Add .gdb_index version 7 support.
[external/binutils.git] / gold / gdb-index.cc
1 // gdb-index.cc -- generate .gdb_index section for fast debug lookup
2
3 // Copyright 2012 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         // Check for languages that require specialized knowledge to
361         // construct fully-qualified names, that we don't yet support.
362         if (this->cu_language_ == elfcpp::DW_LANG_Ada83
363             || this->cu_language_ == elfcpp::DW_LANG_Fortran77
364             || this->cu_language_ == elfcpp::DW_LANG_Fortran90
365             || this->cu_language_ == elfcpp::DW_LANG_Java
366             || this->cu_language_ == elfcpp::DW_LANG_Ada95
367             || this->cu_language_ == elfcpp::DW_LANG_Fortran95)
368           {
369             gold_warning(_("%s: --gdb-index currently supports "
370                            "only C and C++ languages"),
371                          this->object()->name().c_str());
372             return;
373           }
374         if (die->tag() == elfcpp::DW_TAG_compile_unit)
375           this->record_cu_ranges(die);
376         // If there is a pubnames and/or pubtypes section for this
377         // compilation unit, use those; otherwise, parse the DWARF
378         // info to extract the names.
379         if (!this->read_pubnames_and_pubtypes(die))
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
398 // Visit the children of PARENT, looking for symbols to add to the index.
399 // CONTEXT points to the DIE to use for constructing the qualified name --
400 // NULL if PARENT is the top-level DIE; otherwise it is the same as PARENT.
401
402 void
403 Gdb_index_info_reader::visit_children(Dwarf_die* parent, Dwarf_die* context)
404 {
405   off_t next_offset = 0;
406   for (off_t die_offset = parent->child_offset();
407        die_offset != 0;
408        die_offset = next_offset)
409     {
410       Dwarf_die die(this, die_offset, parent);
411       if (die.tag() == 0)
412         break;
413       this->visit_die(&die, context);
414       next_offset = die.sibling_offset();
415     }
416 }
417
418 // Visit a child DIE, looking for symbols to add to the index.
419 // CONTEXT is the parent DIE, used for constructing the qualified name;
420 // it is NULL if the parent DIE is the top-level DIE.
421
422 void
423 Gdb_index_info_reader::visit_die(Dwarf_die* die, Dwarf_die* context)
424 {
425   switch (die->tag())
426     {
427       case elfcpp::DW_TAG_subprogram:
428       case elfcpp::DW_TAG_constant:
429       case elfcpp::DW_TAG_variable:
430       case elfcpp::DW_TAG_enumerator:
431       case elfcpp::DW_TAG_base_type:
432         if (die->is_declaration())
433           this->add_declaration(die, context);
434         else
435           {
436             // If the DIE is not a declaration, add it to the index.
437             std::string full_name = this->get_qualified_name(die, context);
438             if (!full_name.empty())
439               this->gdb_index_->add_symbol(this->cu_index_,
440                                            full_name.c_str(), 0);
441           }
442         break;
443       case elfcpp::DW_TAG_typedef:
444       case elfcpp::DW_TAG_union_type:
445       case elfcpp::DW_TAG_class_type:
446       case elfcpp::DW_TAG_interface_type:
447       case elfcpp::DW_TAG_structure_type:
448       case elfcpp::DW_TAG_enumeration_type:
449       case elfcpp::DW_TAG_subrange_type:
450       case elfcpp::DW_TAG_namespace:
451         {
452           std::string full_name;
453           
454           // For classes at the top level, we need to look for a
455           // member function with a linkage name in order to get
456           // the properly-canonicalized name.
457           if (context == NULL
458               && (die->tag() == elfcpp::DW_TAG_class_type
459                   || die->tag() == elfcpp::DW_TAG_structure_type
460                   || die->tag() == elfcpp::DW_TAG_union_type))
461             full_name.assign(this->guess_full_class_name(die));
462
463           // Because we will visit the children, we need to add this DIE
464           // to the declarations table.
465           if (full_name.empty())
466             this->add_declaration(die, context);
467           else
468             this->add_declaration_with_full_name(die, full_name.c_str());
469
470           // If the DIE is not a declaration, add it to the index.
471           // Gdb stores a namespace in the index even when it is
472           // a declaration.
473           if (die->tag() == elfcpp::DW_TAG_namespace
474               || !die->is_declaration())
475             {
476               if (full_name.empty())
477                 full_name = this->get_qualified_name(die, context);
478               if (!full_name.empty())
479                 this->gdb_index_->add_symbol(this->cu_index_,
480                                              full_name.c_str(), 0);
481             }
482
483           // We're interested in the children only for namespaces and
484           // enumeration types.  For enumeration types, we do not include
485           // the enumeration tag as part of the full name.  For other tags,
486           // visit the children only to collect declarations.
