1// gdb-index.cc -- generate .gdb_index section for fast debug lookup
2
3// Copyright (C) 2012-2017 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
32namespace gold
33{
34
35const int gdb_index_version = 7;
36
37// Sizes of various records in the .gdb_index section.
38const int gdb_index_offset_size = 4;
39const int gdb_index_hdr_size = 6 * gdb_index_offset_size;
40const int gdb_index_cu_size = 16;
41const int gdb_index_tu_size = 24;
42const int gdb_index_addr_size = 16 + gdb_index_offset_size;
43const 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
50template <typename T>
51class 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
158static unsigned int
159mapped_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
176class 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.
310unsigned int Gdb_index_info_reader::dwarf_cu_count = 0;
311// Number of DWARF compilation units without pubnames/pubtypes.
312unsigned int Gdb_index_info_reader::dwarf_cu_nopubnames_count = 0;
313// Total number of DWARF type units processed.
314unsigned int Gdb_index_info_reader::dwarf_tu_count = 0;
315// Number of DWARF type units without pubnames/pubtypes.
316unsigned int Gdb_index_info_reader::dwarf_tu_nopubnames_count = 0;
317
318// Process a compilation unit and parse its child DIE.
319
320void
321Gdb_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
331void
332Gdb_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
350void
351Gdb_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		|| this->cu_language_ == elfcpp::DW_LANG_Fortran03
376		|| this->cu_language_ == elfcpp::DW_LANG_Fortran08)
377	      {
378		gold_warning(_("%s: --gdb-index currently supports "
379			       "only C and C++ languages"),
380			     this->object()->name().c_str());
381		return;
382	      }
383	    if (die->tag() == elfcpp::DW_TAG_compile_unit)
384	      ++Gdb_index_info_reader::dwarf_cu_nopubnames_count;
385	    else
386	      ++Gdb_index_info_reader::dwarf_tu_nopubnames_count;
387	    this->visit_children(die, NULL);
388	  }
389	break;
390      default:
391	// The top level DIE should be one of the above.
392	gold_warning(_("%s: top level DIE is not DW_TAG_compile_unit "
393		       "or DW_TAG_type_unit"),
394		     this->object()->name().c_str());
395	return;
396    }
397}
398
399// Visit the children of PARENT, looking for symbols to add to the index.
400// CONTEXT points to the DIE to use for constructing the qualified name --
401// NULL if PARENT is the top-level DIE; otherwise it is the same as PARENT.
402
403void
404Gdb_index_info_reader::visit_children(Dwarf_die* parent, Dwarf_die* context)
405{
406  off_t next_offset = 0;
407  for (off_t die_offset = parent->child_offset();
408       die_offset != 0;
409       die_offset = next_offset)
410    {
411      Dwarf_die die(this, die_offset, parent);
412      if (die.tag() == 0)
413	break;
414      this->visit_die(&die, context);
415      next_offset = die.sibling_offset();
416    }
417}
418
419// Visit a child DIE, looking for symbols to add to the index.
420// CONTEXT is the parent DIE, used for constructing the qualified name;
421// it is NULL if the parent DIE is the top-level DIE.
422
423void
424Gdb_index_info_reader::visit_die(Dwarf_die* die, Dwarf_die* context)
425{
426  switch (die->tag())
427    {
428      case elfcpp::DW_TAG_subprogram:
429      case elfcpp::DW_TAG_constant:
430      case elfcpp::DW_TAG_variable:
431      case elfcpp::DW_TAG_enumerator:
432      case elfcpp::DW_TAG_base_type:
433	if (die->is_declaration())
434	  this->add_declaration(die, context);
435	else
436	  {
437	    // If the DIE is not a declaration, add it to the index.
438	    std::string full_name = this->get_qualified_name(die, context);
439	    if (!full_name.empty())
440	      this->gdb_index_->add_symbol(this->cu_index_,
441                                           full_name.c_str(), 0);
442	  }
443	break;
444      case elfcpp::DW_TAG_typedef:
445      case elfcpp::DW_TAG_union_type:
446      case elfcpp::DW_TAG_class_type:
447      case elfcpp::DW_TAG_interface_type:
448      case elfcpp::DW_TAG_structure_type:
449      case elfcpp::DW_TAG_enumeration_type:
450      case elfcpp::DW_TAG_subrange_type:
451      case elfcpp::DW_TAG_namespace:
452	{
453	  std::string full_name;
454
455	  // For classes at the top level, we need to look for a
456	  // member function with a linkage name in order to get
457	  // the properly-canonicalized name.
458	  if (context == NULL
459	      && (die->tag() == elfcpp::DW_TAG_class_type
460		  || die->tag() == elfcpp::DW_TAG_structure_type
461		  || die->tag() == elfcpp::DW_TAG_union_type))
462	    full_name.assign(this->guess_full_class_name(die));
463
464	  // Because we will visit the children, we need to add this DIE
465	  // to the declarations table.
