libbacktrace/dwarf.c

/* dwarf.c -- Get file/line information from DWARF for backtraces.
   Copyright (C) 2012-2018 Free Software Foundation, Inc.
   Written by Ian Lance Taylor, Google.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:

    (1) Redistributions of source code must retain the above copyright
    notice, this list of conditions and the following disclaimer.

    (2) Redistributions in binary form must reproduce the above copyright
    notice, this list of conditions and the following disclaimer in
    the documentation and/or other materials provided with the
    distribution.

    (3) The name of the author may not be used to
    endorse or promote products derived from this software without
    specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT,
INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
POSSIBILITY OF SUCH DAMAGE.  */

#include "config.h"

#include <errno.h>
#include <stdlib.h>
#include <string.h>
#include <sys/types.h>

#include "backtrace.h"
#include "internal.h"

/* DWARF constants.  */

enum dwarf_tag {
  DW_TAG_entry_point = 0x3,
  DW_TAG_compile_unit = 0x11,
  DW_TAG_inlined_subroutine = 0x1d,
  DW_TAG_subprogram = 0x2e,
};

enum dwarf_form {
  DW_FORM_addr = 0x1,
  DW_FORM_block2 = 0x3,
  DW_FORM_block4 = 0x4,
  DW_FORM_data2 = 0x5,
  DW_FORM_data4 = 0x6,
  DW_FORM_data8 = 0x07,
  DW_FORM_string = 0x08,
  DW_FORM_block = 0x09,
  DW_FORM_block1 = 0x0a,
  DW_FORM_data1 = 0x0b,
  DW_FORM_flag = 0x0c,
  DW_FORM_sdata = 0x0d,
  DW_FORM_strp = 0x0e,
  DW_FORM_udata = 0x0f,
  DW_FORM_ref_addr = 0x10,
  DW_FORM_ref1 = 0x11,
  DW_FORM_ref2 = 0x12,
  DW_FORM_ref4 = 0x13,
  DW_FORM_ref8 = 0x14,
  DW_FORM_ref_udata = 0x15,
  DW_FORM_indirect = 0x16,
  DW_FORM_sec_offset = 0x17,
  DW_FORM_exprloc = 0x18,
  DW_FORM_flag_present = 0x19,
  DW_FORM_ref_sig8 = 0x20,
  DW_FORM_GNU_addr_index = 0x1f01,
  DW_FORM_GNU_str_index = 0x1f02,
  DW_FORM_GNU_ref_alt = 0x1f20,
  DW_FORM_GNU_strp_alt = 0x1f21,
};

enum dwarf_attribute {
  DW_AT_name = 0x3,
  DW_AT_stmt_list = 0x10,
  DW_AT_low_pc = 0x11,
  DW_AT_high_pc = 0x12,
  DW_AT_comp_dir = 0x1b,
  DW_AT_abstract_origin = 0x31,
  DW_AT_specification = 0x47,
  DW_AT_ranges = 0x55,
  DW_AT_call_file = 0x58,
  DW_AT_call_line = 0x59,
  DW_AT_linkage_name = 0x6e,
  DW_AT_MIPS_linkage_name = 0x2007,
};

enum dwarf_line_number_op {
  DW_LNS_extended_op = 0x0,
  DW_LNS_copy = 0x1,
  DW_LNS_advance_pc = 0x2,
  DW_LNS_advance_line = 0x3,
  DW_LNS_set_file = 0x4,
  DW_LNS_set_column = 0x5,
  DW_LNS_negate_stmt = 0x6,
  DW_LNS_set_basic_block = 0x7,
  DW_LNS_const_add_pc = 0x8,
  DW_LNS_fixed_advance_pc = 0x9,
  DW_LNS_set_prologue_end = 0xa,
  DW_LNS_set_epilogue_begin = 0xb,
  DW_LNS_set_isa = 0xc,
};

enum dwarf_extedned_line_number_op {
  DW_LNE_end_sequence = 0x1,
  DW_LNE_set_address = 0x2,
  DW_LNE_define_file = 0x3,
  DW_LNE_set_discriminator = 0x4,
};

#if defined(__MSDOS__) || defined(_WIN32) || defined(__OS2__) || defined (__CYGWIN__)
# define IS_DIR_SEPARATOR(c) ((c) == '/' || (c) == '\\')
#define HAS_DRIVE_SPEC(f) ((f)[0] && (f)[1] == ':')
# define IS_ABSOLUTE_PATH(f) (IS_DIR_SEPARATOR(f[0]) || HAS_DRIVE_SPEC(f))
#else
# define IS_DIR_SEPARATOR(c) ((c) == '/')
# define IS_ABSOLUTE_PATH(f) IS_DIR_SEPARATOR(f[0])
#endif

