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

[Sagi Grimberg]

>>>>> Thoughts?
>>>>
>>>> It still crashes in the same way with both patches from this
>>>> disucssion applied.
>>>
>>> annoted the __nvme_submit_sync_cmd:
>>>
>>>
>>> [  210.019050] nvme nvme0: rq ffff8881122a0000 bio ffff88810d8cdd00
>>> [  210.022653] nvme nvme0: rq ffff8881122a0200 bio 0000000000000000
>>
>> Why is the admin queue created as a polling hctx?
>> It shouldn't be.
> 
> Isn't this because we use NVME_QID_ANY as qid for the register access?

The admin tagset does not have a polled hctxs map to begin with,
so I'm unclear how any fabrics or admin requests end up polled...