466	  if (full_name.empty())
467	    this->add_declaration(die, context);
468	  else
469	    this->add_declaration_with_full_name(die, full_name.c_str());
470
471	  // If the DIE is not a declaration, add it to the index.
472	  // Gdb stores a namespace in the index even when it is
473	  // a declaration.
474	  if (die->tag() == elfcpp::DW_TAG_namespace
475	      || !die->is_declaration())
476	    {
477	      if (full_name.empty())
478		full_name = this->get_qualified_name(die, context);
479	      if (!full_name.empty())
480		this->gdb_index_->add_symbol(this->cu_index_,
481					     full_name.c_str(), 0);
482	    }
483
484	  // We're interested in the children only for namespaces and
485	  // enumeration types.  For enumeration types, we do not include
486	  // the enumeration tag as part of the full name.  For other tags,
487	  // visit the children only to collect declarations.
488	  if (die->tag() == elfcpp::DW_TAG_namespace
489	      || die->tag() == elfcpp::DW_TAG_enumeration_type)
490	    this->visit_children(die, die);
491	  else
492	    this->visit_children_for_decls(die);
493	}
494	break;
495      default:
496	break;
497    }
498}
499
500// Visit the children of PARENT, looking only for declarations that
501// may be referenced by later specification DIEs.
502
503void
504Gdb_index_info_reader::visit_children_for_decls(Dwarf_die* parent)
505{
506  off_t next_offset = 0;
507  for (off_t die_offset = parent->child_offset();
508       die_offset != 0;
509       die_offset = next_offset)
510    {
511      Dwarf_die die(this, die_offset, parent);
512      if (die.tag() == 0)
513	break;
514      this->visit_die_for_decls(&die, parent);
515      next_offset = die.sibling_offset();
516    }
517}
518
519// Visit a child DIE, looking only for declarations that
520// may be referenced by later specification DIEs.
521
522void
523Gdb_index_info_reader::visit_die_for_decls(Dwarf_die* die, Dwarf_die* context)
524{
525  switch (die->tag())
526    {
527      case elfcpp::DW_TAG_subprogram:
528      case elfcpp::DW_TAG_constant:
529      case elfcpp::DW_TAG_variable:
530      case elfcpp::DW_TAG_enumerator:
531      case elfcpp::DW_TAG_base_type:
532	{
533	  if (die->is_declaration())
534	    this->add_declaration(die, context);
535	}
536	break;
537      case elfcpp::DW_TAG_typedef:
538      case elfcpp::DW_TAG_union_type:
539      case elfcpp::DW_TAG_class_type:
540      case elfcpp::DW_TAG_interface_type:
541      case elfcpp::DW_TAG_structure_type:
542      case elfcpp::DW_TAG_enumeration_type:
543      case elfcpp::DW_TAG_subrange_type:
544      case elfcpp::DW_TAG_namespace:
545	{
546	  if (die->is_declaration())
547	    this->add_declaration(die, context);
548	  this->visit_children_for_decls(die);
549	}
550	break;
551      default:
552	break;
553    }
554}
555
556// Extract the class name from the linkage name of a member function.
557// This code is adapted from ../gdb/cp-support.c.
558
559#define d_left(dc) (dc)->u.s_binary.left
560#define d_right(dc) (dc)->u.s_binary.right
561
562static char*
563class_name_from_linkage_name(const char* linkage_name)
564{
565  void* storage;
566  struct demangle_component* tree =
567      cplus_demangle_v3_components(linkage_name, DMGL_NO_OPTS, &storage);
568  if (tree == NULL)
569    return NULL;
570
571  int done = 0;
572
573  // First strip off any qualifiers, if we have a function or
574  // method.
575  while (!done)
576    switch (tree->type)
577      {
578	case DEMANGLE_COMPONENT_CONST:
579	case DEMANGLE_COMPONENT_RESTRICT:
580	case DEMANGLE_COMPONENT_VOLATILE:
581	case DEMANGLE_COMPONENT_CONST_THIS:
582	case DEMANGLE_COMPONENT_RESTRICT_THIS:
583	case DEMANGLE_COMPONENT_VOLATILE_THIS:
584	case DEMANGLE_COMPONENT_VENDOR_TYPE_QUAL:
585	  tree = d_left(tree);
586	  break;
587	default:
588	  done = 1;
589	  break;
590      }
591
592  // If what we have now is a function, discard the argument list.
593  if (tree->type == DEMANGLE_COMPONENT_TYPED_NAME)
594    tree = d_left(tree);
595
596  // If what we have now is a template, strip off the template
597  // arguments.  The left subtree may be a qualified name.
