Merge remote-tracking branch 'origin/vendor/LIBEDIT'
[dragonfly.git] / contrib / binutils-2.25 / gold / gdb-index.cc
1 // gdb-index.cc -- generate .gdb_index section for fast debug lookup
2
3 // Copyright (C) 2012-2014 Free Software Foundation, Inc.
4 // Written by Cary Coutant <ccoutant@google.com>.
5
6 // This file is part of gold.
7
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.
12
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.
17
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.
22
23 #include "gold.h"
24
25 #include "gdb-index.h"
26 #include "dwarf_reader.h"
27 #include "dwarf.h"
28 #include "object.h"
29 #include "output.h"
30 #include "demangle.h"
31
32 namespace gold
33 {
34
35 const int gdb_index_version = 7;
36
37 // Sizes of various records in the .gdb_index section.
38 const int gdb_index_offset_size = 4;
39 const int gdb_index_hdr_size = 6 * gdb_index_offset_size;
40 const int gdb_index_cu_size = 16;
41 const int gdb_index_tu_size = 24;
42 const int gdb_index_addr_size = 16 + gdb_index_offset_size;
43 const int gdb_index_sym_size = 2 * gdb_index_offset_size;
44
45 // This class manages the hashed symbol table for the .gdb_index section.
46 // It is essentially equivalent to the hashtab implementation in libiberty,
47 // but is copied into gdb sources and here for compatibility because its
48 // data structure is exposed on disk.
49
50 template <typename T>
51 class Gdb_hashtab
52 {
53  public:
54   Gdb_hashtab()
55     : size_(0), capacity_(0), hashtab_(NULL)
56   { }
57
58   ~Gdb_hashtab()
59   {
60     for (size_t i = 0; i < this->capacity_; ++i)
61       if (this->hashtab_[i] != NULL)
62         delete this->hashtab_[i];
63     delete[] this->hashtab_;
64   }
65
66   // Add a symbol.
67   T*
68   add(T* symbol)
69   {
70     // Resize the hash table if necessary.
71     if (4 * this->size_ / 3 >= this->capacity_)
72       this->expand();
73
74     T** slot = this->find_slot(symbol);
75     if (*slot == NULL)
76       {
77         ++this->size_;
78         *slot = symbol;
79       }
80
81     return *slot;
82   }
83
84   // Return the current size.
85   size_t
86   size() const
87   { return this->size_; }
88
89   // Return the current capacity.
90   size_t
91   capacity() const
92   { return this->capacity_; }
93
94   // Return the contents of slot N.
95   T*
96   operator[](size_t n)
97   { return this->hashtab_[n]; }
98
99  private:
100   // Find a symbol in the hash table, or return an empty slot if
101   // the symbol is not in the table.
102   T**
103   find_slot(T* symbol)
104   {
105     unsigned int index = symbol->hash() & (this->capacity_ - 1);
106     unsigned int step = ((symbol->hash() * 17) & (this->capacity_ - 1)) | 1;
107
108     for (;;)
109       {
110         if (this->hashtab_[index] == NULL
111             || this->hashtab_[index]->equal(symbol))
112           return &this->hashtab_[index];
113         index = (index + step) & (this->capacity_ - 1);
114       }
115   }
116
117   // Expand the hash table.
118   void
119   expand()
120   {
121     if (this->capacity_ == 0)
122       {
123         // Allocate the hash table for the first time.
124         this->capacity_ = Gdb_hashtab::initial_size;
125         this->hashtab_ = new T*[this->capacity_];
126         memset(this->hashtab_, 0, this->capacity_ * sizeof(T*));
127       }
128     else
129       {
130         // Expand and rehash.
131         unsigned int old_cap = this->capacity_;
132         T** old_hashtab = this->hashtab_;
133         this->capacity_ *= 2;
134         this->hashtab_ = new T*[this->capacity_];
135         memset(this->hashtab_, 0, this->capacity_ * sizeof(T*));
136         for (size_t i = 0; i < old_cap; ++i)
137           {
138             if (old_hashtab[i] != NULL)
139               {
140                 T** slot = this->find_slot(old_hashtab[i]);
141                 *slot = old_hashtab[i];
142               }
143           }
144         delete[] old_hashtab;
145       }
146   }
147
148   // Initial size of the hash table; must be a power of 2.
149   static const int initial_size = 1024;
150   size_t size_;
151   size_t capacity_;
152   T** hashtab_;
153 };
154
155 // The hash function for strings in the mapped index.  This is copied
156 // directly from gdb/dwarf2read.c.
157
158 static unsigned int
159 mapped_index_string_hash(const unsigned char* str)
160 {
161   unsigned int r = 0;
162   unsigned char c;
163
164   while ((c = *str++) != 0)
165     {
166       if (gdb_index_version >= 5)
167         c = tolower (c);
168       r = r * 67 + c - 113;
169     }
170
171   return r;
172 }
173
174 // A specialization of Dwarf_info_reader, for building the .gdb_index.
175
176 class Gdb_index_info_reader : public Dwarf_info_reader
177 {
178  public:
179   Gdb_index_info_reader(bool is_type_unit,
180                         Relobj* object,
181                         const unsigned char* symbols,
182                         off_t symbols_size,
183                         unsigned int shndx,
184                         unsigned int reloc_shndx,
185                         unsigned int reloc_type,
186                         Gdb_index* gdb_index)
187     : Dwarf_info_reader(is_type_unit, object, symbols, symbols_size, shndx,
188                         reloc_shndx, reloc_type),
189       gdb_index_(gdb_index), cu_index_(0), cu_language_(0)
190   { }
191
192   ~Gdb_index_info_reader()
193   { this->clear_declarations(); }
194
195   // Print usage statistics.
196   static void
197   print_stats();
198
199  protected:
200   // Visit a compilation unit.
201   virtual void
202   visit_compilation_unit(off_t cu_offset, off_t cu_length, Dwarf_die*);
203
204   // Visit a type unit.
205   virtual void
206   visit_type_unit(off_t tu_offset, off_t tu_length, off_t type_offset,
207                   uint64_t signature, Dwarf_die*);
208
209  private:
210   // A map for recording DIEs we've seen that may be referred to be
211   // later DIEs (via DW_AT_specification or DW_AT_abstract_origin).
