[PATCH V2] kernel: locking: rtmutex: Fix a possible sleep-in-atomic-context bug in rt_mutex_handle_deadlock()

[PATCH V2] kernel: locking: rtmutex: Fix a possible sleep-in-atomic-context bug in rt_mutex_handle_deadlock()

[Jia-Ju Bai]

The driver may sleep with holding a spinlock.

The function call paths (from bottom to top) in Linux-4.16 are:

[FUNC] schedule
kernel/locking/rtmutex.c, 1223:
	schedule in rt_mutex_handle_deadlock
kernel/locking/rtmutex.c, 1273:
	rt_mutex_handle_deadlock in rt_mutex_slowlock
kernel/locking/rtmutex.c, 1249:
	_raw_spin_lock_irqsave in rt_mutex_slowlock

To fix the bug, the spinlock is released before the loop of schedule()
This is found by my static analysis tool (DSAC).

Signed-off-by: Jia-Ju Bai <baijiaju1990@gmail.com>
---
v2:
* Release the spinlock before the loop, instead of v1 that releases the
  spinlock before schedule() and then acquires the spinlock again.
  Thank Steven for good advice.
---
 kernel/locking/rtmutex.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/kernel/locking/rtmutex.c b/kernel/locking/rtmutex.c
index 2823d4163a37..8f25a289abe8 100644
--- a/kernel/locking/rtmutex.c
+++ b/kernel/locking/rtmutex.c
@@ -1205,7 +1205,7 @@ __rt_mutex_slowlock(struct rt_mutex *lock, int state,
 }
 
 static void rt_mutex_handle_deadlock(int res, int detect_deadlock,
-				     struct rt_mutex_waiter *w)
+				     struct rt_mutex_waiter *w, struct rt_mutex *lock)
 {
 	/*
 	 * If the result is not -EDEADLOCK or the caller requested
@@ -1218,6 +1218,8 @@ static void rt_mutex_handle_deadlock(int res, int detect_deadlock,
 	 * Yell lowdly and stop the task right here.
 	 */
 	rt_mutex_print_deadlock(w);
+	/* We're not going anywhere, release the wait_lock */
+	raw_spin_unlock_irq(&lock->wait_lock);
 	while (1) {
 		set_current_state(TASK_INTERRUPTIBLE);
 		schedule();
@@ -1269,7 +1271,7 @@ rt_mutex_slowlock(struct rt_mutex *lock, int state,
 	if (unlikely(ret)) {
 		__set_current_state(TASK_RUNNING);
 		remove_waiter(lock, &waiter);
-		rt_mutex_handle_deadlock(ret, chwalk, &waiter);
+		rt_mutex_handle_deadlock(ret, chwalk, &waiter, lock);
 	}
 
 	/*
-- 
2.17.0

Re: [PATCH V2] kernel: locking: rtmutex: Fix a possible sleep-in-atomic-context bug in rt_mutex_handle_deadlock()

[Peter Zijlstra]

On Sat, Aug 11, 2018 at 11:00:37AM +0800, Jia-Ju Bai wrote:

You forgot to Cc the person who wrote this code...

>  kernel/locking/rtmutex.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/locking/rtmutex.c b/kernel/locking/rtmutex.c
> index 2823d4163a37..8f25a289abe8 100644
> --- a/kernel/locking/rtmutex.c
> +++ b/kernel/locking/rtmutex.c
> @@ -1205,7 +1205,7 @@ __rt_mutex_slowlock(struct rt_mutex *lock, int state,
>  }
>  
>  static void rt_mutex_handle_deadlock(int res, int detect_deadlock,
> -				     struct rt_mutex_waiter *w)
> +				     struct rt_mutex_waiter *w, struct rt_mutex *lock)
>  {
>  	/*
>  	 * If the result is not -EDEADLOCK or the caller requested
> @@ -1218,6 +1218,8 @@ static void rt_mutex_handle_deadlock(int res, int detect_deadlock,
>  	 * Yell lowdly and stop the task right here.
>  	 */
>  	rt_mutex_print_deadlock(w);
> +	/* We're not going anywhere, release the wait_lock */
> +	raw_spin_unlock_irq(&lock->wait_lock);
>  	while (1) {
>  		set_current_state(TASK_INTERRUPTIBLE);
>  		schedule();
> @@ -1269,7 +1271,7 @@ rt_mutex_slowlock(struct rt_mutex *lock, int state,
>  	if (unlikely(ret)) {
>  		__set_current_state(TASK_RUNNING);
>  		remove_waiter(lock, &waiter);
> -		rt_mutex_handle_deadlock(ret, chwalk, &waiter);
> +		rt_mutex_handle_deadlock(ret, chwalk, &waiter, lock);
>  	}

The patch is correct; but I can't find myself liking it very much. This
dinly little single use function is growing a lot of arguments.

The alternative is something like the below; not sure myself though.
Thomas?


diff --git a/kernel/locking/rtmutex.c b/kernel/locking/rtmutex.c
index 2823d4163a37..a44d4034e232 100644
--- a/kernel/locking/rtmutex.c
+++ b/kernel/locking/rtmutex.c
@@ -1204,18 +1204,10 @@ __rt_mutex_slowlock(struct rt_mutex *lock, int state,
 	return ret;
 }
 
-static void rt_mutex_handle_deadlock(int res, int detect_deadlock,
-				     struct rt_mutex_waiter *w)
+static void rt_mutex_handle_deadlock(struct rt_mutex_waiter *w)
 {
 	/*
-	 * If the result is not -EDEADLOCK or the caller requested
-	 * deadlock detection, nothing to do here.
-	 */
-	if (res != -EDEADLOCK || detect_deadlock)
-		return;
-
-	/*
-	 * Yell lowdly and stop the task right here.
+	 * Yell loudly and stop the task right here.
 	 */
 	rt_mutex_print_deadlock(w);
 	while (1) {
@@ -1269,7 +1261,10 @@ rt_mutex_slowlock(struct rt_mutex *lock, int state,
 	if (unlikely(ret)) {
 		__set_current_state(TASK_RUNNING);
 		remove_waiter(lock, &waiter);
-		rt_mutex_handle_deadlock(ret, chwalk, &waiter);
+		if (chwalk == RT_MUTEX_MIN_CHAINWALK && ret == -EDEADLOCK) {
+			raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
+			rt_mutex_handle_deadlock(&waiter);
+		}
 	}
 
 	/*