Re: [PATCH RFC srcu] Expedite srcu_schedule_cbs_snp() callback invocation

[Paul E. McKenney]

On Tue, Apr 18, 2017 at 06:43:05PM +0100, Marc Zyngier wrote:
> On 18/04/17 18:33, Paul E. McKenney wrote:
> > Although Tree SRCU does reduce delays when there is at least one
> > synchronize_srcu_expedited() invocation pending, srcu_schedule_cbs_snp()
> > still waits for SRCU_INTERVAL before invoking callbacks.  Since
> > synchronize_srcu_expedited() now posts a callback and waits for
> > that callback to do a wakeup, this destroys the expedited nature of
> > synchronize_srcu_expedited().
> > 
> > This commit therefore invokes callbacks immediately at the end of the
> > grace period when there is at least one synchronize_srcu_expedited()
> > invocation pending.
> > 
> > Reported-by: Marc Zyngier <marc.zyngier@xxxxxxx>
> > Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
> > 
> > diff --git a/kernel/rcu/srcutree.c b/kernel/rcu/srcutree.c
> > index 1e54f91c2ff7..744ffdb79d23 100644
> > --- a/kernel/rcu/srcutree.c
> > +++ b/kernel/rcu/srcutree.c
> > @@ -447,7 +447,8 @@ static void srcu_schedule_cbs_snp(struct srcu_struct *sp, struct srcu_node *snp)
> >  	int cpu;
> >  
> >  	for (cpu = snp->grplo; cpu <= snp->grphi; cpu++)
> > -		srcu_schedule_cbs_sdp(per_cpu_ptr(sp->sda, cpu), SRCU_INTERVAL);
> > +		srcu_schedule_cbs_sdp(per_cpu_ptr(sp->sda, cpu),
> > +				      atomic_read(&sp->srcu_exp_cnt) ? 0 : SRCU_INTERVAL);
> >  }
> >  
> >  /*
> 
> This fixes the issue I was observing, where UEFI in a KVM guest would
> take a x8 slowdown while it messes with some NOR flash emulation on
> arm64 (don't ask, this is horrible...). Anyway:
> 
> Tested-by: Marc Zyngier <marc.zyngier@xxxxxxx>
> 
> Thanks again Paul!

Thank you for your bug-finding and testing efforts!  I will be pushing
this into the upcoming merge window.

							Thanx, Paul

------------------------------------------------------------------------

commit 6eec94fe40e294b04d32c8ef552e28fa6159bdad
Author: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
Date:   Tue Apr 18 10:28:31 2017 -0700

    srcu: Expedite srcu_schedule_cbs_snp() callback invocation
    
    Although Tree SRCU does reduce delays when there is at least one
    synchronize_srcu_expedited() invocation pending, srcu_schedule_cbs_snp()
    still waits for SRCU_INTERVAL before invoking callbacks.  Since
    synchronize_srcu_expedited() now posts a callback and waits for
    that callback to do a wakeup, this destroys the expedited nature of
    synchronize_srcu_expedited().  This destruction became apparent to
    Marc Zyngier in the guise of a guest-OS bootup slowdown from five
    seconds to no fewer than forty seconds.
    
    This commit therefore invokes callbacks immediately at the end of the
    grace period when there is at least one synchronize_srcu_expedited()
    invocation pending.  This brought Marc's guest-OS bootup times back
    into the realm of reason.
    
    Reported-by: Marc Zyngier <marc.zyngier@xxxxxxx>
    Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
    Tested-by: Marc Zyngier <marc.zyngier@xxxxxxx>

diff --git a/kernel/rcu/srcutree.c b/kernel/rcu/srcutree.c
index 1e54f91c2ff7..744ffdb79d23 100644
--- a/kernel/rcu/srcutree.c
+++ b/kernel/rcu/srcutree.c
@@ -447,7 +447,8 @@ static void srcu_schedule_cbs_snp(struct srcu_struct *sp, struct srcu_node *snp)
 	int cpu;
 
 	for (cpu = snp->grplo; cpu <= snp->grphi; cpu++)
-		srcu_schedule_cbs_sdp(per_cpu_ptr(sp->sda, cpu), SRCU_INTERVAL);
+		srcu_schedule_cbs_sdp(per_cpu_ptr(sp->sda, cpu),
+				      atomic_read(&sp->srcu_exp_cnt) ? 0 : SRCU_INTERVAL);
 }
 
 /*