So far I've only lightly tested this patch. I plan on doing a lot
more testing to ensure that I can fail an nvme controller
under load and have this not only work, but also not hit
a race in that path today that I kludged up a fix for
by running multiple threads, but this might be better.