Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH V11 0/4] blk-mq: refactor code of issue directly
To:     Jens Axboe <axboe@kernel.dk>
Cc:     ming.lei@redhat.com, linux-block@vger.kernel.org,
        linux-kernel@vger.kernel.org
References: <1544152185-32667-1-git-send-email-jianchao.w.wang@oracle.com>
 <b2bcbb76-af6f-7660-2bc3-9796338cccff@kernel.dk>
 <0adf3419-bcce-93d8-51fb-aee7cbb5ae17@oracle.com>
 <16205e68-aa5e-c59d-364e-4164a0e51dc7@kernel.dk>
 <f052c016-e521-9f12-f4e5-d65f3814643a@kernel.dk>
 <a59a3a1d-94fb-938d-7ddf-969185cd5b48@oracle.com>
 <1e183b77-2c4d-71ff-b019-2b1070d2ed6b@kernel.dk>
From:   "jianchao.wang" <jianchao.w.wang@oracle.com>
Message-ID: <38eb7c50-dfad-d9cb-f8ab-a8f5250b0ed7@oracle.com>
Date:   Fri, 7 Dec 2018 11:46:43 +0800
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.2.1
MIME-Version: 1.0
In-Reply-To: <1e183b77-2c4d-71ff-b019-2b1070d2ed6b@kernel.dk>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


On 12/7/18 11:42 AM, Jens Axboe wrote:
> On 12/6/18 8:41 PM, jianchao.wang wrote:
>>
>>
>> On 12/7/18 11:34 AM, Jens Axboe wrote:
>>> On 12/6/18 8:32 PM, Jens Axboe wrote:
>>>> On 12/6/18 8:26 PM, jianchao.wang wrote:
>>>>>
>>>>>
>>>>> On 12/7/18 11:16 AM, Jens Axboe wrote:
>>>>>> On 12/6/18 8:09 PM, Jianchao Wang wrote:
>>>>>>> Hi Jens
>>>>>>>
>>>>>>> Please consider this patchset for 4.21.
>>>>>>>
>>>>>>> It refactors the code of issue request directly to unify the interface
>>>>>>> and make the code clearer and more readable.
>>>>>>>
>>>>>>> This patch set is rebased on the recent for-4.21/block and add the 1st
>>>>>>> patch which inserts the non-read-write request to hctx dispatch
>>>>>>> list to avoid to involve merge and io scheduler when bypass_insert
>>>>>>> is true, otherwise, inserting is ignored, BLK_STS_RESOURCE is returned
>>>>>>> and the caller will fail forever.
>>>>>>>
>>>>>>> The 2nd patch refactors the code of issue request directly to unify the
>>>>>>> helper interface which could handle all the cases.
>>>>>>>
>>>>>>> The 3rd patch make blk_mq_sched_insert_requests issue requests directly
>>>>>>> with 'bypass' false, then it needn't to handle the non-issued requests
>>>>>>> any more.
>>>>>>>
>>>>>>> The 4th patch replace and kill the blk_mq_request_issue_directly.
>>>>>>
>>>>>> Sorry to keep iterating on this, but let's default to inserting to
>>>>>> the dispatch list if we ever see busy from a direct dispatch. I'm fine
>>>>>> with doing that for 4.21, as suggested by Ming, I just didn't want to
>>>>>> fiddle with it for 4.20. This will prevent any merging on the request
>>>>>> going forward, which I think is a much safer default.
>>>>>>
>>>>>> You do this already for some cases. Let's do it unconditionally for
>>>>>> a request that was ever subjected to ->queue_rq() and we didn't either
>>>>>> error or finish after the fact.
>>>>>>
>>>>> I have done it in this version if I get your point correctly.
>>>>> Please refer to the following fragment in the 2nd patch.
>>>>>
>>>>> +	/*
>>>>> +	 * If the request is issued unsuccessfully with
>>>>> +	 * BLK_STS_DEV_RESOURCE or BLK_STS_RESOURCE, insert
>>>>> +	 * the request to hctx dispatch list due to attached
>>>>> +	 * lldd resource.
>>>>> +	 */
>>>>> +	force = true;
>>>>> +	ret = __blk_mq_issue_directly(hctx, rq, cookie, last);
>>>>> +out_unlock:
>>>>> +	hctx_unlock(hctx, srcu_idx);
>>>>> +out:
>>>>> +	switch (ret) {
>>>>> +	case BLK_STS_OK:
>>>>> +		break;
>>>>> +	case BLK_STS_DEV_RESOURCE:
>>>>> +	case BLK_STS_RESOURCE:
>>>>> +		if (force) {
>>>>> +			blk_mq_request_bypass_insert(rq, run_queue);
>>>>> +			ret = bypass ? BLK_STS_OK : ret;
>>>>> +		} else if (!bypass) {
>>>>> +			blk_mq_sched_insert_request(rq, false,
>>>>> +						    run_queue, false);
>>>>> +		}
>>>>> +		break;
>>>>> +	default:
>>>>
>>>> You are right, I missed that you set force = true before doing the
>>>> issue. So this looks good to me!
>>>
>>> I applied your series. With this, we should be good to remove the
>>> REQ_NOMERGE logic that was added for the corruption case, and the
>>> blk_rq_can_direct_dispatch() as well?
>>>
>>
>> Yes, it should be that.
>> Every thing rejected by .queue_rq is ended or inserted into hctx dispatch
>> list. And also direct-issue path is unified with normal path.
> 
> Why are we doing that return value dance, depending on whether this
> is a bypass insert or not? That seems confusing.
> 

For the 'bypass == false' case, it need to know whether the request is issued
successfully. This is for the 3rd patch.
I used to use the returned cookie to identify the result, but you don't like it.
So I have to use this return value.

Thanks
Jianchao