#if !defined(HAVE_DECL_STRNLEN) || !HAVE_DECL_STRNLEN

/* If strnlen is not declared, provide our own version.  */

static size_t
xstrnlen (const char *s, size_t maxlen)
{
  size_t i;

  for (i = 0; i < maxlen; ++i)
    if (s[i] == '\0')
      break;
  return i;
}

#define strnlen xstrnlen

#endif

/* A buffer to read DWARF info.  */

struct dwarf_buf
{
  /* Buffer name for error messages.  */
  const char *name;
  /* Start of the buffer.  */
  const unsigned char *start;
  /* Next byte to read.  */
  const unsigned char *buf;
  /* The number of bytes remaining.  */
  size_t left;
  /* Whether the data is big-endian.  */
  int is_bigendian;
  /* Error callback routine.  */
  backtrace_error_callback error_callback;
  /* Data for error_callback.  */
  void *data;
  /* Non-zero if we've reported an underflow error.  */
  int reported_underflow;
};

/* A single attribute in a DWARF abbreviation.  */

struct attr
{
  /* The attribute name.  */
  enum dwarf_attribute name;
  /* The attribute form.  */
  enum dwarf_form form;
};

/* A single DWARF abbreviation.  */

struct abbrev
{
  /* The abbrev code--the number used to refer to the abbrev.  */
  uint64_t code;
  /* The entry tag.  */
  enum dwarf_tag tag;
  /* Non-zero if this abbrev has child entries.  */
  int has_children;
  /* The number of attributes.  */
  size_t num_attrs;
  /* The attributes.  */
  struct attr *attrs;
};

/* The DWARF abbreviations for a compilation unit.  This structure
   only exists while reading the compilation unit.  Most DWARF readers
   seem to a hash table to map abbrev ID's to abbrev entries.
   However, we primarily care about GCC, and GCC simply issues ID's in
   numerical order starting at 1.  So we simply keep a sorted vector,
   and try to just look up the code.  */

struct abbrevs
{
  /* The number of abbrevs in the vector.  */
  size_t num_abbrevs;
  /* The abbrevs, sorted by the code field.  */
  struct abbrev *abbrevs;
};

/* The different kinds of attribute values.  */

enum attr_val_encoding
{
  /* An address.  */
  ATTR_VAL_ADDRESS,
  /* A unsigned integer.  */
  ATTR_VAL_UINT,
  /* A sigd integer.  */
  ATTR_VAL_SINT,
  /* A string.  */
  ATTR_VAL_STRING,
  /* An offset to other data in the containing unit.  */
  ATTR_VAL_REF_UNIT,
  /* An offset to other data within the .dwarf_info section.  */
  ATTR_VAL_REF_INFO,
  /* An offset to data in some other section.  */
  ATTR_VAL_REF_SECTION,
  /* A type signature.  */
  ATTR_VAL_REF_TYPE,
  /* A block of data (not represented).  */
  ATTR_VAL_BLOCK,
  /* An expression (not represented).  */
  ATTR_VAL_EXPR,
};

/* An attribute value.  */

struct attr_val
{
  /* How the value is stored in the field u.  */
  enum attr_val_encoding encoding;
  union
  {
    /* ATTR_VAL_ADDRESS, ATTR_VAL_UINT, ATTR_VAL_REF*.  */
    uint64_t uint;
    /* ATTR_VAL_SINT.  */
    int64_t sint;
    /* ATTR_VAL_STRING.  */
    const char *string;
    /* ATTR_VAL_BLOCK not stored.  */
  } u;
};

/* The line number program header.  */

struct line_header
{
  /* The version of the line number information.  */
  int version;
  /* The minimum instruction length.  */
  unsigned int min_insn_len;
  /* The maximum number of ops per instruction.  */
  unsigned int max_ops_per_insn;
  /* The line base for special opcodes.  */
  int line_base;
  /* The line range for special opcodes.  */
  unsigned int line_range;
  /* The opcode base--the first special opcode.  */
  unsigned int opcode_base;
  /* Opcode lengths, indexed by opcode - 1.  */
  const unsigned char *opcode_lengths;
  /* The number of directory entries.  */
  size_t dirs_count;
  /* The directory entries.  */
  const char **dirs;
  /* The number of filenames.  */
  size_t filenames_count;
  /* The filenames.  */
  const char **filenames;
};

