Subversion Repositories openrisc

// dwarf_reader.cc -- parse dwarf2/3 debug information

// Copyright 2007, 2008, 2009 Free Software Foundation, Inc.

// Written by Ian Lance Taylor <iant@google.com>.

// This file is part of gold.

// This program is free software; you can redistribute it and/or modify

// it under the terms of the GNU General Public License as published by

// the Free Software Foundation; either version 3 of the License, or

// (at your option) any later version.

// This program is distributed in the hope that it will be useful,

// but WITHOUT ANY WARRANTY; without even the implied warranty of

// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the

// GNU General Public License for more details.

// You should have received a copy of the GNU General Public License

// along with this program; if not, write to the Free Software

// Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston,

// MA 02110-1301, USA.

#include "gold.h"

#include <algorithm>

#include <vector>

#include "elfcpp_swap.h"

#include "dwarf.h"

#include "object.h"

#include "parameters.h"

#include "reloc.h"

#include "dwarf_reader.h"

namespace gold {

// Read an unsigned LEB128 number.  Each byte contains 7 bits of

// information, plus one bit saying whether the number continues or

// not.

uint64_t

read_unsigned_LEB_128(const unsigned char* buffer, size_t* len)

{

  uint64_t result = 0;

  size_t num_read = 0;

  unsigned int shift = 0;

  unsigned char byte;

  do

    {

      if (num_read >= 64 / 7)

        {

          gold_warning(_("Unusually large LEB128 decoded, "

                         "debug information may be corrupted"));

          break;

        }

      byte = *buffer++;

      num_read++;

      result |= (static_cast<uint64_t>(byte & 0x7f)) << shift;

      shift += 7;

    }

  while (byte & 0x80);

  *len = num_read;

  return result;

}

// Read a signed LEB128 number.  These are like regular LEB128

// numbers, except the last byte may have a sign bit set.

int64_t

read_signed_LEB_128(const unsigned char* buffer, size_t* len)

{

  int64_t result = 0;

  int shift = 0;

  size_t num_read = 0;

  unsigned char byte;

  do

    {

      if (num_read >= 64 / 7)

        {

          gold_warning(_("Unusually large LEB128 decoded, "

                         "debug information may be corrupted"));

          break;

        }

      byte = *buffer++;

      num_read++;

      result |= (static_cast<uint64_t>(byte & 0x7f) << shift);

      shift += 7;

    }

  while (byte & 0x80);

  if ((shift < 8 * static_cast<int>(sizeof(result))) && (byte & 0x40))

    result |= -((static_cast<int64_t>(1)) << shift);

  *len = num_read;

  return result;

}

// This is the format of a DWARF2/3 line state machine that we process

// opcodes using.  There is no need for anything outside the lineinfo

// processor to know how this works.

struct LineStateMachine

{

  int file_num;

  uint64_t address;

  int line_num;

  int column_num;

  unsigned int shndx;    // the section address refers to

  bool is_stmt;          // stmt means statement.

  bool basic_block;

  bool end_sequence;

};

static void

ResetLineStateMachine(struct LineStateMachine* lsm, bool default_is_stmt)

{

  lsm->file_num = 1;

  lsm->address = 0;

  lsm->line_num = 1;

  lsm->column_num = 0;

  lsm->shndx = -1U;

  lsm->is_stmt = default_is_stmt;

  lsm->basic_block = false;

  lsm->end_sequence = false;

}

template<int size, bool big_endian>

Sized_dwarf_line_info<size, big_endian>::Sized_dwarf_line_info(Object* object,

                                                               unsigned int read_shndx)

  : data_valid_(false), buffer_(NULL), symtab_buffer_(NULL),

    directories_(), files_(), current_header_index_(-1)

{

  unsigned int debug_shndx;

  for (debug_shndx = 0; debug_shndx < object->shnum(); ++debug_shndx)

    // FIXME: do this more efficiently: section_name() isn't super-fast

    if (object->section_name(debug_shndx) == ".debug_line")

      {

        section_size_type buffer_size;

        this->buffer_ = object->section_contents(debug_shndx, &buffer_size,

                                                 false);

        this->buffer_end_ = this->buffer_ + buffer_size;

        break;

      }

  if (this->buffer_ == NULL)

    return;

  // Find the relocation section for ".debug_line".

