Re: [tip:sched/core] sched/numa: Move task_numa_free() to __put_task_struct()

From: Peter Zijlstra <peterz@infradead.org>
To: Mike Galbraith <bitbucket@online.de>
Cc: Sasha Levin <sasha.levin@oracle.com>,
	mingo@kernel.org, hpa@zytor.com, linux-kernel@vger.kernel.org,
	torvalds@linux-foundation.org, mgorman@suse.com,
	akpm@linux-foundation.org, tglx@linutronix.de,
	linux-tip-commits@vger.kernel.org, Dave Jones <davej@redhat.com>
Subject: Re: [tip:sched/core] sched/numa: Move task_numa_free() to __put_task_struct()
Date: Mon, 7 Apr 2014 10:16:44 +0200	[thread overview]
Message-ID: <20140407081644.GD11096@twins.programming.kicks-ass.net> (raw)
In-Reply-To: <1396855830.28539.10.camel@marge.simpson.net>

On Mon, Apr 07, 2014 at 09:30:30AM +0200, Mike Galbraith wrote:
> -	double_lock(&my_grp->lock, &grp->lock);
> +	BUG_ON(irqs_disabled());
> +	double_lock_irq(&my_grp->lock, &grp->lock);

So either make this:

	local_irq_disable();
	double_lock();

or

>  
>  	for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++) {
>  		my_grp->faults[i] -= p->numa_faults_memory[i];
> @@ -1692,6 +1693,7 @@ static void task_numa_group(struct task_
>  
>  	spin_unlock(&my_grp->lock);
>  	spin_unlock(&grp->lock);
> +	local_irq_enable();

use:
	spin_unlock()
	spin_unlock_irq()

or so, but this imbalance is making my itch :-)

>  
>  	rcu_assign_pointer(p->numa_group, grp);
>  