Re: [PATCH v2 07/13] rwsem: use cmpxchg for trying to steal write lock

[Peter Hurley <peter@hurleysoftware.com>]

On Fri, 2013-03-15 at 03:54 -0700, Michel Lespinasse wrote:
> Using rwsem_atomic_update to try stealing the write lock forced us to
> undo the adjustment in the failure path. We can have simpler and faster
> code by using cmpxchg instead.
> 
> Signed-off-by: Michel Lespinasse <walken@google.com>
> 
> ---
>  lib/rwsem.c | 26 ++++++--------------------
>  1 file changed, 6 insertions(+), 20 deletions(-)
> 
> diff --git a/lib/rwsem.c b/lib/rwsem.c
> index 2360bf204098..64c2dc007be2 100644
> --- a/lib/rwsem.c
> +++ b/lib/rwsem.c
> @@ -142,25 +142,6 @@ __rwsem_do_wake(struct rw_semaphore *sem, int wake_type)
>  	return sem;
>  }
>  
> -/* Try to get write sem, caller holds sem->wait_lock: */
> -static int try_get_writer_sem(struct rw_semaphore *sem)
> -{
> -	long oldcount, adjustment;
> -
> -	adjustment = RWSEM_ACTIVE_WRITE_BIAS;
> -	if (list_is_singular(&sem->wait_list))
> -		adjustment -= RWSEM_WAITING_BIAS;
> -
> -try_again_write:
> -	oldcount = rwsem_atomic_update(adjustment, sem) - adjustment;
> -	if (!(oldcount & RWSEM_ACTIVE_MASK))
> -		return 1;
> -	/* some one grabbed the sem already */
> -	if (rwsem_atomic_update(-adjustment, sem) & RWSEM_ACTIVE_MASK)
> -		return 0;
> -	goto try_again_write;
> -}
> -
>  /*
>   * wait for the read lock to be granted
>   */
> @@ -236,7 +217,12 @@ struct rw_semaphore __sched *rwsem_down_write_failed(struct rw_semaphore *sem)
>  	while (true) {
>  		set_task_state(tsk, TASK_UNINTERRUPTIBLE);
>  
> -		if (try_get_writer_sem(sem))
> +		/* Try acquiring the write lock. */
> +		count = RWSEM_ACTIVE_WRITE_BIAS;
> +		if (!list_is_singular(&sem->wait_list))
> +			count += RWSEM_WAITING_BIAS;
> +		if (cmpxchg(&sem->count, RWSEM_WAITING_BIAS, count) ==
> +							RWSEM_WAITING_BIAS)

This is just my opinion but I think this is way more readable if this
stays in try_get_writer_sem() and you change that to static inline
rather than inlining here. Especially because 9/13 adds a scope.

>  			break;
>  
>  		raw_spin_unlock_irq(&sem->wait_lock);