1, file_num = 1, last, line_num = -1}, // 17
2027 // {offset = 5793, header_num = 1, file_num = 1,!last, line_num = 66}, // 18
2028 // {offset = 5795, header_num = 1, file_num = 1, last, line_num = 68}, // 19
2029 // {offset = 5798, header_num = 1, file_num = 1, last, line_num = -1}, // 20
2030 // The entries with line_num == -1 mark the end of a function: the
2031 // associated offset is one past the last instruction in the
2032 // function. This can correspond to the beginning of the next
2033 // function (as is true for offset 3232); alternately, there can be
2034 // a gap between the end of one function and the start of the next
2035 // (as is true for some others, most obviously from 3236->5764).
2037 // Case 1: lookup_key has offset == 10. lower_bound returns
2038 // offsets[0]. Since it's not an exact match and we're
2039 // at the beginning of offsets, we return end() (invalid).
2040 // Case 2: lookup_key has offset 10000. lower_bound returns
2041 // offset[21] (end()). We return end() (invalid).
2042 // Case 3: lookup_key has offset == 3211. lower_bound matches
2043 // offsets[0] exactly, and that's the entry we return.
2044 // Case 4: lookup_key has offset == 3232. lower_bound returns
2045 // offsets[4]. That's an exact match, but indicates
2046 // end-of-function. We check if offsets[5] is also an
2047 // exact match but not end-of-function. It is, so we
2048 // return offsets[5].
2049 // Case 5: lookup_key has offset == 3214. lower_bound returns
2050 // offsets[1]. Since it's not an exact match, we back
2051 // up to the offset that's < lookup_key, offsets[0].
2052 // We note offsets[0] is a valid entry (not end-of-function),
2053 // so that's the entry we return.
2054 // Case 6: lookup_key has offset == 4000. lower_bound returns
2055 // offsets[8]. Since it's not an exact match, we back
2056 // up to offsets[7]. Since offsets[7] indicates
2057 // end-of-function, we know lookup_key is between
2058 // functions, so we return end() (not a valid offset).
2059 // Case 7: lookup_key has offset == 5794. lower_bound returns
2060 // offsets[19]. Since it's not an exact match, we back
2061 // up to offsets[16]. Note we back up to the *first*
2062 // entry with offset 5793, not just offsets[19-1].
2063 // We note offsets[16] is a valid entry, so we return it.
2064 // If offsets[16] had had line_num == -1, we would have
2065 // checked offsets[17]. The reason for this is that
2066 // 16 and 17 can be in an arbitrary order, since we sort
2067 // only by offset and last_line_for_offset. (Note it
2068 // doesn't help to use line_number as a tertiary sort key,
2069 // since sometimes we want the -1 to be first and sometimes
2070 // we want it to be last.)
2072 // This deals with cases (1) and (2).
2073 if ((it
== offsets
->begin() && offset
< it
->offset
)
2074 || it
== offsets
->end())
2075 return offsets
->end();
2077 // This deals with cases (3) and (4).
2078 if (offset
== it
->offset
)
2080 while (it
!= offsets
->end()
2081 && it
->offset
== offset
2082 && it
->line_num
== -1)
2084 if (it
== offsets
->end() || it
->offset
!= offset
)
2085 return offsets
->end();
2090 // This handles the first part of case (7) -- we back up to the
2091 // *first* entry that has the offset that's behind us.
2092 gold_assert(it
!= offsets
->begin());
2093 std::vector
<Offset_to_lineno_entry
>::const_iterator range_end
= it
;
2095 const off_t range_value
= it
->offset
;
2096 while (it
!= offsets
->begin() && (it
-1)->offset
== range_value
)
2099 // This handles cases (5), (6), and (7): if any entry in the
2100 // equal_range [it, range_end) has a line_num != -1, it's a valid
2101 // match. If not, we're not in a function. The line number we saw
2102 // last for an offset will be sorted first, so it'll get returned if
2104 for (; it
!= range_end
; ++it
)
2105 if (it
->line_num
!= -1)
2107 return offsets
->end();
2110 // Returns the canonical filename:lineno for the address passed in.
2111 // If other_lines is not NULL, appends the non-canonical lines
2112 // assigned to the same address.
2114 template<int size
, bool big_endian
>
2116 Sized_dwarf_line_info
<size
, big_endian
>::do_addr2line(
2119 std::vector
<std::string
>* other_lines
)
2121 if (this->data_valid_
== false)
2124 const std::vector
<Offset_to_lineno_entry
>* offsets
;
2125 // If we do not have reloc information, then our input is a .so or
2126 // some similar data structure where all the information is held in
2127 // the offset. In that case, we ignore the input shndx.
2128 if (this->input_is_relobj())
2129 offsets
= &this->line_number_map_
[shndx
];
2131 offsets
= &this->line_number_map_
[-1U];
2132 if (offsets
->empty())
2135 typename
std::vector
<Offset_to_lineno_entry
>::const_iterator it
2136 = offset_to_iterator(offsets
, offset
);
2137 if (it
== offsets
->end())
2140 std::string result
= this->format_file_lineno(*it
);
2141 if (other_lines
!= NULL
)
2142 for (++it
; it
!= offsets
->end() && it
->offset
== offset
; ++it
)
2144 if (it
->line_num
== -1)
2145 continue; // The end of a previous function.
2146 other_lines
->push_back(this->format_file_lineno(*it
));
2151 // Convert the file_num + line_num into a string.