487           if (die->tag() == elfcpp::DW_TAG_namespace
488               || die->tag() == elfcpp::DW_TAG_enumeration_type)
489             this->visit_children(die, die);
490           else
491             this->visit_children_for_decls(die);
492         }
493         break;
494       default:
495         break;
496     }
497 }
498
499 // Visit the children of PARENT, looking only for declarations that
500 // may be referenced by later specification DIEs.
501
502 void
503 Gdb_index_info_reader::visit_children_for_decls(Dwarf_die* parent)
504 {
505   off_t next_offset = 0;
506   for (off_t die_offset = parent->child_offset();
507        die_offset != 0;
508        die_offset = next_offset)
509     {
510       Dwarf_die die(this, die_offset, parent);
511       if (die.tag() == 0)
512         break;
513       this->visit_die_for_decls(&die, parent);
514       next_offset = die.sibling_offset();
515     }
516 }
517
518 // Visit a child DIE, looking only for declarations that
519 // may be referenced by later specification DIEs.
520
521 void
522 Gdb_index_info_reader::visit_die_for_decls(Dwarf_die* die, Dwarf_die* context)
523 {
524   switch (die->tag())
525     {
526       case elfcpp::DW_TAG_subprogram:
527       case elfcpp::DW_TAG_constant:
528       case elfcpp::DW_TAG_variable:
529       case elfcpp::DW_TAG_enumerator:
530       case elfcpp::DW_TAG_base_type:
531         {
532           if (die->is_declaration())
533             this->add_declaration(die, context);
534         }
535         break;
536       case elfcpp::DW_TAG_typedef:
537       case elfcpp::DW_TAG_union_type:
538       case elfcpp::DW_TAG_class_type:
539       case elfcpp::DW_TAG_interface_type:
540       case elfcpp::DW_TAG_structure_type:
541       case elfcpp::DW_TAG_enumeration_type:
542       case elfcpp::DW_TAG_subrange_type:
543       case elfcpp::DW_TAG_namespace:
544         {
545           if (die->is_declaration())
546             this->add_declaration(die, context);
547           this->visit_children_for_decls(die);
548         }
549         break;
550       default:
551         break;
552     }
553 }
554
555 // Extract the class name from the linkage name of a member function.
556 // This code is adapted from ../gdb/cp-support.c.
557
558 #define d_left(dc) (dc)->u.s_binary.left
559 #define d_right(dc) (dc)->u.s_binary.right
560
561 static char*
562 class_name_from_linkage_name(const char* linkage_name)
563 {
564   void* storage;
565   struct demangle_component* tree =
566       cplus_demangle_v3_components(linkage_name, DMGL_NO_OPTS, &storage);
567   if (tree == NULL)
568     return NULL;
569
570   int done = 0;
571
572   // First strip off any qualifiers, if we have a function or
573   // method.
574   while (!done)
575     switch (tree->type)
576       {
577         case DEMANGLE_COMPONENT_CONST:
578         case DEMANGLE_COMPONENT_RESTRICT:
579         case DEMANGLE_COMPONENT_VOLATILE:
580         case DEMANGLE_COMPONENT_CONST_THIS:
581         case DEMANGLE_COMPONENT_RESTRICT_THIS:
582         case DEMANGLE_COMPONENT_VOLATILE_THIS:
583         case DEMANGLE_COMPONENT_VENDOR_TYPE_QUAL:
584           tree = d_left(tree);
585           break;
586         default:
587           done = 1;
588           break;
589       }
590
591   // If what we have now is a function, discard the argument list.
592   if (tree->type == DEMANGLE_COMPONENT_TYPED_NAME)
593     tree = d_left(tree);
594
595   // If what we have now is a template, strip off the template
596   // arguments.  The left subtree may be a qualified name.
597   if (tree->type == DEMANGLE_COMPONENT_TEMPLATE)
598     tree = d_left(tree);
599
600   // What we have now should be a name, possibly qualified.
601   // Additional qualifiers could live in the left subtree or the right
602   // subtree.  Find the last piece.
603   done = 0;
604   struct demangle_component* prev_comp = NULL;
605   struct demangle_component* cur_comp = tree;
606   while (!done)
607     switch (cur_comp->type)
608       {
609         case DEMANGLE_COMPONENT_QUAL_NAME:
610         case DEMANGLE_COMPONENT_LOCAL_NAME:
611           prev_comp = cur_comp;
612           cur_comp = d_right(cur_comp);
613           break;
614         case DEMANGLE_COMPONENT_TEMPLATE:
615         case DEMANGLE_COMPONENT_NAME:
616         case DEMANGLE_COMPONENT_CTOR:
617         case DEMANGLE_COMPONENT_DTOR:
618         case DEMANGLE_COMPONENT_OPERATOR:
619         case DEMANGLE_COMPONENT_EXTENDED_OPERATOR:
620           done = 1;
621           break;
622         default:
623           done = 1;
624           cur_comp = NULL;
625           break;
626       }
627
628   char* ret = NULL;
629   if (cur_comp != NULL && prev_comp != NULL)
630     {
631       // We want to discard the rightmost child of PREV_COMP.