/* Map a single PC value to a file/line.  We will keep a vector of
   these sorted by PC value.  Each file/line will be correct from the
   PC up to the PC of the next entry if there is one.  We allocate one
   extra entry at the end so that we can use bsearch.  */

struct line
{
  /* PC.  */
  uintptr_t pc;
  /* File name.  Many entries in the array are expected to point to
     the same file name.  */
  const char *filename;
  /* Line number.  */
  int lineno;
  /* Index of the object in the original array read from the DWARF
     section, before it has been sorted.  The index makes it possible
     to use Quicksort and maintain stability.  */
  int idx;
};

/* A growable vector of line number information.  This is used while
   reading the line numbers.  */

struct line_vector
{
  /* Memory.  This is an array of struct line.  */
  struct backtrace_vector vec;
  /* Number of valid mappings.  */
  size_t count;
};

/* A function described in the debug info.  */

struct function
{
  /* The name of the function.  */
  const char *name;
  /* If this is an inlined function, the filename of the call
     site.  */
  const char *caller_filename;
  /* If this is an inlined function, the line number of the call
     site.  */
  int caller_lineno;
  /* Map PC ranges to inlined functions.  */
  struct function_addrs *function_addrs;
  size_t function_addrs_count;
};

/* An address range for a function.  This maps a PC value to a
   specific function.  */

struct function_addrs
{
  /* Range is LOW <= PC < HIGH.  */
  uint64_t low;
  uint64_t high;
  /* Function for this address range.  */
  struct function *function;
};

/* A growable vector of function address ranges.  */

struct function_vector
{
  /* Memory.  This is an array of struct function_addrs.  */
  struct backtrace_vector vec;
  /* Number of address ranges present.  */
  size_t count;
};

/* A DWARF compilation unit.  This only holds the information we need
   to map a PC to a file and line.  */

struct unit
{
  /* The first entry for this compilation unit.  */
  const unsigned char *unit_data;
  /* The length of the data for this compilation unit.  */
  size_t unit_data_len;
  /* The offset of UNIT_DATA from the start of the information for
     this compilation unit.  */
  size_t unit_data_offset;
  /* DWARF version.  */
  int version;
  /* Whether unit is DWARF64.  */
  int is_dwarf64;
  /* Address size.  */
  int addrsize;
  /* Offset into line number information.  */
  off_t lineoff;
  /* Primary source file.  */
  const char *filename;
  /* Compilation command working directory.  */
  const char *comp_dir;
  /* Absolute file name, only set if needed.  */
  const char *abs_filename;
  /* The abbreviations for this unit.  */
  struct abbrevs abbrevs;

  /* The fields above this point are read in during initialization and
     may be accessed freely.  The fields below this point are read in
     as needed, and therefore require care, as different threads may
     try to initialize them simultaneously.  */

  /* PC to line number mapping.  This is NULL if the values have not
     been read.  This is (struct line *) -1 if there was an error
     reading the values.  */
  struct line *lines;
  /* Number of entries in lines.  */
  size_t lines_count;
  /* PC ranges to function.  */
  struct function_addrs *function_addrs;
  size_t function_addrs_count;
};

/* An address range for a compilation unit.  This maps a PC value to a
   specific compilation unit.  Note that we invert the representation
   in DWARF: instead of listing the units and attaching a list of
   ranges, we list the ranges and have each one point to the unit.
   This lets us do a binary search to find the unit.  */

struct unit_addrs
{
  /* Range is LOW <= PC < HIGH.  */
  uint64_t low;
  uint64_t high;
  /* Compilation unit for this address range.  */
  struct unit *u;
};

/* A growable vector of compilation unit address ranges.  */

struct unit_addrs_vector
{
  /* Memory.  This is an array of struct unit_addrs.  */
  struct backtrace_vector vec;
  /* Number of address ranges present.  */
  size_t count;
};

/* The information we need to map a PC to a file and line.  */

struct dwarf_data
{
  /* The data for the next file we know about.  */
  struct dwarf_data *next;
  /* The base address for this file.  */
  uintptr_t base_address;
  /* A sorted list of address ranges.  */
  struct unit_addrs *addrs;
  /* Number of address ranges in list.  */
  size_t addrs_count;
  /* The unparsed .debug_info section.  */
  const unsigned char *dwarf_info;
  size_t dwarf_info_size;
  /* The unparsed .debug_line section.  */
  const unsigned char *dwarf_line;
  size_t dwarf_line_size;
  /* The unparsed .debug_ranges section.  */
  const unsigned char *dwarf_ranges;
  size_t dwarf_ranges_size;
  /* The unparsed .debug_str section.  */
  const unsigned char *dwarf_str;
  size_t dwarf_str_size;
  /* Whether the data is big-endian or not.  */
  int is_bigendian;
  /* A vector used for function addresses.  We keep this here so that
     we can grow the vector as we read more functions.  */
  struct function_vector fvec;
};