598  if (tree->type == DEMANGLE_COMPONENT_TEMPLATE)
599    tree = d_left(tree);
600
601  // What we have now should be a name, possibly qualified.
602  // Additional qualifiers could live in the left subtree or the right
603  // subtree.  Find the last piece.
604  done = 0;
605  struct demangle_component* prev_comp = NULL;
606  struct demangle_component* cur_comp = tree;
607  while (!done)
608    switch (cur_comp->type)
609      {
610	case DEMANGLE_COMPONENT_QUAL_NAME:
611	case DEMANGLE_COMPONENT_LOCAL_NAME:
612	  prev_comp = cur_comp;
613	  cur_comp = d_right(cur_comp);
614	  break;
615	case DEMANGLE_COMPONENT_TEMPLATE:
616	case DEMANGLE_COMPONENT_NAME:
617	case DEMANGLE_COMPONENT_CTOR:
618	case DEMANGLE_COMPONENT_DTOR:
619	case DEMANGLE_COMPONENT_OPERATOR:
620	case DEMANGLE_COMPONENT_EXTENDED_OPERATOR:
621	  done = 1;
622	  break;
623	default:
624	  done = 1;
625	  cur_comp = NULL;
626	  break;
627      }
628
629  char* ret = NULL;
630  if (cur_comp != NULL && prev_comp != NULL)
631    {
632      // We want to discard the rightmost child of PREV_COMP.
633      *prev_comp = *d_left(prev_comp);
634      size_t allocated_size;
635      ret = cplus_demangle_print(DMGL_NO_OPTS, tree, 30, &allocated_size);
636    }
637
638  free(storage);
639  return ret;
640}
641
642// Guess a fully-qualified name for a class type, based on member function
643// linkage names.  This is needed for class/struct/union types at the
644// top level, because GCC does not always properly embed them within
645// the namespace.  As in gdb, we look for a member function with a linkage
646// name and extract the qualified name from the demangled name.
647
648std::string
649Gdb_index_info_reader::guess_full_class_name(Dwarf_die* die)
650{
651  std::string full_name;
652  off_t next_offset = 0;
653
654  // This routine scans ahead in the DIE structure, possibly advancing
655  // the relocation tracker beyond the current DIE.  We need to checkpoint
656  // the tracker and reset it when we're done.
657  uint64_t checkpoint = this->get_reloc_checkpoint();
658
659  for (off_t child_offset = die->child_offset();
660       child_offset != 0;
661       child_offset = next_offset)
662    {
663      Dwarf_die child(this, child_offset, die);
664      if (child.tag() == 0)
665	break;
666      if (child.tag() == elfcpp::DW_TAG_subprogram)
667        {
668          const char* linkage_name = child.linkage_name();
669	  if (linkage_name != NULL)
670	    {
671	      char* guess = class_name_from_linkage_name(linkage_name);
672	      if (guess != NULL)
673	        {
674		  full_name.assign(guess);
675		  free(guess);
676		  break;
677	        }
678	    }
679        }
680      next_offset = child.sibling_offset();
681    }
682
683  this->reset_relocs(checkpoint);
684  return full_name;
685}
686
687// Add a declaration DIE to the table of declarations.
688
689void
690Gdb_index_info_reader::add_declaration(Dwarf_die* die, Dwarf_die* context)
691{
692  const char* name = die->name();
693
694  off_t parent_offset = context != NULL ? context->offset() : 0;
695
696  // If this DIE has a DW_AT_specification or DW_AT_abstract_origin
697  // attribute, use the parent and name from the earlier declaration.
698  off_t spec = die->specification();
699  if (spec == 0)
700    spec = die->abstract_origin();
701  if (spec > 0)
702    {
703      Declaration_map::iterator it = this->declarations_.find(spec);
704      if (it != this->declarations_.end())
705        {
706	  parent_offset = it->second.parent_offset_;
707	  name = it->second.name_;
708        }
709    }
710
711  if (name == NULL)
712    {
713      if (die->tag() == elfcpp::DW_TAG_namespace)
714        name = "(anonymous namespace)";
715      else if (die->tag() == elfcpp::DW_TAG_union_type)
716        name = "(anonymous union)";
717      else
718        name = "(unknown)";
719    }
720
721  Declaration_pair decl(parent_offset, name);
722  this->declarations_.insert(std::make_pair(die->offset(), decl));
723}
724
725// Add a declaration whose fully-qualified name is already known.
726// In the case where we had to get the canonical name by demangling
727// a linkage name, this ensures we use that name instead of the one
728// provided in DW_AT_name.
729
730void
731Gdb_index_info_reader::add_declaration_with_full_name(
732    Dwarf_die* die,
733    const char* full_name)
734{
735  // We need to copy the name.
736  int len = strlen(full_name);
737  char* copy = new char[len + 1];
738  memcpy(copy, full_name, len + 1);
739
740  // Flag that we now manage the memory this points to.
