linux-startup
Score config
需要一个程序，能够给一个.config打分，看其能覆盖arch/riscv下的多少代码。
这个程序需要做三件事情：
- 读arch/riscv下的.S, .c, .h文件，分析出哪个文件的哪些行被哪些配置项管理；
- 读.config文件，分析出定义了哪些配置项以及该配置项被设置的值是多少；
- 打分，对于每一个被配置项管理的file:line而言，看值是否被满足。
输出：arch/riscv下共有多少代码行，有多少代码行被配置项管理，当前.config能让多少代码行被编译进内核。
1 统计.c, .S, .h的代码行数
统计 arch/riscv/ 下所有 .c、.S、.h 文件的代码行数 。
这个模块的目标是：
- 递归遍历
arch/riscv/ - 只统计后缀为
.c,.S,.h的文件 - 计算总行数（可以区分空行 / 非空行）
import os
def count_source_lines(root="arch/riscv"):
exts = {".c", ".S", ".h"}
total_lines = 0
total_nonempty_lines = 0
file_line_counts = {}
for dirpath, _, filenames in os.walk(root):
for f in filenames:
_, ext = os.path.splitext(f)
if ext not in exts:
continue
path = os.path.join(dirpath, f)
try:
with open(path, "r", encoding="utf-8", errors="ignore") as fp:
lines = fp.readlines()
except Exception as e:
print(f"[WARN] 无法读取文件 {path}: {e}")
continue
n_lines = len(lines)
n_nonempty = sum(1 for l in lines if l.strip())
file_line_counts[path] = (n_lines, n_nonempty)
total_lines += n_lines
total_nonempty_lines += n_nonempty
return total_lines, total_nonempty_lines, file_line_counts
if __name__ == "__main__":
total, nonempty, detail = count_source_lines("/home/rv/mrvga/linux-6.18-rc3/arch/riscv")
print(f"总代码行数: {total}")
print(f"非空行数: {nonempty}")
print(f"文件数: {len(detail)}")
总代码行数: 65151
非空行数: 55337
文件数: 396
2 分析undertaker -j blockconf path:line的输出结果
对于arch/riscv目录下的每个.c, .S, .h文件的每一行都调用undertaker的blockconf功能，筛选出非#开头的行(叫做CONFIGS))，形成从path:line->CONFIGS的映射。
import os
import subprocess
import multiprocessing
import csv
def run_undertaker_for_line(args):
"""调用 undertaker -j blockconf 并返回非#行输出"""
filepath, line = args
cmd = ["undertaker", "-j", "blockconf", f"{filepath}:{line}"]
try:
output = subprocess.check_output(cmd, stderr=subprocess.DEVNULL)
out_lines = output.decode("utf-8").splitlines()
# 过滤掉全部以 # 开头的行
meaningful_lines = [l for l in out_lines if l.strip() and not l.strip().startswith("#")]
if meaningful_lines:
return f"{filepath}:{line}", meaningful_lines
except subprocess.CalledProcessError:
pass
return None
def analyze_riscv_arch(root="arch/riscv", nprocs=8, csv_file="fileline_output.csv"):
"""
对 arch/riscv 下的每一行调用 undertaker -j blockconf，生成映射。
返回:
fileline_to_output: {FILE:LINE : tuple(output_lines)}
"""
# --- 收集文件行数 ---
exts = {".c", ".S", ".h"}
file_line_counts = {}
for dirpath, _, filenames in os.walk(root):
for f in filenames:
_, ext = os.path.splitext(f)
if ext not in exts:
continue
path = os.path.join(dirpath, f)
try:
with open(path, "r", encoding="utf-8", errors="ignore") as fp:
n_lines = sum(1 for _ in fp)
except Exception as e:
print(f"[WARN] 无法读取文件 {path}: {e}")
continue
file_line_counts[path] = n_lines
# --- 构建任务列表 ---
tasks = [(fp, i) for fp, n in file_line_counts.items() for i in range(1, n+1)]
# --- 多进程调用 undertaker ---
fileline_to_output = {} # FILE:LINE -> [output_line1, ...]
