On Thu, 2019-03-21 at 13:43 +0800, zhengbin wrote: > diff --git a/drivers/scsi/scsi_sysfs.c b/drivers/scsi/scsi_sysfs.c > index 6a9040f..342e513 100644 > --- a/drivers/scsi/scsi_sysfs.c > +++ b/drivers/scsi/scsi_sysfs.c > @@ -771,9 +771,20 @@ store_state_field(struct device *dev, struct > device_attribute *attr, > > mutex_lock(&sdev->state_mutex); > ret = scsi_device_set_state(sdev, state); > + /* If device state changes to SDEV_RUNNING, we need to run hw queue > + * to avoid io hung. > + */ > + if ((ret == 0) && (state == SDEV_RUNNING)) > + goto out_run_hw_queue; > mutex_unlock(&sdev->state_mutex); > > return ret == 0 ? count : -EINVAL; > + > +out_run_hw_queue: > + mutex_unlock(&sdev->state_mutex); > + blk_mq_run_hw_queues(sdev->request_queue, true); > + > + return count; > }
Having two different code paths that call mutex_unlock() is unfortunate because I don't think that is necessary to have two such paths, whether or not blk_mq_run_hw_queues() is called with the state mutex held. If you repost this patch, please remove the superfluous parentheses from the if-condition. Thanks, Bart.
