|
| 1 | +# Copyright (c) Meta Platforms, Inc. and affiliates. |
| 2 | +# All rights reserved. |
| 3 | +# |
| 4 | +# This source code is licensed under the BSD-style license found in the |
| 5 | +# LICENSE file in the root directory of this source tree. |
| 6 | + |
| 7 | +import re |
| 8 | +import subprocess |
| 9 | +import sys |
| 10 | +from dataclasses import dataclass |
| 11 | +from typing import Dict, List, Optional, Union |
| 12 | + |
| 13 | + |
| 14 | +@dataclass |
| 15 | +class Symbol: |
| 16 | + name: str |
| 17 | + addr: int |
| 18 | + size: int |
| 19 | + symbol_type: str |
| 20 | + |
| 21 | + |
| 22 | +class Parser: |
| 23 | + def __init__(self, elf: str, toolchain_prefix: str = "", filter=None): |
| 24 | + self.elf = elf |
| 25 | + self.toolchain_prefix = toolchain_prefix |
| 26 | + self.symbols: Dict[str, Symbol] = self._get_nm_output() |
| 27 | + self.filter = filter |
| 28 | + |
| 29 | + @staticmethod |
| 30 | + def run_nm( |
| 31 | + elf_file_path: str, args: Optional[List[str]] = None, nm: str = "nm" |
| 32 | + ) -> str: |
| 33 | + """ |
| 34 | + Run the nm command on the specified ELF file. |
| 35 | + """ |
| 36 | + args = [] if args is None else args |
| 37 | + cmd = [nm] + args + [elf_file_path] |
| 38 | + try: |
| 39 | + result = subprocess.run(cmd, check=True, capture_output=True, text=True) |
| 40 | + return result.stdout |
| 41 | + except FileNotFoundError: |
| 42 | + print(f"Error: 'nm' command not found. Please ensure it's installed.") |
| 43 | + sys.exit(1) |
| 44 | + except subprocess.CalledProcessError as e: |
| 45 | + print(f"Error running nm on {elf_file_path}: {e}") |
| 46 | + print(f"stderr: {e.stderr}") |
| 47 | + sys.exit(1) |
| 48 | + |
| 49 | + def _get_nm_output(self) -> Dict[str, Symbol]: |
| 50 | + args = [ |
| 51 | + "--print-size", |
| 52 | + "--size-sort", |
| 53 | + "--reverse-sort", |
| 54 | + "--demangle", |
| 55 | + "--format=bsd", |
| 56 | + ] |
| 57 | + output = Parser.run_nm( |
| 58 | + self.elf, |
| 59 | + args, |
| 60 | + nm=self.toolchain_prefix + "nm" if self.toolchain_prefix else "nm", |
| 61 | + ) |
| 62 | + lines = output.splitlines() |
| 63 | + symbols = [] |
| 64 | + symbol_pattern = re.compile( |
| 65 | + r"(?P<addr>[0-9a-fA-F]+)\s+(?P<size>[0-9a-fA-F]+)\s+(?P<type>\w)\s+(?P<name>.+)" |
| 66 | + ) |
| 67 | + |
| 68 | + def parse_line(line: str) -> Optional[Symbol]: |
| 69 | + |
| 70 | + match = symbol_pattern.match(line) |
| 71 | + if match: |
| 72 | + addr = int(match.group("addr"), 16) |
| 73 | + size = int(match.group("size"), 16) |
| 74 | + type_ = match.group("type").strip().strip("\n") |
| 75 | + name = match.group("name").strip().strip("\n") |
| 76 | + return Symbol(name=name, addr=addr, size=size, symbol_type=type_) |
| 77 | + return None |
| 78 | + |
| 79 | + for line in lines: |
| 80 | + symbol = parse_line(line) |
| 81 | + if symbol: |
| 82 | + symbols.append(symbol) |
| 83 | + |
| 84 | + assert len(symbols) > 0, "No symbols found in nm output" |
| 85 | + if len(symbols) != len(lines): |
| 86 | + print( |
| 87 | + "** Warning: Not all lines were parsed, check the output of nm. Parsed {len(symbols)} lines, given {len(lines)}" |
| 88 | + ) |
