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"
31 #include "parameters.h"
33 #include "dwarf_reader.h"
34 #include "int_encoding.h"
35 #include "compressed_output.h"
39 // Class Sized_elf_reloc_mapper
41 // Initialize the relocation tracker for section RELOC_SHNDX.
43 template<int size
, bool big_endian
>
45 Sized_elf_reloc_mapper
<size
, big_endian
>::do_initialize(
46 unsigned int reloc_shndx
, unsigned int reloc_type
)
48 this->reloc_type_
= reloc_type
;
49 return this->track_relocs_
.initialize(this->object_
, reloc_shndx
,
53 // Looks in the symtab to see what section a symbol is in.
55 template<int size
, bool big_endian
>
57 Sized_elf_reloc_mapper
<size
, big_endian
>::symbol_section(
58 unsigned int symndx
, Address
* value
, bool* is_ordinary
)
60 const int symsize
= elfcpp::Elf_sizes
<size
>::sym_size
;
61 gold_assert((symndx
+ 1) * symsize
<= this->symtab_size_
);
62 elfcpp::Sym
<size
, big_endian
> elfsym(this->symtab_
+ symndx
* symsize
);
63 *value
= elfsym
.get_st_value();
64 return this->object_
->adjust_sym_shndx(symndx
, elfsym
.get_st_shndx(),
68 // Return the section index and offset within the section of
69 // the target of the relocation for RELOC_OFFSET.
71 template<int size
, bool big_endian
>
73 Sized_elf_reloc_mapper
<size
, big_endian
>::do_get_reloc_target(
74 off_t reloc_offset
, off_t
* target_offset
)
76 this->track_relocs_
.advance(reloc_offset
);
77 if (reloc_offset
!= this->track_relocs_
.next_offset())
79 unsigned int symndx
= this->track_relocs_
.next_symndx();
80 typename
elfcpp::Elf_types
<size
>::Elf_Addr value
;
82 unsigned int target_shndx
= this->symbol_section(symndx
, &value
,
86 if (this->reloc_type_
== elfcpp::SHT_RELA
)
87 value
+= this->track_relocs_
.next_addend();
88 *target_offset
= value
;
92 static inline Elf_reloc_mapper
*
93 make_elf_reloc_mapper(Object
* object
, const unsigned char* symtab
,
96 switch (parameters
->size_and_endianness())
98 #ifdef HAVE_TARGET_32_LITTLE
99 case Parameters::TARGET_32_LITTLE
:
100 return new Sized_elf_reloc_mapper
<32, false>(object
, symtab
,
103 #ifdef HAVE_TARGET_32_BIG
104 case Parameters::TARGET_32_BIG
:
105 return new Sized_elf_reloc_mapper
<32, true>(object
, symtab
,
108 #ifdef HAVE_TARGET_64_LITTLE
109 case Parameters::TARGET_64_LITTLE
:
110 return new Sized_elf_reloc_mapper
<64, false>(object
, symtab
,
113 #ifdef HAVE_TARGET_64_BIG
114 case Parameters::TARGET_64_BIG
:
115 return new Sized_elf_reloc_mapper
<64, true>(object
, symtab
,
123 // class Dwarf_abbrev_table
126 Dwarf_abbrev_table::clear_abbrev_codes()
128 for (unsigned int code
= 0; code
< this->low_abbrev_code_max_
; ++code
)
130 if (this->low_abbrev_codes_
[code
] != NULL
)
132 delete this->low_abbrev_codes_
[code
];
133 this->low_abbrev_codes_
[code
] = NULL
;
136 for (Abbrev_code_table::iterator it
= this->high_abbrev_codes_
.begin();
137 it
!= this->high_abbrev_codes_
.end();
140 if (it
->second
!= NULL
)
143 this->high_abbrev_codes_
.clear();
146 // Read the abbrev table from an object file.
149 Dwarf_abbrev_table::do_read_abbrevs(
151 unsigned int abbrev_shndx
,
154 this->clear_abbrev_codes();
156 // If we don't have relocations, abbrev_shndx will be 0, and
157 // we'll have to hunt for the .debug_abbrev section.
158 if (abbrev_shndx
== 0 && this->abbrev_shndx_
> 0)
159 abbrev_shndx
= this->abbrev_shndx_
;
160 else if (abbrev_shndx
== 0)
162 for (unsigned int i
= 1; i
< object
->shnum(); ++i
)
164 std::string name
= object
->section_name(i
);
165 if (name
== ".debug_abbrev")
168 // Correct the offset. For incremental update links, we have a
169 // relocated offset that is relative to the output section, but
170 // here we need an offset relative to the input section.
171 abbrev_offset
-= object
->output_section_offset(i
);
175 if (abbrev_shndx
== 0)
179 // Get the section contents and decompress if necessary.
180 if (abbrev_shndx
!= this->abbrev_shndx_
)
182 if (this->owns_buffer_
&& this->buffer_
!= NULL
)
184 delete[] this->buffer_
;
185 this->owns_buffer_
= false;
188 section_size_type buffer_size
;
190 object
->decompressed_section_contents(abbrev_shndx
,
192 &this->owns_buffer_
);
193 this->buffer_end_
= this->buffer_
+ buffer_size
;
194 this->abbrev_shndx_
= abbrev_shndx
;
197 this->buffer_pos_
= this->buffer_
+ abbrev_offset
;
201 // Lookup the abbrev code entry for CODE. This function is called
202 // only when the abbrev code is not in the direct lookup table.
203 // It may be in the hash table, it may not have been read yet,
204 // or it may not exist in the abbrev table.
206 const Dwarf_abbrev_table::Abbrev_code
*
207 Dwarf_abbrev_table::do_get_abbrev(unsigned int code
)
209 // See if the abbrev code is already in the hash table.
210 Abbrev_code_table::const_iterator it
= this->high_abbrev_codes_
.find(code
);
211 if (it
!= this->high_abbrev_codes_
.end())
214 // Read and store abbrev code definitions until we find the
215 // one we're looking for.
218 // Read the abbrev code. A zero here indicates the end of the
221 if (this->buffer_pos_
>= this->buffer_end_
)
223 uint64_t nextcode
= read_unsigned_LEB_128(this->buffer_pos_
, &len
);
226 this->buffer_pos_
= this->buffer_end_
;
229 this->buffer_pos_
+= len
;
232 if (this->buffer_pos_
>= this->buffer_end_
)
234 uint64_t tag
= read_unsigned_LEB_128(this->buffer_pos_
, &len
);
235 this->buffer_pos_
+= len
;
237 // Read the has_children flag.
238 if (this->buffer_pos_
>= this->buffer_end_
)
240 bool has_children
= *this->buffer_pos_
== elfcpp::DW_CHILDREN_yes
;
241 this->buffer_pos_
+= 1;
243 // Read the list of (attribute, form) pairs.
244 Abbrev_code
* entry
= new Abbrev_code(tag
, has_children
);
247 // Read the attribute.
248 if (this->buffer_pos_
>= this->buffer_end_
)
250 uint64_t attr
= read_unsigned_LEB_128(this->buffer_pos_
, &len
);
251 this->buffer_pos_
+= len
;
254 if (this->buffer_pos_
>= this->buffer_end_
)
256 uint64_t form
= read_unsigned_LEB_128(this->buffer_pos_
, &len
);
257 this->buffer_pos_
+= len
;
259 // A (0,0) pair terminates the list.
260 if (attr
== 0 && form
== 0)
263 if (attr
== elfcpp::DW_AT_sibling
)
264 entry
->has_sibling_attribute
= true;
266 entry
->add_attribute(attr
, form
);
269 this->store_abbrev(nextcode
, entry
);
270 if (nextcode
== code
)
277 // class Dwarf_ranges_table
279 // Read the ranges table from an object file.
282 Dwarf_ranges_table::read_ranges_table(
284 const unsigned char* symtab
,
286 unsigned int ranges_shndx
)
288 // If we've already read this abbrev table, return immediately.
289 if (this->ranges_shndx_
> 0
290 && this->ranges_shndx_
== ranges_shndx
)
293 // If we don't have relocations, ranges_shndx will be 0, and
294 // we'll have to hunt for the .debug_ranges section.
295 if (ranges_shndx
== 0 && this->ranges_shndx_
> 0)
296 ranges_shndx
= this->ranges_shndx_
;
297 else if (ranges_shndx
== 0)
299 for (unsigned int i
= 1; i
< object
->shnum(); ++i
)
301 std::string name
= object
->section_name(i
);
302 if (name
== ".debug_ranges")
305 this->output_section_offset_
= object
->output_section_offset(i
);
309 if (ranges_shndx
== 0)
313 // Get the section contents and decompress if necessary.
