xref: /aosp_15_r20/external/angle/build/android/asan_symbolize.py (revision 8975f5c5ed3d1c378011245431ada316dfb6f244)
1#!/usr/bin/env python3
2#
3# Copyright 2013 The Chromium Authors
4# Use of this source code is governed by a BSD-style license that can be
5# found in the LICENSE file.
6
7
8import argparse
9import collections
10import os
11import re
12import sys
13
14from pylib import constants
15from pylib.constants import host_paths
16
17# pylint: disable=wrong-import-order
18# Uses symbol.py from third_party/android_platform, not python's.
19with host_paths.SysPath(
20    host_paths.ANDROID_PLATFORM_DEVELOPMENT_SCRIPTS_PATH,
21    position=0):
22  import symbol
23
24
25_RE_ASAN = re.compile(
26    r"""
27    (?P<prefix>.*?)
28    (?P<pos>\#\S*?)          # position of the call in stack.
29                             # escape the char "#" due to the VERBOSE flag.
30    \s+(\S*?)\s+
31    \(                       # match the char "(".
32        (?P<lib>.*?)         # library path.
33        \+0[xX](?P<addr>.*?) # address of the symbol in hex.
34                             # the prefix "0x" is skipped.
35    \)                       # match the char ")".
36    """, re.VERBOSE)
37
38# This named tuple models a parsed Asan log line.
39AsanParsedLine = collections.namedtuple('AsanParsedLine',
40                                        'prefix,library,pos,rel_address')
41
42# This named tuple models an Asan log line. 'raw' is the raw content
43# while 'parsed' is None or an AsanParsedLine instance.
44AsanLogLine = collections.namedtuple('AsanLogLine', 'raw,parsed')
45
46def _ParseAsanLogLine(line):
47  """Parse line into corresponding AsanParsedLine value, if any, or None."""
48  m = re.match(_RE_ASAN, line)
49  if not m:
50    return None
51  return AsanParsedLine(prefix=m.group('prefix'),
52                        library=m.group('lib'),
53                        pos=m.group('pos'),
54                        rel_address=int(m.group('addr'), 16))
55
56
57def _FindASanLibraries():
58  asan_lib_dir = os.path.join(host_paths.DIR_SOURCE_ROOT,
59                              'third_party', 'llvm-build',
60                              'Release+Asserts', 'lib')
61  asan_libs = []
62  for src_dir, _, files in os.walk(asan_lib_dir):
63    asan_libs += [os.path.relpath(os.path.join(src_dir, f))
64                  for f in files
65                  if f.endswith('.so')]
66  return asan_libs
67
68
69def _TranslateLibPath(library, asan_libs):
70  for asan_lib in asan_libs:
71    if os.path.basename(library) == os.path.basename(asan_lib):
72      return '/' + asan_lib
73  # pylint: disable=no-member
74  return symbol.TranslateLibPath(library)
75
76
77def _PrintSymbolized(asan_input, arch):
78  """Print symbolized logcat output for Asan symbols.
79
80  Args:
81    asan_input: list of input lines.
82    arch: Target CPU architecture.
83  """
84  asan_libs = _FindASanLibraries()
85
86  # Maps library -> [ AsanParsedLine... ]
87  libraries = collections.defaultdict(list)
88
89  asan_log_lines = []
90  for line in asan_input:
91    line = line.rstrip()
92    parsed = _ParseAsanLogLine(line)
93    if parsed:
94      libraries[parsed.library].append(parsed)
95    asan_log_lines.append(AsanLogLine(raw=line, parsed=parsed))
96
97  # Maps library -> { address -> [(symbol, location, obj_sym_with_offset)...] }
98  all_symbols = collections.defaultdict(dict)
99
100  for library, items in libraries.items():
101    libname = _TranslateLibPath(library, asan_libs)
102    lib_relative_addrs = set(i.rel_address for i in items)
103    # pylint: disable=no-member
104    symbols_by_library = symbol.SymbolInformationForSet(libname,
105                                                        lib_relative_addrs,
106                                                        True,
107                                                        cpu_arch=arch)
108    if symbols_by_library:
109      all_symbols[library] = symbols_by_library
110
111  for log_line in asan_log_lines:
112    m = log_line.parsed
113    if (m and m.library in all_symbols and
114        m.rel_address in all_symbols[m.library]):
115      # NOTE: all_symbols[lib][address] is a never-emtpy list of tuples.
116      # NOTE: The documentation for SymbolInformationForSet() indicates
117      # that usually one wants to display the last list item, not the first.
118      # The code below takes the first, is this the best choice here?
119      s = all_symbols[m.library][m.rel_address][0]
120      symbol_name = s[0]
121      symbol_location = s[1]
122      print('%s%s %s %s @ \'%s\'' %
123            (m.prefix, m.pos, hex(m.rel_address), symbol_name, symbol_location))
124    else:
125      print(log_line.raw)
126
127
128def main():
129  parser = argparse.ArgumentParser()
130  parser.add_argument('-l',
131                      '--logcat',
132                      help='File containing adb logcat output with ASan '
133                      'stacks. Use stdin if not specified.')
134  parser.add_argument('--output-directory',
135                      help='Path to the root build directory.')
136  parser.add_argument('--arch', default='arm', help='CPU architecture name')
137  args = parser.parse_args()
138
139  if args.output_directory:
140    constants.SetOutputDirectory(args.output_directory)
141  # Do an up-front test that the output directory is known.
142  constants.CheckOutputDirectory()
143
144  if args.logcat:
145    asan_input = open(args.logcat, 'r')
146  else:
147    asan_input = sys.stdin
148
149  _PrintSymbolized(asan_input.readlines(), args.arch)
150
151
152if __name__ == "__main__":
153  sys.exit(main())
154