[PATCH v2] locking/mutex: Fix the handoff mechanism doesn't take effect

[PATCH v2] locking/mutex: Fix the handoff mechanism doesn't take effect

[Yanfei Xu]

Commit e274795ea7b7 ("locking/mutex: Fix mutex handoff") removes the
judgment of "handoff" in __mutex_trylock_or_owner() as blow, it makes
anyone can clear MUTEX_FLAG_HANDOFF bit when it gets the lock, even it
is the stealing lock. That makes set of MUTEX_FLAG_HANDOFF by the
top-waiter in vain.

-               if (handoff)
-                       flags &= ~MUTEX_FLAG_HANDOFF;
+               flags &= ~MUTEX_FLAG_HANDOFF;

We could fix it by setting MUTEX_FLAG_HANDOFF bit before the top-waiter
in wait_list falls asleep, then It must can grab the lock after being
woken up. Instead of probably being stolen lock by a optimistic spinner,
and being cleared MUTEX_FLAG_HANDOFF bit by the task which stole the lock,
and probably fall to sleep again without MUTEX_FLAG_HANDOFF due to the
task which stole the lock falls asleep.

Note: there still is a very small window that the top-waiter can't get
the lock after being awoken because no MUTEX_FLAG_HANDOFF bit is observed
in unlock path and then wake up the top-waiter. But it doesn't matter,
the top-waiter will optimistically spin on the lock or fall asleep with
MUTEX_FLAG_HANDOFF bit again.

Also correct a obsolete comment in __mutex_trylock_or_owner().

Fixes: e274795ea7b7 ("locking/mutex: Fix mutex handoff")
Suggested-by: Waiman Long <longman@redhat.com>
Signed-off-by: Yanfei Xu <yanfei.xu@windriver.com>
---
v1->v2:
1. Bring the assignment of "first" variable to the front of
   schedule_preempt_disabled() to make the top-waiter can grab the
   lock when it wakes up for the first time.
2. Correct the comments in __mutex_trylock_or_owner by Waiman.
3. Rename this patch name form "locking/mutex: fix the 
   MUTEX_FLAG_HANDOFF bit is cleared unexpected" to "locking/mutex: Fix
   the handoff mechanism doesn't take effect"

 kernel/locking/mutex.c | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/kernel/locking/mutex.c b/kernel/locking/mutex.c
index 013e1b08a1bf..ba36d93e65e8 100644
--- a/kernel/locking/mutex.c
+++ b/kernel/locking/mutex.c
@@ -118,9 +118,9 @@ static inline struct task_struct *__mutex_trylock_or_owner(struct mutex *lock)
 		}
 
 		/*
-		 * We set the HANDOFF bit, we must make sure it doesn't live
-		 * past the point where we acquire it. This would be possible
-		 * if we (accidentally) set the bit on an unlocked mutex.
+		 * Always clear the HANDOFF bit before acquiring the lock.
+		 * Note that if the bit is accidentally set on an unlocked
+		 * mutex, anyone can acquire it.
 		 */
 		flags &= ~MUTEX_FLAG_HANDOFF;
 
@@ -1033,17 +1033,17 @@ __mutex_lock_common(struct mutex *lock, long state, unsigned int subclass,
 		}
 
 		spin_unlock(&lock->wait_lock);
-		schedule_preempt_disabled();
 
 		/*
 		 * ww_mutex needs to always recheck its position since its waiter
 		 * list is not FIFO ordered.
 		 */
-		if (ww_ctx || !first) {
+		if (ww_ctx || !first)
 			first = __mutex_waiter_is_first(lock, &waiter);
-			if (first)
-				__mutex_set_flag(lock, MUTEX_FLAG_HANDOFF);
-		}
+		if (first)
+			__mutex_set_flag(lock, MUTEX_FLAG_HANDOFF);
+
+		schedule_preempt_disabled();
 
 		set_current_state(state);
 		/*
-- 
2.27.0

Re: [PATCH v2] locking/mutex: Fix the handoff mechanism doesn't take effect

[Xu, Yanfei]

Please ignore this.

