gdb/dwarf2read.h

   1 /* DWARF 2 debugging format support for GDB.
   2
   3    Copyright (C) 1994-2020 Free Software Foundation, Inc.
   4
   5    This file is part of GDB.
   6
   7    This program is free software; you can redistribute it and/or modify
   8    it under the terms of the GNU General Public License as published by
   9    the Free Software Foundation; either version 3 of the License, or
  10    (at your option) any later version.
  11
  12    This program is distributed in the hope that it will be useful,
  13    but WITHOUT ANY WARRANTY; without even the implied warranty of
  14    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15    GNU General Public License for more details.
  16
  17    You should have received a copy of the GNU General Public License
  18    along with this program.  If not, see <http://www.gnu.org/licenses/>.  */
  19
  20 #ifndef DWARF2READ_H
  21 #define DWARF2READ_H
  22
  23 #include <unordered_map>
  24 #include "dwarf-index-cache.h"
  25 #include "filename-seen-cache.h"
  26 #include "gdb_obstack.h"
  27 #include "gdbsupport/hash_enum.h"
  28 #include "psympriv.h"
  29
  30 /* Hold 'maintenance (set|show) dwarf' commands.  */
  31 extern struct cmd_list_element *set_dwarf_cmdlist;
  32 extern struct cmd_list_element *show_dwarf_cmdlist;
  33
  34 extern bool dwarf_always_disassemble;
  35
  36 /* A descriptor for dwarf sections.
  37
  38    S.ASECTION, SIZE are typically initialized when the objfile is first
  39    scanned.  BUFFER, READIN are filled in later when the section is read.
  40    If the section contained compressed data then SIZE is updated to record
  41    the uncompressed size of the section.
  42
  43    DWP file format V2 introduces a wrinkle that is easiest to handle by
  44    creating the concept of virtual sections contained within a real section.
  45    In DWP V2 the sections of the input DWO files are concatenated together
  46    into one section, but section offsets are kept relative to the original
  47    input section.
  48    If this is a virtual dwp-v2 section, S.CONTAINING_SECTION is a backlink to
  49    the real section this "virtual" section is contained in, and BUFFER,SIZE
  50    describe the virtual section.  */
  51
  52 struct dwarf2_section_info
  53 {
  54   union
  55   {
  56     /* If this is a real section, the bfd section.  */
  57     asection *section;
  58     /* If this is a virtual section, pointer to the containing ("real")
  59        section.  */
  60     struct dwarf2_section_info *containing_section;
  61   } s;
  62   /* Pointer to section data, only valid if readin.  */
  63   const gdb_byte *buffer;
  64   /* The size of the section, real or virtual.  */
  65   bfd_size_type size;
  66   /* If this is a virtual section, the offset in the real section.
  67      Only valid if is_virtual.  */
  68   bfd_size_type virtual_offset;
  69   /* True if we have tried to read this section.  */
  70   bool readin;
  71   /* True if this is a virtual section, False otherwise.
  72      This specifies which of s.section and s.containing_section to use.  */
  73   bool is_virtual;
  74 };
  75
  76 /* Read the contents of the section INFO.
  77    OBJFILE is the main object file, but not necessarily the file where
  78    the section comes from.  E.g., for DWO files the bfd of INFO is the bfd
  79    of the DWO file.
  80    If the section is compressed, uncompress it before returning.  */
  81
  82 void dwarf2_read_section (struct objfile *objfile, dwarf2_section_info *info);
  83
  84 struct tu_stats
  85 {
  86   int nr_uniq_abbrev_tables;
  87   int nr_symtabs;
  88   int nr_symtab_sharers;
  89   int nr_stmt_less_type_units;
  90   int nr_all_type_units_reallocs;
  91 };
  92
  93 struct dwarf2_debug_sections;
  94 struct mapped_index;
  95 struct mapped_debug_names;
  96 struct signatured_type;
  97 struct die_info;
  98 typedef struct die_info *die_info_ptr;
  99
 100 /* Collection of data recorded per objfile.
