[PATCH] nvme: return the whole CQE through the request passthrough interface

[Christoph Hellwig]

On Mon, Feb 08, 2016 at 08:47:42PM +0100, Matias Bjørling wrote:
> When LightNVM send asynchronous data commands, it would need to allocate 
> extra memory for cqe. Would it make sense to pass cqe directly and assume 
> that it is live through the blk_mq_complete_request call?

The CQE does not survive until ->end_io is called, both in the current
PCI driver, and in the Fabrics drivers.  So we need some caller provided
storage for it.