Skip to contentSkip to conversation

Fix spinlock issue in net/core/sock.c

PATCH In review Vivek Satpute 2 participants 4 messages Jun 25, 2009 – Dec 27, 2009 1 message not in the archive

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Fix spinlock issue in net/core/sock.c

Kernel panic's and reboot while doing network operations such ifconfig

and ping on MIPS architecture after RT-patches applied.

In case of CONFIG_PREEMPT_RT, after releasing the lock with "spin_unlock", context switch might occur before enabling the bottom half with local_bh_enable and this causes the kernel to panic. The issue is resolved by releasing the lock afer acquiring the mutex using spin_unlock_bh.

Tested the fix on MIPS and X86 architecture.

---
net/core/sock.c | 8 ++++++++

1 file changed, 8 insertions(+)
net/core/sock.c +8 -0
Index: linux-2.6.28.4/net/core/sock.c
===================================================================
@@ -1752,12 +1752,20 @@ void lock_sock_nested(struct sock *sk, i
 	if (sk->sk_lock.owned)
 		__lock_sock(sk);
 	sk->sk_lock.owned = 1;
+
+#ifndef CONFIG_PREEMPT_RT
 	spin_unlock(&sk->sk_lock.slock);
+#endif
 	/*
 	 * The sk_lock has mutex_lock() semantics here:
 	 */
 	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
+
+#ifdef CONFIG_PREEMPT_RT
+	spin_unlock_bh(&sk->sk_lock.slock);
+#else
 	local_bh_enable();
+#endif
 }
 
 EXPORT_SYMBOL(lock_sock_nested);

Thanks and Regards,
Vivek Satpute
System Software Engineer
LinSysSoft Technologies, Pune

#1 lore raw
Quoted 30 lines · Vivek Satpute
Kernel panic's and reboot while doing network operations such ifconfig

and ping on MIPS architecture after RT-patches applied.

In case of CONFIG_PREEMPT_RT, after releasing the lock with "spin_unlock", context switch might occur before enabling the bottom half with local_bh_enable and this causes the kernel to panic. The issue is resolved by releasing the lock afer acquiring the mutex using spin_unlock_bh.

Tested the fix on MIPS and X86 architecture.

---
net/core/sock.c |    8 ++++++++

1 file changed, 8 insertions(+)

Index: linux-2.6.28.4/net/core/sock.c
===================================================================

net/core/sock.c +3 -0
@@ -1752,12 +1752,20 @@ void lock_sock_nested(struct sock *sk, i
        if (sk->sk_lock.owned)
                __lock_sock(sk);
        sk->sk_lock.owned = 1;
+
+#ifndef CONFIG_PREEMPT_RT
        spin_unlock(&sk->sk_lock.slock);
+#endif
        /*
         * The sk_lock has mutex_lock() semantics here:

net/core/sock.c void lock_sock_nested(struct sock *sk, i · line 1770

Quoted 10 lines
net/core/sock.c +5 -0
         */
        mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
+
+#ifdef CONFIG_PREEMPT_RT
+       spin_unlock_bh(&sk->sk_lock.slock);
+#else
        local_bh_enable();
+#endif
  }
 

I don't think you should move the unlock after the mutex_acquire(). I guess that this will lead to false positive lockdep warnings.

Anyway, I wonder why using spin_unlock_bh() is fixing the problem that you are having. Do you have more context about the problem or maybe an Oops or so?

Thanks,
Jan

Quoted 10 lines
net/core/sock.c +0 -0
  EXPORT_SYMBOL(lock_sock_nested);

Thanks and Regards,
Vivek Satpute
System Software Engineer
LinSysSoft Technologies, Pune

Quoted 5 lines

--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

A message that is not in the archive (sent off-list or not yet received): [email protected]

#3 lore raw

Hi Jan,

(Sending mail again as first mail to mailing-list got bounced due to Rich Text Format)

Quoted 46 lines · Jan Blunck
Quoted 30 lines · Vivek Satpute
Kernel panic's and reboot while doing network operations such ifconfig

and ping on MIPS architecture after RT-patches applied.

In case of CONFIG_PREEMPT_RT, after releasing the lock with "spin_unlock", context switch might occur before enabling the bottom half with local_bh_enable and this causes the kernel to panic. The issue is resolved by releasing the lock afer acquiring the mutex using spin_unlock_bh.

Tested the fix on MIPS and X86 architecture.

