Re: [PATCH v5 1/3] sched/fair: Introduce the burstable CFS controller

From: Peter Zijlstra <peterz@infradead.org>
To: changhuaixin <changhuaixin@linux.alibaba.com>
Cc: Benjamin Segall <bsegall@google.com>,
	Dietmar Eggemann <dietmar.eggemann@arm.com>,
	dtcccc@linux.alibaba.com, Juri Lelli <juri.lelli@redhat.com>,
	khlebnikov@yandex-team.ru,
	open list <linux-kernel@vger.kernel.org>,
	Mel Gorman <mgorman@suse.de>, Ingo Molnar <mingo@redhat.com>,
	Odin Ugedal <odin@uged.al>, Odin Ugedal <odin@ugedal.com>,
	pauld@redhead.com, Paul Turner <pjt@google.com>,
	Steven Rostedt <rostedt@goodmis.org>,
	Shanpei Chen <shanpeic@linux.alibaba.com>,
	Tejun Heo <tj@kernel.org>,
	Vincent Guittot <vincent.guittot@linaro.org>,
	xiyou.wangcong@gmail.com
Subject: Re: [PATCH v5 1/3] sched/fair: Introduce the burstable CFS controller
Date: Tue, 25 May 2021 12:49:15 +0200	[thread overview]
Message-ID: <YKzWKxD+G4wTd2hf@hirez.programming.kicks-ass.net> (raw)
In-Reply-To: <C759D4F2-FA5F-4EB2-8441-C9ECD79AFD55@linux.alibaba.com>

On Mon, May 24, 2021 at 08:42:03PM +0800, changhuaixin wrote:

> >> --- a/kernel/sysctl.c
> >> +++ b/kernel/sysctl.c
> >> @@ -1816,6 +1816,15 @@ static struct ctl_table kern_table[] = {
> >> 		.proc_handler	= proc_dointvec_minmax,
> >> 		.extra1		= SYSCTL_ONE,
> >> 	},
> >> +	{
> >> +		.procname	= "sched_cfs_bw_burst_enabled",
> >> +		.data		= &sysctl_sched_cfs_bw_burst_enabled,
> >> +		.maxlen		= sizeof(unsigned int),
> >> +		.mode		= 0644,
> >> +		.proc_handler	= proc_dointvec_minmax,
> >> +		.extra1		= SYSCTL_ZERO,
> >> +		.extra2		= SYSCTL_ONE,
> >> +	},
> >> #endif
> >> #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
> >> 	{
> > 
> > What's the purpose of this new sysctl? By default it is disabled because
> > burst==0, only if you set burst to some !0 value does this actually do
> > anything.
> 
> Originally, this is designed to turn burst feature off when the system becomes unstable.
> Guess we can remove this as you have questioned it.

Is stability a concern? This is CFS after all, if we overload, we simply
share time as per usual.

If there is a real use-case for a global knob to limit/disable this I
don't object too much, but then please explicitly mention it.