Re: [PATCH] nvme-core: Fix subsystem instance mismatches

[Logan Gunthorpe]

On 2019-09-04 11:14 a.m., Keith Busch wrote:
> On Wed, Sep 04, 2019 at 11:01:22AM -0600, Logan Gunthorpe wrote:
>> Oh, yes that's simpler than the struct/kref method and looks like it
>> will accomplish the same thing. I did some brief testing with it and it
>> seems to work for me (though I don't have any subsystems with multiple
>> controllers). If you want to make a patch out of it you can add my
>>
>> Reviewed-by: Logan Gunthorpe <logang@xxxxxxxxxxxx>
> 
> Thanks! I'll make it a proper patch and send shortly.
> 
> For testing multi-controller subsystems, I haven't got proper hardware
> either, so I really like the nvme loop target. Here's a very simple json
> defining a two namespace subsystem backed by two real nvme devices:

Cool right, thanks for the tip, I should have thought of that. I just
did some more loop testing with your patch and it behaves roughly as we
expect. The controller and subsystem IDs never overlap unless they are
created at the same time and it doesn't look like any IDs are ever
leaked. With simple non-CMIC devices the ctrl and subsystem always have
the same instance number.

Logan