On Sat, Jan 22, 2022 at 07:10:49PM +0800, Ming Lei wrote: > blk_mq_unfreeze_queue() is used by scsi when releasing disk, so not > necessary to unfreeze into percpu mode, then the following > blk_cleanup_queue doesn't need to freeze queue from percpu mode, and > the implied RCU grace period may be avoided. > > Meantime move clearing QUEUE_FLAG_INIT_DONE into this API, so that > when one disk is added, ->q_usage_counter can be switched to percpu > mode again. Independ of the use case (which I'll need to review in the next patches), exporting lowlevel-helpers like this just seems like a bad idea. If we have a use case add a wrapper that sets the force_atomic flag and document and export that, please.