with multiprocessing.Pool(processes=nprocs) as pool:
for res in pool.imap_unordered(run_undertaker_for_line, tasks):
if res:
file_line, meaningful_lines = res
# === 对 output_lines 进行过滤与修正 ===
processed_lines = []
for line in meaningful_lines:
line = line.strip()
if line.endswith("_MODULE=n"):
continue # 跳过 _MODULE=n
if line.endswith("=m"):
line = line[:-2] + "=y" # 替换 =m → =y
processed_lines.append(line)
if not processed_lines:
continue
# === 路径归一化：去掉绝对路径前缀 ===
if root.endswith("/"):
rel_root = root
else:
rel_root = root + "/"
if file_line.startswith(rel_root):
file_line = file_line.replace(rel_root, "arch/riscv/", 1)
fileline_to_output[file_line] = tuple(processed_lines)
# --- 写 FILE:LINE -> output_lines CSV ---
with open(csv_file, "w", newline="", encoding="utf-8") as f:
writer = csv.writer(f)
writer.writerow(["file_line", "output_lines"])
for fl, lines in fileline_to_output.items():
writer.writerow([fl, ";".join(lines)])
print(f"[INFO] FILE:LINE -> output_lines CSV 写入 {csv_file}")
return fileline_to_output
if __name__ == "__main__":
mapping = analyze_riscv_arch(
root="/home/rv/mrvga/linux-6.18-rc3/arch/riscv",
nprocs=80,
csv_file="fileline_output.csv"
)
print(f"[INFO] 总共有 {len(mapping)} 个行被配置项管理。")
[INFO] FILE:LINE -> output_lines CSV 写入 fileline_output.csv
[INFO] 总共有 13288 个行被配置项管理。
3 分析path:line是否被编进内核
前面分析得到了fileline_to_output，对于每个每个fileline，需要去看其output在.config中是否被满足：
import csv
import re
def parse_config_file(config_path):
"""解析 .config 文件，返回 {CONFIG_KEY: value}"""
config_map = {}
with open(config_path, "r", encoding="utf-8", errors="ignore") as f:
for line in f:
line = line.strip()
if not line or line.startswith("#"):
# 处理 "# CONFIG_X is not set"
m = re.match(r"#\s*(CONFIG_[A-Za-z0-9_]+)\s+is\s+not\s+set", line)
if m:
config_map[m.group(1)] = "n"
continue
if "=" in line:
k, v = line.split("=", 1)
config_map[k.strip()] = v.strip()
return config_map
def evaluate_compilation(fileline_to_output, config_path, csv_compiled, csv_skipped):
"""
判断哪些行被编译进内核。
输入:
fileline_to_output: {file:line -> [CONFIG_*=y/n/...]}
config_path: .config 文件路径
输出:
csv_compiled: 被编译进内核的行
csv_skipped: 未被编译进内核的行
"""
config_map = parse_config_file(config_path)
compiled = []
skipped = []
for fileline, configs in fileline_to_output.items():
conditions = []
for cfg in configs:
if "=" in cfg:
k, v = cfg.split("=", 1)
conditions.append((k.strip(), v.strip()))
all_ok = True
unmet = []
for k, expected in conditions:
actual = config_map.get(k, "n") # 未定义的 CONFIG 视作 n
if actual != expected:
all_ok = False
unmet.append(f"{k}={actual} (need {expected})")
if all_ok:
compiled.append((fileline, ";".join(configs)))
else:
skipped.append((fileline, ";".join(configs), ";".join(unmet)))
# --- 写 CSV ---
with open(csv_compiled, "w", newline="", encoding="utf-8") as f:
w = csv.writer(f)
w.writerow(["file_line", "output_lines"])
for row in compiled:
w.writerow(row)
with open(csv_skipped, "w", newline="", encoding="utf-8") as f:
w = csv.writer(f)
w.writerow(["file_line", "output_lines", "unmet_conditions"])
for row in skipped:
w.writerow(row)
print(f"[INFO] 被编译进内核的行: {len(compiled)}")
print(f"[INFO] 未编译进内核的行: {len(skipped)}")
total = len(compiled) + len(skipped)
if total:
ratio = len(compiled) / total * 100
print(f"[INFO] 编译进内核比例: {ratio:.2f}%")
return compiled, skipped
if __name__ == "__main__":
mapping = analyze_riscv_arch(
root="/home/rv/mrvga/linux-6.18-rc3/arch/riscv",
nprocs=80,