632       *prev_comp = *d_left(prev_comp);
633       size_t allocated_size;
634       ret = cplus_demangle_print(DMGL_NO_OPTS, tree, 30, &allocated_size);
635     }
636
637   free(storage);
638   return ret;
639 }
640
641 // Guess a fully-qualified name for a class type, based on member function
642 // linkage names.  This is needed for class/struct/union types at the
643 // top level, because GCC does not always properly embed them within
644 // the namespace.  As in gdb, we look for a member function with a linkage
645 // name and extract the qualified name from the demangled name.
646
647 std::string
648 Gdb_index_info_reader::guess_full_class_name(Dwarf_die* die)
649 {
650   std::string full_name;
651   off_t next_offset = 0;
652   
653   // This routine scans ahead in the DIE structure, possibly advancing
654   // the relocation tracker beyond the current DIE.  We need to checkpoint
655   // the tracker and reset it when we're done.
656   uint64_t checkpoint = this->get_reloc_checkpoint();
657
658   for (off_t child_offset = die->child_offset();
659        child_offset != 0;
660        child_offset = next_offset)
661     {
662       Dwarf_die child(this, child_offset, die);
663       if (child.tag() == 0)
664         break;
665       if (child.tag() == elfcpp::DW_TAG_subprogram)
666         {
667           const char* linkage_name = child.linkage_name();
668           if (linkage_name != NULL)
669             {
670               char* guess = class_name_from_linkage_name(linkage_name);
671               if (guess != NULL)
672                 {
673                   full_name.assign(guess);
674                   free(guess);
675                   break;
676                 }
677             }
678         }
679       next_offset = child.sibling_offset();
680     }
681
682   this->reset_relocs(checkpoint);
683   return full_name;
684 }
685
686 // Add a declaration DIE to the table of declarations.
687
688 void
689 Gdb_index_info_reader::add_declaration(Dwarf_die* die, Dwarf_die* context)
690 {
691   const char* name = die->name();
692
693   off_t parent_offset = context != NULL ? context->offset() : 0;
694
695   // If this DIE has a DW_AT_specification or DW_AT_abstract_origin
696   // attribute, use the parent and name from the earlier declaration.
697   off_t spec = die->specification();
698   if (spec == 0)
699     spec = die->abstract_origin();
700   if (spec > 0)
701     {
702       Declaration_map::iterator it = this->declarations_.find(spec);
703       if (it != this->declarations_.end())
704         {
705           parent_offset = it->second.parent_offset_;
706           name = it->second.name_;
707         }
708     }
709
710   if (name == NULL)
711     {
712       if (die->tag() == elfcpp::DW_TAG_namespace)
713         name = "(anonymous namespace)";
714       else if (die->tag() == elfcpp::DW_TAG_union_type)
715         name = "(anonymous union)";
716       else
717         name = "(unknown)";
718     }
719
720   Declaration_pair decl(parent_offset, name);
721   this->declarations_.insert(std::make_pair(die->offset(), decl));
722 }
723
724 // Add a declaration whose fully-qualified name is already known.
725 // In the case where we had to get the canonical name by demangling
726 // a linkage name, this ensures we use that name instead of the one
727 // provided in DW_AT_name.
728
729 void
730 Gdb_index_info_reader::add_declaration_with_full_name(
731     Dwarf_die* die,
732     const char* full_name)
733 {
734   // We need to copy the name.
735   int len = strlen(full_name);
736   char* copy = new char[len + 1];
737   memcpy(copy, full_name, len + 1);
738
739   // Flag that we now manage the memory this points to.
740   Declaration_pair decl(-1, copy);
741   this->declarations_.insert(std::make_pair(die->offset(), decl));
742 }
743
744 // Return the context for a DIE whose parent is at DIE_OFFSET.
745
746 std::string
747 Gdb_index_info_reader::get_context(off_t die_offset)
748 {
749   std::string context;
750   Declaration_map::iterator it = this->declarations_.find(die_offset);
751   if (it != this->declarations_.end())
752     {
753       off_t parent_offset = it->second.parent_offset_;
754       if (parent_offset > 0)
755         {
756           context = get_context(parent_offset);
757           context.append("::");
758         }
759       if (it->second.name_ != NULL)
760         context.append(it->second.name_);
761     }
762   return context;
763 }
764
765 // Construct the fully-qualified name for DIE.
