On 2020-10-06 5:55 p.m., Chaitanya Kulkarni wrote:

On 9/30/20 11:54, Logan Gunthorpe wrote:

quoted

This ensures we find the correct nvme loop device if others exist on a
given system (which is generally not expected on test systems).

Additionally, this will be required in the upcomming test nvme/037 which
will have controllers racing with ones being destroyed.

Signed-off-by: Logan Gunthorpe <logang@deltatee.com>

If I create a passthru testcase with :-

1. Create nvme-loop based nvme ctrl backed by null_blk say /dev/nvme1

2. Create a nvme-loop based passthru ctrl backed by /dev/nvme1 say nvme2.


With this patch or this series will I be able to write the testcase ?

This patch helps with that but other helpers introduced in this series
would require minor changes.

As far as I can see, you'd only have to adjust _create_nvmet_passthru()
to take an optional argument because, presently, it always uses
$_test_dev_nvme_ctrl for the backing device.

This can easily be done if and when someone writes such a test.

However, I'm not even sure right now if that test would pass in the
kernel as is -- it seems like an odd thing to do.

Logan