| 89 | + if any(symbol.size == 0 for symbol in symbols): |
| 90 | + print("** Warning: Some symbols have zero size, check the output of nm.") |
| 91 | + |
| 92 | + # TODO: Populate the section and module fields from the linker map if available (-Wl,-Map=linker.map) |
| 93 | + return {symbol.name: symbol for symbol in symbols} |
| 94 | + |
| 95 | + def print(self): |
| 96 | + print(f"Elf: {self.elf}") |
| 97 | + |
| 98 | + def print_table(filter=None, filter_name=None): |
| 99 | + print("\nAddress\t\tSize\tType\tName") |
| 100 | + # Apply filter and sort symbols |
| 101 | + symbols_to_print = { |
| 102 | + name: sym |
| 103 | + for name, sym in self.symbols.items() |
| 104 | + if not filter or filter(sym) |
| 105 | + } |
| 106 | + sorted_symbols = sorted( |
| 107 | + symbols_to_print.items(), key=lambda x: x[1].size, reverse=True |
| 108 | + ) |
| 109 | + |
| 110 | + # Print symbols and calculate total size |
| 111 | + size_total = 0 |
| 112 | + for name, sym in sorted_symbols: |
| 113 | + print(f"{hex(sym.addr)}\t\t{sym.size}\t{sym.symbol_type}\t{sym.name}") |
| 114 | + size_total += sym.size |
| 115 | + |
| 116 | + # Print summary |
| 117 | + symbol_percent = len(symbols_to_print) / len(self.symbols) * 100 |
| 118 | + print("-----") |
| 119 | + print(f"> Total bytes: {size_total}") |
| 120 | + print( |
| 121 | + f"Counted: {len(symbols_to_print)}/{len(self.symbols)}, {symbol_percent:0.2f}% (filter: '{filter_name}')" |
| 122 | + ) |
| 123 | + print("=====\n") |
| 124 | + |
| 125 | + # Print tables with different filters |
| 126 | + def is_executorch_symbol(s): |
| 127 | + return "executorch" in s.name or s.name.startswith("et") |
| 128 | + |
| 129 | + FILTER_NAME_TO_FILTER_AND_LABEL = { |
| 130 | + "all": (None, "All"), |
| 131 | + "executorch": (is_executorch_symbol, "ExecuTorch"), |
| 132 | + "executorch_text": ( |
| 133 | + lambda s: is_executorch_symbol(s) and s.symbol_type.lower() == "t", |
| 134 | + "ExecuTorch .text", |
| 135 | + ), |
| 136 | + } |
| 137 | + |
| 138 | + filter_func, label = FILTER_NAME_TO_FILTER_AND_LABEL.get( |
| 139 | + self.filter, FILTER_NAME_TO_FILTER_AND_LABEL["all"] |
| 140 | + ) |
| 141 | + print_table(filter_func, label) |
| 142 | + |
| 143 | + |
| 144 | +if __name__ == "__main__": |
| 145 | + import argparse |
| 146 | + |
| 147 | + parser = argparse.ArgumentParser( |
| 148 | + description="Process ELF file and linker map file." |
| 149 | + ) |
| 150 | + parser.add_argument( |
| 151 | + "-e", "--elf-file-path", required=True, help="Path to the ELF file" |
| 152 | + ) |
| 153 | + parser.add_argument( |
| 154 | + "-f", |
| 155 | + "--filter", |
| 156 | + required=False, |
| 157 | + default="all", |
| 158 | + help="Filter symbols by pre-defined filters", |
| 159 | + choices=["all", "executorch", "executorch_text"], |
| 160 | + ) |
| 161 | + parser.add_argument( |
| 162 | + "-p", |
| 163 | + "--toolchain-prefix", |
| 164 | + required=False, |
| 165 | + default="", |
| 166 | + help="Optional toolchain prefix for nm", |
| 167 | + ) |
| 168 | + |
| 169 | + args = parser.parse_args() |
| 170 | + p = Parser(args.elf_file_path, args.toolchain_prefix, filter=args.filter) |
| 171 | + p.print() |
0 commit comments