Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Thu, 20 Oct 2022 19:08:13 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Tejun Heo <tj@kernel.org>
Cc:     Chuyi Zhou <zhouchuyi@bytedance.com>, mingo@redhat.com,
        juri.lelli@redhat.com, vincent.guittot@linaro.org,
        linux-kernel@vger.kernel.org, htejun@gmail.com,
        lizefan.x@bytedance.com, vschneid@redhat.com, bsegall@google.com,
        Abel Wu <wuyun.abel@bytedance.com>
Subject: Re: [RESEND] sched/fair: Add min_ratio for cfs bandwidth_control
Message-ID: <Y1GAffm4aHCpvoBB@hirez.programming.kicks-ass.net>
References: <20221019031551.24312-1-zhouchuyi@bytedance.com>
 <Y1BqT59ovFitahJa@slm.duckdns.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <Y1BqT59ovFitahJa@slm.duckdns.org>
Precedence: bulk

On Wed, Oct 19, 2022 at 11:21:19AM -1000, Tejun Heo wrote:
> Hello,
> 
> On Wed, Oct 19, 2022 at 11:15:51AM +0800, Chuyi Zhou wrote:
> > Tasks may be throttled when holding locks for a long time by current
> > cfs bandwidth control mechanism once users set a too small quota/period
> > ratio, which can result whole system get stuck[1].
> > 
> > In order to prevent the above situation from happening, this patch adds
> > sysctl_sched_cfs_bandwidth_min_ratio in /proc/sys/kernel, which indicates
> > the minimum percentage of quota/period users can set. The default value is
> > zero and users can set quota and period without triggering this constraint.
> > 
> > Link[1]:https://lore.kernel.org/lkml/5987be34-b527-4ff5-a17d-5f6f0dc94d6d@huawei.com/T/
> > Signed-off-by: Chuyi Zhou <zhouchuyi@bytedance.com>
> > Suggested-by: Abel Wu <wuyun.abel@bytedance.com>
> 
> This is a bit of a bandaid. I think what we really need to do is only
> throttling when running in userspace. In kernel space, it should just keep
> accumulating used cycles as debt which should be paid back before userspace
> code can run again so that we don't throttle at random places in the kernel.

That's just moving the problem. But yeah; perhaps. Starving random
userspace is less of a problem I suppose.