/* Report an error for a DWARF buffer.  */

static void
dwarf_buf_error (struct dwarf_buf *buf, const char *msg)
{
  char b[200];

  snprintf (b, sizeof b, "%s in %s at %d",
	    msg, buf->name, (int) (buf->buf - buf->start));
  buf->error_callback (buf->data, b, 0);
}

/* Require at least COUNT bytes in BUF.  Return 1 if all is well, 0 on
   error.  */

static int
require (struct dwarf_buf *buf, size_t count)
{
  if (buf->left >= count)
    return 1;

  if (!buf->reported_underflow)
    {
      dwarf_buf_error (buf, "DWARF underflow");
      buf->reported_underflow = 1;
    }

  return 0;
}

/* Advance COUNT bytes in BUF.  Return 1 if all is well, 0 on
   error.  */

static int
advance (struct dwarf_buf *buf, size_t count)
{
  if (!require (buf, count))
    return 0;
  buf->buf += count;
  buf->left -= count;
  return 1;
}

/* Read one byte from BUF and advance 1 byte.  */

static unsigned char
read_byte (struct dwarf_buf *buf)
{
  const unsigned char *p = buf->buf;

  if (!advance (buf, 1))
    return 0;
  return p[0];
}

/* Read a signed char from BUF and advance 1 byte.  */

static signed char
read_sbyte (struct dwarf_buf *buf)
{
  const unsigned char *p = buf->buf;

  if (!advance (buf, 1))
    return 0;
  return (*p ^ 0x80) - 0x80;
}

/* Read a uint16 from BUF and advance 2 bytes.  */

static uint16_t
read_uint16 (struct dwarf_buf *buf)
{
  const unsigned char *p = buf->buf;

  if (!advance (buf, 2))
    return 0;
  if (buf->is_bigendian)
    return ((uint16_t) p[0] << 8) | (uint16_t) p[1];
  else
    return ((uint16_t) p[1] << 8) | (uint16_t) p[0];
}

/* Read a uint32 from BUF and advance 4 bytes.  */

static uint32_t
read_uint32 (struct dwarf_buf *buf)
{
  const unsigned char *p = buf->buf;

  if (!advance (buf, 4))
    return 0;
  if (buf->is_bigendian)
    return (((uint32_t) p[0] << 24) | ((uint32_t) p[1] << 16)
	    | ((uint32_t) p[2] << 8) | (uint32_t) p[3]);
  else
    return (((uint32_t) p[3] << 24) | ((uint32_t) p[2] << 16)
	    | ((uint32_t) p[1] << 8) | (uint32_t) p[0]);
}

/* Read a uint64 from BUF and advance 8 bytes.  */

static uint64_t
read_uint64 (struct dwarf_buf *buf)
{
  const unsigned char *p = buf->buf;

  if (!advance (buf, 8))
    return 0;
  if (buf->is_bigendian)
    return (((uint64_t) p[0] << 56) | ((uint64_t) p[1] << 48)
	    | ((uint64_t) p[2] << 40) | ((uint64_t) p[3] << 32)
	    | ((uint64_t) p[4] << 24) | ((uint64_t) p[5] << 16)
	    | ((uint64_t) p[6] << 8) | (uint64_t) p[7]);
  else
    return (((uint64_t) p[7] << 56) | ((uint64_t) p[6] << 48)
	    | ((uint64_t) p[5] << 40) | ((uint64_t) p[4] << 32)
	    | ((uint64_t) p[3] << 24) | ((uint64_t) p[2] << 16)
	    | ((uint64_t) p[1] << 8) | (uint64_t) p[0]);
}

/* Read an offset from BUF and advance the appropriate number of
   bytes.  */

static uint64_t
read_offset (struct dwarf_buf *buf, int is_dwarf64)
{
  if (is_dwarf64)
    return read_uint64 (buf);
  else
    return read_uint32 (buf);
}

