nvmet panics during high load

[Sagi Grimberg]

>>> Sorry for the lack of response. I havn't tested the patch.
>>> I intended to look into the issue further and suggest a simpler
>>> solution that simply calculates the maximum queue size correctly but
>>> simply didn't have the time.
>>
>> AFAICT there is no maximum calculation, it depends on the congestion
>> that the target may experience..
> 
> So how about providing a few more receive than send buffer by default
> (less than now) and dynamically allocating over the threshold?

We can do that, but it would also introduce a dma map/unmap on the I/O
path. The send buffers are small enough to keep them around when we hit
congestion I think.

We can go either way here I guess...