212   // The map is indexed by a DIE offset within the compile unit.
213   // PARENT_OFFSET_ is the offset of the DIE that represents the
214   // outer context, and NAME_ is a pointer to a component of the
215   // fully-qualified name.
216   // Normally, the names we point to are in a string table, so we don't
217   // have to manage them, but when we have a fully-qualified name
218   // computed, we put it in the table, and set PARENT_OFFSET_ to -1
219   // indicate a string that we are managing.
220   struct Declaration_pair
221   {
222     Declaration_pair(off_t parent_offset, const char* name)
223       : parent_offset_(parent_offset), name_(name)
224     { }
225
226     off_t parent_offset_;
227     const char* name_; 
228   };
229   typedef Unordered_map<off_t, Declaration_pair> Declaration_map;
230
231   // Visit a top-level DIE.
232   void
233   visit_top_die(Dwarf_die* die);
234
235   // Visit the children of a DIE.
236   void
237   visit_children(Dwarf_die* die, Dwarf_die* context);
238
239   // Visit a DIE.
240   void
241   visit_die(Dwarf_die* die, Dwarf_die* context);
242
243   // Visit the children of a DIE.
244   void
245   visit_children_for_decls(Dwarf_die* die);
246
247   // Visit a DIE.
248   void
249   visit_die_for_decls(Dwarf_die* die, Dwarf_die* context);
250
251   // Guess a fully-qualified name for a class type, based on member function
252   // linkage names.
253   std::string
254   guess_full_class_name(Dwarf_die* die);
255
256   // Add a declaration DIE to the table of declarations.
257   void
258   add_declaration(Dwarf_die* die, Dwarf_die* context);
259
260   // Add a declaration whose fully-qualified name is already known.
261   void
262   add_declaration_with_full_name(Dwarf_die* die, const char* full_name);
263
264   // Return the context for a DIE whose parent is at DIE_OFFSET.
265   std::string
266   get_context(off_t die_offset);
267
268   // Construct a fully-qualified name for DIE.
269   std::string
270   get_qualified_name(Dwarf_die* die, Dwarf_die* context);
271
272   // Record the address ranges for a compilation unit.
273   void
274   record_cu_ranges(Dwarf_die* die);
275
276   // Wrapper for read_pubtable.
277   bool
278   read_pubnames_and_pubtypes(Dwarf_die* die);
279
280   // Read the .debug_pubnames and .debug_pubtypes tables.
281   bool
282   read_pubtable(Dwarf_pubnames_table* table, off_t offset);
283
284   // Clear the declarations map.
285   void
286   clear_declarations();
287
288   // The Gdb_index section.
289   Gdb_index* gdb_index_;
290   // The current CU index (negative for a TU).
291   int cu_index_;
292   // The language of the current CU or TU.
293   unsigned int cu_language_;
294   // Map from DIE offset to (parent offset, name) pair,
295   // for DW_AT_specification.
296   Declaration_map declarations_;
297
298   // Statistics.
299   // Total number of DWARF compilation units processed.
300   static unsigned int dwarf_cu_count;
301   // Number of DWARF compilation units with pubnames/pubtypes.
302   static unsigned int dwarf_cu_nopubnames_count;
303   // Total number of DWARF type units processed.
304   static unsigned int dwarf_tu_count;
305   // Number of DWARF type units with pubnames/pubtypes.
306   static unsigned int dwarf_tu_nopubnames_count;
307 };
308
309 // Total number of DWARF compilation units processed.
310 unsigned int Gdb_index_info_reader::dwarf_cu_count = 0;
311 // Number of DWARF compilation units without pubnames/pubtypes.
312 unsigned int Gdb_index_info_reader::dwarf_cu_nopubnames_count = 0;
313 // Total number of DWARF type units processed.
314 unsigned int Gdb_index_info_reader::dwarf_tu_count = 0;
315 // Number of DWARF type units without pubnames/pubtypes.
316 unsigned int Gdb_index_info_reader::dwarf_tu_nopubnames_count = 0;
317
318 // Process a compilation unit and parse its child DIE.
319
320 void
321 Gdb_index_info_reader::visit_compilation_unit(off_t cu_offset, off_t cu_length,
322                                               Dwarf_die* root_die)
323 {
324   ++Gdb_index_info_reader::dwarf_cu_count;
325   this->cu_index_ = this->gdb_index_->add_comp_unit(cu_offset, cu_length);
326   this->visit_top_die(root_die);
327 }
328
329 // Process a type unit and parse its child DIE.
330
331 void
332 Gdb_index_info_reader::visit_type_unit(off_t tu_offset, off_t,
333                                        off_t type_offset, uint64_t signature,
334                                        Dwarf_die* root_die)
335 {
336   ++Gdb_index_info_reader::dwarf_tu_count;
337   // Use a negative index to flag this as a TU instead of a CU.
338   this->cu_index_ = -1 - this->gdb_index_->add_type_unit(tu_offset, type_offset,
339                                                          signature);
340   this->visit_top_die(root_die);
341 }
342
343 // Process a top-level DIE.
344 // For compile_unit DIEs, record the address ranges.  For all
345 // interesting tags, add qualified names to the symbol table
346 // and process interesting children.  We may need to process
347 // certain children just for saving declarations that might be
348 // referenced by later DIEs with a DW_AT_specification attribute.
349
350 void
351 Gdb_index_info_reader::visit_top_die(Dwarf_die* die)
352 {
353   this->clear_declarations();
354
355   switch (die->tag())
356     {
357       case elfcpp::DW_TAG_compile_unit:
358       case elfcpp::DW_TAG_type_unit:
359         this->cu_language_ = die->int_attribute(elfcpp::DW_AT_language);
360         if (die->tag() == elfcpp::DW_TAG_compile_unit)
361           this->record_cu_ranges(die);
362         // If there is a pubnames and/or pubtypes section for this
363         // compilation unit, use those; otherwise, parse the DWARF
364         // info to extract the names.
365         if (!this->read_pubnames_and_pubtypes(die))
366           {
367             // Check for languages that require specialized knowledge to
368             // construct fully-qualified names, that we don't yet support.