/* Read an address from BUF and advance the appropriate number of
   bytes.  */

static uint64_t
read_address (struct dwarf_buf *buf, int addrsize)
{
  switch (addrsize)
    {
    case 1:
      return read_byte (buf);
    case 2:
      return read_uint16 (buf);
    case 4:
      return read_uint32 (buf);
    case 8:
      return read_uint64 (buf);
    default:
      dwarf_buf_error (buf, "unrecognized address size");
      return 0;
    }
}

/* Return whether a value is the highest possible address, given the
   address size.  */

static int
is_highest_address (uint64_t address, int addrsize)
{
  switch (addrsize)
    {
    case 1:
      return address == (unsigned char) -1;
    case 2:
      return address == (uint16_t) -1;
    case 4:
      return address == (uint32_t) -1;
    case 8:
      return address == (uint64_t) -1;
    default:
      return 0;
    }
}

/* Read an unsigned LEB128 number.  */

static uint64_t
read_uleb128 (struct dwarf_buf *buf)
{
  uint64_t ret;
  unsigned int shift;
  int overflow;
  unsigned char b;

  ret = 0;
  shift = 0;
  overflow = 0;
  do
    {
      const unsigned char *p;

      p = buf->buf;
      if (!advance (buf, 1))
	return 0;
      b = *p;
      if (shift < 64)
	ret |= ((uint64_t) (b & 0x7f)) << shift;
      else if (!overflow)
	{
	  dwarf_buf_error (buf, "LEB128 overflows uint64_t");
	  overflow = 1;
	}
      shift += 7;
    }
  while ((b & 0x80) != 0);

  return ret;
}

/* Read a signed LEB128 number.  */

static int64_t
read_sleb128 (struct dwarf_buf *buf)
{
  uint64_t val;
  unsigned int shift;
  int overflow;
  unsigned char b;

  val = 0;
  shift = 0;
  overflow = 0;
  do
    {
      const unsigned char *p;

      p = buf->buf;
      if (!advance (buf, 1))
	return 0;
      b = *p;
      if (shift < 64)
	val |= ((uint64_t) (b & 0x7f)) << shift;
      else if (!overflow)
	{
	  dwarf_buf_error (buf, "signed LEB128 overflows uint64_t");
	  overflow = 1;
	}
      shift += 7;
    }
  while ((b & 0x80) != 0);

  if ((b & 0x40) != 0 && shift < 64)
    val |= ((uint64_t) -1) << shift;

  return (int64_t) val;
}

/* Return the length of an LEB128 number.  */

static size_t
leb128_len (const unsigned char *p)
{
  size_t ret;

  ret = 1;
  while ((*p & 0x80) != 0)
    {
      ++p;
      ++ret;
    }
  return ret;
}

/* Free an abbreviations structure.  */

static void
free_abbrevs (struct backtrace_state *state, struct abbrevs *abbrevs,
	      backtrace_error_callback error_callback, void *data)
{
  size_t i;

  for (i = 0; i < abbrevs->num_abbrevs; ++i)
    backtrace_free (state, abbrevs->abbrevs[i].attrs,
		    abbrevs->abbrevs[i].num_attrs * sizeof (struct attr),
		    error_callback, data);
  backtrace_free (state, abbrevs->abbrevs,
		  abbrevs->num_abbrevs * sizeof (struct abbrev),
		  error_callback, data);
  abbrevs->num_abbrevs = 0;
  abbrevs->abbrevs = NULL;
}

/* Read an attribute value.  Returns 1 on success, 0 on failure.  If
   the value can be represented as a uint64_t, sets *VAL and sets
   *IS_VALID to 1.  We don't try to store the value of other attribute
   forms, because we don't care about them.  */

static int
read_attribute (enum dwarf_form form, struct dwarf_buf *buf,
		int is_dwarf64, int version, int addrsize,
		const unsigned char *dwarf_str, size_t dwarf_str_size,
		struct attr_val *val)
{
  /* Avoid warnings about val.u.FIELD may be used uninitialized if
     this function is inlined.  The warnings aren't valid but can
     occur because the different fields are set and used
     conditionally.  */
  memset (val, 0, sizeof *val);

