Re: [PATCH] locking/rwsem: Disable preemption while trying for rwsem lock

[Peter Zijlstra <peterz@infradead.org>]

On Thu, Sep 01, 2022 at 03:58:10PM +0530, Mukesh Ojha wrote:
> From: Gokul krishna Krishnakumar <quic_gokukris@quicinc.com>
> 
> Make the region inside the rwsem_write_trylock non preemptible.
> 
> We observe RT task is hogging CPU when trying to acquire rwsem lock
> which was acquired by a kworker task but before the rwsem owner was set.
> 
> Here is the scenario:
> 1. CFS task (affined to a particular CPU) takes rwsem lock.
> 
> 2. CFS task gets preempted by a RT task before setting owner.
> 
> 3. RT task (FIFO) is trying to acquire the lock, but spinning until
> RT throttling happens for the lock as the lock was taken by CFS task.
> 
> This patch attempts to fix the above issue by disabling preemption
> until owner is set for the lock. while at it also fix this issue
> at the place where owner being set/cleared.
> 
> Signed-off-by: Gokul krishna Krishnakumar <quic_gokukris@quicinc.com>
> Signed-off-by: Mukesh Ojha <quic_mojha@quicinc.com>

This is not a valid SoB chain.

> ---
>  kernel/locking/rwsem.c | 18 ++++++++++++++----
>  1 file changed, 14 insertions(+), 4 deletions(-)
> 
> diff --git a/kernel/locking/rwsem.c b/kernel/locking/rwsem.c
> index 65f0262..3b4b32e 100644
> --- a/kernel/locking/rwsem.c
> +++ b/kernel/locking/rwsem.c
> @@ -251,13 +251,16 @@ static inline bool rwsem_read_trylock(struct rw_semaphore *sem, long *cntp)
>  static inline bool rwsem_write_trylock(struct rw_semaphore *sem)
>  {
>  	long tmp = RWSEM_UNLOCKED_VALUE;
> +	bool ret = false;
>  
> +	preempt_disable();
>  	if (atomic_long_try_cmpxchg_acquire(&sem->count, &tmp, RWSEM_WRITER_LOCKED)) {
>  		rwsem_set_owner(sem);
> -		return true;
> +		ret = true;
>  	}
>  
> -	return false;
> +	preempt_enable();
> +	return ret;
>  }
>  
>  /*

Yes, this part looks ok.

> @@ -686,16 +689,21 @@ enum owner_state {
>  static inline bool rwsem_try_write_lock_unqueued(struct rw_semaphore *sem)
>  {
>  	long count = atomic_long_read(&sem->count);
> +	bool ret = false;
>  
> +	preempt_disable();
>  	while (!(count & (RWSEM_LOCK_MASK|RWSEM_FLAG_HANDOFF))) {
>  		if (atomic_long_try_cmpxchg_acquire(&sem->count, &count,
>  					count | RWSEM_WRITER_LOCKED)) {
>  			rwsem_set_owner(sem);
>  			lockevent_inc(rwsem_opt_lock);
> -			return true;
> +			ret = true;
> +			break;
>  		}
>  	}
> -	return false;
> +
> +	preempt_enable();
> +	return ret;
>  }
>  

This one I can't follow; afaict this is only called with preemption
already disabled.

>  static inline bool rwsem_can_spin_on_owner(struct rw_semaphore *sem)
> @@ -1352,8 +1360,10 @@ static inline void __up_write(struct rw_semaphore *sem)
>  	DEBUG_RWSEMS_WARN_ON((rwsem_owner(sem) != current) &&
>  			    !rwsem_test_oflags(sem, RWSEM_NONSPINNABLE), sem);
>  
> +	preempt_disable();
>  	rwsem_clear_owner(sem);
>  	tmp = atomic_long_fetch_add_release(-RWSEM_WRITER_LOCKED, &sem->count);
> +	preempt_enable();
>  	if (unlikely(tmp & RWSEM_FLAG_WAITERS))
>  		rwsem_wake(sem);
>  }

Yep, that looks good again.

Perhaps the thing to do would be to add:

  lockdep_assert_preemption_disabled()

to rwsem_{set,clear}_owner() and expand the comment there to explain
that these functions should be in the same preempt-disable section as
the atomic op that changes sem->count.