314 if (ranges_shndx
!= this->ranges_shndx_
)
316 if (this->owns_ranges_buffer_
&& this->ranges_buffer_
!= NULL
)
318 delete[] this->ranges_buffer_
;
319 this->owns_ranges_buffer_
= false;
322 section_size_type buffer_size
;
323 this->ranges_buffer_
=
324 object
->decompressed_section_contents(ranges_shndx
,
326 &this->owns_ranges_buffer_
);
327 this->ranges_buffer_end_
= this->ranges_buffer_
+ buffer_size
;
328 this->ranges_shndx_
= ranges_shndx
;
331 if (this->ranges_reloc_mapper_
!= NULL
)
333 delete this->ranges_reloc_mapper_
;
334 this->ranges_reloc_mapper_
= NULL
;
337 // For incremental objects, we have no relocations.
338 if (object
->is_incremental())
341 // Find the relocation section for ".debug_ranges".
342 unsigned int reloc_shndx
= 0;
343 unsigned int reloc_type
= 0;
344 for (unsigned int i
= 0; i
< object
->shnum(); ++i
)
346 reloc_type
= object
->section_type(i
);
347 if ((reloc_type
== elfcpp::SHT_REL
348 || reloc_type
== elfcpp::SHT_RELA
)
349 && object
->section_info(i
) == ranges_shndx
)
356 this->ranges_reloc_mapper_
= make_elf_reloc_mapper(object
, symtab
,
358 this->ranges_reloc_mapper_
->initialize(reloc_shndx
, reloc_type
);
363 // Read a range list from section RANGES_SHNDX at offset RANGES_OFFSET.
366 Dwarf_ranges_table::read_range_list(
368 const unsigned char* symtab
,
370 unsigned int addr_size
,
371 unsigned int ranges_shndx
,
374 Dwarf_range_list
* ranges
;
376 if (!this->read_ranges_table(object
, symtab
, symtab_size
, ranges_shndx
))
379 // Correct the offset. For incremental update links, we have a
380 // relocated offset that is relative to the output section, but
381 // here we need an offset relative to the input section.
382 offset
-= this->output_section_offset_
;
384 // Read the range list at OFFSET.
385 ranges
= new Dwarf_range_list();
388 this->ranges_buffer_
+ offset
< this->ranges_buffer_end_
;
389 offset
+= 2 * addr_size
)
394 // Read the raw contents of the section.
397 start
= read_from_pointer
<32>(this->ranges_buffer_
+ offset
);
398 end
= read_from_pointer
<32>(this->ranges_buffer_
+ offset
+ 4);
402 start
= read_from_pointer
<64>(this->ranges_buffer_
+ offset
);
403 end
= read_from_pointer
<64>(this->ranges_buffer_
+ offset
+ 8);
406 // Check for relocations and adjust the values.
407 unsigned int shndx1
= 0;
408 unsigned int shndx2
= 0;
409 if (this->ranges_reloc_mapper_
!= NULL
)
412 this->ranges_reloc_mapper_
->get_reloc_target(offset
, &start
);
414 this->ranges_reloc_mapper_
->get_reloc_target(offset
+ addr_size
,
418 // End of list is marked by a pair of zeroes.
419 if (shndx1
== 0 && start
== 0 && end
== 0)
422 // A "base address selection entry" is identified by
423 // 0xffffffff for the first value of the pair. The second
424 // value is used as a base for subsequent range list entries.
425 if (shndx1
== 0 && start
== -1)
427 else if (shndx1
== shndx2
)
429 if (shndx1
== 0 || object
->is_section_included(shndx1
))
430 ranges
->add(shndx1
, base
+ start
, base
+ end
);
433 gold_warning(_("%s: DWARF info may be corrupt; offsets in a "
434 "range list entry are in different sections"),
435 object
->name().c_str());
441 // class Dwarf_pubnames_table
443 // Read the pubnames section SHNDX from the object file.
446 Dwarf_pubnames_table::read_section(Relobj
* object
, unsigned int shndx
)
448 section_size_type buffer_size
;
450 // If we don't have relocations, shndx will be 0, and
451 // we'll have to hunt for the .debug_pubnames/pubtypes section.
454 const char* name
= (this->is_pubtypes_
456 : ".debug_pubnames");
457 for (unsigned int i
= 1; i
< object
->shnum(); ++i
)
459 if (object
->section_name(i
) == name
)
462 this->output_section_offset_
= object
->output_section_offset(i
);
470 this->buffer_
= object
->decompressed_section_contents(shndx
,
472 &this->owns_buffer_
);
473 if (this->buffer_
== NULL
)
475 this->buffer_end_
= this->buffer_
+ buffer_size
;
479 // Read the header for the set at OFFSET.
482 Dwarf_pubnames_table::read_header(off_t offset
)
484 // Correct the offset. For incremental update links, we have a
485 // relocated offset that is relative to the output section, but
486 // here we need an offset relative to the input section.
487 offset
-= this->output_section_offset_
;
489 if (offset
< 0 || offset
+ 14 >= this->buffer_end_
- this->buffer_
)
492 const unsigned char* pinfo
= this->buffer_
+ offset
;
494 // Read the unit_length field.
495 uint32_t unit_length
= read_from_pointer
<32>(pinfo
);
497 if (unit_length
== 0xffffffff)
499 unit_length
= read_from_pointer
<64>(pinfo
);
501 this->offset_size_
= 8;
504 this->offset_size_
= 4;
506 // Check the version.
507 unsigned int version
= read_from_pointer
<16>(pinfo
);
512 // Skip the debug_info_offset and debug_info_size fields.
513 pinfo
+= 2 * this->offset_size_
;
515 if (pinfo
>= this->buffer_end_
)
518 this->pinfo_
= pinfo
;
522 // Read the next name from the set.
525 Dwarf_pubnames_table::next_name()
527 const unsigned char* pinfo
= this->pinfo_
;
529 // Read the offset within the CU. If this is zero, we have reached
530 // the end of the list.
532 if (this->offset_size_
== 4)
533 offset
= read_from_pointer
<32>(&pinfo
);
535 offset
= read_from_pointer
<64>(&pinfo
);
539 // Return a pointer to the string at the current location,
540 // and advance the pointer to the next entry.
541 const char* ret
= reinterpret_cast<const char*>(pinfo
);
542 while (pinfo
< this->buffer_end_
&& *pinfo
!= '\0')
544 if (pinfo
< this->buffer_end_
)
547 this->pinfo_
= pinfo
;
553 Dwarf_die::Dwarf_die(
554 Dwarf_info_reader
* dwinfo
,
557 : dwinfo_(dwinfo
), parent_(parent
), die_offset_(die_offset
),
558 child_offset_(0), sibling_offset_(0), abbrev_code_(NULL
), attributes_(),
559 attributes_read_(false), name_(NULL
), name_off_(-1), linkage_name_(NULL
),
560 linkage_name_off_(-1), string_shndx_(0), specification_(0),
564 const unsigned char* pdie
= dwinfo
->buffer_at_offset(die_offset
);
567 unsigned int code
= read_unsigned_LEB_128(pdie
, &len
);
571 parent
->set_sibling_offset(die_offset
+ len
);
574 this->attr_offset_
= len
;
576 // Lookup the abbrev code in the abbrev table.
577 this->abbrev_code_
= dwinfo
->get_abbrev(code
);
580 // Read all the attributes of the DIE.
