Re: [PATCH can] can: rockchip: rk3576: fix rtnl_lock deadlock during interface down under bus traffic

From: Cheng Liu

Date: Sun Aug 16 2026 - 22:31:41 EST


Hi Marc, Elaine, and linux-can community,

To help anyone easily reproduce and verify this deadlock scenario on
RV1126B / RK3576, below is the full standalone Python SocketCAN test
script used to trigger the race condition:

```python
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
reproduce_can_deadlock.py

Constructs a concurrent kernel-level deadlock scenario:
1. Blocking sender threads: deep sleep wait in can_send() waiting for TX FIFO.
2. Error frame receiver: setsockopt(CAN_RAW_ERR_FILTER) to receive all error frames.
3. Socket churn: rapid open/bind/send/close.
4. Down/up interrupt loop: calls `ip link set can0 down` during active traffic.
"""

import os
import sys
import time
import socket
import struct
import threading
import subprocess

AF_CAN = getattr(socket, "AF_CAN", 29)
PF_CAN = AF_CAN
CAN_RAW = getattr(socket, "CAN_RAW", 1)
SOL_CAN_BASE = 100
CAN_RAW_FILTER = 1
CAN_RAW_ERR_FILTER = 2
CAN_RAW_LOOPBACK = 3
CAN_RAW_RECV_OWN_MSGS = 4
CAN_RAW_FD_FRAMES = 5

CAN_ERR_MASK = 0x1FFFFFFF

INTERFACE = "can0"
STOP_EVENT = threading.Event()
DEADLOCK_DETECTED = threading.Event()

def build_can_frame(can_id, data):
can_dlc = len(data)
data = data.ljust(8, b'\x00')
return struct.pack("=IB3x8s", can_id, can_dlc, data)

def build_canfd_frame(can_id, data, flags=0):
length = len(data)
data = data.ljust(64, b'\x00')
return struct.pack("=IBB2x64s", can_id, length, flags, data)

def sender_thread(thread_id, is_fd=False):
sock = socket.socket(PF_CAN, socket.SOCK_RAW, CAN_RAW)
try:
if is_fd:
try:
sock.setsockopt(SOL_CAN_BASE, CAN_RAW_FD_FRAMES, 1)
except Exception:
pass
sock.bind((INTERFACE,))
sock.setblocking(True)
except Exception:
sock.close()
return

frame_data = b'\x11\x22\x33\x44\x55\x66\x77\x88'
can_id = 0x100 + thread_id
frame = build_can_frame(can_id, frame_data)

while not STOP_EVENT.is_set():
try:
sock.send(frame)
except socket.error:
time.sleep(0.001)
except Exception:
break
sock.close()

def receiver_thread():
try:
sock = socket.socket(PF_CAN, socket.SOCK_RAW, CAN_RAW)
sock.setsockopt(SOL_CAN_BASE, CAN_RAW_ERR_FILTER, CAN_ERR_MASK)
sock.bind((INTERFACE,))
sock.settimeout(0.1)
except Exception:
return

while not STOP_EVENT.is_set():
try:
_ = sock.recv(72)
except socket.timeout:
continue
except Exception:
time.sleep(0.01)
sock.close()

def socket_churn_thread():
while not STOP_EVENT.is_set():
try:
s = socket.socket(PF_CAN, socket.SOCK_RAW, CAN_RAW)
s.bind((INTERFACE,))
s.setblocking(False)
frame = build_can_frame(0x200, b'\xaa\xbb\xcc\xdd')
try:
s.send(frame)
except Exception:
pass
s.close()
except Exception:
pass
time.sleep(0.002)

def main():
if os.geteuid() != 0:
print("[-] Please run as root: sudo python3 reproduce_can_deadlock.py")
sys.exit(1)

print("=" * 65)
print("Starting SocketCAN deadlock reproduction tool")
print("=" * 65)

subprocess.run(["ip", "link", "set", INTERFACE, "down"], stderr=subprocess.DEVNULL)
time.sleep(0.3)
subprocess.run(["ip", "link", "set", INTERFACE, "txqueuelen", "1"], check=False)
subprocess.run([
"ip", "link", "set", INTERFACE, "up", "type", "can",
"bitrate", "500000", "dbitrate", "500000", "fd", "on", "restart-ms", "100"
], check=False)

threads = []
for i in range(3):
t = threading.Thread(target=sender_thread, args=(i, i % 2 == 0), daemon=True)
t.start()
threads.append(t)

t_recv = threading.Thread(target=receiver_thread, daemon=True)
t_recv.start()
threads.append(t_recv)

t_churn = threading.Thread(target=socket_churn_thread, daemon=True)
t_churn.start()
threads.append(t_churn)

loop_count = 0
try:
while True:
loop_count += 1
now_str = time.strftime("%H:%M:%S")
sys.stdout.write(f"[{now_str}] Loop {loop_count:5d}: Bringing can0 down... ")
sys.stdout.flush()

t0 = time.time()
p_down = subprocess.Popen(["ip", "link", "set", INTERFACE, "down"],
stdout=subprocess.PIPE, stderr=subprocess.PIPE)
try:
p_down.communicate(timeout=4.0)
down_cost = time.time() - t0
sys.stdout.write(f"Done ({down_cost*1000:.1f}ms) -> Bringing can0 up... ")
sys.stdout.flush()
except subprocess.TimeoutExpired:
print("\n" + "!" * 65)
print("Captured deadlock! `ip link set can0 down` hung > 4s!")
print("Kernel rtnl_lock is deadlocked!")
print("!" * 65)
DEADLOCK_DETECTED.set()
break

t1 = time.time()
p_up = subprocess.Popen(["ip", "link", "set", INTERFACE, "up"],
stdout=subprocess.PIPE, stderr=subprocess.PIPE)
try:
p_up.communicate(timeout=4.0)
up_cost = time.time() - t1
sys.stdout.write(f"Done ({up_cost*1000:.1f}ms)\n")
sys.stdout.flush()
except subprocess.TimeoutExpired:
print("\n" + "!" * 65)
print("Captured deadlock! `ip link set can0 up` hung > 4s!")
print("!" * 65)
DEADLOCK_DETECTED.set()
break

time.sleep(0.005)

except KeyboardInterrupt:
print("\n[*] Exiting test...")
finally:
STOP_EVENT.set()
if not DEADLOCK_DETECTED.is_set():
subprocess.run(["ip", "link", "set", INTERFACE, "down"], stderr=subprocess.DEVNULL)
print(f"[*] Completed {loop_count} loops.")

if __name__ == "__main__":
main()
```

Also, if maintainers prefer this fix to be split into separate smaller
atomic patches (e.g. 1. close deadlock fix, 2. NAPI budget handling, 3. error
handling NULL check), please let me know and I will gladly send a v2 patch series!

Best regards,
Cheng Liu