766
767 std::string
768 Gdb_index_info_reader::get_qualified_name(Dwarf_die* die, Dwarf_die* context)
769 {
770   std::string full_name;
771   const char* name = die->name();
772
773   off_t parent_offset = context != NULL ? context->offset() : 0;
774
775   // If this DIE has a DW_AT_specification or DW_AT_abstract_origin
776   // attribute, use the parent and name from the earlier declaration.
777   off_t spec = die->specification();
778   if (spec == 0)
779     spec = die->abstract_origin();
780   if (spec > 0)
781     {
782       Declaration_map::iterator it = this->declarations_.find(spec);
783       if (it != this->declarations_.end())
784         {
785           parent_offset = it->second.parent_offset_;
786           name = it->second.name_;
787         }
788     }
789
790   if (name == NULL && die->tag() == elfcpp::DW_TAG_namespace)
791     name = "(anonymous namespace)";
792   else if (name == NULL)
793     return full_name;
794
795   // If this is an enumerator constant, skip the immediate parent,
796   // which is the enumeration tag.
797   if (die->tag() == elfcpp::DW_TAG_enumerator)
798     {
799       Declaration_map::iterator it = this->declarations_.find(parent_offset);
800       if (it != this->declarations_.end())
801         parent_offset = it->second.parent_offset_;
802     }
803
804   if (parent_offset > 0)
805     {
806       full_name.assign(this->get_context(parent_offset));
807       full_name.append("::");
808     }
809   full_name.append(name);
810
811   return full_name;
812 }
813
814 // Record the address ranges for a compilation unit.
815
816 void
817 Gdb_index_info_reader::record_cu_ranges(Dwarf_die* die)
818 {
819   unsigned int shndx;
820   unsigned int shndx2;
821
822   off_t ranges_offset = die->ref_attribute(elfcpp::DW_AT_ranges, &shndx);
823   if (ranges_offset != -1)
824     {
825       Dwarf_range_list* ranges = this->read_range_list(shndx, ranges_offset);
826       if (ranges != NULL)
827         this->gdb_index_->add_address_range_list(this->object(),
828                                                  this->cu_index_, ranges);
829       return;
830     }
831
832   off_t low_pc = die->address_attribute(elfcpp::DW_AT_low_pc, &shndx);
833   off_t high_pc = die->address_attribute(elfcpp::DW_AT_high_pc, &shndx2);
834   if (high_pc == -1)
835     {
836       high_pc = die->uint_attribute(elfcpp::DW_AT_high_pc);
837       high_pc += low_pc;
838       shndx2 = shndx;
839     }
840   if ((low_pc != 0 || high_pc != 0) && low_pc != -1)
841     {
842       if (shndx != shndx2)
843         {
844           gold_warning(_("%s: DWARF info may be corrupt; low_pc and high_pc "
845                          "are in different sections"),
846                        this->object()->name().c_str());
847           return;
848         }
849       if (shndx == 0 || this->object()->is_section_included(shndx))
850         {
851           Dwarf_range_list* ranges = new Dwarf_range_list();
852           ranges->add(shndx, low_pc, high_pc);
853           this->gdb_index_->add_address_range_list(this->object(),
854                                                    this->cu_index_, ranges);
855         }
856     }
857 }
858
859 // Read table and add the relevant names to the index.  Returns true
860 // if any names were added.
861
862 bool
863 Gdb_index_info_reader::read_pubtable(Dwarf_pubnames_table* table, off_t offset)
864 {
865   // If we couldn't read the section when building the cu_pubname_map,
866   // then we won't find any pubnames now.
867   if (table == NULL)
868     return false;
869
870   if (!table->read_header(offset))
871     return false;
872   while (true)
873     {
874       uint8_t flag_byte;
875       const char* name = table->next_name(&flag_byte);
876       if (name == NULL)
877         break;
878
879       this->gdb_index_->add_symbol(this->cu_index_, name, flag_byte);
880     }
881   return true;
882 }
883
884 // Read the .debug_pubnames and .debug_pubtypes tables for the CU or TU.
885 // Returns TRUE if either a pubnames or pubtypes section was found.
886
887 bool
888 Gdb_index_info_reader::read_pubnames_and_pubtypes(Dwarf_die* die)
889 {
890   // If this is a skeleton debug-type die (generated via
891   // -gsplit-dwarf), then the associated pubnames should have been
892   // read along with the corresponding CU.  In any case, there isn't
893   // enough info inside to build a gdb index entry.