583 Dwarf_die::read_attributes()
585 if (this->attributes_read_
)
588 gold_assert(this->abbrev_code_
!= NULL
);
590 const unsigned char* pdie
=
591 this->dwinfo_
->buffer_at_offset(this->die_offset_
);
594 const unsigned char* pattr
= pdie
+ this->attr_offset_
;
596 unsigned int nattr
= this->abbrev_code_
->attributes
.size();
597 this->attributes_
.reserve(nattr
);
598 for (unsigned int i
= 0; i
< nattr
; ++i
)
601 unsigned int attr
= this->abbrev_code_
->attributes
[i
].attr
;
602 unsigned int form
= this->abbrev_code_
->attributes
[i
].form
;
603 if (form
== elfcpp::DW_FORM_indirect
)
605 form
= read_unsigned_LEB_128(pattr
, &len
);
608 off_t attr_off
= this->die_offset_
+ (pattr
- pdie
);
609 bool ref_form
= false;
610 Attribute_value attr_value
;
611 attr_value
.attr
= attr
;
612 attr_value
.form
= form
;
613 attr_value
.aux
.shndx
= 0;
616 case elfcpp::DW_FORM_flag_present
:
617 attr_value
.val
.intval
= 1;
619 case elfcpp::DW_FORM_strp
:
622 if (this->dwinfo_
->offset_size() == 4)
623 str_off
= read_from_pointer
<32>(&pattr
);
625 str_off
= read_from_pointer
<64>(&pattr
);
627 this->dwinfo_
->lookup_reloc(attr_off
, &str_off
);
628 attr_value
.aux
.shndx
= shndx
;
629 attr_value
.val
.refval
= str_off
;
632 case elfcpp::DW_FORM_sec_offset
:
635 if (this->dwinfo_
->offset_size() == 4)
636 sec_off
= read_from_pointer
<32>(&pattr
);
638 sec_off
= read_from_pointer
<64>(&pattr
);
640 this->dwinfo_
->lookup_reloc(attr_off
, &sec_off
);
641 attr_value
.aux
.shndx
= shndx
;
642 attr_value
.val
.refval
= sec_off
;
646 case elfcpp::DW_FORM_addr
:
647 case elfcpp::DW_FORM_ref_addr
:
650 if (this->dwinfo_
->address_size() == 4)
651 sec_off
= read_from_pointer
<32>(&pattr
);
653 sec_off
= read_from_pointer
<64>(&pattr
);
655 this->dwinfo_
->lookup_reloc(attr_off
, &sec_off
);
656 attr_value
.aux
.shndx
= shndx
;
657 attr_value
.val
.refval
= sec_off
;
661 case elfcpp::DW_FORM_block1
:
662 attr_value
.aux
.blocklen
= *pattr
++;
663 attr_value
.val
.blockval
= pattr
;
664 pattr
+= attr_value
.aux
.blocklen
;
666 case elfcpp::DW_FORM_block2
:
667 attr_value
.aux
.blocklen
= read_from_pointer
<16>(&pattr
);
668 attr_value
.val
.blockval
= pattr
;
669 pattr
+= attr_value
.aux
.blocklen
;
671 case elfcpp::DW_FORM_block4
:
672 attr_value
.aux
.blocklen
= read_from_pointer
<32>(&pattr
);
673 attr_value
.val
.blockval
= pattr
;
674 pattr
+= attr_value
.aux
.blocklen
;
676 case elfcpp::DW_FORM_block
:
677 case elfcpp::DW_FORM_exprloc
:
678 attr_value
.aux
.blocklen
= read_unsigned_LEB_128(pattr
, &len
);
679 attr_value
.val
.blockval
= pattr
+ len
;
680 pattr
+= len
+ attr_value
.aux
.blocklen
;
682 case elfcpp::DW_FORM_data1
:
683 case elfcpp::DW_FORM_flag
:
684 attr_value
.val
.intval
= *pattr
++;
686 case elfcpp::DW_FORM_ref1
:
687 attr_value
.val
.refval
= *pattr
++;
690 case elfcpp::DW_FORM_data2
:
691 attr_value
.val
.intval
= read_from_pointer
<16>(&pattr
);
693 case elfcpp::DW_FORM_ref2
:
694 attr_value
.val
.refval
= read_from_pointer
<16>(&pattr
);
697 case elfcpp::DW_FORM_data4
:
700 sec_off
= read_from_pointer
<32>(&pattr
);
702 this->dwinfo_
->lookup_reloc(attr_off
, &sec_off
);
703 attr_value
.aux
.shndx
= shndx
;
704 attr_value
.val
.intval
= sec_off
;
707 case elfcpp::DW_FORM_ref4
:
710 sec_off
= read_from_pointer
<32>(&pattr
);
712 this->dwinfo_
->lookup_reloc(attr_off
, &sec_off
);
713 attr_value
.aux
.shndx
= shndx
;
714 attr_value
.val
.refval
= sec_off
;
718 case elfcpp::DW_FORM_data8
:
721 sec_off
= read_from_pointer
<64>(&pattr
);
723 this->dwinfo_
->lookup_reloc(attr_off
, &sec_off
);
724 attr_value
.aux
.shndx
= shndx
;
725 attr_value
.val
.intval
= sec_off
;
728 case elfcpp::DW_FORM_ref_sig8
:
729 attr_value
.val
.uintval
= read_from_pointer
<64>(&pattr
);
731 case elfcpp::DW_FORM_ref8
:
734 sec_off
= read_from_pointer
<64>(&pattr
);
736 this->dwinfo_
->lookup_reloc(attr_off
, &sec_off
);
737 attr_value
.aux
.shndx
= shndx
;
738 attr_value
.val
.refval
= sec_off
;
742 case elfcpp::DW_FORM_ref_udata
:
743 attr_value
.val
.refval
= read_unsigned_LEB_128(pattr
, &len
);
747 case elfcpp::DW_FORM_udata
:
748 attr_value
.val
.uintval
= read_unsigned_LEB_128(pattr
, &len
);
751 case elfcpp::DW_FORM_sdata
:
752 attr_value
.val
.intval
= read_signed_LEB_128(pattr
, &len
);
755 case elfcpp::DW_FORM_string
:
756 attr_value
.val
.stringval
= reinterpret_cast<const char*>(pattr
);
757 len
= strlen(attr_value
.val
.stringval
);
764 // Cache the most frequently-requested attributes.
767 case elfcpp::DW_AT_name
:
768 if (form
== elfcpp::DW_FORM_string
)
769 this->name_
= attr_value
.val
.stringval
;
770 else if (form
== elfcpp::DW_FORM_strp
)
772 // All indirect strings should refer to the same
773 // string section, so we just save the last one seen.
774 this->string_shndx_
= attr_value
.aux
.shndx
;
775 this->name_off_
= attr_value
.val
.refval
;
778 case elfcpp::DW_AT_linkage_name
:
779 case elfcpp::DW_AT_MIPS_linkage_name
:
780 if (form
== elfcpp::DW_FORM_string
)
781 this->linkage_name_
= attr_value
.val
.stringval
;
782 else if (form
== elfcpp::DW_FORM_strp
)
784 // All indirect strings should refer to the same
785 // string section, so we just save the last one seen.
786 this->string_shndx_
= attr_value
.aux
.shndx
;
787 this->linkage_name_off_
= attr_value
.val
.refval
;
790 case elfcpp::DW_AT_specification
:
792 this->specification_
= attr_value
.val
.refval
;
794 case elfcpp::DW_AT_abstract_origin
:
796 this->abstract_origin_
= attr_value
.val
.refval
;
798 case elfcpp::DW_AT_sibling
:
799 if (ref_form
&& attr_value
.aux
.shndx
== 0)
800 this->sibling_offset_
= attr_value
.val
.refval
;
805 this->attributes_
.push_back(attr_value
);
808 // Now that we know where the next DIE begins, record the offset
809 // to avoid later recalculation.
810 if (this->has_children())
811 this->child_offset_
= this->die_offset_
+ (pattr
- pdie
);
813 this->sibling_offset_
= this->die_offset_
+ (pattr
- pdie
);
815 this->attributes_read_
= true;
819 // Skip all the attributes of the DIE and return the offset of the next DIE.
822 Dwarf_die::skip_attributes()
824 typedef Dwarf_abbrev_table::Attribute Attribute
;
826 gold_assert(this->abbrev_code_
!= NULL
);
828 const unsigned char* pdie
=
829 this->dwinfo_
->buffer_at_offset(this->die_offset_
);
832 const unsigned char* pattr
= pdie
+ this->attr_offset_
;
834 for (unsigned int i
= 0; i
< this->abbrev_code_
->attributes
.size(); ++i
)
837 unsigned int form
= this->abbrev_code_
->attributes
[i
].form
;
838 if (form
== elfcpp::DW_FORM_indirect
)
840 form
= read_unsigned_LEB_128(pattr
, &len
);
845 case elfcpp::DW_FORM_flag_present
:
847 case elfcpp::DW_FORM_strp
:
848 case elfcpp::DW_FORM_sec_offset
:
849 pattr
+= this->dwinfo_
->offset_size();
851 case elfcpp::DW_FORM_addr
:
852 case elfcpp::DW_FORM_ref_addr
:
853 pattr
+= this->dwinfo_
->address_size();
855 case elfcpp::DW_FORM_block1
:
858 case elfcpp::DW_FORM_block2
:
861 block_size
= read_from_pointer
<16>(&pattr
);
865 case elfcpp::DW_FORM_block4
:
868 block_size
= read_from_pointer
<32>(&pattr
);
872 case elfcpp::DW_FORM_block
:
873 case elfcpp::DW_FORM_exprloc
:
876 block_size
= read_unsigned_LEB_128(pattr
, &len
);
877 pattr
+= len
+ block_size
;
880 case elfcpp::DW_FORM_data1
:
881 case elfcpp::DW_FORM_ref1
:
882 case elfcpp::DW_FORM_flag
:
885 case elfcpp::DW_FORM_data2
:
886 case elfcpp::DW_FORM_ref2
:
889 case elfcpp::DW_FORM_data4
:
890 case elfcpp::DW_FORM_ref4
:
893 case elfcpp::DW_FORM_data8
:
894 case elfcpp::DW_FORM_ref8
:
895 case elfcpp::DW_FORM_ref_sig8
:
898 case elfcpp::DW_FORM_ref_udata
:
899 case elfcpp::DW_FORM_udata
:
900 read_unsigned_LEB_128(pattr
, &len
);
903 case elfcpp::DW_FORM_sdata
:
904 read_signed_LEB_128(pattr
, &len
);
907 case elfcpp::DW_FORM_string
:
908 len
= strlen(reinterpret_cast<const char*>(pattr
));
916 return this->die_offset_
+ (pattr
- pdie
);
919 // Get the name of the DIE and cache it.
