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: <5be1cba6-b141-3a05-f801-3af7d2092674@linux.dev>
Date:   Fri, 14 Jul 2023 19:31:46 +0800
MIME-Version: 1.0
Subject: Re: [PATCH v5] blk-mq: fix start_time_ns and alloc_time_ns for
 pre-allocated rq
Content-Language: en-US
To:     Tejun Heo <tj@kernel.org>
Cc:     Jens Axboe <axboe@kernel.dk>, hch@lst.de,
        linux-block@vger.kernel.org, linux-kernel@vger.kernel.org,
        ming.lei@redhat.com, zhouchengming@bytedance.com
References: <20230710105516.2053478-1-chengming.zhou@linux.dev>
 <aa813164-9a6a-53bd-405b-ba4cc1f1b656@kernel.dk>
 <63f93f1c-98da-4c09-b3d8-711f6953d8b7@linux.dev>
 <ZLA7QAfSojxu_FMW@slm.duckdns.org>
From:   Chengming Zhou <chengming.zhou@linux.dev>
In-Reply-To: <ZLA7QAfSojxu_FMW@slm.duckdns.org>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 2023/7/14 01:58, Tejun Heo wrote:
> Hello,
> 
> On Thu, Jul 13, 2023 at 08:25:50PM +0800, Chengming Zhou wrote:
>> Ok, this version will only get time stamp once for one request, it's actually
>> not worse than the current code, which will get start time stamp once for each
>> request even in the batch allocation.
>>
>> But yes, maybe we can also set the start time stamp in the batch mode, and only
>> update the time stamp in the block case, like you said, has better performance.
>>
>> The first version [1] I posted actually just did this, in which use a nr_flush counter
>> in plug to indicate that we blocked & flushed plug. Tejun and I think it seems fragile.
>> So go to this way that only set time stamp once when the request actually used.
>>
>> [1] https://lore.kernel.org/all/20230601053919.3639954-1-chengming.zhou@linux.dev/
>>
>> Another way I can think of is to make rq_qos_throttle() return a bool to indicate
>> if it blocked. Tejun and Jens, how do you think about this way?
>>
>> Although it's better performance, in case of preemption, the time stamp maybe not accurate.
> 
> Trying to manually optimized timestamp reads seems like a bit of fool's
> errand to me. I don't think anyone cares about nanosec accuracy, so there
> are ample opportunities for generically caching timestamp so that we don't
> have to contort code to optimzie timestamp calls.
> 
> It's a bit out of scope for this patchset but I think it might make sense to
> build a timestamp caching infrastructure. The cached timestamp can be
> invalidated on context switches (block layer already hooks into them) and
> issue and other path boundaries (e.g. at the end of plug flush).
> 

Yes, this is a really great idea. It has better performance and is more generic.

Thanks.