  // We expect these for relobjs (.o's) but not dynobjs (.so's).

  bool got_relocs = false;

  for (unsigned int reloc_shndx = 0;

       reloc_shndx < object->shnum();

       ++reloc_shndx)

    {

      unsigned int reloc_sh_type = object->section_type(reloc_shndx);

      if ((reloc_sh_type == elfcpp::SHT_REL

           || reloc_sh_type == elfcpp::SHT_RELA)

          && object->section_info(reloc_shndx) == debug_shndx)

        {

          got_relocs = this->track_relocs_.initialize(object, reloc_shndx,

                                                      reloc_sh_type);

          break;

        }

    }

  // Finally, we need the symtab section to interpret the relocs.

  if (got_relocs)

    {

      unsigned int symtab_shndx;

      for (symtab_shndx = 0; symtab_shndx < object->shnum(); ++symtab_shndx)

        if (object->section_type(symtab_shndx) == elfcpp::SHT_SYMTAB)

          {

            this->symtab_buffer_ = object->section_contents(

                symtab_shndx, &this->symtab_buffer_size_, false);

            break;

          }

      if (this->symtab_buffer_ == NULL)

        return;

    }

  // Now that we have successfully read all the data, parse the debug

  // info.

  this->data_valid_ = true;

  this->read_line_mappings(object, read_shndx);

}

// Read the DWARF header.

template<int size, bool big_endian>

const unsigned char*

Sized_dwarf_line_info<size, big_endian>::read_header_prolog(

    const unsigned char* lineptr)

{

  uint32_t initial_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr);

  lineptr += 4;

  // In DWARF2/3, if the initial length is all 1 bits, then the offset

  // size is 8 and we need to read the next 8 bytes for the real length.

  if (initial_length == 0xffffffff)

    {

      header_.offset_size = 8;

      initial_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr);

      lineptr += 8;

    }

  else

    header_.offset_size = 4;

  header_.total_length = initial_length;

  gold_assert(lineptr + header_.total_length <= buffer_end_);

  header_.version = elfcpp::Swap_unaligned<16, big_endian>::readval(lineptr);

  lineptr += 2;

  if (header_.offset_size == 4)

    header_.prologue_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr);

  else

    header_.prologue_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr);

  lineptr += header_.offset_size;

  header_.min_insn_length = *lineptr;

  lineptr += 1;

  header_.default_is_stmt = *lineptr;

  lineptr += 1;

  header_.line_base = *reinterpret_cast<const signed char*>(lineptr);

  lineptr += 1;

  header_.line_range = *lineptr;

  lineptr += 1;

  header_.opcode_base = *lineptr;

  lineptr += 1;

  header_.std_opcode_lengths.reserve(header_.opcode_base + 1);

  header_.std_opcode_lengths[0] = 0;

  for (int i = 1; i < header_.opcode_base; i++)

    {

      header_.std_opcode_lengths[i] = *lineptr;

      lineptr += 1;

    }

  return lineptr;

}

// The header for a debug_line section is mildly complicated, because

// the line info is very tightly encoded.

template<int size, bool big_endian>

const unsigned char*

Sized_dwarf_line_info<size, big_endian>::read_header_tables(

    const unsigned char* lineptr)

{

  ++this->current_header_index_;

  // Create a new directories_ entry and a new files_ entry for our new

  // header.  We initialize each with a single empty element, because

  // dwarf indexes directory and filenames starting at 1.

  gold_assert(static_cast<int>(this->directories_.size())

