[PATCH 2/2] nvme: pass BLK_MQ_F_NOT_USE_MANAGED_IRQ for fc/rdma/tcp/loop

[Ming Lei]

On Wed, Jun 30, 2021 at 10:15:46AM +0200, Hannes Reinecke wrote:
> On 6/29/21 9:49 AM, Ming Lei wrote:
> > All the 4 host drivers don't use managed irq for completing request, so
> > it is correct to pass the flag to blk-mq.
> > 
> > Secondly with this flag, blk-mq will help us dispatch connect request
> > allocated via blk_mq_alloc_request_hctx() to driver even though all
> > CPU in the specified hctx's cpumask are offline.
> > 
> How is this supposed to work?
> To my understanding, only cpus in the hctx cpumask are eligible to send I/O.
> So if all of these CPUs are offline, who exactly is submitting I/O?
> More to the point, which CPU will be submitting the connect request?

Please see __blk_mq_delay_run_hw_queue(), which will pass WORK_CPU_UNBOUND 
to kblockd_mod_delayed_work_on() for this situation.


Thanks,
Ming