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;
Subject: Re: [PATCH -next 2/4] blk-iocost: don't throttle bio if iocg is
 offlined
To:     Tejun Heo <tj@kernel.org>, Yu Kuai <yukuai1@huaweicloud.com>
Cc:     hch@infradead.org, josef@toxicpanda.com, axboe@kernel.dk,
        cgroups@vger.kernel.org, linux-block@vger.kernel.org,
        linux-kernel@vger.kernel.org, yi.zhang@huawei.com,
        "yukuai (C)" <yukuai3@huawei.com>
References: <20221217030527.1250083-1-yukuai1@huaweicloud.com>
 <20221217030527.1250083-3-yukuai1@huaweicloud.com>
 <Y6DXkLeOmu7VWovz@slm.duckdns.org>
From:   Yu Kuai <yukuai1@huaweicloud.com>
Message-ID: <beb75b37-6088-5f23-6602-5aff22213e41@huaweicloud.com>
Date:   Tue, 20 Dec 2022 17:38:05 +0800
User-Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; rv:60.0) Gecko/20100101
 Thunderbird/60.8.0
MIME-Version: 1.0
In-Reply-To: <Y6DXkLeOmu7VWovz@slm.duckdns.org>
Content-Type: text/plain; charset=gbk; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

Hi,

?? 2022/12/20 5:28, Tejun Heo д??:
> On Sat, Dec 17, 2022 at 11:05:25AM +0800, Yu Kuai wrote:
>> From: Yu Kuai <yukuai3@huawei.com>
>>
>> bio will grab blkg reference, however, blkcg->online_pin is not grabbed,
>> hence cgroup can be removed after thread exit while bio is still in
>> progress. Bypass io in this case since it doesn't make sense to
>> throttle bio while cgroup is removed.
> 
> I don't get it. Why wouldn't that make sense? ISTR some occasions where we
> clear the config to mitigate exits stalling for too long but in general a
> policy being active on a draining cgroup shouldn't be a problem.

The main reason here for patch 2/3 is for patch 4, since bio can still
reach rq_qos after pd_offline_fn is called.

Currently, it's not consistent and seems messy how different policies
implement pd_alloc/free_fn, pd_online/offline_fn, and pd_init_fn. For
iocost, iocg is exited in pd_free_fn, and parent iocg can exits before
child, which will cause many problems.

Patch 2/3 are not necessary if we don't choose to fix such problems by
exiting iocg in ioc_pd_offline() in patch 4.

I'll try to think about how to use refcnting, either from blkg layer or
add refcnting for iocg.

Thanks,
Kuai
> 
> Thanks.
>