369             if (this->cu_language_ == elfcpp::DW_LANG_Ada83
370                 || this->cu_language_ == elfcpp::DW_LANG_Fortran77
371                 || this->cu_language_ == elfcpp::DW_LANG_Fortran90
372                 || this->cu_language_ == elfcpp::DW_LANG_Java
373                 || this->cu_language_ == elfcpp::DW_LANG_Ada95
374                 || this->cu_language_ == elfcpp::DW_LANG_Fortran95)
375               {
376                 gold_warning(_("%s: --gdb-index currently supports "
377                                "only C and C++ languages"),
378                              this->object()->name().c_str());
379                 return;
380               }
381             if (die->tag() == elfcpp::DW_TAG_compile_unit)
382               ++Gdb_index_info_reader::dwarf_cu_nopubnames_count;
383             else
384               ++Gdb_index_info_reader::dwarf_tu_nopubnames_count;
385             this->visit_children(die, NULL);
386           }
387         break;
388       default:
389         // The top level DIE should be one of the above.
390         gold_warning(_("%s: top level DIE is not DW_TAG_compile_unit "
391                        "or DW_TAG_type_unit"),
392                      this->object()->name().c_str());
393         return;
394     }
395 }
396
397 // Visit the children of PARENT, looking for symbols to add to the index.
398 // CONTEXT points to the DIE to use for constructing the qualified name --
399 // NULL if PARENT is the top-level DIE; otherwise it is the same as PARENT.
400
401 void
402 Gdb_index_info_reader::visit_children(Dwarf_die* parent, Dwarf_die* context)
403 {
404   off_t next_offset = 0;
405   for (off_t die_offset = parent->child_offset();
406        die_offset != 0;
407        die_offset = next_offset)
408     {
409       Dwarf_die die(this, die_offset, parent);
410       if (die.tag() == 0)
411         break;
412       this->visit_die(&die, context);
413       next_offset = die.sibling_offset();
414     }
415 }
416
417 // Visit a child DIE, looking for symbols to add to the index.
418 // CONTEXT is the parent DIE, used for constructing the qualified name;
419 // it is NULL if the parent DIE is the top-level DIE.
420
421 void
422 Gdb_index_info_reader::visit_die(Dwarf_die* die, Dwarf_die* context)
423 {
424   switch (die->tag())
425     {
426       case elfcpp::DW_TAG_subprogram:
427       case elfcpp::DW_TAG_constant:
428       case elfcpp::DW_TAG_variable:
429       case elfcpp::DW_TAG_enumerator:
430       case elfcpp::DW_TAG_base_type:
431         if (die->is_declaration())
432           this->add_declaration(die, context);
433         else
434           {
435             // If the DIE is not a declaration, add it to the index.
436             std::string full_name = this->get_qualified_name(die, context);
437             if (!full_name.empty())
438               this->gdb_index_->add_symbol(this->cu_index_,
439                                            full_name.c_str(), 0);
440           }
441         break;
442       case elfcpp::DW_TAG_typedef:
443       case elfcpp::DW_TAG_union_type:
444       case elfcpp::DW_TAG_class_type:
445       case elfcpp::DW_TAG_interface_type:
446       case elfcpp::DW_TAG_structure_type:
447       case elfcpp::DW_TAG_enumeration_type:
448       case elfcpp::DW_TAG_subrange_type:
449       case elfcpp::DW_TAG_namespace:
450         {
451           std::string full_name;
452           
453           // For classes at the top level, we need to look for a
454           // member function with a linkage name in order to get
455           // the properly-canonicalized name.
456           if (context == NULL
457               && (die->tag() == elfcpp::DW_TAG_class_type
458                   || die->tag() == elfcpp::DW_TAG_structure_type
459                   || die->tag() == elfcpp::DW_TAG_union_type))
460             full_name.assign(this->guess_full_class_name(die));
461
462           // Because we will visit the children, we need to add this DIE
463           // to the declarations table.
464           if (full_name.empty())
465             this->add_declaration(die, context);
466           else
467             this->add_declaration_with_full_name(die, full_name.c_str());
468
469           // If the DIE is not a declaration, add it to the index.
470           // Gdb stores a namespace in the index even when it is
471           // a declaration.
472           if (die->tag() == elfcpp::DW_TAG_namespace
473               || !die->is_declaration())
474             {
475               if (full_name.empty())
476                 full_name = this->get_qualified_name(die, context);
477               if (!full_name.empty())
478                 this->gdb_index_->add_symbol(this->cu_index_,
479                                              full_name.c_str(), 0);
480             }
481
482           // We're interested in the children only for namespaces and
483           // enumeration types.  For enumeration types, we do not include
484           // the enumeration tag as part of the full name.  For other tags,
485           // visit the children only to collect declarations.
486           if (die->tag() == elfcpp::DW_TAG_namespace
487               || die->tag() == elfcpp::DW_TAG_enumeration_type)
488             this->visit_children(die, die);
489           else
490             this->visit_children_for_decls(die);
491         }
492         break;
493       default:
494         break;
495     }
496 }
497
498 // Visit the children of PARENT, looking only for declarations that
499 // may be referenced by later specification DIEs.
500
501 void
502 Gdb_index_info_reader::visit_children_for_decls(Dwarf_die* parent)
503 {
504   off_t next_offset = 0;
505   for (off_t die_offset = parent->child_offset();
506        die_offset != 0;
507        die_offset = next_offset)
508     {
509       Dwarf_die die(this, die_offset, parent);
510       if (die.tag() == 0)
511         break;
512       this->visit_die_for_decls(&die, parent);
513       next_offset = die.sibling_offset();
514     }
515 }
516
517 // Visit a child DIE, looking only for declarations that
518 // may be referenced by later specification DIEs.