894   if (die->tag() == elfcpp::DW_TAG_type_unit
895       && die->string_attribute(elfcpp::DW_AT_GNU_dwo_name))
896     return true;
897
898   // We use stmt_list_off as a unique identifier for the
899   // compilation unit and its associated type units.
900   unsigned int shndx;
901   off_t stmt_list_off = die->ref_attribute (elfcpp::DW_AT_stmt_list,
902                                             &shndx);
903   // Look for the attr as either a flag or a ref.
904   off_t offset = die->ref_attribute(elfcpp::DW_AT_GNU_pubnames, &shndx);
905
906   // Newer versions of GCC generate CUs, but not TUs, with
907   // DW_AT_FORM_flag_present.
908   unsigned int flag = die->uint_attribute(elfcpp::DW_AT_GNU_pubnames);
909   if (offset == -1 && flag == 0)
910     {
911       // Didn't find the attribute.
912       if (die->tag() == elfcpp::DW_TAG_type_unit)
913         {
914           // If die is a TU, then it might correspond to a CU which we
915           // have read. If it does, then no need to read the pubnames.
916           // If it doesn't, then the caller will have to parse the
917           // dies manually to find the names.
918           return this->gdb_index_->pubnames_read(this->object(),
919                                                  stmt_list_off);
920         }
921       else
922         {
923           // No attribute on the CU means that no pubnames were read.
924           return false;
925         }
926     }
927
928   // We found the attribute, so we can check if the corresponding
929   // pubnames have been read.
930   if (this->gdb_index_->pubnames_read(this->object(), stmt_list_off))
931     return true;
932
933   this->gdb_index_->set_pubnames_read(this->object(), stmt_list_off);
934
935   // We have an attribute, and the pubnames haven't been read, so read
936   // them.
937   bool names = false;
938   // In some of the cases, we could rely on the previous value of
939   // offset here, but sorting out which cases complicates the logic
940   // enough that it isn't worth it. So just look up the offset again.
941   offset = this->gdb_index_->find_pubname_offset(this->cu_offset());
942   names = this->read_pubtable(this->gdb_index_->pubnames_table(), offset);
943
944   bool types = false;
945   offset = this->gdb_index_->find_pubtype_offset(this->cu_offset());
946   types = this->read_pubtable(this->gdb_index_->pubtypes_table(), offset);
947   return names || types;
948 }
949
950 // Clear the declarations map.
951 void
952 Gdb_index_info_reader::clear_declarations()
953 {
954   // Free strings in memory we manage.
955   for (Declaration_map::iterator it = this->declarations_.begin();
956        it != this->declarations_.end();
957        ++it)
958     {
959       if (it->second.parent_offset_ == -1)
960         delete[] it->second.name_;
961     }
962
963   this->declarations_.clear();
964 }
965
966 // Print usage statistics.
967 void
968 Gdb_index_info_reader::print_stats()
969 {
970   fprintf(stderr, _("%s: DWARF CUs: %u\n"),
971           program_name, Gdb_index_info_reader::dwarf_cu_count);
972   fprintf(stderr, _("%s: DWARF CUs without pubnames/pubtypes: %u\n"),
973           program_name, Gdb_index_info_reader::dwarf_cu_nopubnames_count);
974   fprintf(stderr, _("%s: DWARF TUs: %u\n"),
975           program_name, Gdb_index_info_reader::dwarf_tu_count);
976   fprintf(stderr, _("%s: DWARF TUs without pubnames/pubtypes: %u\n"),
977           program_name, Gdb_index_info_reader::dwarf_tu_nopubnames_count);
978 }
979
980 // Class Gdb_index.
981
982 // Construct the .gdb_index section.
983
984 Gdb_index::Gdb_index(Output_section* gdb_index_section)
985   : Output_section_data(4),
986     pubnames_table_(NULL),
987     pubtypes_table_(NULL),
988     gdb_index_section_(gdb_index_section),
989     comp_units_(),
990     type_units_(),
991     ranges_(),
992     cu_vector_list_(),
993     cu_vector_offsets_(NULL),
994     stringpool_(),
995     tu_offset_(0),
996     addr_offset_(0),
997     symtab_offset_(0),
998     cu_pool_offset_(0),
999     stringpool_offset_(0),
1000     pubnames_object_(NULL),
1001     stmt_list_offset_(-1)
1002 {
1003   this->gdb_symtab_ = new Gdb_hashtab<Gdb_symbol>();
1004 }
1005
1006 Gdb_index::~Gdb_index()
1007 {
1008   // Free the memory used by the symbol table.
1009   delete this->gdb_symtab_;
1010   // Free the memory used by the CU vectors.
