#!/usr/bin/env python3 # # Copyright (c) 2021 Xiaomi Corporation # Copyright (c) 2016, 2020 Intel Corporation # # SPDX-License-Identifier: Apache-2.0 # Based on a script by: # Chereau, Fabien """ Process an ELF file to generate size report on RAM and ROM. """ import argparse import json import os import re import sys from pathlib import Path import elftools from anytree import NodeMixin, RenderTree, findall_by_attr from anytree.exporter import DictExporter from colorama import Fore, init from elftools.dwarf.descriptions import ( describe_DWARF_expr, describe_form_class, set_global_machine_arch, ) from elftools.dwarf.locationlists import LocationExpr, LocationParser from elftools.elf.elffile import ELFFile from elftools.elf.sections import SymbolTableSection from packaging import version if version.parse(elftools.__version__) < version.parse("0.24"): sys.exit("pyelftools is out of date, need version 0.24 or later") # ELF section flags SHF_WRITE = 0x1 SHF_ALLOC = 0x2 SHF_EXEC = 0x4 SHF_WRITE_ALLOC = SHF_WRITE | SHF_ALLOC SHF_ALLOC_EXEC = SHF_ALLOC | SHF_EXEC DT_LOCATION = re.compile(r"\(DW_OP_addr: ([0-9a-f]+)\)") SRC_FILE_EXT = (".h", ".c", ".hpp", ".cpp", ".hxx", ".cxx", ".c++") def get_symbol_addr(sym): """Get the address of a symbol""" return sym["st_value"] def get_symbol_size(sym): """Get the size of a symbol""" return sym["st_size"] def is_symbol_in_ranges(sym, ranges): """ Given a list of start/end addresses, test if the symbol lies within any of these address ranges. """ for bound in ranges: if bound["start"] <= sym["st_value"] <= bound["end"]: return True return False def get_die_mapped_address(die, parser, dwarfinfo): """Get the bounding addresses from a DIE variable or subprogram""" low = None high = None if die.tag == "DW_TAG_variable": if "DW_AT_location" in die.attributes: loc_attr = die.attributes["DW_AT_location"] if parser.attribute_has_location(loc_attr, die.cu["version"]): loc = parser.parse_from_attribute(loc_attr, die.cu["version"]) if isinstance(loc, LocationExpr): addr = describe_DWARF_expr(loc.loc_expr, dwarfinfo.structs) matcher = DT_LOCATION.match(addr) if matcher: low = int(matcher.group(1), 16) high = low + 1 if die.tag == "DW_TAG_subprogram": if "DW_AT_low_pc" in die.attributes: low = die.attributes["DW_AT_low_pc"].value high_pc = die.attributes["DW_AT_high_pc"] high_pc_class = describe_form_class(high_pc.form) if high_pc_class == "address": high = high_pc.value elif high_pc_class == "constant": high = low + high_pc.value return low, high def match_symbol_address(symlist, die, parser, dwarfinfo): """ Find the symbol from a symbol list where it matches the address in DIE variable, or within the range of a DIE subprogram. """ low, high = get_die_mapped_address(die, parser, dwarfinfo) if low is None: return None for sym in symlist: if low <= sym["symbol"]["st_value"] < high: return sym return None def get_symbols(elf, addr_ranges): """ Fetch the symbols from the symbol table and put them into ROM, RAM buckets. """ rom_syms = dict() ram_syms = dict() unassigned_syms = dict() rom_addr_ranges = addr_ranges["rom"] ram_addr_ranges = addr_ranges["ram"] for section in elf.iter_sections(): if isinstance(section, SymbolTableSection): for sym in section.iter_symbols(): # Ignore symbols with size == 0 if get_symbol_size(sym) == 0: continue found_sec = False entry = {"name": sym.name, "symbol": sym, "mapped_files": set()} # If symbol is in ROM area? if is_symbol_in_ranges(sym, rom_addr_ranges): if sym.name not in rom_syms: rom_syms[sym.name] = list() rom_syms[sym.name].append(entry) found_sec = True # If symbol is in RAM area? if is_symbol_in_ranges(sym, ram_addr_ranges): if sym.name not in ram_syms: ram_syms[sym.name] = list() ram_syms[sym.name].append(entry) found_sec = True if not found_sec: unassigned_syms["sym_name"] = entry ret = {"rom": rom_syms, "ram": ram_syms, "unassigned": unassigned_syms} return ret def get_section_ranges(elf): """ Parse ELF header to find out the address ranges of ROM or RAM sections and their total sizes. """ rom_addr_ranges = list() ram_addr_ranges = list() rom_size = 0 ram_size = 0 for section in elf.iter_sections(): size = section["sh_size"] sec_start = section["sh_addr"] sec_end = sec_start + size - 1 bound = {"start": sec_start, "end": sec_end} if section["sh_type"] == "SHT_NOBITS": # BSS and noinit sections ram_addr_ranges.append(bound) ram_size += size elif section["sh_type"] == "SHT_PROGBITS": # Sections to be in flash or memory flags = section["sh_flags"] if (flags & SHF_ALLOC_EXEC) == SHF_ALLOC_EXEC: # Text section rom_addr_ranges.append(bound) rom_size += size elif (flags & SHF_WRITE_ALLOC) == SHF_WRITE_ALLOC: # Data occupies both ROM and RAM # since at boot, content is copied from ROM to RAM rom_addr_ranges.append(bound) rom_size += size ram_addr_ranges.append(bound) ram_size += size elif (flags & SHF_ALLOC) == SHF_ALLOC: # Read only data rom_addr_ranges.append(bound) rom_size += size ret = { "rom": rom_addr_ranges, "rom_total_size": rom_size, "ram": ram_addr_ranges, "ram_total_size": ram_size, } return ret def get_die_filename(die, lineprog): """Get the source code filename associated with a DIE""" file_index = die.attributes["DW_AT_decl_file"].value file_entry = lineprog["file_entry"][file_index - 1] dir_index = file_entry["dir_index"] if dir_index == 0: filename = file_entry.name else: directory = lineprog.header["include_directory"][dir_index - 1] filename = os.path.join(directory, file_entry.name) path = Path(filename.decode()) # Prepend output path to relative path if not path.is_absolute(): output = Path(args.output) path = output.joinpath(path) # Change path to relative to Nuttx base try: path = path.resolve() except OSError as e: # built-ins can't be resolved, so it's not an issue if "" not in str(path): raise e return path def do_simple_name_matching(elf, symbol_dict, processed): """ Sequentially process DIEs in compiler units with direct file mappings within the DIEs themselves, and do simply matching between DIE names and symbol names. """ mapped_symbols = processed["mapped_symbols"] mapped_addresses = processed["mapped_addr"] unmapped_symbols = processed["unmapped_symbols"] newly_mapped_syms = set() dwarfinfo = elf.get_dwarf_info() location_lists = dwarfinfo.location_lists() location_parser = LocationParser(location_lists) unmapped_dies = set() # Loop through all compile units for compile_unit in dwarfinfo.iter_CUs(): lineprog = dwarfinfo.line_program_for_CU(compile_unit) if lineprog is None: continue # Loop through each DIE and find variables and # subprograms (i.e. functions) for die in compile_unit.iter_DIEs(): sym_name = None # Process variables if die.tag == "DW_TAG_variable": # DW_AT_declaration # having "DW_AT_location" means this maps # to an actual address (e.g. not an extern) if "DW_AT_location" in die.attributes: sym_name = die.get_full_path() # Process subprograms (i.e. functions) if they are valid if die.tag == "DW_TAG_subprogram": # Refer to another DIE for name if ("DW_AT_abstract_origin" in die.attributes) or ( "DW_AT_specification" in die.attributes ): unmapped_dies.add(die) # having "DW_AT_low_pc" means it maps to # an actual address elif "DW_AT_low_pc" in die.attributes: # DW_AT_low_pc == 0 is a weak function # which has been overridden if die.attributes["DW_AT_low_pc"].value != 0: sym_name = die.get_full_path() # For mangled function names, the linkage name # is what appears in the symbol list if "DW_AT_linkage_name" in die.attributes: linkage = die.attributes["DW_AT_linkage_name"] sym_name = linkage.value.decode() if sym_name is not None: # Skip DIE with no reference back to a file if "DW_AT_decl_file" not in die.attributes: continue is_die_mapped = False if sym_name in symbol_dict: mapped_symbols.add(sym_name) symlist = symbol_dict[sym_name] symbol = match_symbol_address( symlist, die, location_parser, dwarfinfo ) if symbol is not None: symaddr = symbol["symbol"]["st_value"] if symaddr not in mapped_addresses: is_die_mapped = True path = get_die_filename(die, lineprog) symbol["mapped_files"].add(path) mapped_addresses.add(symaddr) newly_mapped_syms.add(sym_name) if not is_die_mapped: unmapped_dies.add(die) mapped_symbols = mapped_symbols.union(newly_mapped_syms) unmapped_symbols = unmapped_symbols.difference(newly_mapped_syms) processed["mapped_symbols"] = mapped_symbols processed["mapped_addr"] = mapped_addresses processed["unmapped_symbols"] = unmapped_symbols processed["unmapped_dies"] = unmapped_dies def mark_address_aliases(symbol_dict, processed): """ Mark symbol aliases as already mapped to prevent double counting. There are functions and variables which are aliases to other functions/variables. So this marks them as mapped so they will not get counted again when a tree is being built for display. """ mapped_symbols = processed["mapped_symbols"] mapped_addresses = processed["mapped_addr"] unmapped_symbols = processed["unmapped_symbols"] already_mapped_syms = set() for ums in unmapped_symbols: for one_sym in symbol_dict[ums]: symbol = one_sym["symbol"] if symbol["st_value"] in mapped_addresses: already_mapped_syms.add(ums) mapped_symbols = mapped_symbols.union(already_mapped_syms) unmapped_symbols = unmapped_symbols.difference(already_mapped_syms) processed["mapped_symbols"] = mapped_symbols processed["mapped_addr"] = mapped_addresses processed["unmapped_symbols"] = unmapped_symbols def do_address_range_matching(elf, symbol_dict, processed): """ Match symbols indirectly using address ranges. This uses the address ranges of DIEs and map them to symbols residing within those ranges, and works on DIEs that have not been mapped in previous steps. This works on symbol names that do not match the names in DIEs, e.g. "" in DIE, but ".constprop.*" in symbol name list. This also helps with mapping the mangled function names in C++, since the names in DIE are actual function names in source code and not mangled version of them. """ if "unmapped_dies" not in processed: return mapped_symbols = processed["mapped_symbols"] mapped_addresses = processed["mapped_addr"] unmapped_symbols = processed["unmapped_symbols"] newly_mapped_syms = set() dwarfinfo = elf.get_dwarf_info() location_lists = dwarfinfo.location_lists() location_parser = LocationParser(location_lists) unmapped_dies = processed["unmapped_dies"] # Group DIEs by compile units cu_list = dict() for die in unmapped_dies: cu = die.cu if cu not in cu_list: cu_list[cu] = {"dies": set()} cu_list[cu]["dies"].add(die) # Loop through all compile units for cu in cu_list: lineprog = dwarfinfo.line_program_for_CU(cu) # Map offsets from DIEs offset_map = dict() for die in cu.iter_DIEs(): offset_map[die.offset] = die for die in cu_list[cu]["dies"]: if not die.tag == "DW_TAG_subprogram": continue path = None # Has direct reference to file, so use it if "DW_AT_decl_file" in die.attributes: path = get_die_filename(die, lineprog) # Loop through indirect reference until a direct # reference to file is found if ("DW_AT_abstract_origin" in die.attributes) or ( "DW_AT_specification" in die.attributes ): die_ptr = die while path is None: if not (die_ptr.tag == "DW_TAG_subprogram") or not ( ("DW_AT_abstract_origin" in die_ptr.attributes) or ("DW_AT_specification" in die_ptr.attributes) ): break if "DW_AT_abstract_origin" in die_ptr.attributes: ofname = "DW_AT_abstract_origin" elif "DW_AT_specification" in die_ptr.attributes: ofname = "DW_AT_specification" offset = die_ptr.attributes[ofname].value offset += die_ptr.cu.cu_offset # There is nothing to reference so no need to continue if offset not in offset_map: break die_ptr = offset_map[offset] if "DW_AT_decl_file" in die_ptr.attributes: path = get_die_filename(die_ptr, lineprog) # Nothing to map if path is not None: low, high = get_die_mapped_address(die, location_parser, dwarfinfo) if low is None: continue for ums in unmapped_symbols: for one_sym in symbol_dict[ums]: symbol = one_sym["symbol"] symaddr = symbol["st_value"] if symaddr not in mapped_addresses: if low <= symaddr < high: one_sym["mapped_files"].add(path) mapped_addresses.add(symaddr) newly_mapped_syms.add(ums) mapped_symbols = mapped_symbols.union(newly_mapped_syms) unmapped_symbols = unmapped_symbols.difference(newly_mapped_syms) processed["mapped_symbols"] = mapped_symbols