519
520 void
521 Gdb_index_info_reader::visit_die_for_decls(Dwarf_die* die, Dwarf_die* context)
522 {
523   switch (die->tag())
524     {
525       case elfcpp::DW_TAG_subprogram:
526       case elfcpp::DW_TAG_constant:
527       case elfcpp::DW_TAG_variable:
528       case elfcpp::DW_TAG_enumerator:
529       case elfcpp::DW_TAG_base_type:
530         {
531           if (die->is_declaration())
532             this->add_declaration(die, context);
533         }
534         break;
535       case elfcpp::DW_TAG_typedef:
536       case elfcpp::DW_TAG_union_type:
537       case elfcpp::DW_TAG_class_type:
538       case elfcpp::DW_TAG_interface_type:
539       case elfcpp::DW_TAG_structure_type:
540       case elfcpp::DW_TAG_enumeration_type:
541       case elfcpp::DW_TAG_subrange_type:
542       case elfcpp::DW_TAG_namespace:
543         {
544           if (die->is_declaration())
545             this->add_declaration(die, context);
546           this->visit_children_for_decls(die);
547         }
548         break;
549       default:
550         break;
551     }
552 }
553
554 // Extract the class name from the linkage name of a member function.
555 // This code is adapted from ../gdb/cp-support.c.
556
557 #define d_left(dc) (dc)->u.s_binary.left
558 #define d_right(dc) (dc)->u.s_binary.right
559
560 static char*
561 class_name_from_linkage_name(const char* linkage_name)
562 {
563   void* storage;
564   struct demangle_component* tree =
565       cplus_demangle_v3_components(linkage_name, DMGL_NO_OPTS, &storage);
566   if (tree == NULL)
567     return NULL;
568
569   int done = 0;
570
571   // First strip off any qualifiers, if we have a function or
572   // method.
573   while (!done)
574     switch (tree->type)
575       {
576         case DEMANGLE_COMPONENT_CONST:
577         case DEMANGLE_COMPONENT_RESTRICT:
578         case DEMANGLE_COMPONENT_VOLATILE:
579         case DEMANGLE_COMPONENT_CONST_THIS:
580         case DEMANGLE_COMPONENT_RESTRICT_THIS:
581         case DEMANGLE_COMPONENT_VOLATILE_THIS:
582         case DEMANGLE_COMPONENT_VENDOR_TYPE_QUAL:
583           tree = d_left(tree);
584           break;
585         default:
586           done = 1;
587           break;
588       }
589
590   // If what we have now is a function, discard the argument list.
591   if (tree->type == DEMANGLE_COMPONENT_TYPED_NAME)
592     tree = d_left(tree);
593
594   // If what we have now is a template, strip off the template
595   // arguments.  The left subtree may be a qualified name.
596   if (tree->type == DEMANGLE_COMPONENT_TEMPLATE)
597     tree = d_left(tree);
598
599   // What we have now should be a name, possibly qualified.
600   // Additional qualifiers could live in the left subtree or the right
601   // subtree.  Find the last piece.
602   done = 0;
603   struct demangle_component* prev_comp = NULL;
604   struct demangle_component* cur_comp = tree;
605   while (!done)
606     switch (cur_comp->type)
607       {
608         case DEMANGLE_COMPONENT_QUAL_NAME:
609         case DEMANGLE_COMPONENT_LOCAL_NAME:
610           prev_comp = cur_comp;
611           cur_comp = d_right(cur_comp);
612           break;
613         case DEMANGLE_COMPONENT_TEMPLATE:
614         case DEMANGLE_COMPONENT_NAME:
615         case DEMANGLE_COMPONENT_CTOR:
616         case DEMANGLE_COMPONENT_DTOR:
617         case DEMANGLE_COMPONENT_OPERATOR:
618         case DEMANGLE_COMPONENT_EXTENDED_OPERATOR:
619           done = 1;
620           break;
621         default:
622           done = 1;
623           cur_comp = NULL;
624           break;
625       }
626
627   char* ret = NULL;
628   if (cur_comp != NULL && prev_comp != NULL)
629     {
630       // We want to discard the rightmost child of PREV_COMP.
631       *prev_comp = *d_left(prev_comp);
632       size_t allocated_size;
633       ret = cplus_demangle_print(DMGL_NO_OPTS, tree, 30, &allocated_size);
634     }
635
636   free(storage);
637   return ret;
638 }
639
640 // Guess a fully-qualified name for a class type, based on member function
641 // linkage names.  This is needed for class/struct/union types at the
642 // top level, because GCC does not always properly embed them within
643 // the namespace.  As in gdb, we look for a member function with a linkage
644 // name and extract the qualified name from the demangled name.
645
646 std::string
647 Gdb_index_info_reader::guess_full_class_name(Dwarf_die* die)
648 {
649   std::string full_name;
650   off_t next_offset = 0;
651   
652   // This routine scans ahead in the DIE structure, possibly advancing
653   // the relocation tracker beyond the current DIE.  We need to checkpoint
654   // the tracker and reset it when we're done.
655   uint64_t checkpoint = this->get_reloc_checkpoint();
656
657   for (off_t child_offset = die->child_offset();
658        child_offset != 0;
659        child_offset = next_offset)
660     {
661       Dwarf_die child(this, child_offset, die);
662       if (child.tag() == 0)
663         break;
664       if (child.tag() == elfcpp::DW_TAG_subprogram)
665         {
666           const char* linkage_name = child.linkage_name();
667           if (linkage_name != NULL)
668             {
669               char* guess = class_name_from_linkage_name(linkage_name);
670               if (guess != NULL)
671                 {
672                   full_name.assign(guess);
673                   free(guess);
674                   break;
675                 }
676             }
677         }
678       next_offset = child.sibling_offset();
679     }
680
681   this->reset_relocs(checkpoint);
682   return full_name;
683 }
684
685 // Add a declaration DIE to the table of declarations.
686
687 void
688 Gdb_index_info_reader::add_declaration(Dwarf_die* die, Dwarf_die* context)
689 {
690   const char* name = die->name();
691
692   off_t parent_offset = context != NULL ? context->offset() : 0;
693
694   // If this DIE has a DW_AT_specification or DW_AT_abstract_origin
695   // attribute, use the parent and name from the earlier declaration.
696   off_t spec = die->specification();
697   if (spec == 0)
698     spec = die->abstract_origin();
699   if (spec > 0)
700     {
701       Declaration_map::iterator it = this->declarations_.find(spec);
702       if (it != this->declarations_.end())
703         {
704           parent_offset = it->second.parent_offset_;
705           name = it->second.name_;
706         }
707     }
708
709   if (name == NULL)
710     {
711       if (die->tag() == elfcpp::DW_TAG_namespace)
712         name = "(anonymous namespace)";
713       else if (die->tag() == elfcpp::DW_TAG_union_type)
714         name = "(anonymous union)";
715       else
716         name = "(unknown)";
717     }
718
719   Declaration_pair decl(parent_offset, name);
720   this->declarations_.insert(std::make_pair(die->offset(), decl));
721 }
722
723 // Add a declaration whose fully-qualified name is already known.