741  Declaration_pair decl(-1, copy);
742  this->declarations_.insert(std::make_pair(die->offset(), decl));
743}
744
745// Return the context for a DIE whose parent is at DIE_OFFSET.
746
747std::string
748Gdb_index_info_reader::get_context(off_t die_offset)
749{
750  std::string context;
751  Declaration_map::iterator it = this->declarations_.find(die_offset);
752  if (it != this->declarations_.end())
753    {
754      off_t parent_offset = it->second.parent_offset_;
755      if (parent_offset > 0)
756	{
757	  context = get_context(parent_offset);
758	  context.append("::");
759	}
760      if (it->second.name_ != NULL)
761        context.append(it->second.name_);
762    }
763  return context;
764}
765
766// Construct the fully-qualified name for DIE.
767
768std::string
769Gdb_index_info_reader::get_qualified_name(Dwarf_die* die, Dwarf_die* context)
770{
771  std::string full_name;
772  const char* name = die->name();
773
774  off_t parent_offset = context != NULL ? context->offset() : 0;
775
776  // If this DIE has a DW_AT_specification or DW_AT_abstract_origin
777  // attribute, use the parent and name from the earlier declaration.
778  off_t spec = die->specification();
779  if (spec == 0)
780    spec = die->abstract_origin();
781  if (spec > 0)
782    {
783      Declaration_map::iterator it = this->declarations_.find(spec);
784      if (it != this->declarations_.end())
785        {
786	  parent_offset = it->second.parent_offset_;
787	  name = it->second.name_;
788        }
789    }
790
791  if (name == NULL && die->tag() == elfcpp::DW_TAG_namespace)
792    name = "(anonymous namespace)";
793  else if (name == NULL)
794    return full_name;
795
796  // If this is an enumerator constant, skip the immediate parent,
797  // which is the enumeration tag.
798  if (die->tag() == elfcpp::DW_TAG_enumerator)
799    {
800      Declaration_map::iterator it = this->declarations_.find(parent_offset);
801      if (it != this->declarations_.end())
802	parent_offset = it->second.parent_offset_;
803    }
804
805  if (parent_offset > 0)
806    {
807      full_name.assign(this->get_context(parent_offset));
808      full_name.append("::");
809    }
810  full_name.append(name);
811
812  return full_name;
813}
814
815// Record the address ranges for a compilation unit.
816
817void
818Gdb_index_info_reader::record_cu_ranges(Dwarf_die* die)
819{
820  unsigned int shndx;
821  unsigned int shndx2;
822
823  off_t ranges_offset = die->ref_attribute(elfcpp::DW_AT_ranges, &shndx);
824  if (ranges_offset != -1)
825    {
826      Dwarf_range_list* ranges = this->read_range_list(shndx, ranges_offset);
827      if (ranges != NULL)
828	this->gdb_index_->add_address_range_list(this->object(),
829						 this->cu_index_, ranges);
830      return;
831    }
832
833  off_t low_pc = die->address_attribute(elfcpp::DW_AT_low_pc, &shndx);
834  off_t high_pc = die->address_attribute(elfcpp::DW_AT_high_pc, &shndx2);
835  if (high_pc == -1)
836    {
837      high_pc = die->uint_attribute(elfcpp::DW_AT_high_pc);
838      high_pc += low_pc;
839      shndx2 = shndx;
840    }
841  if ((low_pc != 0 || high_pc != 0) && low_pc != -1)
842    {
843      if (shndx != shndx2)
844        {
845	  gold_warning(_("%s: DWARF info may be corrupt; low_pc and high_pc "
846			 "are in different sections"),
847		       this->object()->name().c_str());
848	  return;
849	}
850      if (shndx == 0 || this->object()->is_section_included(shndx))
851        {
852	  Dwarf_range_list* ranges = new Dwarf_range_list();
853	  ranges->add(shndx, low_pc, high_pc);
854	  this->gdb_index_->add_address_range_list(this->object(),
855						   this->cu_index_, ranges);
856        }
857    }
858}
859
860// Read table and add the relevant names to the index.  Returns true
861// if any names were added.
862
863bool
864Gdb_index_info_reader::read_pubtable(Dwarf_pubnames_table* table, off_t offset)
865{
866  // If we couldn't read the section when building the cu_pubname_map,
867  // then we won't find any pubnames now.
868  if (table == NULL)
869    return false;
870
871  if (!table->read_header(offset))
872    return false;
873  while (true)
874    {
875      uint8_t flag_byte;
876      const char* name = table->next_name(&flag_byte);
877      if (name == NULL)
878        break;
879
880      this->gdb_index_->add_symbol(this->cu_index_, name, flag_byte);
881    }
882  return true;
883}
884
885// Read the .debug_pubnames and .debug_pubtypes tables for the CU or TU.
