1 #!/usr/bin/python 2 # 3 # Copyright 2006 Google Inc. All Rights Reserved. 4 5 """Module for looking up symbolic debugging information. 6 7 The information can include symbol names, offsets, and source locations. 8 """ 9 10 import os 11 import re 12 import subprocess 13 14 ANDROID_BUILD_TOP = os.environ["ANDROID_BUILD_TOP"] 15 if not ANDROID_BUILD_TOP: 16 ANDROID_BUILD_TOP = "." 17 18 def FindSymbolsDir(): 19 saveddir = os.getcwd() 20 os.chdir(ANDROID_BUILD_TOP) 21 try: 22 cmd = ("CALLED_FROM_SETUP=true BUILD_SYSTEM=build/core " 23 "SRC_TARGET_DIR=build/target make -f build/core/config.mk " 24 "dumpvar-abs-TARGET_OUT_UNSTRIPPED") 25 stream = subprocess.Popen(cmd, stdout=subprocess.PIPE, shell=True).stdout 26 return os.path.join(ANDROID_BUILD_TOP, stream.read().strip()) 27 finally: 28 os.chdir(saveddir) 29 30 SYMBOLS_DIR = FindSymbolsDir() 31 32 def Uname(): 33 """'uname' for constructing prebuilt/<...> and out/host/<...> paths.""" 34 uname = os.uname()[0] 35 if uname == "Darwin": 36 proc = os.uname()[-1] 37 if proc == "i386" or proc == "x86_64": 38 return "darwin-x86" 39 return "darwin-ppc" 40 if uname == "Linux": 41 return "linux-x86" 42 return uname 43 44 def ToolPath(tool, toolchain_info=None): 45 """Return a full qualified path to the specified tool""" 46 if not toolchain_info: 47 toolchain_info = TOOLCHAIN_INFO 48 (label, target) = toolchain_info 49 return os.path.join(ANDROID_BUILD_TOP, "prebuilts", "gcc", Uname(), "arm", label, "bin", 50 target + "-" + tool) 51 52 def FindToolchain(): 53 """Look for the latest available toolchain 54 55 Args: 56 None 57 58 Returns: 59 A pair of strings containing toolchain label and target prefix. 60 """ 61 62 ## Known toolchains, newer ones in the front. 63 known_toolchains = [ 64 ("arm-eabi-4.6", "arm-eabi"), 65 ("arm-linux-androideabi-4.4.x", "arm-linux-androideabi"), 66 ("arm-eabi-4.4.3", "arm-eabi"), 67 ("arm-eabi-4.4.0", "arm-eabi"), 68 ("arm-eabi-4.3.1", "arm-eabi"), 69 ("arm-eabi-4.2.1", "arm-eabi") 70 ] 71 72 # Look for addr2line to check for valid toolchain path. 73 for (label, target) in known_toolchains: 74 toolchain_info = (label, target); 75 if os.path.exists(ToolPath("addr2line", toolchain_info)): 76 return toolchain_info 77 78 raise Exception("Could not find tool chain") 79 80 TOOLCHAIN_INFO = FindToolchain() 81 82 def SymbolInformation(lib, addr): 83 """Look up symbol information about an address. 84 85 Args: 86 lib: library (or executable) pathname containing symbols 87 addr: string hexidecimal address 88 89 Returns: 90 For a given library and address, return tuple of: (source_symbol, 91 source_location, object_symbol_with_offset) the values may be None 92 if the information was unavailable. 93 94 source_symbol may not be a prefix of object_symbol_with_offset if 95 the source function was inlined in the object code of another 96 function. 97 98 usually you want to display the object_symbol_with_offset and 99 source_location, the source_symbol is only useful to show if the 100 address was from an inlined function. 101 """ 102 info = SymbolInformationForSet(lib, set([addr])) 103 return (info and info.get(addr)) or (None, None, None) 104 105 106 def SymbolInformationForSet(lib, unique_addrs): 107 """Look up symbol information for a set of addresses from the given library. 108 109 Args: 110 lib: library (or executable) pathname containing symbols 111 unique_addrs: set of hexidecimal addresses 112 113 Returns: 114 For a given library and set of addresses, returns a dictionary of the form 115 {addr: (source_symbol, source_location, object_symbol_with_offset)}. The 116 values may be None if the information was unavailable. 117 118 For a given address, source_symbol may not be a prefix of 119 object_symbol_with_offset if the source function was inlined in the 120 object code of another function. 121 122 Usually you want to display the object_symbol_with_offset and 123 source_location; the source_symbol is only useful to show if the 124 address was from an inlined function. 125 """ 126 if not lib: 127 return None 128 129 addr_to_line = CallAddr2LineForSet(lib, unique_addrs) 130 if not addr_to_line: 131 return None 132 133 addr_to_objdump = CallObjdumpForSet(lib, unique_addrs) 134 if not addr_to_objdump: 135 return None 136 137 result = {} 138 for addr in unique_addrs: 139 (source_symbol, source_location) = addr_to_line.get(addr, (None, None)) 140 if addr in addr_to_objdump: 141 (object_symbol, object_offset) = addr_to_objdump.get(addr) 142 object_symbol_with_offset = FormatSymbolWithOffset(object_symbol, 143 object_offset) 144 else: 145 object_symbol_with_offset = None 146 result[addr] = (source_symbol, source_location, object_symbol_with_offset) 147 148 return result 149 150 151 def CallAddr2LineForSet(lib, unique_addrs): 152 """Look up line and symbol information for a set of addresses. 