[PATCH 6/7] nvme: properly handle partially initialized queues in nvme_create_io_queues

[Christoph Hellwig]

On Fri, Oct 02, 2015 at 09:36:26PM +0000, Keith Busch wrote:
> On Fri, 2 Oct 2015, Christoph Hellwig wrote:
>> This avoids having to clean up later in a seemingly unrelated place.
>>
>> Signed-off-by: Christoph Hellwig <hch at lst.de>
>
> Somewhat related to this area of code but not a new concern, I think we're
> in trouble if the controller resets after we have active request_queue,
> but controller allows fewer IO queues than originally configured with
> the blk-mq tagset. On the other hand, we don't have a way to provide
> additional IO queues to the tagset after a reset either, so I think we
> need a new blk-mq API to notify it of changes to number of h/w contexts.

Yes, that's indeed an issue.  I think we need to mark the controller dead
in this case until we have such an API.