On Fri, Sep 17, 2021 at 04:32:15PM +0800, Ming Lei wrote: > > Is it? For the typical case the second free in blk_cleanp_queue will > > be bsically free because there is no pending I/O. The only case > > where that matters if if there is pending passthrough I/O again, > > which can only happen with SCSI, and even there is highly unusual. > > RCU grace period is involved in blk_freeze_queue(). where? > One way you can > avoid it is to keep the percpu ref at atomic mode when running > blk_mq_unfreeze_queue() in del_gendisk(). > > > > Thanks, > Ming ---end quoted text---
