Re: [PATCH v2] sched: Check sched_domain before computing group power.

From: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
To: Preeti Murthy <preeti.lkml@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>,
	mingo@kernel.org, hpa@zytor.com, linux-kernel@vger.kernel.org,
	Thomas Gleixner <tglx@linutronix.de>,
	mikey@neuling.org, linux-tip-commits@vger.kernel.org,
	Preeti U Murthy <preeti@linux.vnet.ibm.com>
Subject: Re: [PATCH v2] sched: Check sched_domain before computing group power.
Date: Wed, 13 Nov 2013 16:53:29 +0530	[thread overview]
Message-ID: <20131113112329.GC543@linux.vnet.ibm.com> (raw)
In-Reply-To: <CAM4v1pNMn=5oZDiX3fUp9uPkZTPJgk=vEKEjevzvpwn=PjTzXg@mail.gmail.com>

* Preeti Murthy <preeti.lkml@gmail.com> [2013-11-13 16:22:37]:

> Hi Srikar,
> 
> update_group_power() is called only during load balancing during
> update_sg_lb_stats().
> Load balancing begins at the base domain of the CPU,rq(cpu)->sd. This is
> checked for
> NULL. So how can update_group_power() be called in a scenario where the
> base domain
> of the CPU is not initialized? I say 'initialized' since you check for NULL
> on rq(cpu)->sd.
> 

update_group_power() also gets called from init_sched_groups_power().
And if you see the oops message, we know that the oops happens from that
path. In build_sched_domains(), we do cpu_attach_domain() what updates
rq->sd after the call to init_sched_groups_power(). So by the time
init_sched_groups_power() is called rq->sd is not yet initialized. 

We only hit oops case, when the sd->flags has SD_OVERLAP set.

> In the changelog, you say 'updated'. Are you saying that it has a stale
> value?

I said, "before the sched_domain for a cpu is updated", so its not yet
updated or has stale value. As I said earlier in this mail, the
initialization happens after we do a update_group_power().

> Please do elaborate on how you observed this.
> 

Does this clarify?

-- 
Thanks and Regards
Srikar Dronamraju