  switch (form)
    {
    case DW_FORM_addr:
      val->encoding = ATTR_VAL_ADDRESS;
      val->u.uint = read_address (buf, addrsize);
      return 1;
    case DW_FORM_block2:
      val->encoding = ATTR_VAL_BLOCK;
      return advance (buf, read_uint16 (buf));
    case DW_FORM_block4:
      val->encoding = ATTR_VAL_BLOCK;
      return advance (buf, read_uint32 (buf));
    case DW_FORM_data2:
      val->encoding = ATTR_VAL_UINT;
      val->u.uint = read_uint16 (buf);
      return 1;
    case DW_FORM_data4:
      val->encoding = ATTR_VAL_UINT;
      val->u.uint = read_uint32 (buf);
      return 1;
    case DW_FORM_data8:
      val->encoding = ATTR_VAL_UINT;
      val->u.uint = read_uint64 (buf);
      return 1;
    case DW_FORM_string:
      val->encoding = ATTR_VAL_STRING;
      val->u.string = (const char *) buf->buf;
      return advance (buf, strnlen ((const char *) buf->buf, buf->left) + 1);
    case DW_FORM_block:
      val->encoding = ATTR_VAL_BLOCK;
      return advance (buf, read_uleb128 (buf));
    case DW_FORM_block1:
      val->encoding = ATTR_VAL_BLOCK;
      return advance (buf, read_byte (buf));
    case DW_FORM_data1:
      val->encoding = ATTR_VAL_UINT;
      val->u.uint = read_byte (buf);
      return 1;
    case DW_FORM_flag:
      val->encoding = ATTR_VAL_UINT;
      val->u.uint = read_byte (buf);
      return 1;
    case DW_FORM_sdata:
      val->encoding = ATTR_VAL_SINT;
      val->u.sint = read_sleb128 (buf);
      return 1;
    case DW_FORM_strp:
      {
	uint64_t offset;

	offset = read_offset (buf, is_dwarf64);
	if (offset >= dwarf_str_size)
	  {
	    dwarf_buf_error (buf, "DW_FORM_strp out of range");
	    return 0;
	  }
	val->encoding = ATTR_VAL_STRING;
	val->u.string = (const char *) dwarf_str + offset;
	return 1;
      }
    case DW_FORM_udata:
      val->encoding = ATTR_VAL_UINT;
      val->u.uint = read_uleb128 (buf);
      return 1;
    case DW_FORM_ref_addr:
      val->encoding = ATTR_VAL_REF_INFO;
      if (version == 2)
	val->u.uint = read_address (buf, addrsize);
      else
	val->u.uint = read_offset (buf, is_dwarf64);
      return 1;
    case DW_FORM_ref1:
      val->encoding = ATTR_VAL_REF_UNIT;
      val->u.uint = read_byte (buf);
      return 1;
    case DW_FORM_ref2:
      val->encoding = ATTR_VAL_REF_UNIT;
      val->u.uint = read_uint16 (buf);
      return 1;
    case DW_FORM_ref4:
      val->encoding = ATTR_VAL_REF_UNIT;
      val->u.uint = read_uint32 (buf);
      return 1;
    case DW_FORM_ref8:
      val->encoding = ATTR_VAL_REF_UNIT;
      val->u.uint = read_uint64 (buf);
      return 1;
    case DW_FORM_ref_udata:
      val->encoding = ATTR_VAL_REF_UNIT;
      val->u.uint = read_uleb128 (buf);
      return 1;
    case DW_FORM_indirect:
      {
	uint64_t form;

	form = read_uleb128 (buf);
	return read_attribute ((enum dwarf_form) form, buf, is_dwarf64,
			       version, addrsize, dwarf_str, dwarf_str_size,
			       val);
      }
    case DW_FORM_sec_offset:
      val->encoding = ATTR_VAL_REF_SECTION;
      val->u.uint = read_offset (buf, is_dwarf64);
      return 1;
    case DW_FORM_exprloc:
      val->encoding = ATTR_VAL_EXPR;
      return advance (buf, read_uleb128 (buf));
    case DW_FORM_flag_present:
      val->encoding = ATTR_VAL_UINT;
      val->u.uint = 1;
      return 1;
    case DW_FORM_ref_sig8:
      val->encoding = ATTR_VAL_REF_TYPE;
      val->u.uint = read_uint64 (buf);
      return 1;
    case DW_FORM_GNU_addr_index:
      val->encoding = ATTR_VAL_REF_SECTION;
      val->u.uint = read_uleb128 (buf);
      return 1;
    case DW_FORM_GNU_str_index:
      val->encoding = ATTR_VAL_REF_SECTION;
      val->u.uint = read_uleb128 (buf);
      return 1;
    case DW_FORM_GNU_ref_alt:
      val->encoding = ATTR_VAL_REF_SECTION;
      val->u.uint = read_offset (buf, is_dwarf64);
      return 1;
    case DW_FORM_GNU_strp_alt:
      val->encoding = ATTR_VAL_REF_SECTION;
      val->u.uint = read_offset (buf, is_dwarf64);
      return 1;
    default:
      dwarf_buf_error (buf, "unrecognized DWARF form");
      return 0;
    }
}