724 // In the case where we had to get the canonical name by demangling
725 // a linkage name, this ensures we use that name instead of the one
726 // provided in DW_AT_name.
727
728 void
729 Gdb_index_info_reader::add_declaration_with_full_name(
730     Dwarf_die* die,
731     const char* full_name)
732 {
733   // We need to copy the name.
734   int len = strlen(full_name);
735   char* copy = new char[len + 1];
736   memcpy(copy, full_name, len + 1);
737
738   // Flag that we now manage the memory this points to.
739   Declaration_pair decl(-1, copy);
740   this->declarations_.insert(std::make_pair(die->offset(), decl));
741 }
742
743 // Return the context for a DIE whose parent is at DIE_OFFSET.
744
745 std::string
746 Gdb_index_info_reader::get_context(off_t die_offset)
747 {
748   std::string context;
749   Declaration_map::iterator it = this->declarations_.find(die_offset);
750   if (it != this->declarations_.end())
751     {
752       off_t parent_offset = it->second.parent_offset_;
753       if (parent_offset > 0)
754         {
755           context = get_context(parent_offset);
756           context.append("::");
757         }
758       if (it->second.name_ != NULL)
759         context.append(it->second.name_);
760     }
761   return context;
762 }
763
764 // Construct the fully-qualified name for DIE.
765
766 std::string
767 Gdb_index_info_reader::get_qualified_name(Dwarf_die* die, Dwarf_die* context)
768 {
769   std::string full_name;
770   const char* name = die->name();
771
772   off_t parent_offset = context != NULL ? context->offset() : 0;
773
774   // If this DIE has a DW_AT_specification or DW_AT_abstract_origin
775   // attribute, use the parent and name from the earlier declaration.
776   off_t spec = die->specification();
777   if (spec == 0)
778     spec = die->abstract_origin();
779   if (spec > 0)
780     {
781       Declaration_map::iterator it = this->declarations_.find(spec);
782       if (it != this->declarations_.end())
783         {
784           parent_offset = it->second.parent_offset_;
785           name = it->second.name_;
786         }
787     }
788
789   if (name == NULL && die->tag() == elfcpp::DW_TAG_namespace)
790     name = "(anonymous namespace)";
791   else if (name == NULL)
792     return full_name;
793
794   // If this is an enumerator constant, skip the immediate parent,
795   // which is the enumeration tag.
796   if (die->tag() == elfcpp::DW_TAG_enumerator)
797     {
798       Declaration_map::iterator it = this->declarations_.find(parent_offset);
799       if (it != this->declarations_.end())
800         parent_offset = it->second.parent_offset_;
801     }
802
803   if (parent_offset > 0)
804     {
805       full_name.assign(this->get_context(parent_offset));
806       full_name.append("::");
807     }
808   full_name.append(name);
809
810   return full_name;
811 }
812
813 // Record the address ranges for a compilation unit.
814
815 void
816 Gdb_index_info_reader::record_cu_ranges(Dwarf_die* die)
817 {
818   unsigned int shndx;
819   unsigned int shndx2;
820
821   off_t ranges_offset = die->ref_attribute(elfcpp::DW_AT_ranges, &shndx);
822   if (ranges_offset != -1)
823     {
824       Dwarf_range_list* ranges = this->read_range_list(shndx, ranges_offset);
825       if (ranges != NULL)
826         this->gdb_index_->add_address_range_list(this->object(),
827                                                  this->cu_index_, ranges);
828       return;
829     }
830
831   off_t low_pc = die->address_attribute(elfcpp::DW_AT_low_pc, &shndx);
832   off_t high_pc = die->address_attribute(elfcpp::DW_AT_high_pc, &shndx2);
833   if (high_pc == -1)
834     {
835       high_pc = die->uint_attribute(elfcpp::DW_AT_high_pc);
836       high_pc += low_pc;
837       shndx2 = shndx;
838     }
839   if ((low_pc != 0 || high_pc != 0) && low_pc != -1)
840     {
841       if (shndx != shndx2)
842         {
843           gold_warning(_("%s: DWARF info may be corrupt; low_pc and high_pc "
844                          "are in different sections"),
845                        this->object()->name().c_str());
846           return;
847         }
848       if (shndx == 0 || this->object()->is_section_included(shndx))
849         {
850           Dwarf_range_list* ranges = new Dwarf_range_list();
851           ranges->add(shndx, low_pc, high_pc);
852           this->gdb_index_->add_address_range_list(this->object(),
853                                                    this->cu_index_, ranges);
854         }
855     }
856 }
857
858 // Read table and add the relevant names to the index.  Returns true
859 // if any names were added.
860
861 bool
862 Gdb_index_info_reader::read_pubtable(Dwarf_pubnames_table* table, off_t offset)
863 {
864   // If we couldn't read the section when building the cu_pubname_map,
865   // then we won't find any pubnames now.
866   if (table == NULL)
867     return false;
868
869   if (!table->read_header(offset))
870     return false;
871   while (true)
872     {
873       uint8_t flag_byte;
874       const char* name = table->next_name(&flag_byte);
875       if (name == NULL)
876         break;
877
878       this->gdb_index_->add_symbol(this->cu_index_, name, flag_byte);
879     }
880   return true;
881 }
882
883 // Read the .debug_pubnames and .debug_pubtypes tables for the CU or TU.
884 // Returns TRUE if either a pubnames or pubtypes section was found.
885
886 bool
887 Gdb_index_info_reader::read_pubnames_and_pubtypes(Dwarf_die* die)
888 {
889   // If this is a skeleton debug-type die (generated via
890   // -gsplit-dwarf), then the associated pubnames should have been
891   // read along with the corresponding CU.  In any case, there isn't
892   // enough info inside to build a gdb index entry.
893   if (die->tag() == elfcpp::DW_TAG_type_unit
894       && die->string_attribute(elfcpp::DW_AT_GNU_dwo_name))
895     return true;
896
897   // We use stmt_list_off as a unique identifier for the
898   // compilation unit and its associated type units.