886// Returns TRUE if either a pubnames or pubtypes section was found.
887
888bool
889Gdb_index_info_reader::read_pubnames_and_pubtypes(Dwarf_die* die)
890{
891  // If this is a skeleton debug-type die (generated via
892  // -gsplit-dwarf), then the associated pubnames should have been
893  // read along with the corresponding CU.  In any case, there isn't
894  // enough info inside to build a gdb index entry.
895  if (die->tag() == elfcpp::DW_TAG_type_unit
896      && die->string_attribute(elfcpp::DW_AT_GNU_dwo_name))
897    return true;
898
899  // We use stmt_list_off as a unique identifier for the
900  // compilation unit and its associated type units.
901  unsigned int shndx;
902  off_t stmt_list_off = die->ref_attribute (elfcpp::DW_AT_stmt_list,
903                                            &shndx);
904  // Look for the attr as either a flag or a ref.
905  off_t offset = die->ref_attribute(elfcpp::DW_AT_GNU_pubnames, &shndx);
906
907  // Newer versions of GCC generate CUs, but not TUs, with
908  // DW_AT_FORM_flag_present.
909  unsigned int flag = die->uint_attribute(elfcpp::DW_AT_GNU_pubnames);
910  if (offset == -1 && flag == 0)
911    {
912      // Didn't find the attribute.
913      if (die->tag() == elfcpp::DW_TAG_type_unit)
914        {
915          // If die is a TU, then it might correspond to a CU which we
916          // have read. If it does, then no need to read the pubnames.
917          // If it doesn't, then the caller will have to parse the
918          // dies manually to find the names.
919          return this->gdb_index_->pubnames_read(this->object(),
920                                                 stmt_list_off);
921        }
922      else
923        {
924          // No attribute on the CU means that no pubnames were read.
925          return false;
926        }
927    }
928
929  // We found the attribute, so we can check if the corresponding
930  // pubnames have been read.
931  if (this->gdb_index_->pubnames_read(this->object(), stmt_list_off))
932    return true;
933
934  this->gdb_index_->set_pubnames_read(this->object(), stmt_list_off);
935
936  // We have an attribute, and the pubnames haven't been read, so read
937  // them.
938  bool names = false;
939  // In some of the cases, we could rely on the previous value of
940  // offset here, but sorting out which cases complicates the logic
941  // enough that it isn't worth it. So just look up the offset again.
942  offset = this->gdb_index_->find_pubname_offset(this->cu_offset());
943  names = this->read_pubtable(this->gdb_index_->pubnames_table(), offset);
944
945  bool types = false;
946  offset = this->gdb_index_->find_pubtype_offset(this->cu_offset());
947  types = this->read_pubtable(this->gdb_index_->pubtypes_table(), offset);
948  return names || types;
949}
950
951// Clear the declarations map.
952void
953Gdb_index_info_reader::clear_declarations()
954{
955  // Free strings in memory we manage.
956  for (Declaration_map::iterator it = this->declarations_.begin();
957       it != this->declarations_.end();
958       ++it)
959    {
960      if (it->second.parent_offset_ == -1)
961	delete[] it->second.name_;
962    }
963
964  this->declarations_.clear();
965}
966
967// Print usage statistics.
968void
969Gdb_index_info_reader::print_stats()
970{
971  fprintf(stderr, _("%s: DWARF CUs: %u\n"),
972          program_name, Gdb_index_info_reader::dwarf_cu_count);
973  fprintf(stderr, _("%s: DWARF CUs without pubnames/pubtypes: %u\n"),
974          program_name, Gdb_index_info_reader::dwarf_cu_nopubnames_count);
975  fprintf(stderr, _("%s: DWARF TUs: %u\n"),
976          program_name, Gdb_index_info_reader::dwarf_tu_count);
977  fprintf(stderr, _("%s: DWARF TUs without pubnames/pubtypes: %u\n"),
978          program_name, Gdb_index_info_reader::dwarf_tu_nopubnames_count);
979}
980
981// Class Gdb_index.
982
983// Construct the .gdb_index section.
984
985Gdb_index::Gdb_index(Output_section* gdb_index_section)
986  : Output_section_data(4),
987    pubnames_table_(NULL),
988    pubtypes_table_(NULL),
989    gdb_index_section_(gdb_index_section),
990    comp_units_(),
991    type_units_(),
992    ranges_(),
993    cu_vector_list_(),
994    cu_vector_offsets_(NULL),
995    stringpool_(),
996    tu_offset_(0),
997    addr_offset_(0),
998    symtab_offset_(0),
999    cu_pool_offset_(0),
1000    stringpool_offset_(0),
1001    pubnames_object_(NULL),
1002    stmt_list_offset_(-1)
1003{
1004  this->gdb_symtab_ = new Gdb_hashtab<Gdb_symbol>();
1005}
1006
1007Gdb_index::~Gdb_index()
1008{
1009  // Free the memory used by the symbol table.
