Re: [PATCH tip/core/rcu 02/23] rcu: Allow RCU grace-period initialization to be preempted

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

On Sat, Sep 01, 2012 at 06:09:35PM -0700, Josh Triplett wrote:
> On Thu, Aug 30, 2012 at 11:18:17AM -0700, Paul E. McKenney wrote:
> > From: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
> > 
> > RCU grace-period initialization is currently carried out with interrupts
> > disabled, which can result in 200-microsecond latency spikes on systems
> > on which RCU has been configured for 4096 CPUs.  This patch therefore
> > makes the RCU grace-period initialization be preemptible, which should
> > eliminate those latency spikes.  Similar spikes from grace-period cleanup
> > and the forcing of quiescent states will be dealt with similarly by later
> > patches.
> > 
> > Reported-by: Mike Galbraith <mgalbraith@suse.de>
> > Reported-by: Dimitri Sivanich <sivanich@sgi.com>
> > Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> 
> Does it make sense to have cond_resched() right before the continues,
> which lead right back up to the wait_event_interruptible at the top of
> the loop?  Or do you expect to usually find that event already
> signalled?

Given that the event might have already signaled, I need to put the
cond_resched() into the loop.  Otherwise, we get back long latencies.

							Thanx, Paul

> In any case:
> 
> Reviewed-by: Josh Triplett <josh@joshtriplett.org>
> 
> >  kernel/rcutree.c |   17 ++++++++++-------
> >  1 files changed, 10 insertions(+), 7 deletions(-)
> > 
> > diff --git a/kernel/rcutree.c b/kernel/rcutree.c
> > index e1c5868..ef56aa3 100644
> > --- a/kernel/rcutree.c
> > +++ b/kernel/rcutree.c
> > @@ -1069,6 +1069,7 @@ static int rcu_gp_kthread(void *arg)
> >  			 * don't start another one.
> >  			 */
> >  			raw_spin_unlock_irqrestore(&rnp->lock, flags);
> > +			cond_resched();
> >  			continue;
> >  		}
> >  
> > @@ -1079,6 +1080,7 @@ static int rcu_gp_kthread(void *arg)
> >  			 */
> >  			rsp->fqs_need_gp = 1;
> >  			raw_spin_unlock_irqrestore(&rnp->lock, flags);
> > +			cond_resched();
> >  			continue;
> >  		}
> >  
> > @@ -1089,10 +1091,10 @@ static int rcu_gp_kthread(void *arg)
> >  		rsp->fqs_state = RCU_GP_INIT; /* Stop force_quiescent_state. */
> >  		rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
> >  		record_gp_stall_check_time(rsp);
> > -		raw_spin_unlock(&rnp->lock);  /* leave irqs disabled. */
> > +		raw_spin_unlock_irqrestore(&rnp->lock, flags);
> >  
> >  		/* Exclude any concurrent CPU-hotplug operations. */
> > -		raw_spin_lock(&rsp->onofflock);  /* irqs already disabled. */
> > +		get_online_cpus();
> >  
> >  		/*
> >  		 * Set the quiescent-state-needed bits in all the rcu_node
> > @@ -1112,7 +1114,7 @@ static int rcu_gp_kthread(void *arg)
> >  		 * due to the fact that we have irqs disabled.
> >  		 */
> >  		rcu_for_each_node_breadth_first(rsp, rnp) {
> > -			raw_spin_lock(&rnp->lock); /* irqs already disabled. */
> > +			raw_spin_lock_irqsave(&rnp->lock, flags);
> >  			rcu_preempt_check_blocked_tasks(rnp);
> >  			rnp->qsmask = rnp->qsmaskinit;
> >  			rnp->gpnum = rsp->gpnum;
> > @@ -1123,15 +1125,16 @@ static int rcu_gp_kthread(void *arg)
> >  			trace_rcu_grace_period_init(rsp->name, rnp->gpnum,
> >  						    rnp->level, rnp->grplo,
> >  						    rnp->grphi, rnp->qsmask);
> > -			raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
> > +			raw_spin_unlock_irqrestore(&rnp->lock, flags);
> > +			cond_resched();
> >  		}
> >  
> >  		rnp = rcu_get_root(rsp);
> > -		raw_spin_lock(&rnp->lock); /* irqs already disabled. */
> > +		raw_spin_lock_irqsave(&rnp->lock, flags);
> >  		/* force_quiescent_state() now OK. */
> >  		rsp->fqs_state = RCU_SIGNAL_INIT;
> > -		raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
> > -		raw_spin_unlock_irqrestore(&rsp->onofflock, flags);
> > +		raw_spin_unlock_irqrestore(&rnp->lock, flags);
> > +		put_online_cpus();
> >  	}
> >  	return 0;
> >  }
> > -- 
> > 1.7.8
> > 
>