922 Dwarf_die::set_name()
924 if (this->name_
!= NULL
|| !this->read_attributes())
926 if (this->name_off_
!= -1)
927 this->name_
= this->dwinfo_
->get_string(this->name_off_
,
928 this->string_shndx_
);
931 // Get the linkage name of the DIE and cache it.
934 Dwarf_die::set_linkage_name()
936 if (this->linkage_name_
!= NULL
|| !this->read_attributes())
938 if (this->linkage_name_off_
!= -1)
939 this->linkage_name_
= this->dwinfo_
->get_string(this->linkage_name_off_
,
940 this->string_shndx_
);
943 // Return the value of attribute ATTR.
945 const Dwarf_die::Attribute_value
*
946 Dwarf_die::attribute(unsigned int attr
)
948 if (!this->read_attributes())
950 for (unsigned int i
= 0; i
< this->attributes_
.size(); ++i
)
952 if (this->attributes_
[i
].attr
== attr
)
953 return &this->attributes_
[i
];
959 Dwarf_die::string_attribute(unsigned int attr
)
961 const Attribute_value
* attr_val
= this->attribute(attr
);
962 if (attr_val
== NULL
)
964 switch (attr_val
->form
)
966 case elfcpp::DW_FORM_string
:
967 return attr_val
->val
.stringval
;
968 case elfcpp::DW_FORM_strp
:
969 return this->dwinfo_
->get_string(attr_val
->val
.refval
,
970 attr_val
->aux
.shndx
);
977 Dwarf_die::int_attribute(unsigned int attr
)
979 const Attribute_value
* attr_val
= this->attribute(attr
);
980 if (attr_val
== NULL
)
982 switch (attr_val
->form
)
984 case elfcpp::DW_FORM_flag_present
:
985 case elfcpp::DW_FORM_data1
:
986 case elfcpp::DW_FORM_flag
:
987 case elfcpp::DW_FORM_data2
:
988 case elfcpp::DW_FORM_data4
:
989 case elfcpp::DW_FORM_data8
:
990 case elfcpp::DW_FORM_sdata
:
991 return attr_val
->val
.intval
;
998 Dwarf_die::uint_attribute(unsigned int attr
)
1000 const Attribute_value
* attr_val
= this->attribute(attr
);
1001 if (attr_val
== NULL
)
1003 switch (attr_val
->form
)
1005 case elfcpp::DW_FORM_flag_present
:
1006 case elfcpp::DW_FORM_data1
:
1007 case elfcpp::DW_FORM_flag
:
1008 case elfcpp::DW_FORM_data4
:
1009 case elfcpp::DW_FORM_data8
:
1010 case elfcpp::DW_FORM_ref_sig8
:
1011 case elfcpp::DW_FORM_udata
:
1012 return attr_val
->val
.uintval
;
1019 Dwarf_die::ref_attribute(unsigned int attr
, unsigned int* shndx
)
1021 const Attribute_value
* attr_val
= this->attribute(attr
);
1022 if (attr_val
== NULL
)
1024 switch (attr_val
->form
)
1026 case elfcpp::DW_FORM_sec_offset
:
1027 case elfcpp::DW_FORM_addr
:
1028 case elfcpp::DW_FORM_ref_addr
:
1029 case elfcpp::DW_FORM_ref1
:
1030 case elfcpp::DW_FORM_ref2
:
1031 case elfcpp::DW_FORM_ref4
:
1032 case elfcpp::DW_FORM_ref8
:
1033 case elfcpp::DW_FORM_ref_udata
:
1034 *shndx
= attr_val
->aux
.shndx
;
1035 return attr_val
->val
.refval
;
1036 case elfcpp::DW_FORM_ref_sig8
:
1037 *shndx
= attr_val
->aux
.shndx
;
1038 return attr_val
->val
.uintval
;
1039 case elfcpp::DW_FORM_data4
:
1040 case elfcpp::DW_FORM_data8
:
1041 *shndx
= attr_val
->aux
.shndx
;
1042 return attr_val
->val
.intval
;
1049 Dwarf_die::address_attribute(unsigned int attr
, unsigned int* shndx
)
1051 const Attribute_value
* attr_val
= this->attribute(attr
);
1052 if (attr_val
== NULL
|| attr_val
->form
!= elfcpp::DW_FORM_addr
)
1055 *shndx
= attr_val
->aux
.shndx
;
1056 return attr_val
->val
.refval
;
1059 // Return the offset of this DIE's first child.
1062 Dwarf_die::child_offset()
1064 gold_assert(this->abbrev_code_
!= NULL
);
1065 if (!this->has_children())
1067 if (this->child_offset_
== 0)
1068 this->child_offset_
= this->skip_attributes();
1069 return this->child_offset_
;
1072 // Return the offset of this DIE's next sibling.
1075 Dwarf_die::sibling_offset()
1077 gold_assert(this->abbrev_code_
!= NULL
);
1079 if (this->sibling_offset_
!= 0)
1080 return this->sibling_offset_
;
1082 if (!this->has_children())
1084 this->sibling_offset_
= this->skip_attributes();
1085 return this->sibling_offset_
;
1088 if (this->has_sibling_attribute())
1090 if (!this->read_attributes())
1092 if (this->sibling_offset_
!= 0)
1093 return this->sibling_offset_
;
1096 // Skip over the children.
1097 off_t child_offset
= this->child_offset();
1098 while (child_offset
> 0)
1100 Dwarf_die
die(this->dwinfo_
, child_offset
, this);
1101 // The Dwarf_die ctor will set this DIE's sibling offset
1102 // when it reads a zero abbrev code.
1105 child_offset
= die
.sibling_offset();
1108 // This should be set by now. If not, there was a problem reading
1109 // the DWARF info, and we return 0.
1110 return this->sibling_offset_
;
1113 // class Dwarf_info_reader
1115 // Check that the pointer P is within the current compilation unit.
1118 Dwarf_info_reader::check_buffer(const unsigned char* p
) const
1120 if (p
> this->buffer_
+ this->cu_offset_
+ this->cu_length_
)
1122 gold_warning(_("%s: corrupt debug info in %s"),
1123 this->object_
->name().c_str(),
1124 this->object_
->section_name(this->shndx_
).c_str());
1130 // Begin parsing the debug info. This calls visit_compilation_unit()
1131 // or visit_type_unit() for each compilation or type unit found in the
1132 // section, and visit_die() for each top-level DIE.
1135 Dwarf_info_reader::parse()
1137 switch (parameters
->size_and_endianness())
1139 #ifdef HAVE_TARGET_32_LITTLE
1140 case Parameters::TARGET_32_LITTLE
:
1141 this->do_parse
<false>();
1144 #ifdef HAVE_TARGET_32_BIG
1145 case Parameters::TARGET_32_BIG
:
1146 this->do_parse
<true>();
1149 #ifdef HAVE_TARGET_64_LITTLE
1150 case Parameters::TARGET_64_LITTLE
:
1151 this->do_parse
<false>();
1154 #ifdef HAVE_TARGET_64_BIG
1155 case Parameters::TARGET_64_BIG
:
1156 this->do_parse
<true>();
1164 template<bool big_endian
>
1166 Dwarf_info_reader::do_parse()
1168 // Get the section contents and decompress if necessary.
1169 section_size_type buffer_size
;
1171 this->buffer_
= this->object_
->decompressed_section_contents(this->shndx_
,
1174 if (this->buffer_
== NULL
|| buffer_size
== 0)
1176 this->buffer_end_
= this->buffer_
+ buffer_size
;
1178 // The offset of this input section in the output section.
1179 off_t section_offset
= this->object_
->output_section_offset(this->shndx_
);
1181 // Start tracking relocations for this section.
1182 this->reloc_mapper_
= make_elf_reloc_mapper(this->object_
, this->symtab_
,
1183 this->symtab_size_
);
1184 this->reloc_mapper_
->initialize(this->reloc_shndx_
, this->reloc_type_
);
1186 // Loop over compilation units (or type units).
1187 unsigned int abbrev_shndx
= 0;
1188 off_t abbrev_offset
= 0;
1189 const unsigned char* pinfo
= this->buffer_
;
1190 while (pinfo
< this->buffer_end_
)
1192 // Read the compilation (or type) unit header.
1193 const unsigned char* cu_start
= pinfo
;
1194 this->cu_offset_
= cu_start
- this->buffer_
;
1195 this->cu_length_
= this->buffer_end_
- cu_start
;
1197 // Read unit_length (4 or 12 bytes).
1198 if (!this->check_buffer(pinfo
+ 4))
1200 uint32_t unit_length
=
1201 elfcpp::Swap_unaligned
<32, big_endian
>::readval(pinfo
);
1203 if (unit_length
== 0xffffffff)
1205 if (!this->check_buffer(pinfo
+ 8))
1207 unit_length
= elfcpp::Swap_unaligned
<64, big_endian
>::readval(pinfo
);
1209 this->offset_size_
= 8;
1212 this->offset_size_
= 4;
1213 if (!this->check_buffer(pinfo
+ unit_length
))
1215 const unsigned char* cu_end
= pinfo
+ unit_length
;
1216 this->cu_length_
= cu_end
- cu_start
;
1217 if (!this->check_buffer(pinfo
+ 2 + this->offset_size_
+ 1))
1220 // Read version (2 bytes).