1010  delete this->gdb_symtab_;
1011  // Free the memory used by the CU vectors.
1012  for (unsigned int i = 0; i < this->cu_vector_list_.size(); ++i)
1013    delete this->cu_vector_list_[i];
1014}
1015
1016
1017// Scan the pubnames and pubtypes sections and build a map of the
1018// various cus and tus they refer to, so we can process the entries
1019// when we encounter the die for that cu or tu.
1020// Return the just-read table so it can be cached.
1021
1022Dwarf_pubnames_table*
1023Gdb_index::map_pubtable_to_dies(unsigned int attr,
1024                                Gdb_index_info_reader* dwinfo,
1025                                Relobj* object,
1026                                const unsigned char* symbols,
1027                                off_t symbols_size)
1028{
1029  uint64_t section_offset = 0;
1030  Dwarf_pubnames_table* table;
1031  Pubname_offset_map* map;
1032
1033  if (attr == elfcpp::DW_AT_GNU_pubnames)
1034    {
1035      table = new Dwarf_pubnames_table(dwinfo, false);
1036      map = &this->cu_pubname_map_;
1037    }
1038  else
1039    {
1040      table = new Dwarf_pubnames_table(dwinfo, true);
1041      map = &this->cu_pubtype_map_;
1042    }
1043
1044  map->clear();
1045  if (!table->read_section(object, symbols, symbols_size))
1046    return NULL;
1047
1048  while (table->read_header(section_offset))
1049    {
1050      map->insert(std::make_pair(table->cu_offset(), section_offset));
1051      section_offset += table->subsection_size();
1052    }
1053
1054  return table;
1055}
1056
1057// Wrapper for map_pubtable_to_dies
1058
1059void
1060Gdb_index::map_pubnames_and_types_to_dies(Gdb_index_info_reader* dwinfo,
1061                                          Relobj* object,
1062                                          const unsigned char* symbols,
1063                                          off_t symbols_size)
1064{
1065  // This is a new object, so reset the relevant variables.
1066  this->pubnames_object_ = object;
1067  this->stmt_list_offset_ = -1;
1068
1069  delete this->pubnames_table_;
1070  this->pubnames_table_
1071      = this->map_pubtable_to_dies(elfcpp::DW_AT_GNU_pubnames, dwinfo,
1072                                   object, symbols, symbols_size);
1073  delete this->pubtypes_table_;
1074  this->pubtypes_table_
1075      = this->map_pubtable_to_dies(elfcpp::DW_AT_GNU_pubtypes, dwinfo,
1076                                   object, symbols, symbols_size);
1077}
1078
1079// Given a cu_offset, find the associated section of the pubnames
1080// table.
1081
1082off_t
1083Gdb_index::find_pubname_offset(off_t cu_offset)
1084{
1085  Pubname_offset_map::iterator it = this->cu_pubname_map_.find(cu_offset);
1086  if (it != this->cu_pubname_map_.end())
1087    return it->second;
1088  return -1;
1089}
1090
1091// Given a cu_offset, find the associated section of the pubnames
1092// table.
1093
1094off_t
1095Gdb_index::find_pubtype_offset(off_t cu_offset)
1096{
1097  Pubname_offset_map::iterator it = this->cu_pubtype_map_.find(cu_offset);
1098  if (it != this->cu_pubtype_map_.end())
1099    return it->second;
1100  return -1;
1101}
1102
1103// Scan a .debug_info or .debug_types input section.
1104
1105void
1106Gdb_index::scan_debug_info(bool is_type_unit,
1107			   Relobj* object,
1108			   const unsigned char* symbols,
1109			   off_t symbols_size,
1110			   unsigned int shndx,
1111			   unsigned int reloc_shndx,
1112			   unsigned int reloc_type)
1113{
1114  Gdb_index_info_reader dwinfo(is_type_unit, object,
1115			       symbols, symbols_size,
1116			       shndx, reloc_shndx,
1117			       reloc_type, this);
1118  if (object != this->pubnames_object_)
1119    map_pubnames_and_types_to_dies(&dwinfo, object, symbols, symbols_size);
1120  dwinfo.parse();
1121}
1122
1123// Add a symbol.
1124
1125void
1126Gdb_index::add_symbol(int cu_index, const char* sym_name, uint8_t flags)
1127{
1128  unsigned int hash = mapped_index_string_hash(
1129      reinterpret_cast<const unsigned char*>(sym_name));
1130  Gdb_symbol* sym = new Gdb_symbol();
1131  this->stringpool_.add(sym_name, true, &sym->name_key);
1132  sym->hashval = hash;
1133  sym->cu_vector_index = 0;
1134
1135  Gdb_symbol* found = this->gdb_symtab_->add(sym);
1136  if (found == sym)
1137    {
1138      // New symbol -- allocate a new CU index vector.
