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;
MIME-Version: 1.0
References: <DM6PR04MB5754D8E261B4200AA62D442D860D0@DM6PR04MB5754.namprd04.prod.outlook.com>
 <20200121201014.52345-1-muraliraja.muniraju@rubrik.com> <19d3397e-f820-bae0-7e4f-93bafe7ce166@acm.org>
In-Reply-To: <19d3397e-f820-bae0-7e4f-93bafe7ce166@acm.org>
From:   Muraliraja Muniraju <muraliraja.muniraju@rubrik.com>
Date:   Wed, 22 Jan 2020 14:14:46 -0800
Message-ID: <CAByjrT-08OQPtf0ejsaBni6VGkSNGHmFmT_yp1_4H0oi-TBsPQ@mail.gmail.com>
Subject: Re: Re [PATCH] Adding multiple workers to the loop device.
To:     Bart Van Assche <bvanassche@acm.org>
Cc:     Chaitanya Kulkarni <Chaitanya.Kulkarni@wdc.com>,
        Jens Axboe <axboe@kernel.dk>, linux-block@vger.kernel.org,
        linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

I see there was a submission that was done sometime ago to use
workqueues and was reverted
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=e03a3d7a94e2485b6e2fa3fb630b9b3a30b65718
Hence used the approach of multiple workers this way. The current
change by default uses the default behaviour of 1 worker and one can
program the number of workers needed as a ioctl to scale as needed.

On Tue, Jan 21, 2020 at 8:01 PM Bart Van Assche <bvanassche@acm.org> wrote:
>
> On 2020-01-21 12:10, muraliraja.muniraju wrote:
> > +     for (i = 0; i < lo->num_loop_workers; i++) {
> > +             kthread_init_worker(&(lo->workers[i]));
> > +             lo->worker_tasks[i] = kthread_run(
> > +                             loop_kthread_worker_fn, &(lo->workers[i]),
> > +                             "loop%d(%d)", lo->lo_number, i);
> > +             if (IS_ERR((lo->worker_tasks[i])))
> > +                     goto err;
> > +             set_user_nice(lo->worker_tasks[i], MIN_NICE);
> > +     }
>
> Unless if there is a really good reason, the workqueue mechanism should
> be used instead of creating kthreads. And again unless if there is a
> really good reason, one of the system workqueues (e.g. system_wq) should
> be used instead of creating dedicated workqueues.
>
> Bart.