bfq: dispatch request to prevent queue stalling after the request completion
authorHou Tao <houtao1@huawei.com>
Tue, 11 Jul 2017 13:58:15 +0000 (21:58 +0800)
committerJens Axboe <axboe@kernel.dk>
Wed, 12 Jul 2017 14:32:04 +0000 (08:32 -0600)
commit3f7cb4f4130ca5693f0698211bf1e42945efbb8c
tree7aecf67d6851b4213a89bb1e632b42878ee3b2c8
parent38c9140740eb0993924b676a111c164903163b1e
bfq: dispatch request to prevent queue stalling after the request completion

There are mq devices (eg., virtio-blk, nbd and loopback) which don't
invoke blk_mq_run_hw_queues() after the completion of a request.
If bfq is enabled on these devices and the slice_idle attribute or
strict_guarantees attribute is set as zero, it is possible that
after a request completion the remaining requests of busy bfq queue
will stalled in the bfq schedule until a new request arrives.

To fix the scheduler latency problem, we need to check whether or not
all issued requests have completed and dispatch more requests to driver
if there is no request in driver.

The problem can be reproduced by running the following script
on a virtio-blk device with nr_hw_queues as 1:

#!/bin/sh

dev=vdb
# mount point for dev
mp=/tmp/mnt
cd $mp

job=strict.job
cat <<EOF > $job
[global]
direct=1
bs=4k
size=256M
rw=write
ioengine=libaio
iodepth=128
runtime=5
time_based

[1]
filename=1.data

[2]
new_group
filename=2.data
EOF

echo bfq > /sys/block/$dev/queue/scheduler
echo 1 > /sys/block/$dev/queue/iosched/strict_guarantees
fio $job

Signed-off-by: Hou Tao <houtao1@huawei.com>
Reviewed-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
block/bfq-iosched.c