[PATCH v2 06/10] nvmet_fc: Reduce work_q count

Christoph Hellwig hch at infradead.org
Tue May 16 05:48:21 PDT 2017


On Mon, May 15, 2017 at 05:10:20PM -0700, James Smart wrote:
> Instead of a work_q per controller queue, make 1 per cpu and have
> controller queues post work elements to the work_q.

Having per-cpu workqueues doesn't make any sense at all as workqueues
already are per-cpu underneath.  Please use a global workqueue
instead.  That also avoids the sort of messy interactions with
CPU hotplug that would have to be added to this patch otherwise.



More information about the Linux-nvme mailing list