On Wed, Dec 07, 2016 at 05:46:34PM -0500, Dan Streetman wrote: > > Is there a reason you want to share the interrupt between the queues?
The admin queue is hardly ever used (if at all) compared to an IO queue's usage, so why waste the resource? I bet you can't measure a preformance difference on queues that don't share, and it's reasonable to expect an nvme controller's MSI-x count matches the IO queue count, which forces us to share a vector.