2153 template<int size
, bool big_endian
>
2155 Sized_dwarf_line_info
<size
, big_endian
>::format_file_lineno(
2156 const Offset_to_lineno_entry
& loc
) const
2160 gold_assert(loc
.header_num
< static_cast<int>(this->files_
.size()));
2161 gold_assert(loc
.file_num
2162 < static_cast<unsigned int>(this->files_
[loc
.header_num
].size()));
2163 const std::pair
<int, std::string
>& filename_pair
2164 = this->files_
[loc
.header_num
][loc
.file_num
];
2165 const std::string
& filename
= filename_pair
.second
;
2167 gold_assert(loc
.header_num
< static_cast<int>(this->directories_
.size()));
2168 gold_assert(filename_pair
.first
2169 < static_cast<int>(this->directories_
[loc
.header_num
].size()));
2170 const std::string
& dirname
2171 = this->directories_
[loc
.header_num
][filename_pair
.first
];
2173 if (!dirname
.empty())
2182 char buffer
[64]; // enough to hold a line number
2183 snprintf(buffer
, sizeof(buffer
), "%d", loc
.line_num
);
2190 // Dwarf_line_info routines.
2192 static unsigned int next_generation_count
= 0;
2194 struct Addr2line_cache_entry
2198 Dwarf_line_info
* dwarf_line_info
;
2199 unsigned int generation_count
;
2200 unsigned int access_count
;
2202 Addr2line_cache_entry(Object
* o
, unsigned int s
, Dwarf_line_info
* d
)
2203 : object(o
), shndx(s
), dwarf_line_info(d
),
2204 generation_count(next_generation_count
), access_count(0)
2206 if (next_generation_count
< (1U << 31))
2207 ++next_generation_count
;
2210 // We expect this cache to be small, so don't bother with a hashtable
2211 // or priority queue or anything: just use a simple vector.
2212 static std::vector
<Addr2line_cache_entry
> addr2line_cache
;
2215 Dwarf_line_info::one_addr2line(Object
* object
,
2216 unsigned int shndx
, off_t offset
,
2218 std::vector
<std::string
>* other_lines
)
2220 Dwarf_line_info
* lineinfo
= NULL
;
2221 std::vector
<Addr2line_cache_entry
>::iterator it
;
2223 // First, check the cache. If we hit, update the counts.
2224 for (it
= addr2line_cache
.begin(); it
!= addr2line_cache
.end(); ++it
)
2226 if (it
->object
== object
&& it
->shndx
== shndx
)
2228 lineinfo
= it
->dwarf_line_info
;
2229 it
->generation_count
= next_generation_count
;
2230 // We cap generation_count at 2^31 -1 to avoid overflow.
2231 if (next_generation_count
< (1U << 31))
2232 ++next_generation_count
;
2233 // We cap access_count at 31 so 2^access_count doesn't overflow
2234 if (it
->access_count
< 31)
2240 // If we don't hit the cache, create a new object and insert into the
2242 if (lineinfo
== NULL
)
2244 switch (parameters
->size_and_endianness())
2246 #ifdef HAVE_TARGET_32_LITTLE
2247 case Parameters::TARGET_32_LITTLE
:
2248 lineinfo
= new Sized_dwarf_line_info
<32, false>(object
, shndx
); break;
2250 #ifdef HAVE_TARGET_32_BIG
2251 case Parameters::TARGET_32_BIG
:
2252 lineinfo
= new Sized_dwarf_line_info
<32, true>(object
, shndx
); break;
2254 #ifdef HAVE_TARGET_64_LITTLE
2255 case Parameters::TARGET_64_LITTLE
:
2256 lineinfo
= new Sized_dwarf_line_info
<64, false>(object
, shndx
); break;
2258 #ifdef HAVE_TARGET_64_BIG
2259 case Parameters::TARGET_64_BIG
:
2260 lineinfo
= new Sized_dwarf_line_info
<64, true>(object
, shndx
); break;
2265 addr2line_cache
.push_back(Addr2line_cache_entry(object
, shndx
, lineinfo
));
2268 // Now that we have our object, figure out the answer
2269 std::string retval
= lineinfo
->addr2line(shndx
, offset
, other_lines
);
2271 // Finally, if our cache has grown too big, delete old objects. We
2272 // assume the common (probably only) case is deleting only one object.
2273 // We use a pretty simple scheme to evict: function of LRU and MFU.
2274 while (addr2line_cache
.size() > cache_size
)
2276 unsigned int lowest_score
= ~0U;
2277 std::vector
<Addr2line_cache_entry
>::iterator lowest
2278 = addr2line_cache
.end();
2279 for (it
= addr2line_cache
.begin(); it
!= addr2line_cache
.end(); ++it
)
2281 const unsigned int score
= (it
->generation_count
2282 + (1U << it
->access_count
));
2283 if (score
< lowest_score
)
2285 lowest_score
= score
;
2289 if (lowest
!= addr2line_cache
.end())
2291 delete lowest
->dwarf_line_info
;
2292 addr2line_cache
.erase(lowest
);
2300 Dwarf_line_info::clear_addr2line_cache()
2302 for (std::vector
<Addr2line_cache_entry
>::iterator it
= addr2line_cache
.begin();
2303 it
!= addr2line_cache
.end();
2305 delete it
->dwarf_line_info
;
2306 addr2line_cache
.clear();
2309 #ifdef HAVE_TARGET_32_LITTLE
2311 class Sized_dwarf_line_info
<32, false>;
2314 #ifdef HAVE_TARGET_32_BIG
2316 class Sized_dwarf_line_info
<32, true>;
2319 #ifdef HAVE_TARGET_64_LITTLE
2321 class Sized_dwarf_line_info
<64, false>;
2324 #ifdef HAVE_TARGET_64_BIG
2326 class Sized_dwarf_line_info
<64, true>;
2329 } // End namespace gold.