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;
Sender: Tejun Heo <htejun@gmail.com>
Date:   Mon, 30 May 2022 14:57:17 -1000
From:   Tejun Heo <tj@kernel.org>
To:     Frederic Weisbecker <frederic@kernel.org>
Cc:     Peter Zijlstra <peterz@infradead.org>,
        Juri Lelli <juri.lelli@redhat.com>,
        Waiman Long <longman@redhat.com>,
        LKML <linux-kernel@vger.kernel.org>,
        "Paul E . McKenney" <paulmck@kernel.org>,
        Paul Gortmaker <paul.gortmaker@windriver.com>,
        Johannes Weiner <hannes@cmpxchg.org>,
        Marcelo Tosatti <mtosatti@redhat.com>,
        Phil Auld <pauld@redhat.com>,
        Zefan Li <lizefan.x@bytedance.com>,
        Daniel Bristot de Oliveira <bristot@kernel.org>,
        Nicolas Saenz Julienne <nsaenz@kernel.org>,
        rcu@vger.kernel.org
Subject: Re: [RFC PATCH 4/4] cpuset: Support RCU-NOCB toggle on v2 root
 partitions
Message-ID: <YpVn7XXE20vwbS3Q@slm.duckdns.org>
References: <YpAHEt0j30vBw9au@slm.duckdns.org>
 <9e44bb00-955a-dbc6-a863-be649e0c701f@redhat.com>
 <YpAdSW8JXVPOoNJl@slm.duckdns.org>
 <20220527083018.n43nc73vuuzm5ixo@localhost.localdomain>
 <YpIwsiaY2IPK96WO@hirez.programming.kicks-ass.net>
 <20220530004049.GA1251147@lothringen>
 <YpR8PUlIraYE2+5L@worktop.programming.kicks-ass.net>
 <20220530105650.GA1257179@lothringen>
 <YpTDq6Z/+hp+CHwf@worktop.programming.kicks-ass.net>
 <20220530213556.GD1257179@lothringen>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20220530213556.GD1257179@lothringen>
Precedence: bulk

On Mon, May 30, 2022 at 11:35:56PM +0200, Frederic Weisbecker wrote:
> That makes sense. So there would be a simple cpuset.isolation that can
> be either 1 or 0 where 1 has all possible isolation stuff on. Then
> if the need arises we can provide more tuning through a new specific
> cgroup controller, right?

Given that there isn't much that is hierarchical about them, I'm pretty
skeptical about introducing a new controller or fancy hierarchical interface
for it. If isolation is intertwined with cpuset partitioning and a simple
knob for it fits well with the rest of configuration, yeah, but let's please
try to avoid maximizing the interface. We want the interface to encode
users' intentions (e.g., here, I want these cpus isolated) not the
implementation details to make that happen. Of course, there are gradients
but it becomes really ugly when you try to expose low level details on
cgroups because of the implied flexibility (I can organize however I want
hierarchically and the controls must nest and be delegatable properly).

So, If you think isolation feature will need lots of low level knobs
exposed, cgroup isn't the right place. It should be something simpler and
lower level. This probably is a good time to spend some time thinking how
it'd look like, say, five years down the line. If it's gonna be the "I want
isolation" knob + maybe some obscure system wide knobs that most people
don't need to think about, it's gonna be fine. Otherwise, we shouldn't put
this in cgroup until we have better ideas on what the interface should look
like.

Thanks.

-- 
tejun