153 154 Args: 155 lib: library (or executable) pathname containing symbols 156 unique_addrs: set of string hexidecimal addresses look up. 157 158 Returns: 159 A dictionary of the form {addr: (symbol, file:line)}. The values may 160 be (None, None) if the address could not be looked up. 161 """ 162 if not lib: 163 return None 164 165 166 symbols = SYMBOLS_DIR + lib 167 if not os.path.exists(symbols): 168 return None 169 170 (label, target) = TOOLCHAIN_INFO 171 cmd = [ToolPath("addr2line"), "--functions", "--demangle", "--exe=" + symbols] 172 child = subprocess.Popen(cmd, stdin=subprocess.PIPE, stdout=subprocess.PIPE) 173 174 result = {} 175 addrs = sorted(unique_addrs) 176 for addr in addrs: 177 child.stdin.write("0x%s\n" % addr) 178 child.stdin.flush() 179 symbol = child.stdout.readline().strip() 180 if symbol == "??": 181 symbol = None 182 location = child.stdout.readline().strip() 183 if location == "??:0": 184 location = None 185 result[addr] = (symbol, location) 186 child.stdin.close() 187 child.stdout.close() 188 return result 189 190 191 def CallObjdumpForSet(lib, unique_addrs): 192 """Use objdump to find out the names of the containing functions. 193 194 Args: 195 lib: library (or executable) pathname containing symbols 196 unique_addrs: set of string hexidecimal addresses to find the functions for. 197 198 Returns: 199 A dictionary of the form {addr: (string symbol, offset)}. 200 """ 201 if not lib: 202 return None 203 204 symbols = SYMBOLS_DIR + lib 205 if not os.path.exists(symbols): 206 return None 207 208 symbols = SYMBOLS_DIR + lib 209 if not os.path.exists(symbols): 210 return None 211 212 addrs = sorted(unique_addrs) 213 start_addr_hex = addrs[0] 214 stop_addr_dec = str(int(addrs[-1], 16) + 8) 215 cmd = [ToolPath("objdump"), 216 "--section=.text", 217 "--demangle", 218 "--disassemble", 219 "--start-address=0x" + start_addr_hex, 220 "--stop-address=" + stop_addr_dec, 221 symbols] 222 223 # Function lines look like: 224 # 000177b0 <android::IBinder::~IBinder()+0x2c>: 225 # We pull out the address and function first. Then we check for an optional 226 # offset. This is tricky due to functions that look like "operator+(..)+0x2c" 227 func_regexp = re.compile("(^[a-f0-9]*) \<(.*)\>:$") 228 offset_regexp = re.compile("(.*)\+0x([a-f0-9]*)") 229 230 # A disassembly line looks like: 231 # 177b2: b510 push {r4, lr} 232 asm_regexp = re.compile("(^[ a-f0-9]*):[ a-f0-0]*.*$") 233 234 current_symbol = None # The current function symbol in the disassembly. 235 current_symbol_addr = 0 # The address of the current function. 236 addr_index = 0 # The address that we are currently looking for. 237 238 stream = subprocess.Popen(cmd, stdout=subprocess.PIPE).stdout 239 result = {} 240 for line in stream: 241 # Is it a function line like: 242 # 000177b0 <android::IBinder::~IBinder()>: 243 components = func_regexp.match(line) 244 if components: 245 # This is a new function, so record the current function and its address. 246 current_symbol_addr = int(components.group(1), 16) 247 current_symbol = components.group(2) 248 249 # Does it have an optional offset like: "foo(..)+0x2c"? 250 components = offset_regexp.match(current_symbol) 251 if components: 252 current_symbol = components.group(1) 253 offset = components.group(2) 254 if offset: 255 current_symbol_addr -= int(offset, 16) 256 257 # Is it an disassembly line like: 258 # 177b2: b510 push {r4, lr} 259 components = asm_regexp.match(line) 260 if components: 261 addr = components.group(1) 262 target_addr = addrs[addr_index] 263 i_addr = int(addr, 16) 264 i_target = int(target_addr, 16) 265 if i_addr == i_target: 266 result[target_addr] = (current_symbol, i_target - current_symbol_addr) 267 addr_index += 1 268 if addr_index >= len(addrs): 269 break 270 stream.close() 271 272 return result 273 274 275 def CallCppFilt(mangled_symbol): 276 cmd = [ToolPath("c++filt")] 277 process = subprocess.Popen(cmd, stdin=subprocess.PIPE, stdout=subprocess.PIPE) 278 process.stdin.write(mangled_symbol) 279 process.stdin.write("\n") 280 process.stdin.close() 281 demangled_symbol = process.stdout.readline().strip() 282 process.stdout.close() 283 return demangled_symbol 284 285 def FormatSymbolWithOffset(symbol, offset): 286 if offset == 0: 287 return symbol 288 return "%s+%d" % (symbol, offset) 289