On Thu, 2019-03-21 at 13:43 +-0800, zhengbin wrote: +AD4 diff --git a/drivers/scsi/scsi+AF8-sysfs.c b/drivers/scsi/scsi+AF8-sysfs.c +AD4 index 6a9040f..342e513 100644 +AD4 --- a/drivers/scsi/scsi+AF8-sysfs.c +AD4 +-+-+- b/drivers/scsi/scsi+AF8-sysfs.c +AD4 +AEAAQA -771,9 +-771,20 +AEAAQA store+AF8-state+AF8-field(struct device +ACo-dev, struct device+AF8-attribute +ACo-attr, +AD4 +AD4 mutex+AF8-lock(+ACY-sdev-+AD4-state+AF8-mutex)+ADs +AD4 ret +AD0 scsi+AF8-device+AF8-set+AF8-state(sdev, state)+ADs +AD4 +- /+ACo If device state changes to SDEV+AF8-RUNNING, we need to run hw queue +AD4 +- +ACo to avoid io hung. +AD4 +- +ACo-/ +AD4 +- if ((ret +AD0APQ 0) +ACYAJg (state +AD0APQ SDEV+AF8-RUNNING)) +AD4 +- goto out+AF8-run+AF8-hw+AF8-queue+ADs +AD4 mutex+AF8-unlock(+ACY-sdev-+AD4-state+AF8-mutex)+ADs +AD4 +AD4 return ret +AD0APQ 0 ? count : -EINVAL+ADs +AD4 +- +AD4 +-out+AF8-run+AF8-hw+AF8-queue: +AD4 +- mutex+AF8-unlock(+ACY-sdev-+AD4-state+AF8-mutex)+ADs +AD4 +- blk+AF8-mq+AF8-run+AF8-hw+AF8-queues(sdev-+AD4-request+AF8-queue, true)+ADs +AD4 +- +AD4 +- return count+ADs +AD4 +AH0 Having two different code paths that call mutex+AF8-unlock() is unfortunate because I don't think that is necessary to have two such paths, whether or not blk+AF8-mq+AF8-run+AF8-hw+AF8-queues() is called with the state mutex held. If you repost this patch, please remove the superfluous parentheses from the if-condition. Thanks, Bart.
