Re: [PATCH V4 1/2] sched/fair: Fix load_balance() affinity redo path

[Jeffrey Hugo]

On 6/5/2017 11:23 AM, Jeffrey Hugo wrote:
> On 6/2/2017 4:27 PM, Jeffrey Hugo wrote:
> > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > index d711093..84255ab 100644
> > --- a/kernel/sched/fair.c
> > +++ b/kernel/sched/fair.c
> > @@ -6737,10 +6737,10 @@ int can_migrate_task(struct task_struct *p, 
> > struct lb_env *env)
> >            * our sched_group. We may want to revisit it if we couldn't
> >            * meet load balance goals by pulling other tasks on src_cpu.
> >            *
> > -         * Also avoid computing new_dst_cpu if we have already computed
> > -         * one in current iteration.
> > +         * Avoid computing new_dst_cpu for NEWLY_IDLE or if we have
> > +         * already computed one in current iteration.
> >            */
> > -        if (!env->dst_grpmask || (env->flags & LBF_DST_PINNED))
> > +        if (env->idle == CPU_NEWLY_IDLE || (env->flags & 
> > LBF_DST_PINNED))
> >               return 0;
>
> Self NACK.  This breaks active_load_balance_cpu_stop().  Looks like 
> env->idle == CPU_IDLE, but env->dst_grpmask is uninitialized, so it can 
> be NULL, which causes a null pointer dereference a few lines later.
>
> I'm still having a look to see what makes sense to address the issue.

As far as I can see, there appears to be two options to resolve the issue -

1. Update active_load_balance_cpu_stop() to initialize dst_grpmask to a 
sane value

2. Undo the proposed changes in load_balance() to "ensure" dst_grpmask 
is valid, and calculate the value on demand when checking to see if the 
redo path needs to be done.

The downside to #1 is that dst_grpmask is not needed in the 
active_load_balance_cpu_stop() path, and the loop to calculate a new 
dst_cpu will be used.  Extra code is evaluated, but there appears to be 
no side effects.

The downside to #2 is that dst_grpmask is valid the majority of the time 
in load_balance(), so calculating it on demand is redundant most of the 
time, but again there appears to be no side effects.

It somewhat feels like a choice of which option is less bad.

Peter/Dietmar, any preferences?

--
Jeffrey Hugo
Qualcomm Datacenter Technologies as an affiliate of Qualcomm 
Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the
Code Aurora Forum, a Linux Foundation Collaborative Project.