Re: [PATCH v2 1/2] sched/cputime: Use only pi_lock to protect sum_exec_runtime read

[Stanislaw Gruszka <sgruszka@redhat.com>]

On Mon, Sep 05, 2016 at 11:13:01AM +0200, Stanislaw Gruszka wrote:
> Currently we protect 64bit sum_exec_runtime read on 32bit cpus using
> task_rq_lock() which internally takes t->pi_lock and rq->lock. Taking
> rq->lock is not needed in this case.

I looked more at kernel/sched/ code and now I'm not sure about this.
I assumed that update_curr() is called with rq->curr->pi_lock, but
looks like it can be called with some other task->pi_lock not
necessary the rq->curr, hence looks that we need rq->lock to assure
protection

Stanislaw

> Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
> ---
>  kernel/sched/cputime.c | 7 +++----
>  1 file changed, 3 insertions(+), 4 deletions(-)
> 
> diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c
> index b93c72d..5535774 100644
> --- a/kernel/sched/cputime.c
> +++ b/kernel/sched/cputime.c
> @@ -315,12 +315,11 @@ static inline u64 read_sum_exec_runtime(struct task_struct *t)
>  static u64 read_sum_exec_runtime(struct task_struct *t)
>  {
>  	u64 ns;
> -	struct rq_flags rf;
> -	struct rq *rq;
> +	unsigned long flags;
>  
> -	rq = task_rq_lock(t, &rf);
> +	raw_spin_lock_irqsave(&t->pi_lock, flags);
>  	ns = t->se.sum_exec_runtime;
> -	task_rq_unlock(rq, t, &rf);
> +	raw_spin_unlock_irqrestore(&t->pi_lock, flags);
>  
>  	return ns;
>  }
> -- 
> 1.8.3.1
>