Thanks,
Yanfei

On 6/30/21 1:42 PM, Yanfei Xu wrote:
> Commit e274795ea7b7 ("locking/mutex: Fix mutex handoff") removes the
> judgment of "handoff" in __mutex_trylock_or_owner() as blow, it makes
> anyone can clear MUTEX_FLAG_HANDOFF bit when it gets the lock, even it
> is the stealing lock. That makes set of MUTEX_FLAG_HANDOFF by the
> top-waiter in vain.
> 
> -               if (handoff)
> -                       flags &= ~MUTEX_FLAG_HANDOFF;
> +               flags &= ~MUTEX_FLAG_HANDOFF;
> 
> We could fix it by setting MUTEX_FLAG_HANDOFF bit before the top-waiter
> in wait_list falls asleep, then It must can grab the lock after being
> woken up. Instead of probably being stolen lock by a optimistic spinner,
> and being cleared MUTEX_FLAG_HANDOFF bit by the task which stole the lock,
> and probably fall to sleep again without MUTEX_FLAG_HANDOFF due to the
> task which stole the lock falls asleep.
> 
> Note: there still is a very small window that the top-waiter can't get
> the lock after being awoken because no MUTEX_FLAG_HANDOFF bit is observed
> in unlock path and then wake up the top-waiter. But it doesn't matter,
> the top-waiter will optimistically spin on the lock or fall asleep with
> MUTEX_FLAG_HANDOFF bit again.
> 
> Also correct a obsolete comment in __mutex_trylock_or_owner().
> 
> Fixes: e274795ea7b7 ("locking/mutex: Fix mutex handoff")
> Suggested-by: Waiman Long <longman@redhat.com>
> Signed-off-by: Yanfei Xu <yanfei.xu@windriver.com>
> ---
> v1->v2:
> 1. Bring the assignment of "first" variable to the front of
>     schedule_preempt_disabled() to make the top-waiter can grab the
>     lock when it wakes up for the first time.
> 2. Correct the comments in __mutex_trylock_or_owner by Waiman.
> 3. Rename this patch name form "locking/mutex: fix the
>     MUTEX_FLAG_HANDOFF bit is cleared unexpected" to "locking/mutex: Fix
>     the handoff mechanism doesn't take effect"
> 
>   kernel/locking/mutex.c | 16 ++++++++--------
>   1 file changed, 8 insertions(+), 8 deletions(-)
> 
> diff --git a/kernel/locking/mutex.c b/kernel/locking/mutex.c
> index 013e1b08a1bf..ba36d93e65e8 100644
> --- a/kernel/locking/mutex.c
> +++ b/kernel/locking/mutex.c
> @@ -118,9 +118,9 @@ static inline struct task_struct *__mutex_trylock_or_owner(struct mutex *lock)
>   		}
>   
>   		/*
> -		 * We set the HANDOFF bit, we must make sure it doesn't live
> -		 * past the point where we acquire it. This would be possible
> -		 * if we (accidentally) set the bit on an unlocked mutex.
> +		 * Always clear the HANDOFF bit before acquiring the lock.
> +		 * Note that if the bit is accidentally set on an unlocked
> +		 * mutex, anyone can acquire it.
>   		 */
>   		flags &= ~MUTEX_FLAG_HANDOFF;
>   
> @@ -1033,17 +1033,17 @@ __mutex_lock_common(struct mutex *lock, long state, unsigned int subclass,
>   		}
>   
>   		spin_unlock(&lock->wait_lock);
> -		schedule_preempt_disabled();
>   
>   		/*
>   		 * ww_mutex needs to always recheck its position since its waiter
>   		 * list is not FIFO ordered.
>   		 */
> -		if (ww_ctx || !first) {
> +		if (ww_ctx || !first)
>   			first = __mutex_waiter_is_first(lock, &waiter);
> -			if (first)
> -				__mutex_set_flag(lock, MUTEX_FLAG_HANDOFF);
> -		}
> +		if (first)
> +			__mutex_set_flag(lock, MUTEX_FLAG_HANDOFF);
> +
> +		schedule_preempt_disabled();
>   
>   		set_current_state(state);
>   		/*
>