---
net/core/sock.c |    8 ++++++++

1 file changed, 8 insertions(+)

Index: linux-2.6.28.4/net/core/sock.c
===================================================================

net/core/sock.c +3 -0
@@ -1752,12 +1752,20 @@ void lock_sock_nested(struct sock *sk, i
        if (sk->sk_lock.owned)
                __lock_sock(sk);
        sk->sk_lock.owned = 1;
+
+#ifndef CONFIG_PREEMPT_RT
        spin_unlock(&sk->sk_lock.slock);
+#endif
        /*
         * The sk_lock has mutex_lock() semantics here:

net/core/sock.c void lock_sock_nested(struct sock *sk, i · line 1770

Quoted 10 lines
net/core/sock.c +5 -0
         */
        mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
+
+#ifdef CONFIG_PREEMPT_RT
+       spin_unlock_bh(&sk->sk_lock.slock);
+#else
        local_bh_enable();
+#endif
  }
 

I don't think you should move the unlock after the mutex_acquire(). I guess that this will lead to false positive lockdep warnings.

Quoted 3 lines

Anyway, I wonder why using spin_unlock_bh() is fixing the problem that you are having. Do you have more context about the problem or maybe an Oops or so?

On firing command "ifconfig", I get following messages:

--------------------------------<snip>--------------------------------

# ifconfig
Kernel panic - not syncing: Aiee, killing interrupt handler!
Rebooting in 1 seconds..
--------------------------------<snip>--------------------------------

If kernel compiled with Lock Debugging options then it gives following call trace:

--------------------------------<snip>--------------------------------

Call Trace:
[<80111768>] dump_stack+0x8/0x34
[<8012e9a0>] warn_on_slowpath+0x60/0x88
[<80135b48>] local_bh_enable+0x40/0x11c
[<8035d05c>] lock_sock_nested+0xf8/0x11c
[<803bbba8>] inet_bind+0x100/0x210
[<803d7888>] xs_bind4+0x70/0x158
[<803d9ac0>] xs_udp_connect_worker4+0x120/0x1a4
[<801440f8>] run_workqueue+0x1d0/0x264
[<801450ac>] worker_thread+0x7c/0xec
[<80148f70>] kthread+0x58/0x94
[<8010cc2c>] kernel_thread_helper+0x10/0x18

--------------------------------<snip>--------------------------------

Quoted 19 lines

Thanks,
Jan

Quoted 15 lines
net/core/sock.c +0 -0
  EXPORT_SYMBOL(lock_sock_nested);

Thanks and Regards,
Vivek Satpute
System Software Engineer
LinSysSoft Technologies, Pune

--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Thanks and Regards,
Vivek Satpute
System Software Engineer
LinSysSoft Technologies, Pune
--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#4 lore raw
In reply to Vivek Satpute

Hi,

I want to keep this thread alive. Has anyone encountered/tested this issue ?

Thanks and Regards,
Vivek Satpute
System Software Engineer
LinSysSoft Technologies, Pune

Quoted 50 lines · Vivek Satpute
Kernel panic's and reboot while doing network operations such ifconfig

and ping on MIPS architecture after RT-patches applied.

In case of CONFIG_PREEMPT_RT, after releasing the lock with "spin_unlock", context switch might occur before enabling the bottom half with local_bh_enable and this causes the kernel to panic. The issue is resolved by releasing the lock afer acquiring the mutex using spin_unlock_bh.

Tested the fix on MIPS and X86 architecture.

---
net/core/sock.c | 8 ++++++++

1 file changed, 8 insertions(+)

Index: linux-2.6.28.4/net/core/sock.c
===================================================================

net/core/sock.c +8 -0
@@ -1752,12 +1752,20 @@ void lock_sock_nested(struct sock *sk, i
 	if (sk->sk_lock.owned)
 		__lock_sock(sk);
 	sk->sk_lock.owned = 1;
+
+#ifndef CONFIG_PREEMPT_RT
 	spin_unlock(&sk->sk_lock.slock);
+#endif
 	/*
 	 * The sk_lock has mutex_lock() semantics here:
 	 */
 	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
+
+#ifdef CONFIG_PREEMPT_RT
+	spin_unlock_bh(&sk->sk_lock.slock);
+#else
 	local_bh_enable();
+#endif
 }
 
 EXPORT_SYMBOL(lock_sock_nested);

Thanks and Regards,
Vivek Satpute
System Software Engineer
LinSysSoft Technologies, Pune