899   unsigned int shndx;
900   off_t stmt_list_off = die->ref_attribute (elfcpp::DW_AT_stmt_list,
901                                             &shndx);
902   // Look for the attr as either a flag or a ref.
903   off_t offset = die->ref_attribute(elfcpp::DW_AT_GNU_pubnames, &shndx);
904
905   // Newer versions of GCC generate CUs, but not TUs, with
906   // DW_AT_FORM_flag_present.
907   unsigned int flag = die->uint_attribute(elfcpp::DW_AT_GNU_pubnames);
908   if (offset == -1 && flag == 0)
909     {
910       // Didn't find the attribute.
911       if (die->tag() == elfcpp::DW_TAG_type_unit)
912         {
913           // If die is a TU, then it might correspond to a CU which we
914           // have read. If it does, then no need to read the pubnames.
915           // If it doesn't, then the caller will have to parse the
916           // dies manually to find the names.
917           return this->gdb_index_->pubnames_read(this->object(),
918                                                  stmt_list_off);
919         }
920       else
921         {
922           // No attribute on the CU means that no pubnames were read.
923           return false;
924         }
925     }
926
927   // We found the attribute, so we can check if the corresponding
928   // pubnames have been read.
929   if (this->gdb_index_->pubnames_read(this->object(), stmt_list_off))
930     return true;
931
932   this->gdb_index_->set_pubnames_read(this->object(), stmt_list_off);
933
934   // We have an attribute, and the pubnames haven't been read, so read
935   // them.
936   bool names = false;
937   // In some of the cases, we could rely on the previous value of
938   // offset here, but sorting out which cases complicates the logic
939   // enough that it isn't worth it. So just look up the offset again.
940   offset = this->gdb_index_->find_pubname_offset(this->cu_offset());
941   names = this->read_pubtable(this->gdb_index_->pubnames_table(), offset);
942
943   bool types = false;
944   offset = this->gdb_index_->find_pubtype_offset(this->cu_offset());
945   types = this->read_pubtable(this->gdb_index_->pubtypes_table(), offset);
946   return names || types;
947 }
948
949 // Clear the declarations map.
950 void
951 Gdb_index_info_reader::clear_declarations()
952 {
953   // Free strings in memory we manage.
954   for (Declaration_map::iterator it = this->declarations_.begin();
955        it != this->declarations_.end();
956        ++it)
957     {
958       if (it->second.parent_offset_ == -1)
959         delete[] it->second.name_;
960     }
961
962   this->declarations_.clear();
963 }
964
965 // Print usage statistics.
966 void
967 Gdb_index_info_reader::print_stats()
968 {
969   fprintf(stderr, _("%s: DWARF CUs: %u\n"),
970           program_name, Gdb_index_info_reader::dwarf_cu_count);
971   fprintf(stderr, _("%s: DWARF CUs without pubnames/pubtypes: %u\n"),
972           program_name, Gdb_index_info_reader::dwarf_cu_nopubnames_count);
973   fprintf(stderr, _("%s: DWARF TUs: %u\n"),
974           program_name, Gdb_index_info_reader::dwarf_tu_count);
975   fprintf(stderr, _("%s: DWARF TUs without pubnames/pubtypes: %u\n"),
976           program_name, Gdb_index_info_reader::dwarf_tu_nopubnames_count);
977 }
978
979 // Class Gdb_index.
980
981 // Construct the .gdb_index section.
982
983 Gdb_index::Gdb_index(Output_section* gdb_index_section)
984   : Output_section_data(4),
985     pubnames_table_(NULL),
986     pubtypes_table_(NULL),
987     gdb_index_section_(gdb_index_section),
988     comp_units_(),
989     type_units_(),
990     ranges_(),
991     cu_vector_list_(),
992     cu_vector_offsets_(NULL),
993     stringpool_(),
994     tu_offset_(0),
995     addr_offset_(0),
996     symtab_offset_(0),
997     cu_pool_offset_(0),
998     stringpool_offset_(0),
999     pubnames_object_(NULL),
1000     stmt_list_offset_(-1)
1001 {
1002   this->gdb_symtab_ = new Gdb_hashtab<Gdb_symbol>();
1003 }
1004
1005 Gdb_index::~Gdb_index()
1006 {
1007   // Free the memory used by the symbol table.
1008   delete this->gdb_symtab_;
1009   // Free the memory used by the CU vectors.
1010   for (unsigned int i = 0; i < this->cu_vector_list_.size(); ++i)
1011     delete this->cu_vector_list_[i];
1012 }
1013
1014
1015 // Scan the pubnames and pubtypes sections and build a map of the
1016 // various cus and tus they refer to, so we can process the entries
1017 // when we encounter the die for that cu or tu.
1018 // Return the just-read table so it can be cached.
1019
1020 Dwarf_pubnames_table*
1021 Gdb_index::map_pubtable_to_dies(unsigned int attr,
1022                                 Gdb_index_info_reader* dwinfo,
1023                                 Relobj* object,
1024                                 const unsigned char* symbols,
1025                                 off_t symbols_size)
1026 {
1027   uint64_t section_offset = 0;
1028   Dwarf_pubnames_table* table;
1029   Pubname_offset_map* map;
1030
1031   if (attr == elfcpp::DW_AT_GNU_pubnames)
1032     {
1033       table = new Dwarf_pubnames_table(dwinfo, false);
1034       map = &this->cu_pubname_map_;
1035     }
1036   else
1037     {
1038       table = new Dwarf_pubnames_table(dwinfo, true);
1039       map = &this->cu_pubtype_map_;
1040     }
1041
1042   map->clear();
1043   if (!table->read_section(object, symbols, symbols_size))
1044     return NULL;
1045
1046   while (table->read_header(section_offset))
1047     {
1048       map->insert(std::make_pair(table->cu_offset(), section_offset));
1049       section_offset += table->subsection_size();
1050     }
1051
1052   return table;
1053 }
1054
1055 // Wrapper for map_pubtable_to_dies
1056
1057 void
1058 Gdb_index::map_pubnames_and_types_to_dies(Gdb_index_info_reader* dwinfo,
1059                                           Relobj* object,
1060                                           const unsigned char* symbols,
1061                                           off_t symbols_size)
1062 {
1063   // This is a new object, so reset the relevant variables.
