Corrupted BMS-Can values on Cerbo GX traced to Ethernet duplex mismatch

Sharing this in case it saves someone a month of troubleshooting. In my case a network problem on the Cerbo’s Ethernet port showed up as corrupted battery data on the BMS-Can port.

System

  • Cerbo GX, Venus OS v3.75 → v3.80 (no difference between versions)
  • 9 × Pylontech US5000 on the BMS-Can port (BMS firmware v1.6 as reported by the GX)
  • MultiPlus-II 3-phase, ESS with Dynamic ESS
  • Home Assistant reading the GX over Modbus TCP, later also over MQTT

Symptom

Every now and then a single sample from the battery was physically impossible, while the readings right before and after were normal:

  • Min/max cell temperature −273 °C, i.e. a raw value of 0 K
  • Max cell voltage at the same moment 0.27–0.76 V or 12.7–14.5 V, with the max cell ID jumping to 0101
  • Separately, battery current spikes of ~1300–2040 A

The same spikes were visible in VRM, so it was not a Home Assistant or MQTT issue. My first suspect was the battery BMS, and I opened a ticket with Pylontech.

It got worse as network traffic grew

Counting hours with at least one anomaly, from Home Assistant long-term statistics:

Period Anomaly hours
Dec 2024 – Jul 2025 0
Aug – Nov 2025 a few, plus one cluster in late Nov
Dec 2025 – May 2026 0–1 per month
Jun 2026 (MQTT added alongside Modbus on 22 Jun) 2
Jul 2026 5
Aug 2026 8
Sep 2026 (two more MQTT clients added mid-month) 27, plus 16 with current spikes
1–7 Oct 2026 15
After 7 Oct 2026 (fix applied) 0

Each time network traffic to the GX increased, the anomalies became more frequent. Looking back, the extra traffic made an existing Ethernet problem worse rather than causing the anomalies itself.

The BMS-Can counters looked almost healthy

Settings → Connectivity → BMS-Can port → Network status showed no bus-off, error-passive or bus-warning events. RX overruns grew slowly and steadily at about 1.3 per hour (about 0.006 % of frames). Because the rate was so low, I first dismissed overruns as a cause.

The actual cause was only visible over SSH

Nothing in the GUI pointed to the network. With root SSH access, dmesg was full of Ethernet TX abort messages:

[804212.769746] sun4i-emac 1c0b000.ethernet eth0:  ab : 8
[804212.775000] sun4i-emac 1c0b000.ethernet eth0:  ab : 40104
[804212.782851] sun4i-emac 1c0b000.ethernet eth0:  ab : 8
[804212.789304] sun4i-emac 1c0b000.ethernet eth0:  ab : 8

The UniFi switch port had been forced to 100 Mbps full duplex. The Cerbo uses autonegotiation, so it falls back to half duplex. The result is a classic duplex mismatch with collisions and aborted transmits. I don’t know when or why the port was forced to this setting.

After I set the switch port back to auto:

[804470.049368] sun4i-emac 1c0b000.ethernet eth0: Link is Down
[804471.189313] sun4i-emac 1c0b000.ethernet eth0: Link is Up - 100Mbps/Full - flow control off

Result

  • RX overruns on the BMS-Can port stopped (238 → 290 over 12 days; the previous rate would have predicted ~620)
  • No corrupted battery values at all for 12 days and counting

My interpretation (not verified at driver level)

The Cerbo GX is based on an Allwinner A20, and its sun4i-emac Ethernet driver moves frames using the CPU. A constant stream of Ethernet errors therefore costs CPU time in interrupt context. When the CPU couldn’t service the CAN controller in time, the receive buffer overflowed. Some of those overflows seem to have produced a corrupted frame instead of a cleanly dropped one, with zeros or bytes mixed from different frames. That would explain why overruns were about 20× more common than visible anomalies: only some overflows corrupted a frame. If someone from Victron can confirm how the BMS-Can controller handles RX overflow, I’d be interested.

Side note: when the link came back up, hub4control crashed once with a segfault (signal 11). daemontools restarted it right away, and it has not happened again.

[804473.639298] hub4control: hub4control: potentially unexpected fatal signal 11.
[804473.646803] CPU: 0 UID: 0 PID: 3690 Comm: hub4control Tainted: G           O       6.12.90-venus-4 #1
[804473.660020] Hardware name: Allwinner sun7i (A20) Family
[804473.665430] PC is at 0xc8

If you see similar phantom BMS values

  1. Check the switch port for the Cerbo: use autonegotiation, never forced speed or duplex.
  2. Enable SSH (Settings → General, access level Superuser, SSH on LAN) and run dmesg | grep eth0. Look for ab : lines or repeated link changes.
  3. Run ip -s link show eth0 and look for errors or collisions.
  4. Check the overrun counter under BMS-Can port → Network status and see whether it keeps growing.

A suggestion for Victron: a duplex or link error indicator in the GUI, or a notification when the Ethernet driver logs errors, would have made this findable without SSH.

2 Likes