On Wed, May 13, 2020 at 01:58:32PM +0200, Christoph Hellwig wrote: > I think the flag should be inverted, indicated managed irqs if set. > And we should fine a a way to automatically set it from the managed > IRQ blk_mq_*_map_queues helpers instead of leaving the decisions to blk_mq_*_map_queues doesn't tell us if managed IRQ is used, only the driver knows that. > the driver author that is most likely going to get it wrong, especially > for SCSI, where the actual driver can't even get at the current flag. > BLK_MQ_F_NO_MANAGED_IRQ is just for avoiding deadlock in draining IO during cpu hotplug. So far, only stacking blk-mq drivers need this flag. That is why the flag is named as NO_MANAGED_IRQ, or we may add flag of BLK_MQ_F_STACKING for this purpose too. thanks, Ming