              == this->current_header_index_);

  gold_assert(static_cast<int>(this->files_.size())

              == this->current_header_index_);

  this->directories_.push_back(std::vector<std::string>(1));

  this->files_.push_back(std::vector<std::pair<int, std::string> >(1));

  // It is legal for the directory entry table to be empty.

  if (*lineptr)

    {

      int dirindex = 1;

      while (*lineptr)

        {

          const char* dirname = reinterpret_cast<const char*>(lineptr);

          gold_assert(dirindex

                      == static_cast<int>(this->directories_.back().size()));

          this->directories_.back().push_back(dirname);

          lineptr += this->directories_.back().back().size() + 1;

          dirindex++;

        }

    }

  lineptr++;

  // It is also legal for the file entry table to be empty.

  if (*lineptr)

    {

      int fileindex = 1;

      size_t len;

      while (*lineptr)

        {

          const char* filename = reinterpret_cast<const char*>(lineptr);

          lineptr += strlen(filename) + 1;

          uint64_t dirindex = read_unsigned_LEB_128(lineptr, &len);

          lineptr += len;

          if (dirindex >= this->directories_.back().size())

            dirindex = 0;

          int dirindexi = static_cast<int>(dirindex);

          read_unsigned_LEB_128(lineptr, &len);   // mod_time

          lineptr += len;

          read_unsigned_LEB_128(lineptr, &len);   // filelength

          lineptr += len;

          gold_assert(fileindex

                      == static_cast<int>(this->files_.back().size()));

          this->files_.back().push_back(std::make_pair(dirindexi, filename));

          fileindex++;

        }

    }

  lineptr++;

  return lineptr;

}

// Process a single opcode in the .debug.line structure.

// Templating on size and big_endian would yield more efficient (and

// simpler) code, but would bloat the binary.  Speed isn't important

// here.

template<int size, bool big_endian>

bool

Sized_dwarf_line_info<size, big_endian>::process_one_opcode(

    const unsigned char* start, struct LineStateMachine* lsm, size_t* len)

{

  size_t oplen = 0;

  size_t templen;

  unsigned char opcode = *start;

  oplen++;

  start++;

  // If the opcode is great than the opcode_base, it is a special

  // opcode. Most line programs consist mainly of special opcodes.

  if (opcode >= header_.opcode_base)

    {

      opcode -= header_.opcode_base;

      const int advance_address = ((opcode / header_.line_range)

                                   * header_.min_insn_length);

      lsm->address += advance_address;

      const int advance_line = ((opcode % header_.line_range)

                                + header_.line_base);

      lsm->line_num += advance_line;

      lsm->basic_block = true;

      *len = oplen;

      return true;

    }

  // Otherwise, we have the regular opcodes

  switch (opcode)

    {

    case elfcpp::DW_LNS_copy:

      lsm->basic_block = false;

      *len = oplen;

      return true;

    case elfcpp::DW_LNS_advance_pc:

      {

        const uint64_t advance_address

            = read_unsigned_LEB_128(start, &templen);

        oplen += templen;

        lsm->address += header_.min_insn_length * advance_address;

      }

      break;

    case elfcpp::DW_LNS_advance_line:

      {

        const uint64_t advance_line = read_signed_LEB_128(start, &templen);

        oplen += templen;

        lsm->line_num += advance_line;

      }

      break;

    case elfcpp::DW_LNS_set_file:

      {

        const uint64_t fileno = read_unsigned_LEB_128(start, &templen);

        oplen += templen;

        lsm->file_num = fileno;

      }

      break;

    case elfcpp::DW_LNS_set_column:

      {

        const uint64_t colno = read_unsigned_LEB_128(start, &templen);

        oplen += templen;

        lsm->column_num = colno;

