Re: [PATCH] cgroup/cpuset: simplify the percpu kthreads check in update_tasks_cpumask()

[Waiman Long]

On 7/5/23 01:56, Miaohe Lin wrote:
> On 2023/7/5 11:14, Waiman Long wrote:
> > On 7/4/23 07:30, Miaohe Lin wrote:
> > > kthread_is_per_cpu() can be called directly without checking whether
> > > PF_KTHREAD is set in task->flags. So remove PF_KTHREAD check to make
> > > code more concise.
> > >
> > > Signed-off-by: Miaohe Lin <linmiaohe@xxxxxxxxxx>
> > > ---
> > >    kernel/cgroup/cpuset.c | 2 +-
> > >    1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
> > > index 58e6f18f01c1..601c40da8e03 100644
> > > --- a/kernel/cgroup/cpuset.c
> > > +++ b/kernel/cgroup/cpuset.c
> > > @@ -1230,7 +1230,7 @@ static void update_tasks_cpumask(struct cpuset *cs, struct cpumask *new_cpus)
> > >                /*
> > >                 * Percpu kthreads in top_cpuset are ignored
> > >                 */
> > > -            if ((task->flags & PF_KTHREAD) && kthread_is_per_cpu(task))
> > > +            if (kthread_is_per_cpu(task))
> > >                    continue;
> > >                cpumask_andnot(new_cpus, possible_mask, cs->subparts_cpus);
> > >            } else {
> > The initial intention was to ignore only percpu kthreads. The current code likely ignore all the kthreads. Removing the PF_KTHREAD flag, however, may introduce unexpected regression at this point. I would like to hold off for now until more investigation are done.
> IMHO, the current code will ignore only percpu kthreads:
>    1.If PF_KTHREAD is set in task->flags, this patch doesn't make any difference.
>    2.If PF_KTHREAD is not set in task->flags, kthread_is_per_cpu will *always return false*. So this patch doesn't make any functional change.
>
>      bool kthread_is_per_cpu(struct task_struct *p)
>      {
>          struct kthread *kthread = __to_kthread(p);
> 	if (!kthread)
> 		return false;
>          ....
>      }
>
>      static inline struct kthread *__to_kthread(struct task_struct *p)
>      {
> 	void *kthread = p->worker_private;
> 	if (kthread && !(p->flags & PF_KTHREAD))
> 			 ^^^^^^^^^^^^^^^^^^^^^^
> 			 PF_KTHREAD is not set, so kthread = NULL.
> 		kthread = NULL;
> 	return kthread;
>      }
>
> Or am I miss something? Thanks for comment and review.

Yes, you are right. I was that conscious when I reviewed the patch last 
night :-)

Reviewed-by: Waiman Long <longman@xxxxxxxxxx>