[bug report] blktests nvme/tcp triggered WARNING at kernel/workqueue.c:2628 check_flush_dependency+0x110/0x14c

[Sagi Grimberg]

>>> Hello
>>>
>>> I reproduced this issue on the linux-block/for-next, pls help check
>>> it, feel free to let me know if you need info/test, thanks.
>>
>>
>> These reports are making me tired... Should we just remove
>> MEM_RECLAIM from all nvme/nvmet workqueues are be done with
>> it?
>>
>> The only downside is that nvme reset/error-recovery will
>> also be susceptible for low-memory situation...
> 
> We can't just do that.  We need to sort out the dependency chains
> properly.

The problem is that nvme_wq is MEM_RECLAIM, and nvme_tcp_wq is
for the socket threads, that does not need to be MEM_RECLAIM workqueue.
But reset/error-recovery that take place on nvme_wq, stop nvme-tcp
queues, and that must involve flushing queue->io_work in order to
fence concurrent execution.

So what is the solution? make nvme_tcp_wq MEM_RECLAIM?