Re: [patch] sched: fix next_interval determination in idle_balance()

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

On Tue, Jun 19, 2007 at 09:39:03PM +0200, Ingo Molnar wrote:
> 
> 2.6.22 must-have item - perhaps suitable for -stable too, because it was 
> reproduced on 2.6.21.5 too.
> 
> ---------------------->
> From: Christoph Lameter <clameter@sgi.com>
> Subject: [patch] sched: fix next_interval determination in idle_balance()
> 
> Fix massive SMP imbalance on NUMA nodes observed on 2.6.21.5 with CFS. 
> (and later on reproduced without CFS as well).
> 
> The intervals of domains that do not have SD_BALANCE_NEWIDLE must be 
> considered for the calculation of the time of the next balance. 
> Otherwise we may defer rebalancing forever and nodes might stay idle for 
> very long times.
> 
> Siddha also spotted that the conversion of the balance interval to 
> jiffies is missing. Fix that to.
> 
> From: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
> 
> also continue the loop if !(sd->flags & SD_LOAD_BALANCE).
> 
> Tested-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>

Retested-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>

Spread evenly in well under a minute on the machine that previously
kept all of the rcutorture tasks on a single CPU forever, which is most
definitely not the way to torture RCU.

Good stuff!!!

						Thanx, Paul

> It did in fact trigger under all three of mainline, CFS, and -rt 
> including CFS -- see below for a couple of emails from last Friday 
> giving results for these three on the AMD box (where it happened) and on 
> a single-quad NUMA-Q system (where it did not, at least not with such 
> severity).
> 
> Signed-off-by: Christoph Lameter <clameter@sgi.com>
> Signed-off-by: Ingo Molnar <mingo@elte.hu>
> ---
>  kernel/sched.c |   22 +++++++++++++---------
>  1 file changed, 13 insertions(+), 9 deletions(-)
> 
> Index: v/kernel/sched.c
> ===================================================================
> --- v.orig/kernel/sched.c
> +++ v/kernel/sched.c
> @@ -2938,17 +2938,21 @@ static void idle_balance(int this_cpu, s
>  	unsigned long next_balance = jiffies + 60 *  HZ;
> 
>  	for_each_domain(this_cpu, sd) {
> -		if (sd->flags & SD_BALANCE_NEWIDLE) {
> +		unsigned long interval;
> +
> +		if (!(sd->flags & SD_LOAD_BALANCE))
> +			continue;
> +
> +		if (sd->flags & SD_BALANCE_NEWIDLE)
>  			/* If we've pulled tasks over stop searching: */
>  			pulled_task = load_balance_newidle(this_cpu,
> -							this_rq, sd);
> -			if (time_after(next_balance,
> -				  sd->last_balance + sd->balance_interval))
> -				next_balance = sd->last_balance
> -						+ sd->balance_interval;
> -			if (pulled_task)
> -				break;
> -		}
> +								this_rq, sd);
> +
> +		interval = msecs_to_jiffies(sd->balance_interval);
> +		if (time_after(next_balance, sd->last_balance + interval))
> +			next_balance = sd->last_balance + interval;
> +		if (pulled_task)
> +			break;
>  	}
>  	if (!pulled_task)
>  		/*