 101    This hangs off of dwarf2_objfile_data_key.  */
 102
 103 struct dwarf2_per_objfile
 104 {
 105   /* Construct a dwarf2_per_objfile for OBJFILE.  NAMES points to the
 106      dwarf2 section names, or is NULL if the standard ELF names are
 107      used.  CAN_COPY is true for formats where symbol
 108      interposition is possible and so symbol values must follow copy
 109      relocation rules.  */
 110   dwarf2_per_objfile (struct objfile *objfile,
 111                       const dwarf2_debug_sections *names,
 112                       bool can_copy);
 113
 114   ~dwarf2_per_objfile ();
 115
 116   DISABLE_COPY_AND_ASSIGN (dwarf2_per_objfile);
 117
 118   /* Return the CU/TU given its index.
 119
 120      This is intended for loops like:
 121
 122      for (i = 0; i < (dwarf2_per_objfile->n_comp_units
 123                       + dwarf2_per_objfile->n_type_units); ++i)
 124        {
 125          dwarf2_per_cu_data *per_cu = dwarf2_per_objfile->get_cutu (i);
 126
 127          ...;
 128        }
 129   */
 130   dwarf2_per_cu_data *get_cutu (int index);
 131
 132   /* Return the CU given its index.
 133      This differs from get_cutu in that it's for when you know INDEX refers to a
 134      CU.  */
 135   dwarf2_per_cu_data *get_cu (int index);
 136
 137   /* Return the TU given its index.
 138      This differs from get_cutu in that it's for when you know INDEX refers to a
 139      TU.  */
 140   signatured_type *get_tu (int index);
 141
 142   /* Free all cached compilation units.  */
 143   void free_cached_comp_units ();
 144 private:
 145   /* This function is mapped across the sections and remembers the
 146      offset and size of each of the debugging sections we are
 147      interested in.  */
 148   void locate_sections (bfd *abfd, asection *sectp,
 149                         const dwarf2_debug_sections &names);
 150
 151 public:
 152   dwarf2_section_info info {};
 153   dwarf2_section_info abbrev {};
 154   dwarf2_section_info line {};
 155   dwarf2_section_info loc {};
 156   dwarf2_section_info loclists {};
 157   dwarf2_section_info macinfo {};
 158   dwarf2_section_info macro {};
 159   dwarf2_section_info str {};
 160   dwarf2_section_info str_offsets {};
 161   dwarf2_section_info line_str {};
 162   dwarf2_section_info ranges {};
 163   dwarf2_section_info rnglists {};
 164   dwarf2_section_info addr {};
 165   dwarf2_section_info frame {};
 166   dwarf2_section_info eh_frame {};
 167   dwarf2_section_info gdb_index {};
 168   dwarf2_section_info debug_names {};
 169   dwarf2_section_info debug_aranges {};
 170
 171   std::vector<dwarf2_section_info> types;
 172
 173   /* Back link.  */
 174   struct objfile *objfile = NULL;
 175
 176   /* Table of all the compilation units.  This is used to locate
 177      the target compilation unit of a particular reference.  */
 178   std::vector<dwarf2_per_cu_data *> all_comp_units;
 179
 180   /* The .debug_types-related CUs (TUs).  */
 181   std::vector<signatured_type *> all_type_units;
 182
 183   /* Table of struct type_unit_group objects.
 184      The hash key is the DW_AT_stmt_list value.  */
 185   htab_t type_unit_groups {};
 186
 187   /* A table mapping .debug_types signatures to its signatured_type entry.
 188      This is NULL if the .debug_types section hasn't been read in yet.  */
 189   htab_t signatured_types {};
 190
 191   /* Type unit statistics, to see how well the scaling improvements
 192      are doing.  */
 193   struct tu_stats tu_stats {};
 194
 195   /* A chain of compilation units that are currently read in, so that
 196      they can be freed later.  */
 197   dwarf2_per_cu_data *read_in_chain = NULL;
 198
 199   /* A table mapping DW_AT_dwo_name values to struct dwo_file objects.
