Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH v3 3/7] sched: rotate the cpu search window for better
 spread
To:     subhra mazumdar <subhra.mazumdar@oracle.com>,
        linux-kernel@vger.kernel.org
Cc:     peterz@infradead.org, mingo@redhat.com, tglx@linutronix.de,
        steven.sistare@oracle.com, dhaval.giani@oracle.com,
        daniel.lezcano@linaro.org, vincent.guittot@linaro.org,
        viresh.kumar@linaro.org, tim.c.chen@linux.intel.com,
        mgorman@techsingularity.net
References: <20190627012919.4341-1-subhra.mazumdar@oracle.com>
 <20190627012919.4341-4-subhra.mazumdar@oracle.com>
From:   Parth Shah <parth@linux.ibm.com>
Date:   Sat, 29 Jun 2019 00:06:58 +0530
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.0
MIME-Version: 1.0
In-Reply-To: <20190627012919.4341-4-subhra.mazumdar@oracle.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Message-Id: <f8b185ca-ae3e-8c54-5381-e9104be4954c@linux.ibm.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi Subhra,

I ran your patch series on IBM POWER systems and this is what I have observed.

On 6/27/19 6:59 AM, subhra mazumdar wrote:
> Rotate the cpu search window for better spread of threads. This will ensure
> an idle cpu will quickly be found if one exists.
> 
> Signed-off-by: subhra mazumdar <subhra.mazumdar@oracle.com>
> ---
>  kernel/sched/fair.c | 10 ++++++++--
>  1 file changed, 8 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index b58f08f..c1ca88e 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -6188,7 +6188,7 @@ static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, int t
>  	u64 avg_cost, avg_idle;
>  	u64 time, cost;
>  	s64 delta;
> -	int cpu, limit, floor, nr = INT_MAX;
> +	int cpu, limit, floor, target_tmp, nr = INT_MAX;
> 
>  	this_sd = rcu_dereference(*this_cpu_ptr(&sd_llc));
>  	if (!this_sd)
> @@ -6219,9 +6219,15 @@ static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, int t
>  		}
>  	}
> 
> +	if (per_cpu(next_cpu, target) != -1)
> +		target_tmp = per_cpu(next_cpu, target);
> +	else
> +		target_tmp = target;
> +
>  	time = local_clock();
> 
> -	for_each_cpu_wrap(cpu, sched_domain_span(sd), target) {
> +	for_each_cpu_wrap(cpu, sched_domain_span(sd), target_tmp) {
> +		per_cpu(next_cpu, target) = cpu;

This leads to a problem of cache hotness.
AFAIU, in most cases, `target = prev_cpu` of the task being woken up and
selecting an idle CPU nearest to the prev_cpu is favorable.
But since this doesn't keep track of last idle cpu per task, it fails to find the
nearest possible idle CPU in cases when the task is being woken up after other
scheduled task.

Consider below scenario:
=======================
- System: 44 cores, 88 CPUs
- 44 CPU intensive tasks pinned to any CPU in each core. This makes 'select_idle_core' return -1;
- Consider below shown timeline:
- Task T1 runs for time 0-5 on CPU0
- Then task T2 runs for time 6-10 on CPU0
- T1 wakes at time 7, with target=0, and setting
  per_cpu(next_cpu,0)= 4 (let's say cpu 0-3 are busy at the time)
- So T1 runs for time 7-12 on CPU4.
- Meanwhile, T2 wakes at time 11, with target=0, but per_cpu(next_cpu, 0) is 4. So starts
  searching from CPU4 and ends up at CPU 8 or so even though CPU0 is free at that time.
- This goes on further far away from the prev_cpu on each such iteration unless it wraps around after 44 CPUs.


              ^T1           T1$   ^T2          T2$
CPU 0          |             |     |            |      
       -----------------------------------------------------------------------------
               0             5     6            10                       time----->

                                       ^T1                T1$
CPU 4                                   |                  |
       -----------------------------------------------------------------------------
                                        7                  12            time----->

                                                      ^T2          T2$
CPU 8                                                  |            |
       -----------------------------------------------------------------------------
                                                       11                time------>

Symbols: ^Tn: Task Tn wake-up,  Tn$: task Tn sleeps

Above example indicates the both the task T1 and T2 suffers from cache hotness in further iterations.


>  		if (!--nr)
>  			return -1;
>  		if (!cpumask_test_cpu(cpu, &p->cpus_allowed))
> 


Best
Parth