processed["mapped_addr"] = mapped_addresses processed["unmapped_symbols"] = unmapped_symbols def set_root_path_for_unmapped_symbols(symbol_dict, addr_range, processed): """ Set root path for unmapped symbols. Any unmapped symbols are added under the root node if those symbols reside within the desired memory address ranges (e.g. ROM or RAM). """ mapped_symbols = processed["mapped_symbols"] mapped_addresses = processed["mapped_addr"] unmapped_symbols = processed["unmapped_symbols"] newly_mapped_syms = set() for ums in unmapped_symbols: for one_sym in symbol_dict[ums]: symbol = one_sym["symbol"] symaddr = symbol["st_value"] if is_symbol_in_ranges(symbol, addr_range): if symaddr not in mapped_addresses: path = Path(":") one_sym["mapped_files"].add(path) mapped_addresses.add(symaddr) newly_mapped_syms.add(ums) mapped_symbols = mapped_symbols.union(newly_mapped_syms) unmapped_symbols = unmapped_symbols.difference(newly_mapped_syms) processed["mapped_symbols"] = mapped_symbols processed["mapped_addr"] = mapped_addresses processed["unmapped_symbols"] = unmapped_symbols def find_common_path_prefix(symbol_dict): """ Find the common path prefix of all mapped files. Must be called before set_root_path_for_unmapped_symbols(). """ paths = list() for _, sym in symbol_dict.items(): for symbol in sym: for file in symbol["mapped_files"]: paths.append(file) return os.path.commonpath(paths) class TreeNode(NodeMixin): """ A symbol node. """ def __init__(self, name, identifier, size=0, parent=None, children=None): super().__init__() self.name = name self.size = size self.parent = parent self.identifier = identifier if children: self.children = children def __repr__(self): return self.name def sum_node_children_size(node): """ Calculate the sum of symbol size of all direct children. """ size = 0 for child in node.children: size += child.size return size def generate_any_tree(symbol_dict, total_size, path_prefix): """ Generate a symbol tree for output. """ root = TreeNode("Root", "root") node_no_paths = TreeNode("(no paths)", ":", parent=root) if Path(path_prefix) == Path(args.nuttxbase): # All source files are under nuttx_base so there is # no need for another level. node_nuttx_base = root node_output_dir = root node_workspace = root node_others = root else: node_nuttx_base = TreeNode("nuttx_base", args.nuttxbase) node_output_dir = TreeNode("OUTPUT_DIR", args.output) node_others = TreeNode("/", "/") if args.workspace: node_workspace = TreeNode("WORKSPACE", args.workspace) else: node_workspace = node_others # A set of helper function for building a simple tree with a path-like # hierarchy. def _insert_one_elem(root, path, size): cur = None node = None parent = root for part in path.parts: if cur is None: cur = part else: cur = str(Path(cur, part)) results = findall_by_attr(root, cur, name="identifier") if results: item = results[0] item.size += size parent = item else: if node: parent = node node = TreeNode( name=str(part), identifier=cur, size=size, parent=parent ) # Mapping paths to tree nodes path_node_map = [ [Path(args.nuttxbase), node_nuttx_base], [Path(args.output), node_output_dir], ] if args.workspace: path_node_map.append([Path(args.workspace), node_workspace]) for name, sym in symbol_dict.items(): for symbol in sym: size = get_symbol_size(symbol["symbol"]) for file in symbol["mapped_files"]: path = Path(file, name) if path.is_absolute(): has_node = False for one_path in path_node_map: if one_path[0] in path.parents: path = path.relative_to(one_path[0]) dest_node = one_path[1] has_node = True break if not has_node: dest_node = node_others else: dest_node = node_no_paths _insert_one_elem(dest_node, path, size) if node_nuttx_base is not root: # nuttx_base and OUTPUT_DIR nodes don't have sum of symbol size # so calculate them here. node_nuttx_base.size = sum_node_children_size(node_nuttx_base) node_output_dir.size = sum_node_children_size(node_output_dir) # Find out which nodes need to be in the tree. # "(no path)", nuttx_base nodes are essential. children = [node_no_paths, node_nuttx_base] if node_output_dir.height != 0: # OUTPUT_DIR may