/* Compare function_addrs for qsort.  When ranges are nested, make the
   smallest one sort last.  */

static int
function_addrs_compare (const void *v1, const void *v2)
{
  const struct function_addrs *a1 = (const struct function_addrs *) v1;
  const struct function_addrs *a2 = (const struct function_addrs *) v2;

  if (a1->low < a2->low)
    return -1;
  if (a1->low > a2->low)
    return 1;
  if (a1->high < a2->high)
    return 1;
  if (a1->high > a2->high)
    return -1;
  return strcmp (a1->function->name, a2->function->name);
}

/* Compare a PC against a function_addrs for bsearch.  Note that if
   there are multiple ranges containing PC, which one will be returned
   is unpredictable.  We compensate for that in dwarf_fileline.  */

static int
function_addrs_search (const void *vkey, const void *ventry)
{
  const uintptr_t *key = (const uintptr_t *) vkey;
  const struct function_addrs *entry = (const struct function_addrs *) ventry;
  uintptr_t pc;

  pc = *key;
  if (pc < entry->low)
    return -1;
  else if (pc >= entry->high)
    return 1;
  else
    return 0;
}

/* Add a new compilation unit address range to a vector.  Returns 1 on
   success, 0 on failure.  */

static int
add_unit_addr (struct backtrace_state *state, uintptr_t base_address,
	       struct unit_addrs addrs,
	       backtrace_error_callback error_callback, void *data,
	       struct unit_addrs_vector *vec)
{
  struct unit_addrs *p;

  /* Add in the base address of the module here, so that we can look
     up the PC directly.  */
  addrs.low += base_address;
  addrs.high += base_address;

  /* Try to merge with the last entry.  */
  if (vec->count > 0)
    {
      p = (struct unit_addrs *) vec->vec.base + (vec->count - 1);
      if ((addrs.low == p->high || addrs.low == p->high + 1)
	  && addrs.u == p->u)
	{
	  if (addrs.high > p->high)
	    p->high = addrs.high;
	  return 1;
	}
    }

  p = ((struct unit_addrs *)
       backtrace_vector_grow (state, sizeof (struct unit_addrs),
			      error_callback, data, &vec->vec));
  if (p == NULL)
    return 0;

  *p = addrs;
  ++vec->count;
  return 1;
}

/* Free a unit address vector.  */

static void
free_unit_addrs_vector (struct backtrace_state *state,
			struct unit_addrs_vector *vec,
			backtrace_error_callback error_callback, void *data)
{
  struct unit_addrs *addrs;
  size_t i;

  addrs = (struct unit_addrs *) vec->vec.base;
  for (i = 0; i < vec->count; ++i)
    free_abbrevs (state, &addrs[i].u->abbrevs, error_callback, data);
}

/* Compare unit_addrs for qsort.  When ranges are nested, make the
   smallest one sort last.  */

static int
unit_addrs_compare (const void *v1, const void *v2)
{
  const struct unit_addrs *a1 = (const struct unit_addrs *) v1;
  const struct unit_addrs *a2 = (const struct unit_addrs *) v2;

  if (a1->low < a2->low)
    return -1;
  if (a1->low > a2->low)
    return 1;
  if (a1->high < a2->high)
    return 1;
  if (a1->high > a2->high)
    return -1;
  if (a1->u->lineoff < a2->u->lineoff)
    return -1;
  if (a1->u->lineoff > a2->u->lineoff)
    return 1;
  return 0;
}

/* Compare a PC against a unit_addrs for bsearch.  Note that if there
   are multiple ranges containing PC, which one will be returned is
   unpredictable.  We compensate for that in dwarf_fileline.  */

static int
unit_addrs_search (const void *vkey, const void *ventry)
{
  const uintptr_t *key = (const uintptr_t *) vkey;
  const struct unit_addrs *entry = (const struct unit_addrs *) ventry;
  uintptr_t pc;

  pc = *key;
  if (pc < entry->low)
    return -1;
  else if (pc >= entry->high)
    return 1;
  else
    return 0;
}

/* Sort the line vector by PC.  We want a stable sort here to maintain
   the order of lines for the same PC values.  Since the sequence is
   being sorted in place, their addresses cannot be relied on to
   maintain stability.  That is the purpose of the index member.  */

static int
line_compare (const void *v1, const void *v2)
{
  const struct line *ln1 = (const struct line *) v1;
  const struct line *ln2 = (const struct line *) v2;

  if (ln1->pc < ln2->pc)
    return -1;
  else if (ln1->pc > ln2->pc)
    return 1;
  else if (ln1->idx < ln2->idx)
    return -1;
  else if (ln1->idx > ln2->idx)
    return 1;
  else
    return 0;
}