 200      This is NULL if the table hasn't been allocated yet.  */
 201   htab_up dwo_files;
 202
 203   /* True if we've checked for whether there is a DWP file.  */
 204   bool dwp_checked = false;
 205
 206   /* The DWP file if there is one, or NULL.  */
 207   std::unique_ptr<struct dwp_file> dwp_file;
 208
 209   /* The shared '.dwz' file, if one exists.  This is used when the
 210      original data was compressed using 'dwz -m'.  */
 211   std::unique_ptr<struct dwz_file> dwz_file;
 212
 213   /* Whether copy relocations are supported by this object format.  */
 214   bool can_copy;
 215
 216   /* A flag indicating whether this objfile has a section loaded at a
 217      VMA of 0.  */
 218   bool has_section_at_zero = false;
 219
 220   /* True if we are using the mapped index,
 221      or we are faking it for OBJF_READNOW's sake.  */
 222   bool using_index = false;
 223
 224   /* The mapped index, or NULL if .gdb_index is missing or not being used.  */
 225   std::unique_ptr<mapped_index> index_table;
 226
 227   /* The mapped index, or NULL if .debug_names is missing or not being used.  */
 228   std::unique_ptr<mapped_debug_names> debug_names_table;
 229
 230   /* When using index_table, this keeps track of all quick_file_names entries.
 231      TUs typically share line table entries with a CU, so we maintain a
 232      separate table of all line table entries to support the sharing.
 233      Note that while there can be way more TUs than CUs, we've already
 234      sorted all the TUs into "type unit groups", grouped by their
 235      DW_AT_stmt_list value.  Therefore the only sharing done here is with a
 236      CU and its associated TU group if there is one.  */
 237   htab_t quick_file_names_table {};
 238
 239   /* Set during partial symbol reading, to prevent queueing of full
 240      symbols.  */
 241   bool reading_partial_symbols = false;
 242
 243   /* Table mapping type DIEs to their struct type *.
 244      This is NULL if not allocated yet.
 245      The mapping is done via (CU/TU + DIE offset) -> type.  */
 246   htab_t die_type_hash {};
 247
 248   /* The CUs we recently read.  */
 249   std::vector<dwarf2_per_cu_data *> just_read_cus;
 250
 251   /* Table containing line_header indexed by offset and offset_in_dwz.  */
 252   htab_t line_header_hash {};
 253
 254   /* Table containing all filenames.  This is an optional because the
 255      table is lazily constructed on first access.  */
 256   gdb::optional<filename_seen_cache> filenames_cache;
 257
 258   /* If we loaded the index from an external file, this contains the
 259      resources associated to the open file, memory mapping, etc.  */
 260   std::unique_ptr<index_cache_resource> index_cache_res;
 261
 262   /* Mapping from abstract origin DIE to concrete DIEs that reference it as
 263      DW_AT_abstract_origin.  */
 264   std::unordered_map<sect_offset, std::vector<sect_offset>,
 265                      gdb::hash_enum<sect_offset>>
 266     abstract_to_concrete;
 267 };
 268
 269 /* Get the dwarf2_per_objfile associated to OBJFILE.  */
 270
 271 dwarf2_per_objfile *get_dwarf2_per_objfile (struct objfile *objfile);
 272
 273 /* A partial symtab specialized for DWARF.  */
 274 struct dwarf2_psymtab : public partial_symtab
 275 {
 276   dwarf2_psymtab (const char *filename, struct objfile *objfile)
 277     : partial_symtab (filename, objfile)
 278   {
 279   }
 280
 281   dwarf2_psymtab (const char *filename, struct objfile *objfile,
 282                   CORE_ADDR addr)
 283     : partial_symtab (filename, objfile, addr)
 284   {
 285   }
 286
 287   void read_symtab (struct objfile *) override;
 288   void expand_psymtab (struct objfile *) override;
 289
 290   struct dwarf2_per_cu_data *per_cu_data;
 291 };
 292
 293 /* Persistent data held for a compilation unit, even when not
 294    processing it.  We put a pointer to this structure in the
 295    psymtab.  */
 296
 297 struct dwarf2_per_cu_data
 298 {
 299   /* The start offset and length of this compilation unit.
 300      NOTE: Unlike comp_unit_head.length, this length includes
 301      initial_length_size.
 302      If the DIE refers to a DWO file, this is always of the original die,
 303      not the DWO file.  */
 304   sect_offset sect_off;
 305   unsigned int length;
 306
 307   /* DWARF standard version this data has been read from (such as 4 or 5).  */
 308   short dwarf_version;
 309
 310   /* Flag indicating this compilation unit will be read in before
 311      any of the current compilation units are processed.  */
 312   unsigned int queued : 1;
 313
 314   /* This flag will be set when reading partial DIEs if we need to load
 315      absolutely all DIEs for this compilation unit, instead of just the ones
 316      we think are interesting.  It gets set if we look for a DIE in the
 317      hash table and don't find it.  */
 318   unsigned int load_all_dies : 1;
 319
 320   /* Non-zero if this CU is from .debug_types.
