Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.32 as permitted sender) client-ip=23.128.96.32;
Message-ID: <8de482b5-1942-4312-8de4-6f54565ab517@redhat.com>
Date:   Tue, 28 Nov 2023 13:32:53 -0500
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH-cgroup 2/2] cgroup/cpuset: Include isolated cpuset CPUs in
 cpu_is_isolated() check
Content-Language: en-US
To:     Tejun Heo <tj@kernel.org>
Cc:     Zefan Li <lizefan.x@bytedance.com>,
        Johannes Weiner <hannes@cmpxchg.org>,
        Andrew Morton <akpm@linux-foundation.org>,
        Michal Hocko <mhocko@suse.com>,
        Frederic Weisbecker <frederic@kernel.org>,
        cgroups@vger.kernel.org, linux-kernel@vger.kernel.org,
        Mrunal Patel <mpatel@redhat.com>,
        Ryan Phillips <rphillips@redhat.com>,
        Brent Rowsell <browsell@redhat.com>,
        Peter Hunt <pehunt@redhat.com>
References: <20231127041956.266026-1-longman@redhat.com>
 <20231127041956.266026-3-longman@redhat.com>
 <ZWYbqNnnt6gQOssK@slm.duckdns.org>
From:   Waiman Long <longman@redhat.com>
In-Reply-To: <ZWYbqNnnt6gQOssK@slm.duckdns.org>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk


On 11/28/23 11:56, Tejun Heo wrote:
> Hello,
>
> On Sun, Nov 26, 2023 at 11:19:56PM -0500, Waiman Long wrote:
>> +bool cpuset_cpu_is_isolated(int cpu)
>> +{
>> +	unsigned int seq;
>> +	bool ret;
>> +
>> +	do {
>> +		seq = read_seqcount_begin(&isolcpus_seq);
>> +		ret = cpumask_test_cpu(cpu, isolated_cpus);
>> +	} while (read_seqcount_retry(&isolcpus_seq, seq));
>> +	return ret;
>> +}
>> +EXPORT_SYMBOL_GPL(cpuset_cpu_is_isolated);
> We're testing a bit in a bitmask. I don't think we need to worry about value
> integrity from RMW updates being broken up. ie. We can just test the bit
> without seqlock and drop the first patch?

My concern is that if we have an isolated partition with a set of 
isolated CPUs (say 2-4), I don't want any addition, deletion of changes 
made to another isolated partition affects the test of the pre-existing 
one. Testing result of the partition being change is fair game.

Depending on how the cpumask operators are implemented, we may not have 
a guarantee that testing CPU 2, for instance, will always return true. 
That is why I am adding some synchronization primitive to prevent 
racing. My original plan was to take the callback_lock. However, that 
can be somewhat costly if this API is used rather frequently, especially 
on systems with large # of CPUs. So I change it to use seqcount for read 
protection which has a much lower cost.

Regarding patch 1 on converting callback_lock to raw_spinlock_t, I can 
drop it if you have concern about that change. I just need to surround 
the write_seqcount_begin()/write_seqcount_end() calls with 
preempt_disabled()/preempt_enabled().

Cheers,
Longman