nvme-tcp: kernel NULL pointer dereference, address: 0000000000000034

[Sagi Grimberg]

>>>> admin_tagset.nr_maps = 1 (only the default map, no read, no poll)
>>>
>>> Indeed, that would be to easy.
>>>
>>> I've just triggered a crash where we are passing in a non-null bio. Some
>>> more annotation. This time I am printing from blk_rq_is_poll() and
>>> we see that that is also the case where we have a valid bio but
>>> want to use the poll context:
>>
>> That is not a crash, but a WARN stack dump.
> 
> Not sure how you get to this conclusion.
> 
>> It is still unclear to me
>> how exactly you get to poll for a bio-less request.
> 
> I don't do anything special here. The only thing which is special is that I am
> testing against Linux soft target with Hannes TP8013 patches.
> 
>> See my other reply, While I removed the below warning, and allow
>> bio-less request polling, I was not able to observe any bio-less
>> requests actually being polled.
> 
> Unfortunatly, somehow all your inline patches receive my inbox whitespace
> damaged. Takes a few minutes to patch it manually. Anyway, with your patch
> the crash is gone.

Can you verify that you never see a bio-less requests is being polled?