/* Find a PC in a line vector.  We always allocate an extra entry at
   the end of the lines vector, so that this routine can safely look
   at the next entry.  Note that when there are multiple mappings for
   the same PC value, this will return the last one.  */

static int
line_search (const void *vkey, const void *ventry)
{
  const uintptr_t *key = (const uintptr_t *) vkey;
  const struct line *entry = (const struct line *) ventry;
  uintptr_t pc;

  pc = *key;
  if (pc < entry->pc)
    return -1;
  else if (pc >= (entry + 1)->pc)
    return 1;
  else
    return 0;
}

/* Sort the abbrevs by the abbrev code.  This function is passed to
   both qsort and bsearch.  */

static int
abbrev_compare (const void *v1, const void *v2)
{
  const struct abbrev *a1 = (const struct abbrev *) v1;
  const struct abbrev *a2 = (const struct abbrev *) v2;

  if (a1->code < a2->code)
    return -1;
  else if (a1->code > a2->code)
    return 1;
  else
    {
      /* This really shouldn't happen.  It means there are two
	 different abbrevs with the same code, and that means we don't
	 know which one lookup_abbrev should return.  */
      return 0;
    }
}

/* Read the abbreviation table for a compilation unit.  Returns 1 on
   success, 0 on failure.  */

static int
read_abbrevs (struct backtrace_state *state, uint64_t abbrev_offset,
	      const unsigned char *dwarf_abbrev, size_t dwarf_abbrev_size,
	      int is_bigendian, backtrace_error_callback error_callback,
	      void *data, struct abbrevs *abbrevs)
{
  struct dwarf_buf abbrev_buf;
  struct dwarf_buf count_buf;
  size_t num_abbrevs;

  abbrevs->num_abbrevs = 0;
  abbrevs->abbrevs = NULL;

  if (abbrev_offset >= dwarf_abbrev_size)
    {
      error_callback (data, "abbrev offset out of range", 0);
      return 0;
    }

  abbrev_buf.name = ".debug_abbrev";
  abbrev_buf.start = dwarf_abbrev;
  abbrev_buf.buf = dwarf_abbrev + abbrev_offset;
  abbrev_buf.left = dwarf_abbrev_size - abbrev_offset;
  abbrev_buf.is_bigendian = is_bigendian;
  abbrev_buf.error_callback = error_callback;
  abbrev_buf.data = data;
  abbrev_buf.reported_underflow = 0;

  /* Count the number of abbrevs in this list.  */

  count_buf = abbrev_buf;
  num_abbrevs = 0;
  while (read_uleb128 (&count_buf) != 0)
    {
      if (count_buf.reported_underflow)
	return 0;
      ++num_abbrevs;
      // Skip tag.
      read_uleb128 (&count_buf);
      // Skip has_children.
      read_byte (&count_buf);
      // Skip attributes.
      while (read_uleb128 (&count_buf) != 0)
	read_uleb128 (&count_buf);
      // Skip form of last attribute.
      read_uleb128 (&count_buf);
    }

  if (count_buf.reported_underflow)
    return 0;

  if (num_abbrevs == 0)
    return 1;

  abbrevs->num_abbrevs = num_abbrevs;
  abbrevs->abbrevs = ((struct abbrev *)
		      backtrace_alloc (state,
				       num_abbrevs * sizeof (struct abbrev),
				       error_callback, data));
  if (abbrevs->abbrevs == NULL)
    return 0;
  memset (abbrevs->abbrevs, 0, num_abbrevs * sizeof (struct abbrev));

  num_abbrevs = 0;
  while (1)
    {
      uint64_t code;
      struct abbrev a;
      size_t num_attrs;
      struct attr *attrs;

      if (abbrev_buf.reported_underflow)
	goto fail;

      code = read_uleb128 (&abbrev_buf);
      if (code == 0)
	break;

      a.code = code;
      a.tag = (enum dwarf_tag) read_uleb128 (&abbrev_buf);
      a.has_children = read_byte (&abbrev_buf);

      count_buf = abbrev_buf;
      num_attrs = 0;
      while (read_uleb128 (&count_buf) != 0)
	{
	  ++num_attrs;
	  read_uleb128 (&count_buf);
	}