Re: [PATCH 07/16] rcu: Save and restore irq flags in rcu_gp_cleanup()

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Wed, Jul 23, 2014 at 01:09:44AM -0400, Pranith Kumar wrote:
> We use raw_spin_lock_irqsave/restore() family of functions throughout the code
> but for two locations. This commit replaces raw_spin_lock_irq()/unlock_irq()
> with irqsave/restore() in one such location. This is not strictly necessary,
> so I did not change the other location. I will update the other location if
> this is accepted :)
> 
> This commit changes raw_spin_lock_irq()/unlock_irq() to lock_irqsave()/restore().
> 
> Signed-off-by: Pranith Kumar <bobby.prani@gmail.com>

I sympathize, as I used to take the approach that you are advocating.

The reason that I changed is that we -know- that interrupts are enabled
at this point in the code, so there is no point in incurring the extra
cognitive and machine overhead of the _irqsave() variant.  Plus the
current code has documentation benefits -- it tells you that the author
felt that irqs could not possible be disabled here.

So sorry, but no.

							Thanx, Paul

> ---
>  kernel/rcu/tree.c | 14 +++++++-------
>  1 file changed, 7 insertions(+), 7 deletions(-)
> 
> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> index b14cecd..5dcbf36 100644
> --- a/kernel/rcu/tree.c
> +++ b/kernel/rcu/tree.c
> @@ -1706,13 +1706,13 @@ static int rcu_gp_fqs(struct rcu_state *rsp, int fqs_state_in)
>   */
>  static void rcu_gp_cleanup(struct rcu_state *rsp)
>  {
> -	unsigned long gp_duration;
> +	unsigned long gp_duration, flags;
>  	bool needgp = false;
>  	int nocb = 0;
>  	struct rcu_data *rdp;
>  	struct rcu_node *rnp = rcu_get_root(rsp);
> 
> -	raw_spin_lock_irq(&rnp->lock);
> +	raw_spin_lock_irqsave(&rnp->lock, flags);
>  	smp_mb__after_unlock_lock();
>  	gp_duration = jiffies - rsp->gp_start;
>  	if (gp_duration > rsp->gp_max)
> @@ -1726,7 +1726,7 @@ static void rcu_gp_cleanup(struct rcu_state *rsp)
>  	 * safe for us to drop the lock in order to mark the grace
>  	 * period as completed in all of the rcu_node structures.
>  	 */
> -	raw_spin_unlock_irq(&rnp->lock);
> +	raw_spin_unlock_irqrestore(&rnp->lock, flags);
> 
>  	/*
>  	 * Propagate new ->completed value to rcu_node structures so
> @@ -1738,7 +1738,7 @@ static void rcu_gp_cleanup(struct rcu_state *rsp)
>  	 * grace period is recorded in any of the rcu_node structures.
>  	 */
>  	rcu_for_each_node_breadth_first(rsp, rnp) {
> -		raw_spin_lock_irq(&rnp->lock);
> +		raw_spin_lock_irqsave(&rnp->lock, flags);
>  		smp_mb__after_unlock_lock();
>  		ACCESS_ONCE(rnp->completed) = rsp->gpnum;
>  		rdp = this_cpu_ptr(rsp->rda);
> @@ -1746,11 +1746,11 @@ static void rcu_gp_cleanup(struct rcu_state *rsp)
>  			needgp = __note_gp_changes(rsp, rnp, rdp) || needgp;
>  		/* smp_mb() provided by prior unlock-lock pair. */
>  		nocb += rcu_future_gp_cleanup(rsp, rnp);
> -		raw_spin_unlock_irq(&rnp->lock);
> +		raw_spin_unlock_irqrestore(&rnp->lock, flags);
>  		cond_resched();
>  	}
>  	rnp = rcu_get_root(rsp);
> -	raw_spin_lock_irq(&rnp->lock);
> +	raw_spin_lock_irqsave(&rnp->lock, flags);
>  	smp_mb__after_unlock_lock(); /* Order GP before ->completed update. */
>  	rcu_nocb_gp_set(rnp, nocb);
> 
> @@ -1767,7 +1767,7 @@ static void rcu_gp_cleanup(struct rcu_state *rsp)
>  				       ACCESS_ONCE(rsp->gpnum),
>  				       TPS("newreq"));
>  	}
> -	raw_spin_unlock_irq(&rnp->lock);
> +	raw_spin_unlock_irqrestore(&rnp->lock, flags);
>  }
> 
>  /*
> -- 
> 2.0.0.rc2
>