      }

      break;

    case elfcpp::DW_LNS_negate_stmt:

      lsm->is_stmt = !lsm->is_stmt;

      break;

    case elfcpp::DW_LNS_set_basic_block:

      lsm->basic_block = true;

      break;

    case elfcpp::DW_LNS_fixed_advance_pc:

      {

        int advance_address;

        advance_address = elfcpp::Swap_unaligned<16, big_endian>::readval(start);

        oplen += 2;

        lsm->address += advance_address;

      }

      break;

    case elfcpp::DW_LNS_const_add_pc:

      {

        const int advance_address = (header_.min_insn_length

                                     * ((255 - header_.opcode_base)

                                        / header_.line_range));

        lsm->address += advance_address;

      }

      break;

    case elfcpp::DW_LNS_extended_op:

      {

        const uint64_t extended_op_len

            = read_unsigned_LEB_128(start, &templen);

        start += templen;

        oplen += templen + extended_op_len;

        const unsigned char extended_op = *start;

        start++;

        switch (extended_op)

          {

          case elfcpp::DW_LNE_end_sequence:

            // This means that the current byte is the one immediately

            // after a set of instructions.  Record the current line

            // for up to one less than the current address.

            lsm->line_num = -1;

            lsm->end_sequence = true;

            *len = oplen;

            return true;

          case elfcpp::DW_LNE_set_address:

            {

              lsm->address = elfcpp::Swap_unaligned<size, big_endian>::readval(start);

              typename Reloc_map::const_iterator it

                  = reloc_map_.find(start - this->buffer_);

              if (it != reloc_map_.end())

                {

                  // value + addend.

                  lsm->address += it->second.second;

                  lsm->shndx = it->second.first;

                }

              else

                {

                  // If we're a normal .o file, with relocs, every

                  // set_address should have an associated relocation.

                  if (this->input_is_relobj())

                    this->data_valid_ = false;

                }

              break;

            }

          case elfcpp::DW_LNE_define_file:

            {

              const char* filename  = reinterpret_cast<const char*>(start);

              templen = strlen(filename) + 1;

              start += templen;

              uint64_t dirindex = read_unsigned_LEB_128(start, &templen);

              oplen += templen;

              if (dirindex >= this->directories_.back().size())

                dirindex = 0;

              int dirindexi = static_cast<int>(dirindex);

              read_unsigned_LEB_128(start, &templen);   // mod_time

              oplen += templen;

              read_unsigned_LEB_128(start, &templen);   // filelength

              oplen += templen;

              this->files_.back().push_back(std::make_pair(dirindexi,

                                                           filename));

            }

            break;

          }

      }

      break;

    default:

      {

        // Ignore unknown opcode  silently

        for (int i = 0; i < header_.std_opcode_lengths[opcode]; i++)

          {

            size_t templen;

            read_unsigned_LEB_128(start, &templen);

            start += templen;

            oplen += templen;

          }

      }

      break;

  }

  *len = oplen;

  return false;

}

// Read the debug information at LINEPTR and store it in the line

// number map.

template<int size, bool big_endian>

unsigned const char*

Sized_dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr,

                                                    unsigned int shndx)

{

  struct LineStateMachine lsm;

  // LENGTHSTART is the place the length field is based on.  It is the

  // point in the header after the initial length field.

  const unsigned char* lengthstart = buffer_;

  // In 64 bit dwarf, the initial length is 12 bytes, because of the

  // 0xffffffff at the start.

  if (header_.offset_size == 8)

    lengthstart += 12;

  else

    lengthstart += 4;

  while (lineptr < lengthstart + header_.total_length)

    {

      ResetLineStateMachine(&lsm, header_.default_is_stmt);

      while (!lsm.end_sequence)

        {

          size_t oplength;

          bool add_line = this->process_one_opcode(lineptr, &lsm, &oplength);

          if (add_line

              && (shndx == -1U || lsm.shndx == -1U || shndx == lsm.shndx))