1222 elfcpp::Swap_unaligned
<16, big_endian
>::readval(pinfo
);
1225 // Read debug_abbrev_offset (4 or 8 bytes).
1226 if (this->offset_size_
== 4)
1227 abbrev_offset
= elfcpp::Swap_unaligned
<32, big_endian
>::readval(pinfo
);
1229 abbrev_offset
= elfcpp::Swap_unaligned
<64, big_endian
>::readval(pinfo
);
1230 if (this->reloc_shndx_
> 0)
1232 off_t reloc_offset
= pinfo
- this->buffer_
;
1235 this->reloc_mapper_
->get_reloc_target(reloc_offset
, &value
);
1236 if (abbrev_shndx
== 0)
1238 if (this->reloc_type_
== elfcpp::SHT_REL
)
1239 abbrev_offset
+= value
;
1241 abbrev_offset
= value
;
1243 pinfo
+= this->offset_size_
;
1245 // Read address_size (1 byte).
1246 this->address_size_
= *pinfo
++;
1248 // For type units, read the two extra fields.
1249 uint64_t signature
= 0;
1250 off_t type_offset
= 0;
1251 if (this->is_type_unit_
)
1253 if (!this->check_buffer(pinfo
+ 8 + this->offset_size_
))
1256 // Read type_signature (8 bytes).
1257 signature
= elfcpp::Swap_unaligned
<64, big_endian
>::readval(pinfo
);
1260 // Read type_offset (4 or 8 bytes).
1261 if (this->offset_size_
== 4)
1263 elfcpp::Swap_unaligned
<32, big_endian
>::readval(pinfo
);
1266 elfcpp::Swap_unaligned
<64, big_endian
>::readval(pinfo
);
1267 pinfo
+= this->offset_size_
;
1270 // Read the .debug_abbrev table.
1271 this->abbrev_table_
.read_abbrevs(this->object_
, abbrev_shndx
,
1274 // Visit the root DIE.
1275 Dwarf_die
root_die(this,
1276 pinfo
- (this->buffer_
+ this->cu_offset_
),
1278 if (root_die
.tag() != 0)
1280 // Visit the CU or TU.
1281 if (this->is_type_unit_
)
1282 this->visit_type_unit(section_offset
+ this->cu_offset_
,
1283 type_offset
, signature
, &root_die
);
1285 this->visit_compilation_unit(section_offset
+ this->cu_offset_
,
1286 cu_end
- cu_start
, &root_die
);
1289 // Advance to the next CU.
1295 delete[] this->buffer_
;
1296 this->buffer_
= NULL
;
1300 // Read the DWARF string table.
1303 Dwarf_info_reader::do_read_string_table(unsigned int string_shndx
)
1305 Relobj
* object
= this->object_
;
1307 // If we don't have relocations, string_shndx will be 0, and
1308 // we'll have to hunt for the .debug_str section.
1309 if (string_shndx
== 0)
1311 for (unsigned int i
= 1; i
< this->object_
->shnum(); ++i
)
1313 std::string name
= object
->section_name(i
);
1314 if (name
== ".debug_str")
1317 this->string_output_section_offset_
=
1318 object
->output_section_offset(i
);
1322 if (string_shndx
== 0)
1326 if (this->owns_string_buffer_
&& this->string_buffer_
!= NULL
)
1328 delete[] this->string_buffer_
;
1329 this->owns_string_buffer_
= false;
1332 // Get the secton contents and decompress if necessary.
1333 section_size_type buffer_size
;
1334 const unsigned char* buffer
=
1335 object
->decompressed_section_contents(string_shndx
,
1337 &this->owns_string_buffer_
);
1338 this->string_buffer_
= reinterpret_cast<const char*>(buffer
);
1339 this->string_buffer_end_
= this->string_buffer_
+ buffer_size
;
1340 this->string_shndx_
= string_shndx
;
1344 // Look for a relocation at offset ATTR_OFF in the dwarf info,
1345 // and return the section index and offset of the target.
1348 Dwarf_info_reader::lookup_reloc(off_t attr_off
, off_t
* target_off
)
1351 attr_off
+= this->cu_offset_
;
1352 unsigned int shndx
= this->reloc_mapper_
->get_reloc_target(attr_off
, &value
);
1355 if (this->reloc_type_
== elfcpp::SHT_REL
)
1356 *target_off
+= value
;
1358 *target_off
= value
;
1362 // Return a string from the DWARF string table.
1365 Dwarf_info_reader::get_string(off_t str_off
, unsigned int string_shndx
)
1367 if (!this->read_string_table(string_shndx
))
1370 // Correct the offset. For incremental update links, we have a
1371 // relocated offset that is relative to the output section, but
1372 // here we need an offset relative to the input section.
1373 str_off
-= this->string_output_section_offset_
;
1375 const char* p
= this->string_buffer_
+ str_off
;
1377 if (p
< this->string_buffer_
|| p
>= this->string_buffer_end_
)
1383 // The following are default, do-nothing, implementations of the
1384 // hook methods normally provided by a derived class. We provide
1385 // default implementations rather than no implementation so that
1386 // a derived class needs to implement only the hooks that it needs
1389 // Process a compilation unit and parse its child DIE.
1392 Dwarf_info_reader::visit_compilation_unit(off_t
, off_t
, Dwarf_die
*)
1396 // Process a type unit and parse its child DIE.
1399 Dwarf_info_reader::visit_type_unit(off_t
, off_t
, uint64_t, Dwarf_die
*)
1403 // class Sized_dwarf_line_info
1405 struct LineStateMachine
1411 unsigned int shndx
; // the section address refers to
1412 bool is_stmt
; // stmt means statement.
1418 ResetLineStateMachine(struct LineStateMachine
* lsm
, bool default_is_stmt
)
1423 lsm
->column_num
= 0;
1425 lsm
->is_stmt
= default_is_stmt
;
1426 lsm
->basic_block
= false;
1427 lsm
->end_sequence
= false;
1430 template<int size
, bool big_endian
>
1431 Sized_dwarf_line_info
<size
, big_endian
>::Sized_dwarf_line_info(
1433 unsigned int read_shndx
)
1434 : data_valid_(false), buffer_(NULL
), buffer_start_(NULL
),
1435 reloc_mapper_(NULL
), symtab_buffer_(NULL
), directories_(), files_(),
1436 current_header_index_(-1)
1438 unsigned int debug_shndx
;
1440 for (debug_shndx
= 1; debug_shndx
< object
->shnum(); ++debug_shndx
)
1442 // FIXME: do this more efficiently: section_name() isn't super-fast
1443 std::string name
= object
->section_name(debug_shndx
);
1444 if (name
== ".debug_line" || name
== ".zdebug_line")
1446 section_size_type buffer_size
;
1447 bool is_new
= false;
1448 this->buffer_
= object
->decompressed_section_contents(debug_shndx
,
1452 this->buffer_start_
= this->buffer_
;
1453 this->buffer_end_
= this->buffer_
+ buffer_size
;
1457 if (this->buffer_
== NULL
)
1460 // Find the relocation section for ".debug_line".
1461 // We expect these for relobjs (.o's) but not dynobjs (.so's).
1462 unsigned int reloc_shndx
= 0;
1463 for (unsigned int i
= 0; i
< object
->shnum(); ++i
)
1465 unsigned int reloc_sh_type
= object
->section_type(i
);
1466 if ((reloc_sh_type
== elfcpp::SHT_REL
1467 || reloc_sh_type
== elfcpp::SHT_RELA
)
1468 && object
->section_info(i
) == debug_shndx
)
1471 this->track_relocs_type_
= reloc_sh_type
;
1476 // Finally, we need the symtab section to interpret the relocs.
1477 if (reloc_shndx
!= 0)
1479 unsigned int symtab_shndx
;
1480 for (symtab_shndx
= 0; symtab_shndx
< object
->shnum(); ++symtab_shndx
)
1481 if (object
->section_type(symtab_shndx
) == elfcpp::SHT_SYMTAB
)
1483 this->symtab_buffer_
= object
->section_contents(
1484 symtab_shndx
, &this->symtab_buffer_size_
, false);
1487 if (this->symtab_buffer_
== NULL
)
1491 this->reloc_mapper_
=
1492 new Sized_elf_reloc_mapper
<size
, big_endian
>(object
,
1493 this->symtab_buffer_
,
1494 this->symtab_buffer_size_
);
1495 if (!this->reloc_mapper_
->initialize(reloc_shndx
, this->track_relocs_type_
))
1498 // Now that we have successfully read all the data, parse the debug
1500 this->data_valid_
= true;
1501 this->read_line_mappings(read_shndx
);
1504 // Read the DWARF header.