1139      found->cu_vector_index = this->cu_vector_list_.size();
1140      this->cu_vector_list_.push_back(new Cu_vector());
1141    }
1142  else
1143    {
1144      // Found an existing symbol -- append to the existing
1145      // CU index vector.
1146      delete sym;
1147    }
1148
1149  // Add the CU index to the vector list for this symbol,
1150  // if it's not already on the list.  We only need to
1151  // check the last added entry.
1152  Cu_vector* cu_vec = this->cu_vector_list_[found->cu_vector_index];
1153  if (cu_vec->size() == 0
1154      || cu_vec->back().first != cu_index
1155      || cu_vec->back().second != flags)
1156    cu_vec->push_back(std::make_pair(cu_index, flags));
1157}
1158
1159// Return TRUE if we have already processed the pubnames associated
1160// with the statement list at the given OFFSET.
1161
1162bool
1163Gdb_index::pubnames_read(const Relobj* object, off_t offset)
1164{
1165  bool ret = (this->pubnames_object_ == object
1166	      && this->stmt_list_offset_ == offset);
1167  return ret;
1168}
1169
1170// Record that we have processed the pubnames associated with the
1171// statement list for OBJECT at the given OFFSET.
1172
1173void
1174Gdb_index::set_pubnames_read(const Relobj* object, off_t offset)
1175{
1176  this->pubnames_object_ = object;
1177  this->stmt_list_offset_ = offset;
1178}
1179
1180// Set the size of the .gdb_index section.
1181
1182void
1183Gdb_index::set_final_data_size()
1184{
1185  // Finalize the string pool.
1186  this->stringpool_.set_string_offsets();
1187
1188  // Compute the total size of the CU vectors.
1189  // For each CU vector, include one entry for the count at the
1190  // beginning of the vector.
1191  unsigned int cu_vector_count = this->cu_vector_list_.size();
1192  unsigned int cu_vector_size = 0;
1193  this->cu_vector_offsets_ = new off_t[cu_vector_count];
1194  for (unsigned int i = 0; i < cu_vector_count; ++i)
1195    {
1196      Cu_vector* cu_vec = this->cu_vector_list_[i];
1197      cu_vector_offsets_[i] = cu_vector_size;
1198      cu_vector_size += gdb_index_offset_size * (cu_vec->size() + 1);
1199    }
1200
1201  // Assign relative offsets to each portion of the index,
1202  // and find the total size of the section.
1203  section_size_type data_size = gdb_index_hdr_size;
1204  data_size += this->comp_units_.size() * gdb_index_cu_size;
1205  this->tu_offset_ = data_size;
1206  data_size += this->type_units_.size() * gdb_index_tu_size;
1207  this->addr_offset_ = data_size;
1208  for (unsigned int i = 0; i < this->ranges_.size(); ++i)
1209    data_size += this->ranges_[i].ranges->size() * gdb_index_addr_size;
1210  this->symtab_offset_ = data_size;
1211  data_size += this->gdb_symtab_->capacity() * gdb_index_sym_size;
1212  this->cu_pool_offset_ = data_size;
1213  data_size += cu_vector_size;
1214  this->stringpool_offset_ = data_size;
1215  data_size += this->stringpool_.get_strtab_size();
1216
1217  this->set_data_size(data_size);
1218}
1219
1220// Write the data to the file.
1221
1222void
1223Gdb_index::do_write(Output_file* of)
1224{
1225  const off_t off = this->offset();
1226  const off_t oview_size = this->data_size();
1227  unsigned char* const oview = of->get_output_view(off, oview_size);
1228  unsigned char* pov = oview;
1229
1230  // Write the file header.
1231  // (1) Version number.
1232  elfcpp::Swap<32, false>::writeval(pov, gdb_index_version);
1233  pov += 4;
1234  // (2) Offset of the CU list.
1235  elfcpp::Swap<32, false>::writeval(pov, gdb_index_hdr_size);
1236  pov += 4;
1237  // (3) Offset of the types CU list.
1238  elfcpp::Swap<32, false>::writeval(pov, this->tu_offset_);
1239  pov += 4;
1240  // (4) Offset of the address area.
1241  elfcpp::Swap<32, false>::writeval(pov, this->addr_offset_);
1242  pov += 4;
1243  // (5) Offset of the symbol table.
1244  elfcpp::Swap<32, false>::writeval(pov, this->symtab_offset_);
1245  pov += 4;
1246  // (6) Offset of the constant pool.
1247  elfcpp::Swap<32, false>::writeval(pov, this->cu_pool_offset_);
1248  pov += 4;
1249
1250  gold_assert(pov - oview == gdb_index_hdr_size);
1251
1252  // Write the CU list.