1064   this->pubnames_object_ = object;
1065   this->stmt_list_offset_ = -1;
1066
1067   delete this->pubnames_table_;
1068   this->pubnames_table_
1069       = this->map_pubtable_to_dies(elfcpp::DW_AT_GNU_pubnames, dwinfo,
1070                                    object, symbols, symbols_size);
1071   delete this->pubtypes_table_;
1072   this->pubtypes_table_
1073       = this->map_pubtable_to_dies(elfcpp::DW_AT_GNU_pubtypes, dwinfo,
1074                                    object, symbols, symbols_size);
1075 }
1076
1077 // Given a cu_offset, find the associated section of the pubnames
1078 // table.
1079
1080 off_t
1081 Gdb_index::find_pubname_offset(off_t cu_offset)
1082 {
1083   Pubname_offset_map::iterator it = this->cu_pubname_map_.find(cu_offset);
1084   if (it != this->cu_pubname_map_.end())
1085     return it->second;
1086   return -1;
1087 }
1088
1089 // Given a cu_offset, find the associated section of the pubnames
1090 // table.
1091
1092 off_t
1093 Gdb_index::find_pubtype_offset(off_t cu_offset)
1094 {
1095   Pubname_offset_map::iterator it = this->cu_pubtype_map_.find(cu_offset);
1096   if (it != this->cu_pubtype_map_.end())
1097     return it->second;
1098   return -1;
1099 }
1100
1101 // Scan a .debug_info or .debug_types input section.
1102
1103 void
1104 Gdb_index::scan_debug_info(bool is_type_unit,
1105                            Relobj* object,
1106                            const unsigned char* symbols,
1107                            off_t symbols_size,
1108                            unsigned int shndx,
1109                            unsigned int reloc_shndx,
1110                            unsigned int reloc_type)
1111 {
1112   Gdb_index_info_reader dwinfo(is_type_unit, object,
1113                                symbols, symbols_size,
1114                                shndx, reloc_shndx,
1115                                reloc_type, this);
1116   if (object != this->pubnames_object_)
1117     map_pubnames_and_types_to_dies(&dwinfo, object, symbols, symbols_size);
1118   dwinfo.parse();
1119 }
1120
1121 // Add a symbol.
1122
1123 void
1124 Gdb_index::add_symbol(int cu_index, const char* sym_name, uint8_t flags)
1125 {
1126   unsigned int hash = mapped_index_string_hash(
1127       reinterpret_cast<const unsigned char*>(sym_name));
1128   Gdb_symbol* sym = new Gdb_symbol();
1129   this->stringpool_.add(sym_name, true, &sym->name_key);
1130   sym->hashval = hash;
1131   sym->cu_vector_index = 0;
1132
1133   Gdb_symbol* found = this->gdb_symtab_->add(sym);
1134   if (found == sym)
1135     {
1136       // New symbol -- allocate a new CU index vector.
1137       found->cu_vector_index = this->cu_vector_list_.size();
1138       this->cu_vector_list_.push_back(new Cu_vector());
1139     }
1140   else
1141     {
1142       // Found an existing symbol -- append to the existing
1143       // CU index vector.
1144       delete sym;
1145     }
1146
1147   // Add the CU index to the vector list for this symbol,
1148   // if it's not already on the list.  We only need to
1149   // check the last added entry.
1150   Cu_vector* cu_vec = this->cu_vector_list_[found->cu_vector_index];
1151   if (cu_vec->size() == 0
1152       || cu_vec->back().first != cu_index
1153       || cu_vec->back().second != flags)
1154     cu_vec->push_back(std::make_pair(cu_index, flags));
1155 }
1156
1157 // Return TRUE if we have already processed the pubnames associated
1158 // with the statement list at the given OFFSET.
1159
1160 bool
1161 Gdb_index::pubnames_read(const Relobj* object, off_t offset)
1162 {
1163   bool ret = (this->pubnames_object_ == object
1164               && this->stmt_list_offset_ == offset);
1165   return ret;
1166 }
1167
1168 // Record that we have processed the pubnames associated with the
1169 // statement list for OBJECT at the given OFFSET.
1170
1171 void
1172 Gdb_index::set_pubnames_read(const Relobj* object, off_t offset)
1173 {
1174   this->pubnames_object_ = object;
1175   this->stmt_list_offset_ = offset;
1176 }
1177
1178 // Set the size of the .gdb_index section.
1179
1180 void
1181 Gdb_index::set_final_data_size()
1182 {
1183   // Finalize the string pool.
1184   this->stringpool_.set_string_offsets();
1185
1186   // Compute the total size of the CU vectors.
1187   // For each CU vector, include one entry for the count at the
1188   // beginning of the vector.
1189   unsigned int cu_vector_count = this->cu_vector_list_.size();
1190   unsigned int cu_vector_size = 0;
1191   this->cu_vector_offsets_ = new off_t[cu_vector_count];
1192   for (unsigned int i = 0; i < cu_vector_count; ++i)
1193     {
1194       Cu_vector* cu_vec = this->cu_vector_list_[i];
1195       cu_vector_offsets_[i] = cu_vector_size;
1196       cu_vector_size += gdb_index_offset_size * (cu_vec->size() + 1);
1197     }
1198
1199   // Assign relative offsets to each portion of the index,
1200   // and find the total size of the section.
1201   section_size_type data_size = gdb_index_hdr_size;
1202   data_size += this->comp_units_.size() * gdb_index_cu_size;
1203   this->tu_offset_ = data_size;
1204   data_size += this->type_units_.size() * gdb_index_tu_size;
1205   this->addr_offset_ = data_size;
1206   for (unsigned int i = 0; i < this->ranges_.size(); ++i)
1207     data_size += this->ranges_[i].ranges->size() * gdb_index_addr_size;
1208   this->symtab_offset_ = data_size;
1209   data_size += this->gdb_symtab_->capacity() * gdb_index_sym_size;
1210   this->cu_pool_offset_ = data_size;
1211   data_size += cu_vector_size;
1212   this->stringpool_offset_ = data_size;
1213   data_size += this->stringpool_.get_strtab_size();
1214
1215   this->set_data_size(data_size);
1216 }
1217
1218 // Write the data to the file.
