Re: Q: lockdep_assert_held_read() after downgrade_write()

[Waiman Long]

On 01/31/2017 06:25 AM, Peter Zijlstra wrote:
> On Tue, Jan 31, 2017 at 11:36:20AM +0100, Peter Zijlstra wrote:
>> On Mon, Jan 30, 2017 at 02:30:45PM -0700, Jens Axboe wrote:
>>> I don't think you understand how it works. downgrade_write() turns a write
>>> lock into read held. To make that last sequence valid, you'd need:
>> Correct, and I'm surprised that didn't explode in different ways.
>>
>>> 	down_write(&rw);
>>> 	downgrade_write(&rw);
>>> 	lockdep_assert_held_read(&rw)
>>> 	up_read(&rw);
>>>
>>> or just not drop up_write() from the last section.
>> Right, but also, there seems to be a missing lockdep annotation to make
>> that work. That is, downgrade_write() doesn't have a lockdep annotation,
>> so it (lockdep) will still think its a write lock.
>>
>>
>> Let me try and fix both issues.
> Something like so I suppose,... completely untested.
>
> There could be a good reason for the current lockdep behaviour, but I
> cannot remember.
>
> ---
> diff --git a/kernel/locking/rwsem.c b/kernel/locking/rwsem.c
> index 45ba475d4be3..dfa9e40f83d5 100644
> --- a/kernel/locking/rwsem.c
> +++ b/kernel/locking/rwsem.c
> @@ -123,10 +123,9 @@ EXPORT_SYMBOL(up_write);
>   */
>  void downgrade_write(struct rw_semaphore *sem)
>  {
> -	/*
> -	 * lockdep: a downgraded write will live on as a write
> -	 * dependency.
> -	 */
> +	rwsem_release(&sem->dep_map, 1, _RET_IP_);
> +	rwsem_acquire_read(&sem->dep_map, 0, 0, _RET_IP_);
> +
>  	rwsem_set_reader_owned(sem);
>  	__downgrade_write(sem);
>  }
> diff --git a/kernel/locking/rwsem.h b/kernel/locking/rwsem.h
> index a699f4048ba1..3bd584c81b0b 100644
> --- a/kernel/locking/rwsem.h
> +++ b/kernel/locking/rwsem.h
> @@ -40,8 +40,10 @@ static inline void rwsem_set_reader_owned(struct rw_semaphore *sem)
>  	 * do a write to the rwsem cacheline when it is really necessary
>  	 * to minimize cacheline contention.
>  	 */
> -	if (sem->owner != RWSEM_READER_OWNED)
> +	if (sem->owner != RWSEM_READER_OWNED) {
> +		WARN_ON_ONCE(sem->owner != current);
>  		WRITE_ONCE(sem->owner, RWSEM_READER_OWNED);
> +	}
>  }
>  
>  static inline bool rwsem_owner_is_writer(struct task_struct *owner)

I don't think you can do a WARN_ON_ONCE() check for sem->owner !=
current here. If the rwsem starts from an unlock state, sem->owner will
be NULL and an incorrect warning message will be printed.

Cheers,
Longman