1506 template<int size
, bool big_endian
>
1507 const unsigned char*
1508 Sized_dwarf_line_info
<size
, big_endian
>::read_header_prolog(
1509 const unsigned char* lineptr
)
1511 uint32_t initial_length
= elfcpp::Swap_unaligned
<32, big_endian
>::readval(lineptr
);
1514 // In DWARF2/3, if the initial length is all 1 bits, then the offset
1515 // size is 8 and we need to read the next 8 bytes for the real length.
1516 if (initial_length
== 0xffffffff)
1518 header_
.offset_size
= 8;
1519 initial_length
= elfcpp::Swap_unaligned
<64, big_endian
>::readval(lineptr
);
1523 header_
.offset_size
= 4;
1525 header_
.total_length
= initial_length
;
1527 gold_assert(lineptr
+ header_
.total_length
<= buffer_end_
);
1529 header_
.version
= elfcpp::Swap_unaligned
<16, big_endian
>::readval(lineptr
);
1532 if (header_
.offset_size
== 4)
1533 header_
.prologue_length
= elfcpp::Swap_unaligned
<32, big_endian
>::readval(lineptr
);
1535 header_
.prologue_length
= elfcpp::Swap_unaligned
<64, big_endian
>::readval(lineptr
);
1536 lineptr
+= header_
.offset_size
;
1538 header_
.min_insn_length
= *lineptr
;
1541 header_
.default_is_stmt
= *lineptr
;
1544 header_
.line_base
= *reinterpret_cast<const signed char*>(lineptr
);
1547 header_
.line_range
= *lineptr
;
1550 header_
.opcode_base
= *lineptr
;
1553 header_
.std_opcode_lengths
.resize(header_
.opcode_base
+ 1);
1554 header_
.std_opcode_lengths
[0] = 0;
1555 for (int i
= 1; i
< header_
.opcode_base
; i
++)
1557 header_
.std_opcode_lengths
[i
] = *lineptr
;
1564 // The header for a debug_line section is mildly complicated, because
1565 // the line info is very tightly encoded.
1567 template<int size
, bool big_endian
>
1568 const unsigned char*
1569 Sized_dwarf_line_info
<size
, big_endian
>::read_header_tables(
1570 const unsigned char* lineptr
)
1572 ++this->current_header_index_
;
1574 // Create a new directories_ entry and a new files_ entry for our new
1575 // header. We initialize each with a single empty element, because
1576 // dwarf indexes directory and filenames starting at 1.
1577 gold_assert(static_cast<int>(this->directories_
.size())
1578 == this->current_header_index_
);
1579 gold_assert(static_cast<int>(this->files_
.size())
1580 == this->current_header_index_
);
1581 this->directories_
.push_back(std::vector
<std::string
>(1));
1582 this->files_
.push_back(std::vector
<std::pair
<int, std::string
> >(1));
1584 // It is legal for the directory entry table to be empty.
1590 const char* dirname
= reinterpret_cast<const char*>(lineptr
);
1591 gold_assert(dirindex
1592 == static_cast<int>(this->directories_
.back().size()));
1593 this->directories_
.back().push_back(dirname
);
1594 lineptr
+= this->directories_
.back().back().size() + 1;
1600 // It is also legal for the file entry table to be empty.
1607 const char* filename
= reinterpret_cast<const char*>(lineptr
);
1608 lineptr
+= strlen(filename
) + 1;
1610 uint64_t dirindex
= read_unsigned_LEB_128(lineptr
, &len
);
1613 if (dirindex
>= this->directories_
.back().size())
1615 int dirindexi
= static_cast<int>(dirindex
);
1617 read_unsigned_LEB_128(lineptr
, &len
); // mod_time
1620 read_unsigned_LEB_128(lineptr
, &len
); // filelength
1623 gold_assert(fileindex
1624 == static_cast<int>(this->files_
.back().size()));
1625 this->files_
.back().push_back(std::make_pair(dirindexi
, filename
));
1634 // Process a single opcode in the .debug.line structure.
1636 template<int size
, bool big_endian
>
1638 Sized_dwarf_line_info
<size
, big_endian
>::process_one_opcode(
1639 const unsigned char* start
, struct LineStateMachine
* lsm
, size_t* len
)
1643 unsigned char opcode
= *start
;
1647 // If the opcode is great than the opcode_base, it is a special
1648 // opcode. Most line programs consist mainly of special opcodes.
1649 if (opcode
>= header_
.opcode_base
)
1651 opcode
-= header_
.opcode_base
;
1652 const int advance_address
= ((opcode
/ header_
.line_range
)
1653 * header_
.min_insn_length
);
1654 lsm
->address
+= advance_address
;
1656 const int advance_line
= ((opcode
% header_
.line_range
)
1657 + header_
.line_base
);
1658 lsm
->line_num
+= advance_line
;
1659 lsm
->basic_block
= true;
1664 // Otherwise, we have the regular opcodes
1667 case elfcpp::DW_LNS_copy
:
1668 lsm
->basic_block
= false;
1672 case elfcpp::DW_LNS_advance_pc
:
1674 const uint64_t advance_address
1675 = read_unsigned_LEB_128(start
, &templen
);
1677 lsm
->address
+= header_
.min_insn_length
* advance_address
;
1681 case elfcpp::DW_LNS_advance_line
:
1683 const uint64_t advance_line
= read_signed_LEB_128(start
, &templen
);
1685 lsm
->line_num
+= advance_line
;
1689 case elfcpp::DW_LNS_set_file
:
1691 const uint64_t fileno
= read_unsigned_LEB_128(start
, &templen
);
1693 lsm
->file_num
= fileno
;
1697 case elfcpp::DW_LNS_set_column
:
1699 const uint64_t colno
= read_unsigned_LEB_128(start
, &templen
);
1701 lsm
->column_num
= colno
;
1705 case elfcpp::DW_LNS_negate_stmt
:
1706 lsm
->is_stmt
= !lsm
->is_stmt
;
1709 case elfcpp::DW_LNS_set_basic_block
:
1710 lsm
->basic_block
= true;
1713 case elfcpp::DW_LNS_fixed_advance_pc
:
1715 int advance_address
;
1716 advance_address
= elfcpp::Swap_unaligned
<16, big_endian
>::readval(start
);
1718 lsm
->address
+= advance_address
;
1722 case elfcpp::DW_LNS_const_add_pc
:
1724 const int advance_address
= (header_
.min_insn_length
1725 * ((255 - header_
.opcode_base
)
1726 / header_
.line_range
));
1727 lsm
->address
+= advance_address
;
1731 case elfcpp::DW_LNS_extended_op
:
1733 const uint64_t extended_op_len
1734 = read_unsigned_LEB_128(start
, &templen
);
1736 oplen
+= templen
+ extended_op_len
;
1738 const unsigned char extended_op
= *start
;
1741 switch (extended_op
)
1743 case elfcpp::DW_LNE_end_sequence
:
1744 // This means that the current byte is the one immediately
1745 // after a set of instructions. Record the current line
1746 // for up to one less than the current address.
1748 lsm
->end_sequence
= true;
1752 case elfcpp::DW_LNE_set_address
:
1755 elfcpp::Swap_unaligned
<size
, big_endian
>::readval(start
);
1756 typename
Reloc_map::const_iterator it
1757 = this->reloc_map_
.find(start
- this->buffer_
);
1758 if (it
!= reloc_map_
.end())
1760 // If this is a SHT_RELA section, then ignore the
1761 // section contents. This assumes that this is a
1762 // straight reloc which just uses the reloc addend.
1763 // The reloc addend has already been included in the
1765 if (this->track_relocs_type_
== elfcpp::SHT_RELA
)
1767 // Add in the symbol value.
1768 lsm
->address
+= it
->second
.second
;
1769 lsm
->shndx
= it
->second
.first
;
1773 // If we're a normal .o file, with relocs, every
1774 // set_address should have an associated relocation.
1775 if (this->input_is_relobj())
1776 this->data_valid_
= false;
1780 case elfcpp::DW_LNE_define_file
:
1782 const char* filename
= reinterpret_cast<const char*>(start
);
1783 templen
= strlen(filename
) + 1;
1786 uint64_t dirindex
= read_unsigned_LEB_128(start
, &templen
);
1788 if (dirindex
>= this->directories_
.back().size())
1790 int dirindexi
= static_cast<int>(dirindex
);
1792 // This opcode takes two additional ULEB128 parameters
1793 // (mod_time and filelength), but we don't use those
1794 // values. Because OPLEN already tells us how far to
1795 // skip to the next opcode, we don't need to read
1798 this->files_
.back().push_back(std::make_pair(dirindexi
,
1808 // Ignore unknown opcode silently
1809 for (int i
= 0; i
< header_
.std_opcode_lengths
[opcode
]; i
++)
1812 read_unsigned_LEB_128(start
, &templen
);
1823 // Read the debug information at LINEPTR and store it in the line
1826 template<int size
, bool big_endian
>
1827 unsigned const char*
1828 Sized_dwarf_line_info
<size
, big_endian
>::read_lines(unsigned const char* lineptr
,
1831 struct LineStateMachine lsm
;
1833 // LENGTHSTART is the place the length field is based on. It is the
1834 // point in the header after the initial length field.