1011   for (unsigned int i = 0; i < this->cu_vector_list_.size(); ++i)
1012     delete this->cu_vector_list_[i];
1013 }
1014
1015
1016 // Scan the pubnames and pubtypes sections and build a map of the
1017 // various cus and tus they refer to, so we can process the entries
1018 // when we encounter the die for that cu or tu.
1019 // Return the just-read table so it can be cached.
1020
1021 Dwarf_pubnames_table*
1022 Gdb_index::map_pubtable_to_dies(unsigned int attr,
1023                                 Gdb_index_info_reader* dwinfo,
1024                                 Relobj* object,
1025                                 const unsigned char* symbols,
1026                                 off_t symbols_size)
1027 {
1028   uint64_t section_offset = 0;
1029   Dwarf_pubnames_table* table;
1030   Pubname_offset_map* map;
1031
1032   if (attr == elfcpp::DW_AT_GNU_pubnames)
1033     {
1034       table = new Dwarf_pubnames_table(dwinfo, false);
1035       map = &this->cu_pubname_map_;
1036     }
1037   else
1038     {
1039       table = new Dwarf_pubnames_table(dwinfo, true);
1040       map = &this->cu_pubtype_map_;
1041     }
1042
1043   map->clear();
1044   if (!table->read_section(object, symbols, symbols_size))
1045     return NULL;
1046
1047   while (table->read_header(section_offset))
1048     {
1049       map->insert(std::make_pair(table->cu_offset(), section_offset));
1050       section_offset += table->subsection_size();
1051     }
1052
1053   return table;
1054 }
1055
1056 // Wrapper for map_pubtable_to_dies
1057
1058 void
1059 Gdb_index::map_pubnames_and_types_to_dies(Gdb_index_info_reader* dwinfo,
1060                                           Relobj* object,
1061                                           const unsigned char* symbols,
1062                                           off_t symbols_size)
1063 {
1064   // This is a new object, so reset the relevant variables.
1065   this->pubnames_object_ = object;
1066   this->stmt_list_offset_ = -1;
1067
1068   delete this->pubnames_table_;
1069   this->pubnames_table_
1070       = this->map_pubtable_to_dies(elfcpp::DW_AT_GNU_pubnames, dwinfo,
1071                                    object, symbols, symbols_size);
1072   delete this->pubtypes_table_;
1073   this->pubtypes_table_
1074       = this->map_pubtable_to_dies(elfcpp::DW_AT_GNU_pubtypes, dwinfo,
1075                                    object, symbols, symbols_size);
1076 }
1077
1078 // Given a cu_offset, find the associated section of the pubnames
1079 // table.
1080
1081 off_t
1082 Gdb_index::find_pubname_offset(off_t cu_offset)
1083 {
1084   Pubname_offset_map::iterator it = this->cu_pubname_map_.find(cu_offset);
1085   if (it != this->cu_pubname_map_.end())
1086     return it->second;
1087   return -1;
1088 }
1089
1090 // Given a cu_offset, find the associated section of the pubnames
1091 // table.
1092
1093 off_t
1094 Gdb_index::find_pubtype_offset(off_t cu_offset)
1095 {
1096   Pubname_offset_map::iterator it = this->cu_pubtype_map_.find(cu_offset);
1097   if (it != this->cu_pubtype_map_.end())
1098     return it->second;
1099   return -1;
1100 }
1101
1102 // Scan a .debug_info or .debug_types input section.
1103
1104 void
1105 Gdb_index::scan_debug_info(bool is_type_unit,
1106                            Relobj* object,
1107                            const unsigned char* symbols,
1108                            off_t symbols_size,
1109                            unsigned int shndx,
1110                            unsigned int reloc_shndx,
1111                            unsigned int reloc_type)
1112 {
1113   Gdb_index_info_reader dwinfo(is_type_unit, object,
1114                                symbols, symbols_size,
1115                                shndx, reloc_shndx,
1116                                reloc_type, this);
1117   if (object != this->pubnames_object_)
1118     map_pubnames_and_types_to_dies(&dwinfo, object, symbols, symbols_size);
1119   dwinfo.parse();
1120 }
1121
1122 // Add a symbol.
1123
1124 void
1125 Gdb_index::add_symbol(int cu_index, const char* sym_name, uint8_t flags)
1126 {
1127   unsigned int hash = mapped_index_string_hash(
1128       reinterpret_cast<const unsigned char*>(sym_name));
1129   Gdb_symbol* sym = new Gdb_symbol();
1130   this->stringpool_.add(sym_name, true, &sym->name_key);
1131   sym->hashval = hash;
1132   sym->cu_vector_index = 0;
1133
1134   Gdb_symbol* found = this->gdb_symtab_->add(sym);
1135   if (found == sym)
1136     {
1137       // New symbol -- allocate a new CU index vector.
