Re: [PATCH] RFC: loop: Avoid calling blk_mq_freeze_queue() when possible.

From: Martijn Coenen <maco@android.com>
To: Ming Lei <ming.lei@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>,
	linux-block <linux-block@vger.kernel.org>,
	LKML <linux-kernel@vger.kernel.org>,
	Greg KH <gregkh@linuxfoundation.org>,
	kernel-team@android.com, Narayan Kamath <narayan@google.com>,
	Dario Freni <dariofreni@google.com>,
	Nikita Ioffe <ioffe@google.com>, Jiyong Park <jiyong@google.com>,
	Martijn Coenen <maco@google.com>
Subject: Re: [PATCH] RFC: loop: Avoid calling blk_mq_freeze_queue() when possible.
Date: Mon, 19 Aug 2019 11:06:18 +0200	[thread overview]
Message-ID: <CAB0TPYFzgm7pJvXfYYER6qqHM3J8dU7WXWv8ct51e2CGctydzw@mail.gmail.com> (raw)
In-Reply-To: <20190814113348.GA525@ming.t460p>

On Wed, Aug 14, 2019 at 1:34 PM Ming Lei <ming.lei@redhat.com> wrote:
> Another candidate is to not switch to q_usage_counter's percpu mode
> until loop becomes Lo_bound, and this way may be more clean.

I was thinking about this more; our current sequence is:

1) LOOP_SET_FD
2) LOOP_SET_STATUS64 // for lo_offset/lo_sizelimit
3) LOOP_SET_BLOCK_SIZE // to make sure block size is correct for DIO
4) LOOP_SET_DIRECT_IO // to enable DIO

I noticed that LOOP_SET_FD already tries to configure DIO if the
underlying file is opened with O_DIRECT; but in this case, it will
still fail, because the logical block size of the loop queue must
exceed the logical I/O size of the backing device. But the default
logical block size of mq is 512.

One idea to fix is to call blk_queue_logical_block_size() as part of
LOOP_SET_FD, to match the block size of the backing fs in case the
backing file is opened with O_DIRECT; you could argue that if the
backing file is opened with O_DIRECT, this is what the user wanted
anyway. This would allow us to get rid of the latter two ioctl's and
already save quite some time.

Thanks,
Martijn

>
> Something like the following patch:
>
> diff --git a/drivers/block/loop.c b/drivers/block/loop.c
> index a7461f482467..8791f9242583 100644
> --- a/drivers/block/loop.c
> +++ b/drivers/block/loop.c
> @@ -1015,6 +1015,9 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
>          */
>         bdgrab(bdev);
>         mutex_unlock(&loop_ctl_mutex);
> +
> +       percpu_ref_switch_to_percpu(&lo->lo_queue->q_usage_counter);
> +
>         if (partscan)
>                 loop_reread_partitions(lo, bdev);
>         if (claimed_bdev)
> @@ -1171,6 +1174,8 @@ static int __loop_clr_fd(struct loop_device *lo, bool release)
>         lo->lo_state = Lo_unbound;
>         mutex_unlock(&loop_ctl_mutex);
>
> +       percpu_ref_switch_to_atomic(&lo->lo_queue->q_usage_counter, NULL);
> +
>         /*
>          * Need not hold loop_ctl_mutex to fput backing file.
>          * Calling fput holding loop_ctl_mutex triggers a circular
> @@ -2003,6 +2008,12 @@ static int loop_add(struct loop_device **l, int i)
>         }
>         lo->lo_queue->queuedata = lo;
>
> +       /*
> +        * cheat block layer for not switching to q_usage_counter's
> +        * percpu mode before loop becomes Lo_bound
> +        */
> +       blk_queue_flag_set(QUEUE_FLAG_INIT_DONE, lo->lo_queue);
> +
>         blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
>
>         /*
>
>
> thanks,
> Ming