[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <3ef65dbf-6a75-45f5-b9e4-af2d6603c581@xiaomi.com>
Date: Thu, 5 Sep 2024 03:05:09 +0000
From: 章辉 <zhanghui31@...omi.com>
To: Bart Van Assche <bvanassche@....org>, "axboe@...nel.dk" <axboe@...nel.dk>
CC: "linux-block@...r.kernel.org" <linux-block@...r.kernel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
方翔 <fangxiang@...omi.com>, 王辉
<wanghui33@...omi.com>
Subject: Re: [External Mail]Re: [PATCH v3] block: move non sync requests
complete flow to softirq
On 2024/9/5 1:22, Bart Van Assche wrote:
> On 9/3/24 8:35 PM, 章辉 wrote:
>> Does set rq_affinity to 2 mean QUEUE_FLAG_SAME_COMP?
>
> From block/blk-sysfs.c:
>
> if (val == 2) {
> blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
> blk_queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
> }
>
>> This seems to determine on which core the current request is
>> completed, rather than in the interrupt top or bottom half?
>
> That's correct. I suggested this because I was wondering whether
> spreading the I/O completion workload over more CPU cores would help?
>
> Thanks,
>
> Bart.
>
hi Bart,
Increasing the CPU frequency or directly binding the interrupt to the
big core processing will definitely help the end io processing delay,
depending on the system's requirements for IO response delay.
But it does not seem to conflict with this modification to put the
time-consuming operation in the lower half.
Thanks
Zhang
Powered by blists - more mailing lists