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;
Message-ID: <c2053491-abb6-dc75-923d-bfea81431afa@grimberg.me>
Date:   Mon, 4 Jul 2022 19:19:09 +0300
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.9.1
Subject: Re: [PATCH V3 1/1] ublk: add io_uring based userspace block driver
Content-Language: en-US
To:     Gabriel Krisman Bertazi <krisman@collabora.com>
Cc:     Ming Lei <ming.lei@redhat.com>, Jens Axboe <axboe@kernel.dk>,
        linux-block@vger.kernel.org,
        Harris James R <james.r.harris@intel.com>,
        linux-kernel@vger.kernel.org, io-uring@vger.kernel.org,
        ZiyangZhang <ZiyangZhang@linux.alibaba.com>,
        Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com>,
        Stefan Hajnoczi <stefanha@redhat.com>
References: <20220628160807.148853-1-ming.lei@redhat.com>
 <20220628160807.148853-2-ming.lei@redhat.com>
 <da861bbb-1506-7598-fa06-32201456967d@grimberg.me> <YsLeR1QWPmqfNAQY@T590>
 <8cf1aef0-ea5b-a3df-266d-ae67674c96ae@grimberg.me>
 <87a69oamap.fsf@collabora.com>
From:   Sagi Grimberg <sagi@grimberg.me>
In-Reply-To: <87a69oamap.fsf@collabora.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk


>>>>> diff --git a/drivers/block/Kconfig b/drivers/block/Kconfig
>>>>> index fdb81f2794cd..d218089cdbec 100644
>>>>> --- a/drivers/block/Kconfig
>>>>> +++ b/drivers/block/Kconfig
>>>>> @@ -408,6 +408,12 @@ config BLK_DEV_RBD
>>>>>     	  If unsure, say N.
>>>>> +config BLK_DEV_UBLK
>>>>> +	bool "Userspace block driver"
>>>>
>>>> Really? why compile this to the kernel and not tristate as loadable
>>>> module?
>>> So far, this is only one reason: task_work_add() is required, which
>>> isn't exported for modules.
>>
>> So why not exporting it?
>> Doesn't seem like a good justification to build it into the kernel.
> 
> Sagi,
> 
> If I understand correctly, the task_work_add function is quite a core
> API that we probably want to avoid exposing directly to (out-of-tree)
> modules?  I agree, though, it would be great to have this buildable as a
> module for general use cases.  Would it make sense to have it exposed
> through a thin built-in wrapper, specific to UBD, which is exported, and
> therefore able to invoke that function?  Is it a reasonable approach?

All I'm saying is that either we should expose it (or an interface to
it) if it has merit, or use something else (use a workqueue).
Having a block driver driver builtin is probably not the answer.