            {

              Offset_to_lineno_entry entry

                  = { lsm.address, this->current_header_index_,

                      lsm.file_num, lsm.line_num };

              line_number_map_[lsm.shndx].push_back(entry);

            }

          lineptr += oplength;

        }

    }

  return lengthstart + header_.total_length;

}

// Looks in the symtab to see what section a symbol is in.

template<int size, bool big_endian>

unsigned int

Sized_dwarf_line_info<size, big_endian>::symbol_section(

    Object* object,

    unsigned int sym,

    typename elfcpp::Elf_types<size>::Elf_Addr* value,

    bool* is_ordinary)

{

  const int symsize = elfcpp::Elf_sizes<size>::sym_size;

  gold_assert(sym * symsize < this->symtab_buffer_size_);

  elfcpp::Sym<size, big_endian> elfsym(this->symtab_buffer_ + sym * symsize);

  *value = elfsym.get_st_value();

  return object->adjust_sym_shndx(sym, elfsym.get_st_shndx(), is_ordinary);

}

// Read the relocations into a Reloc_map.

template<int size, bool big_endian>

void

Sized_dwarf_line_info<size, big_endian>::read_relocs(Object* object)

{

  if (this->symtab_buffer_ == NULL)

    return;

  typename elfcpp::Elf_types<size>::Elf_Addr value;

  off_t reloc_offset;

  while ((reloc_offset = this->track_relocs_.next_offset()) != -1)

    {

      const unsigned int sym = this->track_relocs_.next_symndx();

      bool is_ordinary;

      const unsigned int shndx = this->symbol_section(object, sym, &value,

                                                      &is_ordinary);

      // There is no reason to record non-ordinary section indexes, or

      // SHN_UNDEF, because they will never match the real section.

      if (is_ordinary && shndx != elfcpp::SHN_UNDEF)

        this->reloc_map_[reloc_offset] = std::make_pair(shndx, value);

      this->track_relocs_.advance(reloc_offset + 1);

    }

}

// Read the line number info.

template<int size, bool big_endian>

void

Sized_dwarf_line_info<size, big_endian>::read_line_mappings(Object* object,

                                                            unsigned int shndx)

{

  gold_assert(this->data_valid_ == true);

  this->read_relocs(object);

  while (this->buffer_ < this->buffer_end_)

    {

      const unsigned char* lineptr = this->buffer_;

      lineptr = this->read_header_prolog(lineptr);

      lineptr = this->read_header_tables(lineptr);

      lineptr = this->read_lines(lineptr, shndx);

      this->buffer_ = lineptr;

    }

  // Sort the lines numbers, so addr2line can use binary search.

  for (typename Lineno_map::iterator it = line_number_map_.begin();

       it != line_number_map_.end();

       ++it)

    // Each vector needs to be sorted by offset.

    std::sort(it->second.begin(), it->second.end());

}

// Some processing depends on whether the input is a .o file or not.

// For instance, .o files have relocs, and have .debug_lines

// information on a per section basis.  .so files, on the other hand,

// lack relocs, and offsets are unique, so we can ignore the section

// information.

template<int size, bool big_endian>

bool

Sized_dwarf_line_info<size, big_endian>::input_is_relobj()

{

  // Only .o files have relocs and the symtab buffer that goes with them.

  return this->symtab_buffer_ != NULL;

}

// Given an Offset_to_lineno_entry vector, and an offset, figure out

// if the offset points into a function according to the vector (see

// comments below for the algorithm).  If it does, return an iterator

// into the vector that points to the line-number that contains that

// offset.  If not, it returns vector::end().

static std::vector<Offset_to_lineno_entry>::const_iterator

offset_to_iterator(const std::vector<Offset_to_lineno_entry>* offsets,

                   off_t offset)

{

  const Offset_to_lineno_entry lookup_key = { offset, 0, 0, 0 };

  // lower_bound() returns the smallest offset which is >= lookup_key.