 321      Struct dwarf2_per_cu_data is contained in struct signatured_type iff
 322      this is non-zero.  */
 323   unsigned int is_debug_types : 1;
 324
 325   /* Non-zero if this CU is from the .dwz file.  */
 326   unsigned int is_dwz : 1;
 327
 328   /* Non-zero if reading a TU directly from a DWO file, bypassing the stub.
 329      This flag is only valid if is_debug_types is true.
 330      We can't read a CU directly from a DWO file: There are required
 331      attributes in the stub.  */
 332   unsigned int reading_dwo_directly : 1;
 333
 334   /* Non-zero if the TU has been read.
 335      This is used to assist the "Stay in DWO Optimization" for Fission:
 336      When reading a DWO, it's faster to read TUs from the DWO instead of
 337      fetching them from random other DWOs (due to comdat folding).
 338      If the TU has already been read, the optimization is unnecessary
 339      (and unwise - we don't want to change where gdb thinks the TU lives
 340      "midflight").
 341      This flag is only valid if is_debug_types is true.  */
 342   unsigned int tu_read : 1;
 343
 344   /* The section this CU/TU lives in.
 345      If the DIE refers to a DWO file, this is always the original die,
 346      not the DWO file.  */
 347   struct dwarf2_section_info *section;
 348
 349   /* Set to non-NULL iff this CU is currently loaded.  When it gets freed out
 350      of the CU cache it gets reset to NULL again.  This is left as NULL for
 351      dummy CUs (a CU header, but nothing else).  */
 352   struct dwarf2_cu *cu;
 353
 354   /* The corresponding dwarf2_per_objfile.  */
 355   struct dwarf2_per_objfile *dwarf2_per_objfile;
 356
 357   /* When dwarf2_per_objfile->using_index is true, the 'quick' field
 358      is active.  Otherwise, the 'psymtab' field is active.  */
 359   union
 360   {
 361     /* The partial symbol table associated with this compilation unit,
 362        or NULL for unread partial units.  */
 363     dwarf2_psymtab *psymtab;
 364
 365     /* Data needed by the "quick" functions.  */
 366     struct dwarf2_per_cu_quick_data *quick;
 367   } v;
 368
 369   /* Return true of IMPORTED_SYMTABS is empty or not yet allocated.  */
 370   bool imported_symtabs_empty () const
 371   {
 372     return (imported_symtabs == nullptr || imported_symtabs->empty ());
 373   }
 374
 375   /* Push P to the back of IMPORTED_SYMTABS, allocated IMPORTED_SYMTABS
 376      first if required.  */
 377   void imported_symtabs_push (dwarf2_per_cu_data *p)
 378   {
 379     if (imported_symtabs == nullptr)
 380       imported_symtabs = new std::vector <dwarf2_per_cu_data *>;
 381     imported_symtabs->push_back (p);
 382   }
 383
 384   /* Return the size of IMPORTED_SYMTABS if it is allocated, otherwise
 385      return 0.  */
 386   size_t imported_symtabs_size () const
 387   {
 388     if (imported_symtabs == nullptr)
 389       return 0;
 390     return imported_symtabs->size ();
 391   }
 392
 393   /* Delete IMPORTED_SYMTABS and set the pointer back to nullptr.  */
 394   void imported_symtabs_free ()
 395   {
 396     delete imported_symtabs;
 397     imported_symtabs = nullptr;
 398   }
 399
 400   /* The CUs we import using DW_TAG_imported_unit.  This is filled in
 401      while reading psymtabs, used to compute the psymtab dependencies,
 402      and then cleared.  Then it is filled in again while reading full
 403      symbols, and only deleted when the objfile is destroyed.
