[PATCH 3.13, 3.14, 3.16, 3.17] rcu: Reduce rcu kthread wakeups

[Kamal Mostafa]

On Mon, 2014-10-13 at 12:29 -0400, Pranith Kumar wrote:
> Backport 2aa792e and parts of 48a7639 to 3.13 which is Ubuntu LTS kernel.
> 
> This commit reduces the number of rcu kthread wakeups. Tested for over a week on
> my desktop 14.04 system.
> 
> Signed-off-by: Pranith Kumar <bobby.prani at gmail.com>
> CC: Paul E. McKenney <paulmck at linux.vnet.ibm.com>


Hi Pranith and Paul-

This looks good, but how about lets split it back into two commits (the
second a clean cherry-pick) and submit it for 3.{14,16,17}-stable also:

3.13 and 3.14 could apply the attached 48a7639 mini-backport (just
supplying rcu_gp_kthread_wake), and then cherry-pick 2aa792e.

3.16 and 3.17 carry 48a7639 already, so could just cherry-pick 2aa792e.

 -Kamal


>  kernel/rcu/tree.c | 22 +++++++++++++++++++---
>  1 file changed, 19 insertions(+), 3 deletions(-)
> 
> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> index dd08198..9b3beea 100644
> --- a/kernel/rcu/tree.c
> +++ b/kernel/rcu/tree.c
> @@ -1193,6 +1193,22 @@ static int rcu_future_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp)
>  }
>  
>  /*
> + * Awaken the grace-period kthread for the specified flavor of RCU.
> + * Don't do a self-awaken, and don't bother awakening when there is
> + * nothing for the grace-period kthread to do (as in several CPUs
> + * raced to awaken, and we lost), and finally don't try to awaken
> + * a kthread that has not yet been created.
> + */
> +static void rcu_gp_kthread_wake(struct rcu_state *rsp)
> +{
> +	if (current == rsp->gp_kthread ||
> +	    !ACCESS_ONCE(rsp->gp_flags) ||
> +	    !rsp->gp_kthread)
> +		return;
> +	wake_up(&rsp->gp_wq);
> +}
> +
> +/*
>   * If there is room, assign a ->completed number to any callbacks on
>   * this CPU that have not already been assigned.  Also accelerate any
>   * callbacks that were previously assigned a ->completed number that has
> @@ -1625,7 +1641,7 @@ static void rsp_wakeup(struct irq_work *work)
>  	struct rcu_state *rsp = container_of(work, struct rcu_state, wakeup_work);
>  
>  	/* Wake up rcu_gp_kthread() to start the grace period. */
> -	wake_up(&rsp->gp_wq);
> +	rcu_gp_kthread_wake(rsp);
>  }
>  
>  /*
> @@ -1701,7 +1717,7 @@ static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags)
>  {
>  	WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
>  	raw_spin_unlock_irqrestore(&rcu_get_root(rsp)->lock, flags);
> -	wake_up(&rsp->gp_wq);  /* Memory barrier implied by wake_up() path. */
> +	rcu_gp_kthread_wake(rsp);
>  }
>  
>  /*
> @@ -2271,7 +2287,7 @@ static void force_quiescent_state(struct rcu_state *rsp)
>  	}
>  	rsp->gp_flags |= RCU_GP_FLAG_FQS;
>  	raw_spin_unlock_irqrestore(&rnp_old->lock, flags);
> -	wake_up(&rsp->gp_wq);  /* Memory barrier implied by wake_up() path. */
> +	rcu_gp_kthread_wake(rsp);
>  }
>  
>  /*

-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-rcu-Make-callers-awaken-grace-period-kthread.patch
Type: text/x-patch
Size: 4508 bytes
Desc: not available
URL: <https://lists.ubuntu.com/archives/kernel-team/attachments/20141013/a568c6e1/attachment.bin>