[MinnowBoard] MinnowBoard MAX 0.93 64-Bit & network lost

[John Hawley]

Inline as there's some detailed thoughts that context would be better
for, so I'm going to chop this message back some too -JH

> With 0.93 firmware on the Minnowboard Max (8111gs) I was unable to reproduce
> this issue running 16.04. 

Ok so that matches (roughly) what I was seeing, so that's "good".

> I actually have a second drive loaded with Ubuntu 14.04 (which uses kernel
> 3.16) and was able reproduce on our custom board using the iperf tests. So
> I'm not sure you'd be more likely to see this under a newer kernel. Whether
> it becomes more common is unclear -- I had to run my test for a bit longer
> (30 min test) on 14.04 before the disconnect message.

Ok I was only running for 5 minutes (as you had).  I'm going to set this
up to run overnight and we'll see if it's keeled over in the morning.
So if you were able to reproduce on an older kernel I'd *GUESS* this is
kernel agnostic, since we are talking a huge separation in kernel
versions between 16.04 and 14.04 in your setup.

> I'm a bit worried that this is a HW problem on our end, but it's still odd
> to me that 0.92 works without PCIe disconnects.

So one of the changes in 0.93 is that spread spectrum was turned on by
default in the firmware vs. 0.92 (where it was off if memory serves me
correctly).  It's possible that with 0.93 having it on is exacerbating a
problem on your board, where 0.92 wasn't.

Right now, that's my best guess - and it's testable in 0.92 if you go to:

http://wiki.minnowboard.org/UEFI#Miscellaneous_Configuration

in the UEFI menu, and set spread spectrum on, save, and reboot - you
should have the same setup as 0.93.  It might be worth testing as it
will indicate what's going on with your board better, or at the very
least negate at least one more possibility of what's going on.

> It would be nice to see Abdelghani's dmesg when the connection fails to see
> if we're running into similar problems. Unlike Abdelghani, when the NIC
> disconnects the only way I'm able to save it is a full reboot. Perhaps the
> 8111gs driver is better at recovering from this condition or he isn't seeing
> PCIe disconnects at all?

Not sure, going to have to wait for Abdelghani to respond to see if we
are looking at the same issue or not.

- John