csv_file="fileline_output.csv"
)
compiled, skipped = evaluate_compilation(
fileline_to_output=mapping,
config_path="/home/rv/mrvga/syzkaller-config",
csv_compiled="compiled_lines.csv",
csv_skipped="skipped_lines.csv"
)
# 对于syzbot提供的和优化后的config作比较
rv@userid-07:~/mrvga/1106$ python test.py
[INFO] FILE:LINE -> output_lines CSV 写入 fileline_output.csv
[INFO] 被编译进内核的行: 11420
[INFO] 未编译进内核的行: 1868
[INFO] 编译进内核比例: 85.94%
rv@userid-07:~/mrvga/1106$ python test.py
[INFO] FILE:LINE -> output_lines CSV 写入 fileline_output.csv
[INFO] 被编译进内核的行: 10523
[INFO] 未编译进内核的行: 2765
[INFO] 编译进内核比例: 79.19%
4 makefile/kbuild对文件的影响
有的配置项是在makefile/kbuild中直接控制某个文件是否被编进内核，而这是undertaker无法处理的。所以需要借助kbuildparser。
import subprocess
import os
import re
def update_mapping_from_kbuildparser_live(fileline_to_output, root="arch/riscv", debug_file="kbuildparser_debug.log"):
"""
调用 kbuildparser -a riscv，解析输出并更新 fileline_to_output。
增加调试信息，写入 debug_file。
"""
updated_mapping = fileline_to_output.copy()
with open(debug_file, "w", encoding="utf-8") as debug:
debug.write(f"[INFO] 调用 kbuildparser 在 {root}\n")
# 执行 kbuildparser
try:
result = subprocess.run(
["kbuildparser", "-a", "riscv"],
cwd=root,
stdout=subprocess.PIPE,
stderr=subprocess.DEVNULL,
text=True,
check=True
)
kbuild_lines = result.stdout.splitlines()
debug.write(f"[INFO] kbuildparser 输出共 {len(kbuild_lines)} 行\n")
debug.write("[INFO] 原始输出开始:\n")
for l in kbuild_lines:
debug.write(l + "\n")
debug.write("[INFO] 原始输出结束\n\n")
except subprocess.CalledProcessError as e:
debug.write(f"[WARN] kbuildparser 执行失败: {e}\n")
return updated_mapping
def parse_kbuild_line(line):
"""解析 kbuildparser 输出行为 (filepath, [(CONFIG_KEY, value), ...])"""
if "<-" not in line:
return line.strip(), []
path, cond = line.split("<-", 1)
path = path.strip()
cond = cond.strip()
configs = []
for token in re.split(r"\s*&&\s*", cond):
token = token.strip()
if token.startswith("!"):
configs.append((token[1:], "n"))
else:
configs.append((token, "y"))
return path, configs
# 遍历 kbuildparser 输出
for line in kbuild_lines:
line = line.strip()
if not line or line.startswith("#"):
continue
filepath, configs = parse_kbuild_line(line)
debug.write(f"[DEBUG] 解析行: '{line}' -> 文件: {filepath}, 配置: {configs}\n")
if not configs:
continue
# 获取文件所有行
file_lines = [k for k in updated_mapping if k.startswith(filepath + ":")]
if not file_lines:
# mapping 中没有文件，尝试读取文件生成每行 key
try:
with open(os.path.join(root, filepath), "r", encoding="utf-8", errors="ignore") as f:
n_lines = sum(1 for _ in f)
file_lines = [f"{filepath}:{i}" for i in range(1, n_lines + 1)]
debug.write(f"[DEBUG] 文件 {filepath} 不在 mapping 中，读取文件行数 {n_lines}\n")
except Exception:
debug.write(f"[WARN] 无法读取文件 {filepath}，跳过\n")
continue
# 更新 mapping
for fl in file_lines:
existing = list(updated_mapping.get(fl, ()))
added = []
for cfg, val in configs:
entry = f"{cfg}={val}"
if entry not in existing:
existing.append(entry)
added.append(entry)
updated_mapping[fl] = tuple(existing)
if added:
debug.write(f"[INFO] 更新 {fl} -> 新增配置: {added}\n")
debug.write(f"\n[INFO] 总共更新了 {len(updated_mapping)} 个行条目\n")
return updated_mapping
5 效果
rv@userid-07:~/mrvga/1106$ python test.py
[INFO] 总代码行数: 66861
[INFO] 非空行数: 57030
[INFO] 文件数: 461
[INFO] 被编译进内核的行: 11220
[INFO] 未编译进内核的行: 2085
[INFO] 编译进内核比例: 84.33%
rv@userid-07:~/mrvga/1106$ python test.py
[INFO] 总代码行数: 66861
[INFO] 非空行数: 57030
[INFO] 文件数: 461
[INFO] 被编译进内核的行: 9596
[INFO] 未编译进内核的行: 3709
[INFO] 编译进内核比例: 72.12%