LVS
lvs-devel
Google
 
Web LinuxVirtualServer.org

Re: [PATCH 2/2] ipvs: Use cond_resched_rcu_lock() helper when dumping co

To: Julian Anastasov <ja@xxxxxx>
Subject: Re: [PATCH 2/2] ipvs: Use cond_resched_rcu_lock() helper when dumping connections
Cc: Eric Dumazet <eric.dumazet@xxxxxxxxx>, Peter Zijlstra <peterz@xxxxxxxxxxxxx>, Simon Horman <horms@xxxxxxxxxxxx>, Ingo Molnar <mingo@xxxxxxxxxx>, lvs-devel@xxxxxxxxxxxxxxx, netdev@xxxxxxxxxxxxxxx, netfilter-devel@xxxxxxxxxxxxxxx, linux-kernel@xxxxxxxxxxxxxxx, Pablo Neira Ayuso <pablo@xxxxxxxxxxxxx>, Dipankar Sarma <dipankar@xxxxxxxxxx>, dhaval.giani@xxxxxxxxx
From: "Paul E. McKenney" <paulmck@xxxxxxxxxxxxxxxxxx>
Date: Mon, 29 Apr 2013 14:30:02 -0700
On Tue, Apr 30, 2013 at 12:08:18AM +0300, Julian Anastasov wrote:
> 
>       Hello,
> 
> On Sat, 27 Apr 2013, Paul E. McKenney wrote:
> 
> > On Sat, Apr 27, 2013 at 02:32:48PM +0300, Julian Anastasov wrote:
> > > 
> > >   So, I assume, to help realtime kernels and rcu_barrier
> > > it is not a good idea to guard rcu_read_unlock with checks.
> > > I see that rcu_read_unlock will try to reschedule in the 
> > > !CONFIG_PREEMPT_RCU case (via preempt_enable), can we
> > > use ifdefs to avoid double TIF_NEED_RESCHED check?:
> > > 
> > >   rcu_read_unlock();
> > > #if !defined(CONFIG_PREEMPT) || defined(CONFIG_PREEMPT_RCU)
> > 
> > I would instead suggest something like:
> > 
> > #ifndef CONFIG_PREEMPT_RCU
> > 
> > But yes, in the CONFIG_PREEMPT_RCU case, the cond_resched() is not
> > needed.
> 
>       Hm, is this correct? If I follow the ifdefs
> preempt_schedule is called when CONFIG_PREEMPT is
> defined _and_ CONFIG_PREEMPT_RCU is not defined.
> Your example for CONFIG_PREEMPT_RCU is the opposite to this?

Yep, I really did intend to say "#ifndef CONFIG_PREEMPT_RCU".

A couple of things to keep in mind:

1.      Although rcu_read_unlock() does map to preempt_enable() for
        CONFIG_TINY_RCU and CONFIG_TREE_RCU, the current Kconfig refuses
        to allow either CONFIG_TINY_RCU or CONFIG_TREE_RCU to be selected
        if CONFIG_PREEMPT=y.

2.      In the CONFIG_PREEMPT_RCU case, __rcu_read_unlock() will check
        to see if the RCU core needs to be informed, so there is no
        need to invoke cond_resched() in that case.

3.      If we drop your "|| defined(CONFIG_PREEMPT_RCU)", we get an
        almost-synonym for my "#ifndef CONFIG_PREEMPT_RCU".  The "almost"
        applies to older kernels due to the possibility of having a
        CONFIG_TINY_PREEMPT_RCU kernel -- but this possibility is going
        away soon.

Make sense?

                                                        Thanx, Paul

> > >   cond_resched();
> > > #endif
> > >   rcu_read_lock();
> 
> Regards
> 
> --
> Julian Anastasov <ja@xxxxxx>
> 

--
To unsubscribe from this list: send the line "unsubscribe lvs-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html

<Prev in Thread] Current Thread [Next in Thread>