2018-01-18 10:35:23

by Liu, Changcheng

Subject: [PATCH] rcu: refine structure rcu_node field for rcu boost

Do not allocate space for rcu boost field when
RCU BOOST is not configured.

Signed-off-by: Liu Changcheng <[email protected]>

diff --git a/kernel/rcu/tree.h b/kernel/rcu/tree.h
index 46a5d19..88f087e 100644
--- a/kernel/rcu/tree.h
+++ b/kernel/rcu/tree.h
@@ -129,6 +129,7 @@ struct rcu_node {
/* if there is no such task. If there */
/* is no current expedited grace period, */
/* then there can cannot be any such task. */
+#ifdef CONFIG_RCU_BOOST
struct list_head *boost_tasks;
/* Pointer to first task that needs to be */
/* priority boosted, or NULL if no priority */
@@ -153,6 +154,8 @@ struct rcu_node {
/* Number of tasks boosted for expedited GP. */
unsigned long n_normal_boosts;
/* Number of tasks boosted for normal GP. */
+#endif/* #ifdef CONFIG_RCU_BOOST*/
+
#ifdef CONFIG_RCU_NOCB_CPU
struct swait_queue_head nocb_gp_wq[2];
/* Place for rcu_nocb_kthread() to wait GP. */
diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
index db85ca3..fee0b1e 100644
--- a/kernel/rcu/tree_plugin.h
+++ b/kernel/rcu/tree_plugin.h
@@ -506,8 +506,10 @@ void rcu_read_unlock_special(struct task_struct *t)
if (IS_ENABLED(CONFIG_RCU_BOOST)) {
/* Snapshot ->boost_mtx ownership w/rnp->lock held. */
drop_boost_mutex = rt_mutex_owner(&rnp->boost_mtx) == t;
+#ifdef CONFIG_RCU_BOOST
if (&t->rcu_node_entry == rnp->boost_tasks)
rnp->boost_tasks = np;
+#endif
}

/*
--
2.7.4

2018-01-18 17:40:26

by Paul E. McKenney

Subject: Re: [PATCH] rcu: refine structure rcu_node field for rcu boost

On Thu, Jan 18, 2018 at 06:33:43PM +0800, Liu, Changcheng wrote:
> Do not allocate space for rcu boost field when
> RCU BOOST is not configured.
>
> Signed-off-by: Liu Changcheng <[email protected]>

The added #ifdef in rcu_read_unlock_special() is a deal-breaker.
Just out of curiosity, is this decrease in storage measurable at
the system level?

Thanx, Paul

> diff --git a/kernel/rcu/tree.h b/kernel/rcu/tree.h
> index 46a5d19..88f087e 100644
> --- a/kernel/rcu/tree.h
> +++ b/kernel/rcu/tree.h
> @@ -129,6 +129,7 @@ struct rcu_node {
> /* if there is no such task. If there */
> /* is no current expedited grace period, */
> /* then there can cannot be any such task. */
> +#ifdef CONFIG_RCU_BOOST
> struct list_head *boost_tasks;
> /* Pointer to first task that needs to be */
> /* priority boosted, or NULL if no priority */
> @@ -153,6 +154,8 @@ struct rcu_node {
> /* Number of tasks boosted for expedited GP. */
> unsigned long n_normal_boosts;
> /* Number of tasks boosted for normal GP. */
> +#endif/* #ifdef CONFIG_RCU_BOOST*/
> +
> #ifdef CONFIG_RCU_NOCB_CPU
> struct swait_queue_head nocb_gp_wq[2];
> /* Place for rcu_nocb_kthread() to wait GP. */
> diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
> index db85ca3..fee0b1e 100644
> --- a/kernel/rcu/tree_plugin.h
> +++ b/kernel/rcu/tree_plugin.h
> @@ -506,8 +506,10 @@ void rcu_read_unlock_special(struct task_struct *t)
> if (IS_ENABLED(CONFIG_RCU_BOOST)) {
> /* Snapshot ->boost_mtx ownership w/rnp->lock held. */
> drop_boost_mutex = rt_mutex_owner(&rnp->boost_mtx) == t;
> +#ifdef CONFIG_RCU_BOOST
> if (&t->rcu_node_entry == rnp->boost_tasks)
> rnp->boost_tasks = np;
> +#endif
> }
>
> /*
> --
> 2.7.4
>

2018-01-19 01:23:39

by Liu, Changcheng

Subject: Re: [PATCH] rcu: refine structure rcu_node field for rcu boost

Hi Paul,
This patch is optional.

1. It's true that the added #ifdef in rcu_read_unlock_special is a
bad breaker. The patch follows the current code e.g.
CONFIG_PROVE_RCU

2. Haven't evaluated the performance in system.
Studying RCU and found that boost feature field is allocated in
in the rcu_node structure even rcu boost isn't enabled.

B.R.
Changcheng

On 09:38 Thu 18 Jan, Paul E. McKenney wrote:
> On Thu, Jan 18, 2018 at 06:33:43PM +0800, Liu, Changcheng wrote:
> > Do not allocate space for rcu boost field when
> > RCU BOOST is not configured.
> >
> > Signed-off-by: Liu Changcheng <[email protected]>
>
> The added #ifdef in rcu_read_unlock_special() is a deal-breaker.
> Just out of curiosity, is this decrease in storage measurable at
> the system level?
>
> Thanx, Paul
>
> > diff --git a/kernel/rcu/tree.h b/kernel/rcu/tree.h
> > index 46a5d19..88f087e 100644
> > --- a/kernel/rcu/tree.h
> > +++ b/kernel/rcu/tree.h
> > @@ -129,6 +129,7 @@ struct rcu_node {
> > /* if there is no such task. If there */
> > /* is no current expedited grace period, */
> > /* then there can cannot be any such task. */
> > +#ifdef CONFIG_RCU_BOOST
> > struct list_head *boost_tasks;
> > /* Pointer to first task that needs to be */
> > /* priority boosted, or NULL if no priority */
> > @@ -153,6 +154,8 @@ struct rcu_node {
> > /* Number of tasks boosted for expedited GP. */
> > unsigned long n_normal_boosts;
> > /* Number of tasks boosted for normal GP. */
> > +#endif/* #ifdef CONFIG_RCU_BOOST*/
> > +
> > #ifdef CONFIG_RCU_NOCB_CPU
> > struct swait_queue_head nocb_gp_wq[2];
> > /* Place for rcu_nocb_kthread() to wait GP. */
> > diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
> > index db85ca3..fee0b1e 100644
> > --- a/kernel/rcu/tree_plugin.h
> > +++ b/kernel/rcu/tree_plugin.h
> > @@ -506,8 +506,10 @@ void rcu_read_unlock_special(struct task_struct *t)
> > if (IS_ENABLED(CONFIG_RCU_BOOST)) {
> > /* Snapshot ->boost_mtx ownership w/rnp->lock held. */
> > drop_boost_mutex = rt_mutex_owner(&rnp->boost_mtx) == t;
> > +#ifdef CONFIG_RCU_BOOST
> > if (&t->rcu_node_entry == rnp->boost_tasks)
> > rnp->boost_tasks = np;
> > +#endif
> > }
> >
> > /*
> > --
> > 2.7.4
> >
>
>