On Mon, Oct 24, 2016 at 10:57:16AM -0700, Dan Williams wrote:

...but it still might be the better way to go. I.e. push the hacks to
the legacy driver and not burden nvme with this new "nvme_dev_ops"
concept.

Let's get some coherent text on how this device even operates out
first and then we can decide if and how we can support it.