 404
 405      This is also used to work around a difference between the way gold
 406      generates .gdb_index version <=7 and the way gdb does.  Arguably this
 407      is a gold bug.  For symbols coming from TUs, gold records in the index
 408      the CU that includes the TU instead of the TU itself.  This breaks
 409      dw2_lookup_symbol: It assumes that if the index says symbol X lives
 410      in CU/TU Y, then one need only expand Y and a subsequent lookup in Y
 411      will find X.  Alas TUs live in their own symtab, so after expanding CU Y
 412      we need to look in TU Z to find X.  Fortunately, this is akin to
 413      DW_TAG_imported_unit, so we just use the same mechanism: For
 414      .gdb_index version <=7 this also records the TUs that the CU referred
 415      to.  Concurrently with this change gdb was modified to emit version 8
 416      indices so we only pay a price for gold generated indices.
 417      http://sourceware.org/bugzilla/show_bug.cgi?id=15021.
 418
 419      This currently needs to be a public member due to how
 420      dwarf2_per_cu_data is allocated and used.  Ideally in future things
 421      could be refactored to make this private.  Until then please try to
 422      avoid direct access to this member, and instead use the helper
 423      functions above.  */
 424   std::vector <dwarf2_per_cu_data *> *imported_symtabs;
 425 };
 426
 427 /* Entry in the signatured_types hash table.  */
 428
 429 struct signatured_type
 430 {
 431   /* The "per_cu" object of this type.
 432      This struct is used iff per_cu.is_debug_types.
 433      N.B.: This is the first member so that it's easy to convert pointers
 434      between them.  */
 435   struct dwarf2_per_cu_data per_cu;
 436
 437   /* The type's signature.  */
 438   ULONGEST signature;
 439
 440   /* Offset in the TU of the type's DIE, as read from the TU header.
 441      If this TU is a DWO stub and the definition lives in a DWO file
 442      (specified by DW_AT_GNU_dwo_name), this value is unusable.  */
 443   cu_offset type_offset_in_tu;
 444
 445   /* Offset in the section of the type's DIE.
 446      If the definition lives in a DWO file, this is the offset in the
 447      .debug_types.dwo section.
 448      The value is zero until the actual value is known.
 449      Zero is otherwise not a valid section offset.  */
 450   sect_offset type_offset_in_section;
 451
 452   /* Type units are grouped by their DW_AT_stmt_list entry so that they
 453      can share them.  This points to the containing symtab.  */
 454   struct type_unit_group *type_unit_group;
 455
 456   /* The type.
 457      The first time we encounter this type we fully read it in and install it
 458      in the symbol tables.  Subsequent times we only need the type.  */
 459   struct type *type;
 460
 461   /* Containing DWO unit.
 462      This field is valid iff per_cu.reading_dwo_directly.  */
 463   struct dwo_unit *dwo_unit;
 464 };
 465
 466 ULONGEST read_unsigned_leb128 (bfd *, const gdb_byte *, unsigned int *);
 467
 468 /* This represents a '.dwz' file.  */
 469
 470 struct dwz_file
 471 {
 472   dwz_file (gdb_bfd_ref_ptr &&bfd)
 473     : dwz_bfd (std::move (bfd))
 474   {
 475   }
 476
 477   const char *filename () const
 478   {
 479     return bfd_get_filename (this->dwz_bfd.get ());
 480   }
 481
 482   /* A dwz file can only contain a few sections.  */
 483   struct dwarf2_section_info abbrev {};
 484   struct dwarf2_section_info info {};
 485   struct dwarf2_section_info str {};
 486   struct dwarf2_section_info line {};
 487   struct dwarf2_section_info macro {};
 488   struct dwarf2_section_info gdb_index {};
 489   struct dwarf2_section_info debug_names {};
 490
 491   /* The dwz's BFD.  */
 492   gdb_bfd_ref_ptr dwz_bfd;
 493
 494   /* If we loaded the index from an external file, this contains the
 495      resources associated to the open file, memory mapping, etc.  */
 496   std::unique_ptr<index_cache_resource> index_cache_res;
 497 };
 498
 499 /* Open the separate '.dwz' debug file, if needed.  Return NULL if
 500    there is no .gnu_debugaltlink section in the file.  Error if there
 501    is such a section but the file cannot be found.  */
 502
 503 extern struct dwz_file *dwarf2_get_dwz_file
 504     (struct dwarf2_per_objfile *dwarf2_per_objfile);
 505
 506 #endif /* DWARF2READ_H */