1138       found->cu_vector_index = this->cu_vector_list_.size();
1139       this->cu_vector_list_.push_back(new Cu_vector());
1140     }
1141   else
1142     {
1143       // Found an existing symbol -- append to the existing
1144       // CU index vector.
1145       delete sym;
1146     }
1147
1148   // Add the CU index to the vector list for this symbol,
1149   // if it's not already on the list.  We only need to
1150   // check the last added entry.
1151   Cu_vector* cu_vec = this->cu_vector_list_[found->cu_vector_index];
1152   if (cu_vec->size() == 0
1153       || cu_vec->back().first != cu_index
1154       || cu_vec->back().second != flags)
1155     cu_vec->push_back(std::make_pair(cu_index, flags));
1156 }
1157
1158 // Return TRUE if we have already processed the pubnames associated
1159 // with the statement list at the given OFFSET.
1160
1161 bool
1162 Gdb_index::pubnames_read(const Relobj* object, off_t offset)
1163 {
1164   bool ret = (this->pubnames_object_ == object
1165               && this->stmt_list_offset_ == offset);
1166   return ret;
1167 }
1168
1169 // Record that we have processed the pubnames associated with the
1170 // statement list for OBJECT at the given OFFSET.
1171
1172 void
1173 Gdb_index::set_pubnames_read(const Relobj* object, off_t offset)
1174 {
1175   this->pubnames_object_ = object;
1176   this->stmt_list_offset_ = offset;
1177 }
1178
1179 // Set the size of the .gdb_index section.
1180
1181 void
1182 Gdb_index::set_final_data_size()
1183 {
1184   // Finalize the string pool.
1185   this->stringpool_.set_string_offsets();
1186
1187   // Compute the total size of the CU vectors.
1188   // For each CU vector, include one entry for the count at the
1189   // beginning of the vector.
1190   unsigned int cu_vector_count = this->cu_vector_list_.size();
1191   unsigned int cu_vector_size = 0;
1192   this->cu_vector_offsets_ = new off_t[cu_vector_count];
1193   for (unsigned int i = 0; i < cu_vector_count; ++i)
1194     {
1195       Cu_vector* cu_vec = this->cu_vector_list_[i];
1196       cu_vector_offsets_[i] = cu_vector_size;
1197       cu_vector_size += gdb_index_offset_size * (cu_vec->size() + 1);
1198     }
1199
1200   // Assign relative offsets to each portion of the index,
1201   // and find the total size of the section.
1202   section_size_type data_size = gdb_index_hdr_size;
1203   data_size += this->comp_units_.size() * gdb_index_cu_size;
1204   this->tu_offset_ = data_size;
1205   data_size += this->type_units_.size() * gdb_index_tu_size;
1206   this->addr_offset_ = data_size;
1207   for (unsigned int i = 0; i < this->ranges_.size(); ++i)
1208     data_size += this->ranges_[i].ranges->size() * gdb_index_addr_size;
1209   this->symtab_offset_ = data_size;
1210   data_size += this->gdb_symtab_->capacity() * gdb_index_sym_size;
1211   this->cu_pool_offset_ = data_size;
1212   data_size += cu_vector_size;
1213   this->stringpool_offset_ = data_size;
1214   data_size += this->stringpool_.get_strtab_size();
1215
1216   this->set_data_size(data_size);
1217 }
1218
1219 // Write the data to the file.
1220
1221 void
1222 Gdb_index::do_write(Output_file* of)
1223 {
1224   const off_t off = this->offset();
1225   const off_t oview_size = this->data_size();
1226   unsigned char* const oview = of->get_output_view(off, oview_size);
1227   unsigned char* pov = oview;
1228
1229   // Write the file header.
1230   // (1) Version number.
1231   elfcpp::Swap<32, false>::writeval(pov, gdb_index_version);
1232   pov += 4;
1233   // (2) Offset of the CU list.
1234   elfcpp::Swap<32, false>::writeval(pov, gdb_index_hdr_size);
1235   pov += 4;
1236   // (3) Offset of the types CU list.
1237   elfcpp::Swap<32, false>::writeval(pov, this->tu_offset_);
1238   pov += 4;
1239   // (4) Offset of the address area.
1240   elfcpp::Swap<32, false>::writeval(pov, this->addr_offset_);
1241   pov += 4;
1242   // (5) Offset of the symbol table.
1243   elfcpp::Swap<32, false>::writeval(pov, this->symtab_offset_);
1244   pov += 4;
1245   // (6) Offset of the constant pool.
1246   elfcpp::Swap<32, false>::writeval(pov, this->cu_pool_offset_);
1247   pov += 4;
1248
1249   gold_assert(pov - oview == gdb_index_hdr_size);
1250
1251   // Write the CU list.
