Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Tue, 6 Oct 2020 16:09:26 +0200
From:   Juri Lelli <juri.lelli@redhat.com>
To:     Peter Zijlstra <peterz@infradead.org>
Cc:     tglx@linutronix.de, mingo@kernel.org, linux-kernel@vger.kernel.org,
        bigeasy@linutronix.de, qais.yousef@arm.com, swood@redhat.com,
        valentin.schneider@arm.com, vincent.guittot@linaro.org,
        dietmar.eggemann@arm.com, rostedt@goodmis.org, bsegall@google.com,
        mgorman@suse.de, bristot@redhat.com, vincent.donnefort@arm.com,
        tj@kernel.org
Subject: Re: [PATCH -v2 12/17] sched,rt: Use cpumask_any*_distribute()
Message-ID: <20201006140926.GI4352@localhost.localdomain>
References: <20201005145717.346020688@infradead.org>
 <20201005150922.156231026@infradead.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20201005150922.156231026@infradead.org>
Precedence: bulk

Hi,

On 05/10/20 16:57, Peter Zijlstra wrote:
> Replace a bunch of cpumask_any*() instances with
> cpumask_any*_distribute(), by injecting this little bit of random in
> cpu selection, we reduce the chance two competing balance operations
> working off the same lowest_mask pick the same CPU.
> 
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> ---
>  include/linux/cpumask.h |    6 ++++++
>  kernel/sched/cpupri.c   |    4 ++--
>  kernel/sched/deadline.c |    2 +-
>  kernel/sched/rt.c       |    6 +++---
>  lib/cpumask.c           |   18 ++++++++++++++++++
>  5 files changed, 30 insertions(+), 6 deletions(-)
> 
> --- a/include/linux/cpumask.h
> +++ b/include/linux/cpumask.h
> @@ -199,6 +199,11 @@ static inline int cpumask_any_and_distri
>  	return cpumask_next_and(-1, src1p, src2p);
>  }
>  
> +static inline int cpumask_any_distribute(const struct cpumask *srcp)
> +{
> +	return cpumask_first(srcp);
> +}
> +
>  #define for_each_cpu(cpu, mask)			\
>  	for ((cpu) = 0; (cpu) < 1; (cpu)++, (void)mask)
>  #define for_each_cpu_not(cpu, mask)		\
> @@ -252,6 +257,7 @@ int cpumask_any_but(const struct cpumask
>  unsigned int cpumask_local_spread(unsigned int i, int node);
>  int cpumask_any_and_distribute(const struct cpumask *src1p,
>  			       const struct cpumask *src2p);
> +int cpumask_any_distribute(const struct cpumask *srcp);
>  
>  /**
>   * for_each_cpu - iterate over every cpu in a mask
> --- a/kernel/sched/deadline.c
> +++ b/kernel/sched/deadline.c
> @@ -2001,7 +2001,7 @@ static int find_later_rq(struct task_str
>  	if (this_cpu != -1)
>  		return this_cpu;
>  
> -	cpu = cpumask_any(later_mask);
> +	cpu = cpumask_any_distribute(later_mask);
>  	if (cpu < nr_cpu_ids)
>  		return cpu;

Think we can use cpumask_any_and_distribute() with later_mask for
deadline as well inside the for_each_domain loop as you do for rt below.

Best,
Juri

> --- a/kernel/sched/rt.c
> +++ b/kernel/sched/rt.c
> @@ -1752,8 +1752,8 @@ static int find_lowest_rq(struct task_st
>  				return this_cpu;
>  			}
>  
> -			best_cpu = cpumask_first_and(lowest_mask,
> -						     sched_domain_span(sd));
> +			best_cpu = cpumask_any_and_distribute(lowest_mask,
> +							      sched_domain_span(sd));
>  			if (best_cpu < nr_cpu_ids) {
>  				rcu_read_unlock();
>  				return best_cpu;
> @@ -1770,7 +1770,7 @@ static int find_lowest_rq(struct task_st
>  	if (this_cpu != -1)
>  		return this_cpu;
>  
> -	cpu = cpumask_any(lowest_mask);
> +	cpu = cpumask_any_distribute(lowest_mask);
>  	if (cpu < nr_cpu_ids)
>  		return cpu;