I wish we could have blktests for tearing down device-drivers; doing a regression test here will be really hard.The problem with a remove is that while we have a generic device remove attribute, it: a) isn't always in the same place relatively to the disk b) once removed we have no generic way to add the device back for further testing nvme/032 has basic remove testing for nvme, and I think I can also wire up my scsi bind/unbind testing for blktests using scsi_debug without too much effort. But that still isn't exactly a generic test.
It should be trivial to add a similar test for fabrics with toggling nvmet $ns/enable config knob...