1253  unsigned int comp_units_count = this->comp_units_.size();
1254  for (unsigned int i = 0; i < comp_units_count; ++i)
1255    {
1256      const Comp_unit& cu = this->comp_units_[i];
1257      elfcpp::Swap<64, false>::writeval(pov, cu.cu_offset);
1258      elfcpp::Swap<64, false>::writeval(pov + 8, cu.cu_length);
1259      pov += 16;
1260    }
1261
1262  gold_assert(pov - oview == this->tu_offset_);
1263
1264  // Write the types CU list.
1265  for (unsigned int i = 0; i < this->type_units_.size(); ++i)
1266    {
1267      const Type_unit& tu = this->type_units_[i];
1268      elfcpp::Swap<64, false>::writeval(pov, tu.tu_offset);
1269      elfcpp::Swap<64, false>::writeval(pov + 8, tu.type_offset);
1270      elfcpp::Swap<64, false>::writeval(pov + 16, tu.type_signature);
1271      pov += 24;
1272    }
1273
1274  gold_assert(pov - oview == this->addr_offset_);
1275
1276  // Write the address area.
1277  for (unsigned int i = 0; i < this->ranges_.size(); ++i)
1278    {
1279      int cu_index = this->ranges_[i].cu_index;
1280      // Translate negative indexes, which refer to a TU, to a
1281      // logical index into a concatenated CU/TU list.
1282      if (cu_index < 0)
1283        cu_index = comp_units_count + (-1 - cu_index);
1284      Relobj* object = this->ranges_[i].object;
1285      const Dwarf_range_list& ranges = *this->ranges_[i].ranges;
1286      for (unsigned int j = 0; j < ranges.size(); ++j)
1287        {
1288	  const Dwarf_range_list::Range& range = ranges[j];
1289	  uint64_t base = 0;
1290	  if (range.shndx > 0)
1291	    {
1292	      const Output_section* os = object->output_section(range.shndx);
1293	      base = (os->address()
1294		      + object->output_section_offset(range.shndx));
1295	    }
1296	  elfcpp::Swap_aligned32<64, false>::writeval(pov, base + range.start);
1297	  elfcpp::Swap_aligned32<64, false>::writeval(pov + 8,
1298						      base + range.end);
1299	  elfcpp::Swap<32, false>::writeval(pov + 16, cu_index);
1300	  pov += 20;
1301	}
1302    }
1303
1304  gold_assert(pov - oview == this->symtab_offset_);
1305
1306  // Write the symbol table.
1307  for (unsigned int i = 0; i < this->gdb_symtab_->capacity(); ++i)
1308    {
1309      const Gdb_symbol* sym = (*this->gdb_symtab_)[i];
1310      section_offset_type name_offset = 0;
1311      unsigned int cu_vector_offset = 0;
1312      if (sym != NULL)
1313	{
1314	  name_offset = (this->stringpool_.get_offset_from_key(sym->name_key)
1315			 + this->stringpool_offset_ - this->cu_pool_offset_);
1316	  cu_vector_offset = this->cu_vector_offsets_[sym->cu_vector_index];
1317	}
1318      elfcpp::Swap<32, false>::writeval(pov, name_offset);
1319      elfcpp::Swap<32, false>::writeval(pov + 4, cu_vector_offset);
1320      pov += 8;
1321    }
1322
1323  gold_assert(pov - oview == this->cu_pool_offset_);
1324
1325  // Write the CU vectors into the constant pool.
1326  for (unsigned int i = 0; i < this->cu_vector_list_.size(); ++i)
1327    {
1328      Cu_vector* cu_vec = this->cu_vector_list_[i];
1329      elfcpp::Swap<32, false>::writeval(pov, cu_vec->size());
1330      pov += 4;
1331      for (unsigned int j = 0; j < cu_vec->size(); ++j)
1332	{
1333	  int cu_index = (*cu_vec)[j].first;
1334          uint8_t flags = (*cu_vec)[j].second;
1335	  if (cu_index < 0)
1336	    cu_index = comp_units_count + (-1 - cu_index);
1337          cu_index |= flags << 24;
1338	  elfcpp::Swap<32, false>::writeval(pov, cu_index);
1339	  pov += 4;
1340	}
1341    }
1342
1343  gold_assert(pov - oview == this->stringpool_offset_);
1344
1345  // Write the strings into the constant pool.
1346  this->stringpool_.write_to_buffer(pov, oview_size - this->stringpool_offset_);
1347
1348  of->write_output_view(off, oview_size, oview);
1349}
1350
1351// Print usage statistics.
1352void
1353Gdb_index::print_stats()
1354{
1355  if (parameters->options().gdb_index())
1356    Gdb_index_info_reader::print_stats();
1357}
1358
1359} // End namespace gold.
1360