netPower 7R Lite (CSS610-1Gi-7R-2S+) - constant "Tx Deferred" errors after temporary duplex mismatch

I believe I’ve found a bug in SwOS, unless it’s some strange underlying hardware problem with the switch chipset.

I have a netPower 7R Lite (CSS610-1Gi-7R-2S+) that is powering a 60GHz radio on its Port8 / PoE out interface. The netPower 7R Lite is running SwOS 2.14, the latest available release.

On the netPower 7R Lite, the Port8 interface is showing constantly increasing “Tx Deferred” count on the “Errors” tab of SwOS Lite, at a rate of about 50 “Tx Deferred” errors per second while I have the webUI loaded and showing the Errors tab.

I was trying to figure out the cause of these errors, and correct if possible.

When initially connected to the switch, the radio on Port8 had itself been hardcoded to 100/fdx (long story). The switch was set to auto-negotiate. Obviously this would cause a duplex mis-match on the switch side. The radio was reconfigured back to auto-negotiation, and was linking to Port8 at 1Gb / FDX while I was observing this error counter increase.

More facts:

  • The radio connected to Port8 did not show any further Ethernet errors from its perspective. There were some stale counters from the initial mis-match, but they did not return after the interface was set to auto and the counters were cleared.
  • After the initial mis-match, the radio connected to Port8 was set to auto-negotiate, and was linking at 1Gb as expected. (The RF interface on that radio is actually faster than 1Gb, and is full-duplex, so the radio itself shouldn’t be a bottleneck.)
  • There are no other error counters on any other interface on the switch, and no other error counters are increasing on Port8 – only “Tx Deferred.”
  • There are only two other ports connected to the switch, and total throughput is only about 50-100Kbps as I have been watching the “Tx Deferred” increase

As best I can tell, “Tx Deferred” should be the same as other switches “deferred-tx,” which is only supposed to occur if a port is in half-duplex mode. None of the ports on the switch are reporting half-duplex.

After rebooting the switch, the error counter stopped increasing.
There seems to be some sort of bug in SwOS Lite or the way the switch chip is controlled on this device that caused it to somehow think the port was still half-duplex, I guess?

I can confirm the same issue in 2.17 - the port where it happens was reconfigured to forced 100Mb temporarily to work around link down between two gigabit ports through a 100Mb 2-pair PoE adapter (since replaced with proper gigabit 4-pair one). Back to 1Gb autonegotiation, Tx deferred counter still increasing, fixed by reboot. Some port state is apparently not reset properly on link down/up. Not testing 2.18 - waiting with upgrade until the Hosts table sytax errors are fixed.