dwarf-assembly/compare_sizes.py

201 lines
5.9 KiB
Python
Raw Normal View History

#!/usr/bin/env python3
""" Compare the sizes of the .eh_frame section in the original binary and of
the .text in the generated .eh_elf.so. """
import argparse
import os
import subprocess
from collections import namedtuple
2018-05-09 17:55:57 +02:00
from shared_python import elf_so_deps
''' An ELF object, including the path to the ELF itself, and the path to its
matching eh_elf '''
ElfObject = namedtuple('ElfObject', 'elf eh_elf')
def format_size(size):
''' Format a size to a human-readable string '''
units = ['B', 'KiB', 'MiB', 'GiB'] # We'll never go over that
cur_unit = 0
while cur_unit < len(units) and size >= 1024:
size /= 1024
cur_unit += 1
return '{:.1f} {}'.format(size, units[cur_unit])
2018-05-15 16:14:17 +02:00
def invoke_objdump_headers(elf_loc):
''' Call objdump -h, returning the list of lines outputted '''
if not os.path.isfile(elf_loc):
raise FileNotFoundError
try:
objdump_out = subprocess.check_output(['objdump', '-h', elf_loc]) \
.decode('utf-8')
except subprocess.CalledProcessError as exn:
2018-05-15 16:14:17 +02:00
raise Exception(("Cannot run objdump on {}: objdump "
"terminated with exit code {}.").format(
2018-05-15 16:14:17 +02:00
elf_loc, exn.returncode))
2018-05-15 16:14:17 +02:00
return objdump_out.split('\n')
def get_elf_sections(elf_loc):
''' List the ELF sections of the given ELF '''
sections = {}
for line in invoke_objdump_headers(elf_loc):
line = line.strip()
2018-05-09 17:55:57 +02:00
if not line or not '0' <= line[0] <= '9': # not a section line
continue
spl = line.split()
2018-05-15 16:14:17 +02:00
sections[spl[1]] = {
'name': spl[1],
'size': int(spl[2], 0x10),
}
2018-05-15 16:14:17 +02:00
return sections
def matching_eh_elf(eh_locs, elf_name):
''' Get the .eh_elf.so file matching elf_name in the list of directories
eh_locs.
Raises FileNotFoundError if there is no such file '''
basename = os.path.basename(elf_name) + '.eh_elf.so'
for prefix in eh_locs:
eh_elf_path = os.path.join(prefix, basename)
if os.path.isfile(eh_elf_path):
return eh_elf_path
raise FileNotFoundError("No such file {}".format(basename))
def objects_list(args):
''' Get the list of elf objects to process '''
out = []
if args.deps:
objects = set(args.object)
for obj in args.object:
objects = objects.union(elf_so_deps(obj))
objects = list(objects)
objects.sort()
else:
objects = args.object
for obj in objects:
out.append(ElfObject(obj, matching_eh_elf(args.eh_elfs, obj)))
return out
def process_args():
''' Process `sys.argv` arguments '''
parser = argparse.ArgumentParser(
description=("Compare the sizes of the .eh_frame section in the "
"original binary and of the .text in the generated "
".eh_elf.so."),
)
parser.add_argument('--deps', action='store_true',
help=("Also compare the shared objects this object "
"depends on"))
parser.add_argument('--eh-elfs', required=True, action='append',
help=("Indicate the directory in which eh_elfs are "
"located"))
parser.add_argument('object', nargs='+',
help="The ELF object(s) to process")
return parser.parse_args()
def main():
args = process_args()
objs = objects_list(args)
col_names = [
'Shared object',
'Orig eh_frame',
2018-05-15 16:14:17 +02:00
'Gen eh_elf .text',
'+ .rodata',
'Growth',
]
col_len = []
displayed_name_filter = lambda x: os.path.basename(x.elf)
max_elf_name = max(map(lambda x: len(displayed_name_filter(x)), objs))
col_len.append(max(max_elf_name, len(col_names[0])))
2018-05-15 16:14:17 +02:00
for i in range(1, len(col_names)):
col_len.append(len(col_names[i]) + 1)
col_len = list(map(str, col_len))
header_format = ('{:<' + col_len[0] + '} '
'{:<' + col_len[1] + '} '
'{:<' + col_len[2] + '} '
2018-05-15 16:14:17 +02:00
'{:<' + col_len[3] + '} '
'{:<' + col_len[4] + '}')
row_format = ('{:>' + col_len[0] + '} '
'{:>' + col_len[1] + '} '
'{:>' + col_len[2] + '} '
2018-05-15 16:14:17 +02:00
'{:>' + col_len[3] + '} '
'{:>' + col_len[4] + '}')
print(header_format.format(
col_names[0],
col_names[1],
col_names[2],
col_names[3],
2018-05-15 16:14:17 +02:00
col_names[4],
))
total_eh_frame_size = 0
2018-05-15 16:14:17 +02:00
total_eh_elf_text_size = 0
total_eh_elf_size = 0
for obj in objs:
2018-05-15 16:14:17 +02:00
elf_sections = get_elf_sections(obj.elf)
eh_elf_sections = get_elf_sections(obj.eh_elf)
eh_frame_size = elf_sections['.eh_frame']['size']
eh_elf_text_size = eh_elf_sections['.text']['size']
eh_elf_size = eh_elf_text_size + eh_elf_sections['.rodata']['size']
total_eh_frame_size += eh_frame_size
2018-05-15 16:14:17 +02:00
total_eh_elf_text_size += eh_elf_text_size
total_eh_elf_size += eh_elf_size
print(row_format.format(
displayed_name_filter(obj),
format_size(eh_frame_size),
2018-05-15 16:14:17 +02:00
format_size(eh_elf_text_size),
format_size(eh_elf_size),
'{:.2f}'.format(eh_elf_size / eh_frame_size)))
2018-05-15 16:14:17 +02:00
# Checking for missed big sections
for section in eh_elf_sections:
if section == '.text' or section == '.rodata':
continue
if eh_elf_sections[section]['size'] > eh_elf_size / 2:
print("\t\t/!\\ Section {} is big ({}) in the eh_elf".format(
section, format_size(eh_elf_sections[section]['size'])))
print(row_format.format(
'Total',
format_size(total_eh_frame_size),
format_size(total_eh_elf_size),
2018-05-15 16:14:17 +02:00
format_size(total_eh_elf_text_size),
'{:.2f}'.format(total_eh_elf_size / total_eh_frame_size)))
if __name__ == '__main__':
main()