1252   unsigned int comp_units_count = this->comp_units_.size();
1253   for (unsigned int i = 0; i < comp_units_count; ++i)
1254     {
1255       const Comp_unit& cu = this->comp_units_[i];
1256       elfcpp::Swap<64, false>::writeval(pov, cu.cu_offset);
1257       elfcpp::Swap<64, false>::writeval(pov + 8, cu.cu_length);
1258       pov += 16;
1259     }
1260
1261   gold_assert(pov - oview == this->tu_offset_);
1262
1263   // Write the types CU list.
1264   for (unsigned int i = 0; i < this->type_units_.size(); ++i)
1265     {
1266       const Type_unit& tu = this->type_units_[i];
1267       elfcpp::Swap<64, false>::writeval(pov, tu.tu_offset);
1268       elfcpp::Swap<64, false>::writeval(pov + 8, tu.type_offset);
1269       elfcpp::Swap<64, false>::writeval(pov + 16, tu.type_signature);
1270       pov += 24;
1271     }
1272
1273   gold_assert(pov - oview == this->addr_offset_);
1274
1275   // Write the address area.
1276   for (unsigned int i = 0; i < this->ranges_.size(); ++i)
1277     {
1278       int cu_index = this->ranges_[i].cu_index;
1279       // Translate negative indexes, which refer to a TU, to a
1280       // logical index into a concatenated CU/TU list.
1281       if (cu_index < 0)
1282         cu_index = comp_units_count + (-1 - cu_index);
1283       Relobj* object = this->ranges_[i].object;
1284       const Dwarf_range_list& ranges = *this->ranges_[i].ranges;
1285       for (unsigned int j = 0; j < ranges.size(); ++j)
1286         {
1287           const Dwarf_range_list::Range& range = ranges[j];
1288           uint64_t base = 0;
1289           if (range.shndx > 0)
1290             {
1291               const Output_section* os = object->output_section(range.shndx);
1292               base = (os->address()
1293                       + object->output_section_offset(range.shndx));
1294             }
1295           elfcpp::Swap_aligned32<64, false>::writeval(pov, base + range.start);
1296           elfcpp::Swap_aligned32<64, false>::writeval(pov + 8,
1297                                                       base + range.end);
1298           elfcpp::Swap<32, false>::writeval(pov + 16, cu_index);
1299           pov += 20;
1300         }
1301     }
1302
1303   gold_assert(pov - oview == this->symtab_offset_);
1304
1305   // Write the symbol table.
1306   for (unsigned int i = 0; i < this->gdb_symtab_->capacity(); ++i)
1307     {
1308       const Gdb_symbol* sym = (*this->gdb_symtab_)[i];
1309       section_offset_type name_offset = 0;
1310       unsigned int cu_vector_offset = 0;
1311       if (sym != NULL)
1312         {
1313           name_offset = (this->stringpool_.get_offset_from_key(sym->name_key)
1314                          + this->stringpool_offset_ - this->cu_pool_offset_);
1315           cu_vector_offset = this->cu_vector_offsets_[sym->cu_vector_index];
1316         }
1317       elfcpp::Swap<32, false>::writeval(pov, name_offset);
1318       elfcpp::Swap<32, false>::writeval(pov + 4, cu_vector_offset);
1319       pov += 8;
1320     }
1321
1322   gold_assert(pov - oview == this->cu_pool_offset_);
1323
1324   // Write the CU vectors into the constant pool.
1325   for (unsigned int i = 0; i < this->cu_vector_list_.size(); ++i)
1326     {
1327       Cu_vector* cu_vec = this->cu_vector_list_[i];
1328       elfcpp::Swap<32, false>::writeval(pov, cu_vec->size());
1329       pov += 4;
1330       for (unsigned int j = 0; j < cu_vec->size(); ++j)
1331         {
1332           int cu_index = (*cu_vec)[j].first;
1333           uint8_t flags = (*cu_vec)[j].second;
1334           if (cu_index < 0)
1335             cu_index = comp_units_count + (-1 - cu_index);
1336           cu_index |= flags << 24;
1337           elfcpp::Swap<32, false>::writeval(pov, cu_index);
1338           pov += 4;
1339         }
1340     }
1341
1342   gold_assert(pov - oview == this->stringpool_offset_);
1343
1344   // Write the strings into the constant pool.
1345   this->stringpool_.write_to_buffer(pov, oview_size - this->stringpool_offset_);
1346
1347   of->write_output_view(off, oview_size, oview);
1348 }
1349
1350 // Print usage statistics.
1351 void
1352 Gdb_index::print_stats()
1353 {
1354   if (parameters->options().gdb_index())
1355     Gdb_index_info_reader::print_stats();
1356 }
1357
1358 } // End namespace gold.