Network issue with specific switch

I have discovered a network issue where the Jetson TX2’s network stack becomes unresponsive as soon as a specific switch is introduced on the network. This issue affects versions r30.1, r31.1, r32.1 and r32.2 but not r28.3.1 and lower. The switch that triggers this issue is a Zyxel GS1900 (48 ports).

In order to reproduce the issue one might replicate the following setup.

https://imgur.com/X7fVJ1D

Once that’s done, start pinging the TX2 from 192.168.0.1, this should work just fine. Then plug the Zyxel switch into the network as follows.

https://imgur.com/dR9y2ZE

Shortly after you have introduced the Zyxel switch on the network, the ping command should drop out provided that you’re running one of the affected versions of the rootfs on the Jetson board.

Note that the Jetson TX2 won’t come back even if you take the Zyxel switch off the network unless you take the eth0 interface down and back up again.

Ways to mitigate this issue include downgrading the rootfs to r28.x or removing the problematic switch from the network.

I’m happy to provide more details as necessary or help test potential fixes.

You may want to add the output of “ifconfig” on both the system providing the ping and the TX2. The interest is to see what shows up on each for RX and TX errors, dropped, overruns, carrier, and collisions. If these don’t show problems the debugging will be different than if they do show problems.