1835 const unsigned char* lengthstart
= buffer_
;
1837 // In 64 bit dwarf, the initial length is 12 bytes, because of the
1838 // 0xffffffff at the start.
1839 if (header_
.offset_size
== 8)
1844 while (lineptr
< lengthstart
+ header_
.total_length
)
1846 ResetLineStateMachine(&lsm
, header_
.default_is_stmt
);
1847 while (!lsm
.end_sequence
)
1850 bool add_line
= this->process_one_opcode(lineptr
, &lsm
, &oplength
);
1852 && (shndx
== -1U || lsm
.shndx
== -1U || shndx
== lsm
.shndx
))
1854 Offset_to_lineno_entry entry
1855 = { static_cast<off_t
>(lsm
.address
),
1856 this->current_header_index_
,
1857 static_cast<unsigned int>(lsm
.file_num
),
1858 true, lsm
.line_num
};
1859 std::vector
<Offset_to_lineno_entry
>&
1860 map(this->line_number_map_
[lsm
.shndx
]);
1861 // If we see two consecutive entries with the same
1862 // offset and a real line number, then mark the first
1863 // one as non-canonical.
1865 && (map
.back().offset
== static_cast<off_t
>(lsm
.address
))
1866 && lsm
.line_num
!= -1
1867 && map
.back().line_num
!= -1)
1868 map
.back().last_line_for_offset
= false;
1869 map
.push_back(entry
);
1871 lineptr
+= oplength
;
1875 return lengthstart
+ header_
.total_length
;
1878 // Read the relocations into a Reloc_map.
1880 template<int size
, bool big_endian
>
1882 Sized_dwarf_line_info
<size
, big_endian
>::read_relocs()
1884 if (this->symtab_buffer_
== NULL
)
1889 while ((reloc_offset
= this->reloc_mapper_
->next_offset()) != -1)
1891 const unsigned int shndx
=
1892 this->reloc_mapper_
->get_reloc_target(reloc_offset
, &value
);
1894 // There is no reason to record non-ordinary section indexes, or
1895 // SHN_UNDEF, because they will never match the real section.
1897 this->reloc_map_
[reloc_offset
] = std::make_pair(shndx
, value
);
1899 this->reloc_mapper_
->advance(reloc_offset
+ 1);
1903 // Read the line number info.
1905 template<int size
, bool big_endian
>
1907 Sized_dwarf_line_info
<size
, big_endian
>::read_line_mappings(unsigned int shndx
)
1909 gold_assert(this->data_valid_
== true);
1911 this->read_relocs();
1912 while (this->buffer_
< this->buffer_end_
)
1914 const unsigned char* lineptr
= this->buffer_
;
1915 lineptr
= this->read_header_prolog(lineptr
);
1916 lineptr
= this->read_header_tables(lineptr
);
1917 lineptr
= this->read_lines(lineptr
, shndx
);
1918 this->buffer_
= lineptr
;
1921 // Sort the lines numbers, so addr2line can use binary search.
1922 for (typename
Lineno_map::iterator it
= line_number_map_
.begin();
1923 it
!= line_number_map_
.end();
1925 // Each vector needs to be sorted by offset.
1926 std::sort(it
->second
.begin(), it
->second
.end());
1929 // Some processing depends on whether the input is a .o file or not.
1930 // For instance, .o files have relocs, and have .debug_lines
1931 // information on a per section basis. .so files, on the other hand,
1932 // lack relocs, and offsets are unique, so we can ignore the section
1935 template<int size
, bool big_endian
>
1937 Sized_dwarf_line_info
<size
, big_endian
>::input_is_relobj()
1939 // Only .o files have relocs and the symtab buffer that goes with them.
1940 return this->symtab_buffer_
!= NULL
;
1943 // Given an Offset_to_lineno_entry vector, and an offset, figure out
1944 // if the offset points into a function according to the vector (see
1945 // comments below for the algorithm). If it does, return an iterator
1946 // into the vector that points to the line-number that contains that
1947 // offset. If not, it returns vector::end().
1949 static std::vector
<Offset_to_lineno_entry
>::const_iterator
1950 offset_to_iterator(const std::vector
<Offset_to_lineno_entry
>* offsets
,
1953 const Offset_to_lineno_entry lookup_key
= { offset
, 0, 0, true, 0 };
1955 // lower_bound() returns the smallest offset which is >= lookup_key.
1956 // If no offset in offsets is >= lookup_key, returns end().
1957 std::vector
<Offset_to_lineno_entry
>::const_iterator it
1958 = std::lower_bound(offsets
->begin(), offsets
->end(), lookup_key
);
1960 // This code is easiest to understand with a concrete example.
1961 // Here's a possible offsets array:
1962 // {{offset = 3211, header_num = 0, file_num = 1, last, line_num = 16}, // 0
1963 // {offset = 3224, header_num = 0, file_num = 1, last, line_num = 20}, // 1
1964 // {offset = 3226, header_num = 0, file_num = 1, last, line_num = 22}, // 2
1965 // {offset = 3231, header_num = 0, file_num = 1, last, line_num = 25}, // 3
1966 // {offset = 3232, header_num = 0, file_num = 1, last, line_num = -1}, // 4
1967 // {offset = 3232, header_num = 0, file_num = 1, last, line_num = 65}, // 5
1968 // {offset = 3235, header_num = 0, file_num = 1, last, line_num = 66}, // 6
1969 // {offset = 3236, header_num = 0, file_num = 1, last, line_num = -1}, // 7
1970 // {offset = 5764, header_num = 0, file_num = 1, last, line_num = 48}, // 8
1971 // {offset = 5764, header_num = 0, file_num = 1,!last, line_num = 47}, // 9
1972 // {offset = 5765, header_num = 0, file_num = 1, last, line_num = 49}, // 10
1973 // {offset = 5767, header_num = 0, file_num = 1, last, line_num = 50}, // 11
1974 // {offset = 5768, header_num = 0, file_num = 1, last, line_num = 51}, // 12
1975 // {offset = 5773, header_num = 0, file_num = 1, last, line_num = -1}, // 13
1976 // {offset = 5787, header_num = 1, file_num = 1, last, line_num = 19}, // 14
1977 // {offset = 5790, header_num = 1, file_num = 1, last, line_num = 20}, // 15
1978 // {offset = 5793, header_num = 1, file_num = 1, last, line_num = 67}, // 16
1979 // {offset = 5793, header_num = 1, file_num = 1, last, line_num = -1}, // 17
1980 // {offset = 5793, header_num = 1, file_num = 1,!last, line_num = 66}, // 18
1981 // {offset = 5795, header_num = 1, file_num = 1, last, line_num = 68}, // 19
1982 // {offset = 5798, header_num = 1, file_num = 1, last, line_num = -1}, // 20
1983 // The entries with line_num == -1 mark the end of a function: the
1984 // associated offset is one past the last instruction in the
1985 // function. This can correspond to the beginning of the next
1986 // function (as is true for offset 3232); alternately, there can be
1987 // a gap between the end of one function and the start of the next
1988 // (as is true for some others, most obviously from 3236->5764).
1990 // Case 1: lookup_key has offset == 10. lower_bound returns
1991 // offsets[0]. Since it's not an exact match and we're
1992 // at the beginning of offsets, we return end() (invalid).
1993 // Case 2: lookup_key has offset 10000. lower_bound returns
1994 // offset[21] (end()). We return end() (invalid).
1995 // Case 3: lookup_key has offset == 3211. lower_bound matches
1996 // offsets[0] exactly, and that's the entry we return.
1997 // Case 4: lookup_key has offset == 3232. lower_bound returns
1998 // offsets[4]. That's an exact match, but indicates
1999 // end-of-function. We check if offsets[5] is also an
2000 // exact match but not end-of-function. It is, so we
2001 // return offsets[5].
2002 // Case 5: lookup_key has offset == 3214. lower_bound returns
2003 // offsets[1]. Since it's not an exact match, we back
2004 // up to the offset that's < lookup_key, offsets[0].
2005 // We note offsets[0] is a valid entry (not end-of-function),
2006 // so that's the entry we return.
2007 // Case 6: lookup_key has offset == 4000. lower_bound returns
2008 // offsets[8]. Since it's not an exact match, we back
2009 // up to offsets[7]. Since offsets[7] indicates
2010 // end-of-function, we know lookup_key is between
2011 // functions, so we return end() (not a valid offset).
2012 // Case 7: lookup_key has offset == 5794. lower_bound returns
2013 // offsets[19]. Since it's not an exact match, we back
2014 // up to offsets[16]. Note we back up to the *first*
2015 // entry with offset 5793, not just offsets[19-1].
