Re: [PATCH 2/2] sched/fair: Fix use of NULL with find_idlest_group

[Peter Zijlstra]

On Mon, Aug 21, 2017 at 06:59:28PM +0100, Brendan Jackman wrote:
> Hi Josef,
> 
> Thanks for taking a look.
> 
> On Mon, Aug 21 2017 at 17:26, Josef Bacik wrote:
> > On Mon, Aug 21, 2017 at 04:21:28PM +0100, Brendan Jackman wrote:
> [...]
> >> -		local_group = cpumask_test_cpu(this_cpu,
> >> -					       sched_group_span(group));
> >> -
> >
> > This isn't right is it?  cpu isn't necessarily in the very first group of a sd
> > right?
> 
> I think it is - I haven't grokked the sd/sg setup code in
> kernel/sched/topology.c but there is a comment[1] that I interpret as
> saying this. I'll take a more careful look tomorrow.
> 
> [1] http://elixir.free-electrons.com/linux/v4.13-rc6/source/kernel/sched/topology.c#L786
> 
> If I'm wrong, this can be rewritten not to use that assumption - I did
> it this way in the caller ("else if (group == sd->groups)") because I
> didn't want to use cpumask_test_cpu, and then changed it inside
> find_idlest_group so there weren't two ways of doing the same thing in
> the same neighbourhood of code.

No you are quite correct. The sched_domain of a CPU always includes that
CPU and the first group of a sched_domain is it's child domain and
therefore also includes that CPU.