1219
1220 void
1221 Gdb_index::do_write(Output_file* of)
1222 {
1223   const off_t off = this->offset();
1224   const off_t oview_size = this->data_size();
1225   unsigned char* const oview = of->get_output_view(off, oview_size);
1226   unsigned char* pov = oview;
1227
1228   // Write the file header.
1229   // (1) Version number.
1230   elfcpp::Swap<32, false>::writeval(pov, gdb_index_version);
1231   pov += 4;
1232   // (2) Offset of the CU list.
1233   elfcpp::Swap<32, false>::writeval(pov, gdb_index_hdr_size);
1234   pov += 4;
1235   // (3) Offset of the types CU list.
1236   elfcpp::Swap<32, false>::writeval(pov, this->tu_offset_);
1237   pov += 4;
1238   // (4) Offset of the address area.
1239   elfcpp::Swap<32, false>::writeval(pov, this->addr_offset_);
1240   pov += 4;
1241   // (5) Offset of the symbol table.
1242   elfcpp::Swap<32, false>::writeval(pov, this->symtab_offset_);
1243   pov += 4;
1244   // (6) Offset of the constant pool.
1245   elfcpp::Swap<32, false>::writeval(pov, this->cu_pool_offset_);
1246   pov += 4;
1247
1248   gold_assert(pov - oview == gdb_index_hdr_size);
1249
1250   // Write the CU list.
1251   unsigned int comp_units_count = this->comp_units_.size();
1252   for (unsigned int i = 0; i < comp_units_count; ++i)
1253     {
1254       const Comp_unit& cu = this->comp_units_[i];
1255       elfcpp::Swap<64, false>::writeval(pov, cu.cu_offset);
1256       elfcpp::Swap<64, false>::writeval(pov + 8, cu.cu_length);
1257       pov += 16;
1258     }
1259
1260   gold_assert(pov - oview == this->tu_offset_);
1261
1262   // Write the types CU list.
1263   for (unsigned int i = 0; i < this->type_units_.size(); ++i)
1264     {
1265       const Type_unit& tu = this->type_units_[i];
1266       elfcpp::Swap<64, false>::writeval(pov, tu.tu_offset);
1267       elfcpp::Swap<64, false>::writeval(pov + 8, tu.type_offset);
1268       elfcpp::Swap<64, false>::writeval(pov + 16, tu.type_signature);
1269       pov += 24;
1270     }
1271
1272   gold_assert(pov - oview == this->addr_offset_);
1273
1274   // Write the address area.
1275   for (unsigned int i = 0; i < this->ranges_.size(); ++i)
1276     {
1277       int cu_index = this->ranges_[i].cu_index;
1278       // Translate negative indexes, which refer to a TU, to a
1279       // logical index into a concatenated CU/TU list.
1280       if (cu_index < 0)
1281         cu_index = comp_units_count + (-1 - cu_index);
1282       Relobj* object = this->ranges_[i].object;
1283       const Dwarf_range_list& ranges = *this->ranges_[i].ranges;
1284       for (unsigned int j = 0; j < ranges.size(); ++j)
1285         {
1286           const Dwarf_range_list::Range& range = ranges[j];
1287           uint64_t base = 0;
1288           if (range.shndx > 0)
1289             {
1290               const Output_section* os = object->output_section(range.shndx);
1291               base = (os->address()
1292                       + object->output_section_offset(range.shndx));
1293             }
1294           elfcpp::Swap_aligned32<64, false>::writeval(pov, base + range.start);
1295           elfcpp::Swap_aligned32<64, false>::writeval(pov + 8,
1296                                                       base + range.end);
1297           elfcpp::Swap<32, false>::writeval(pov + 16, cu_index);
1298           pov += 20;
1299         }
1300     }
1301
1302   gold_assert(pov - oview == this->symtab_offset_);
1303
1304   // Write the symbol table.
1305   for (unsigned int i = 0; i < this->gdb_symtab_->capacity(); ++i)
1306     {
1307       const Gdb_symbol* sym = (*this->gdb_symtab_)[i];
1308       section_offset_type name_offset = 0;
1309       unsigned int cu_vector_offset = 0;
1310       if (sym != NULL)
1311         {
1312           name_offset = (this->stringpool_.get_offset_from_key(sym->name_key)
1313                          + this->stringpool_offset_ - this->cu_pool_offset_);
1314           cu_vector_offset = this->cu_vector_offsets_[sym->cu_vector_index];
1315         }
1316       elfcpp::Swap<32, false>::writeval(pov, name_offset);
1317       elfcpp::Swap<32, false>::writeval(pov + 4, cu_vector_offset);
1318       pov += 8;
1319     }
1320
1321   gold_assert(pov - oview == this->cu_pool_offset_);
1322
1323   // Write the CU vectors into the constant pool.
1324   for (unsigned int i = 0; i < this->cu_vector_list_.size(); ++i)
1325     {
1326       Cu_vector* cu_vec = this->cu_vector_list_[i];
1327       elfcpp::Swap<32, false>::writeval(pov, cu_vec->size());
1328       pov += 4;
1329       for (unsigned int j = 0; j < cu_vec->size(); ++j)
1330         {
1331           int cu_index = (*cu_vec)[j].first;
1332           uint8_t flags = (*cu_vec)[j].second;
1333           if (cu_index < 0)
1334             cu_index = comp_units_count + (-1 - cu_index);
1335           cu_index |= flags << 24;
1336           elfcpp::Swap<32, false>::writeval(pov, cu_index);
1337           pov += 4;
1338         }
1339     }
1340
1341   gold_assert(pov - oview == this->stringpool_offset_);
1342
1343   // Write the strings into the constant pool.
1344   this->stringpool_.write_to_buffer(pov, oview_size - this->stringpool_offset_);
1345
1346   of->write_output_view(off, oview_size, oview);
1347 }
1348
1349 // Print usage statistics.
1350 void
1351 Gdb_index::print_stats()
1352 {
1353   if (parameters->options().gdb_index())
1354     Gdb_index_info_reader::print_stats();
1355 }
1356
1357 } // End namespace gold.