2016 // We note offsets[16] is a valid entry, so we return it.
2017 // If offsets[16] had had line_num == -1, we would have
2018 // checked offsets[17]. The reason for this is that
2019 // 16 and 17 can be in an arbitrary order, since we sort
2020 // only by offset and last_line_for_offset. (Note it
2021 // doesn't help to use line_number as a tertiary sort key,
2022 // since sometimes we want the -1 to be first and sometimes
2023 // we want it to be last.)
2025 // This deals with cases (1) and (2).
2026 if ((it
== offsets
->begin() && offset
< it
->offset
)
2027 || it
== offsets
->end())
2028 return offsets
->end();
2030 // This deals with cases (3) and (4).
2031 if (offset
== it
->offset
)
2033 while (it
!= offsets
->end()
2034 && it
->offset
== offset
2035 && it
->line_num
== -1)
2037 if (it
== offsets
->end() || it
->offset
!= offset
)
2038 return offsets
->end();
2043 // This handles the first part of case (7) -- we back up to the
2044 // *first* entry that has the offset that's behind us.
2045 gold_assert(it
!= offsets
->begin());
2046 std::vector
<Offset_to_lineno_entry
>::const_iterator range_end
= it
;
2048 const off_t range_value
= it
->offset
;
2049 while (it
!= offsets
->begin() && (it
-1)->offset
== range_value
)
2052 // This handles cases (5), (6), and (7): if any entry in the
2053 // equal_range [it, range_end) has a line_num != -1, it's a valid
2054 // match. If not, we're not in a function. The line number we saw
2055 // last for an offset will be sorted first, so it'll get returned if
2057 for (; it
!= range_end
; ++it
)
2058 if (it
->line_num
!= -1)
2060 return offsets
->end();
2063 // Returns the canonical filename:lineno for the address passed in.
2064 // If other_lines is not NULL, appends the non-canonical lines
2065 // assigned to the same address.
2067 template<int size
, bool big_endian
>
2069 Sized_dwarf_line_info
<size
, big_endian
>::do_addr2line(
2072 std::vector
<std::string
>* other_lines
)
2074 if (this->data_valid_
== false)
2077 const std::vector
<Offset_to_lineno_entry
>* offsets
;
2078 // If we do not have reloc information, then our input is a .so or
2079 // some similar data structure where all the information is held in
2080 // the offset. In that case, we ignore the input shndx.
2081 if (this->input_is_relobj())
2082 offsets
= &this->line_number_map_
[shndx
];
2084 offsets
= &this->line_number_map_
[-1U];
2085 if (offsets
->empty())
2088 typename
std::vector
<Offset_to_lineno_entry
>::const_iterator it
2089 = offset_to_iterator(offsets
, offset
);
2090 if (it
== offsets
->end())
2093 std::string result
= this->format_file_lineno(*it
);
2094 if (other_lines
!= NULL
)
2095 for (++it
; it
!= offsets
->end() && it
->offset
== offset
; ++it
)
2097 if (it
->line_num
== -1)
2098 continue; // The end of a previous function.
2099 other_lines
->push_back(this->format_file_lineno(*it
));
2104 // Convert the file_num + line_num into a string.
2106 template<int size
, bool big_endian
>
2108 Sized_dwarf_line_info
<size
, big_endian
>::format_file_lineno(
2109 const Offset_to_lineno_entry
& loc
) const
2113 gold_assert(loc
.header_num
< static_cast<int>(this->files_
.size()));
2114 gold_assert(loc
.file_num
2115 < static_cast<unsigned int>(this->files_
[loc
.header_num
].size()));
2116 const std::pair
<int, std::string
>& filename_pair
2117 = this->files_
[loc
.header_num
][loc
.file_num
];
2118 const std::string
& filename
= filename_pair
.second
;
2120 gold_assert(loc
.header_num
< static_cast<int>(this->directories_
.size()));
2121 gold_assert(filename_pair
.first
2122 < static_cast<int>(this->directories_
[loc
.header_num
].size()));
2123 const std::string
& dirname
2124 = this->directories_
[loc
.header_num
][filename_pair
.first
];
2126 if (!dirname
.empty())
2135 char buffer
[64]; // enough to hold a line number
2136 snprintf(buffer
, sizeof(buffer
), "%d", loc
.line_num
);
2143 // Dwarf_line_info routines.
2145 static unsigned int next_generation_count
= 0;
2147 struct Addr2line_cache_entry
2151 Dwarf_line_info
* dwarf_line_info
;
2152 unsigned int generation_count
;
2153 unsigned int access_count
;
2155 Addr2line_cache_entry(Object
* o
, unsigned int s
, Dwarf_line_info
* d
)
2156 : object(o
), shndx(s
), dwarf_line_info(d
),
2157 generation_count(next_generation_count
), access_count(0)
2159 if (next_generation_count
< (1U << 31))
2160 ++next_generation_count
;
2163 // We expect this cache to be small, so don't bother with a hashtable
2164 // or priority queue or anything: just use a simple vector.
2165 static std::vector
<Addr2line_cache_entry
> addr2line_cache
;
2168 Dwarf_line_info::one_addr2line(Object
* object
,
2169 unsigned int shndx
, off_t offset
,
2171 std::vector
<std::string
>* other_lines
)
2173 Dwarf_line_info
* lineinfo
= NULL
;
2174 std::vector
<Addr2line_cache_entry
>::iterator it
;
2176 // First, check the cache. If we hit, update the counts.
2177 for (it
= addr2line_cache
.begin(); it
!= addr2line_cache
.end(); ++it
)
2179 if (it
->object
== object
&& it
->shndx
== shndx
)
2181 lineinfo
= it
->dwarf_line_info
;
2182 it
->generation_count
= next_generation_count
;
2183 // We cap generation_count at 2^31 -1 to avoid overflow.
2184 if (next_generation_count
< (1U << 31))
2185 ++next_generation_count
;
2186 // We cap access_count at 31 so 2^access_count doesn't overflow
2187 if (it
->access_count
< 31)
2193 // If we don't hit the cache, create a new object and insert into the
2195 if (lineinfo
== NULL
)
2197 switch (parameters
->size_and_endianness())
2199 #ifdef HAVE_TARGET_32_LITTLE
2200 case Parameters::TARGET_32_LITTLE
:
2201 lineinfo
= new Sized_dwarf_line_info
<32, false>(object
, shndx
); break;
2203 #ifdef HAVE_TARGET_32_BIG
2204 case Parameters::TARGET_32_BIG
:
2205 lineinfo
= new Sized_dwarf_line_info
<32, true>(object
, shndx
); break;
2207 #ifdef HAVE_TARGET_64_LITTLE
2208 case Parameters::TARGET_64_LITTLE
:
2209 lineinfo
= new Sized_dwarf_line_info
<64, false>(object
, shndx
); break;
2211 #ifdef HAVE_TARGET_64_BIG
2212 case Parameters::TARGET_64_BIG
:
2213 lineinfo
= new Sized_dwarf_line_info
<64, true>(object
, shndx
); break;
2218 addr2line_cache
.push_back(Addr2line_cache_entry(object
, shndx
, lineinfo
));
2221 // Now that we have our object, figure out the answer
2222 std::string retval
= lineinfo
->addr2line(shndx
, offset
, other_lines
);
2224 // Finally, if our cache has grown too big, delete old objects. We
2225 // assume the common (probably only) case is deleting only one object.
2226 // We use a pretty simple scheme to evict: function of LRU and MFU.
2227 while (addr2line_cache
.size() > cache_size
)
2229 unsigned int lowest_score
= ~0U;
2230 std::vector
<Addr2line_cache_entry
>::iterator lowest
2231 = addr2line_cache
.end();
2232 for (it
= addr2line_cache
.begin(); it
!= addr2line_cache
.end(); ++it
)
2234 const unsigned int score
= (it
->generation_count
2235 + (1U << it
->access_count
));
2236 if (score
< lowest_score
)
2238 lowest_score
= score
;
2242 if (lowest
!= addr2line_cache
.end())
2244 delete lowest
->dwarf_line_info
;
2245 addr2line_cache
.erase(lowest
);
2253 Dwarf_line_info::clear_addr2line_cache()
2255 for (std::vector
<Addr2line_cache_entry
>::iterator it
= addr2line_cache
.begin();
2256 it
!= addr2line_cache
.end();
2258 delete it
->dwarf_line_info
;
2259 addr2line_cache
.clear();
2262 #ifdef HAVE_TARGET_32_LITTLE
2264 class Sized_dwarf_line_info
<32, false>;
2267 #ifdef HAVE_TARGET_32_BIG
2269 class Sized_dwarf_line_info
<32, true>;
2272 #ifdef HAVE_TARGET_64_LITTLE
2274 class Sized_dwarf_line_info
<64, false>;
2277 #ifdef HAVE_TARGET_64_BIG
2279 class Sized_dwarf_line_info
<64, true>;
2282 } // End namespace gold.