be under nuttx_base. children.append(node_output_dir) if node_others.height != 0: # Only include "others" node if there is something. children.append(node_others) if args.workspace: node_workspace.size = sum_node_children_size(node_workspace) if node_workspace.height != 0: children.append(node_workspace) root.children = children root.size = total_size # Need to account for code and data where there are not emitted # symbols associated with them. node_hidden_syms = TreeNode("(hidden)", "(hidden)", parent=root) node_hidden_syms.size = root.size - sum_node_children_size(root) return root def node_sort(items): """ Node sorting used with RenderTree. """ return sorted(items, key=lambda item: item.name) def print_any_tree(root, total_size, depth): """ Print the symbol tree. """ print("{:101s} {:7s} {:8s}".format(Fore.YELLOW + "Path", "Size", "%" + Fore.RESET)) print("=" * 110) for row in RenderTree(root, childiter=node_sort, maxlevel=depth): f = len(row.pre) + len(row.node.name) s = str(row.node.size).rjust(100 - f) percent = 100 * float(row.node.size) / float(total_size) cc = cr = "" if not row.node.children: if row.node.name != "(hidden)": cc = Fore.CYAN cr = Fore.RESET elif row.node.name.endswith(SRC_FILE_EXT): cc = Fore.GREEN cr = Fore.RESET print( f"{row.pre}{cc}{row.node.name} {s} {cr}{Fore.BLUE}{percent:6.2f}%{Fore.RESET}" ) print("=" * 110) print(f"{total_size:>101}") def parse_args(): """ Parse command line arguments. """ global args parser = argparse.ArgumentParser() parser.add_argument("-k", "--kernel", required=True, help="Nuttx ELF binary") parser.add_argument("-z", "--nuttxbase", required=True, help="Nuttx base path") parser.add_argument( "-q", "--quiet", action="store_true", help="Do not output anything on the screen.", ) parser.add_argument("-o", "--output", required=True, help="Output path") parser.add_argument( "-w", "--workspace", default=None, help="Workspace path (Usually the same as TOPDIR)", ) parser.add_argument("target", choices=["rom", "ram", "all"]) parser.add_argument( "-d", "--depth", dest="depth", type=int, default=None, help="How deep should we go into the tree", metavar="DEPTH", ) parser.add_argument( "-v", "--verbose", action="store_true", help="Print extra debugging information" ) parser.add_argument("--json", help="store results in a JSON file.") args = parser.parse_args() def main(): """ Main program. """ parse_args() # Init colorama init() assert os.path.exists(args.kernel), "{0} does not exist.".format(args.kernel) if args.target == "ram": targets = ["ram"] elif args.target == "rom": targets = ["rom"] elif args.target == "all": targets = ["rom", "ram"] for t in targets: elf = ELFFile(open(args.kernel, "rb")) assert elf.has_dwarf_info(), "ELF file has no DWARF information" set_global_machine_arch(elf.get_machine_arch()) addr_ranges = get_section_ranges(elf) symbols = get_symbols(elf, addr_ranges) for sym in symbols["unassigned"].values(): print("WARN: Symbol '{0}' is not in RAM or ROM".format(sym["name"])) symbol_dict = None if args.json: jsonout = args.json else: jsonout = os.path.join(args.output, f"{t}.json") symbol_dict = symbols[t] symsize = addr_ranges[f"{t}_total_size"] ranges = addr_ranges[t] if symbol_dict is not None: processed = { "mapped_symbols": set(), "mapped_addr": set(), "unmapped_symbols": set(symbol_dict.keys()), } do_simple_name_matching(elf, symbol_dict, processed) mark_address_aliases(symbol_dict, processed) do_address_range_matching(elf, symbol_dict, processed) mark_address_aliases(symbol_dict, processed) common_path_prefix = find_common_path_prefix(symbol_dict) set_root_path_for_unmapped_symbols(symbol_dict, ranges, processed) if args.verbose: for sym in processed["unmapped_symbols"]: print("INFO: Unmapped symbol: {0}".format(sym)) root = generate_any_tree(symbol_dict, symsize, common_path_prefix) if not args.quiet: print_any_tree(root, symsize, args.depth) exporter = DictExporter() data = dict() data["symbols"] = exporter.export(root) data["total_size"] = symsize with open(jsonout, "w") as fp: json.dump(data, fp, indent=4) if __name__ == "__main__": main()