Re: [PATCH] sched_groups are expected to be circular linked list, make it so right after allocation

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Wed, 2012-05-09 at 12:38 +0200, Igor Mammedov wrote:
> init_sched_groups_power() that expects sched_groups to be
> circular linked list. However it is not always true, since sched_groups
> preallocated in __sdt_alloc are initialized in build_sched_groups and it
> may exit early
> 
>         if (cpu != cpumask_first(sched_domain_span(sd)))
>                 return 0;
> 
> without initializing sd->groups->next field.

The only way I can see that happen is if the arch code is lying to us.
We build the sched_domain_span() like:

  cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu));

and the above != first_cpumask() can only happen if the topology mask
provided by the architecture includes a cpu that isn't actually there.

(and equally, how did it get in the active_mask if its not there).

Jiang, how did your IA64 arrive in this state?

> Fix bug by initializing next field right after sched_group was allocated.

I'd not call it a bug, the bug is the arch being broken, its a
robustification of the code to better handle broken input.

Ideally we'd also add a WARN someplace to notify us of this situation.

Still, nice catch..

Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>

> Signed-off-by: Igor Mammedov <imammedo@redhat.com>
> ---
>  kernel/sched/core.c |    2 ++
>  1 files changed, 2 insertions(+), 0 deletions(-)
> 
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 0533a68..e5212ae 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -6382,6 +6382,8 @@ static int __sdt_alloc(const struct cpumask *cpu_map)
>                         if (!sg)
>                                 return -ENOMEM;
>  
> +                       sg->next = sg;
> +
>                         *per_cpu_ptr(sdd->sg, j) = sg;
>  
>                         sgp = kzalloc_node(sizeof(struct sched_group_power),