Chronic 2.5GbE link flapping + autoneg downshift-latching on quad Intel I226-V (N5105 board, Proxmox, igc) — has anyone actually fixed this with a driver?
retained text
TL;DR: my 4x I226-V 2.5GbE ports keep dropping/retraining the link and occasionally downshift to 100 Mbps for days. Ruled out cables, EEE, router, and driver version. Has anyone fixed this at 2.5G — including with a patched or self-written driver? **Hardware:** one of the Topton N5105 mini-ITX boards with quad Intel I226-V 2.5GbE (rev 04 — the Topton/CWWK-style ones), running Proxmox VE (stock `igc` driver, kernel 7.0.14-15-pve). All four ports in one VLAN-aware Linux bridge; one port is the LAN uplink to a 2.5GbE router. **Symptoms** (weeks now, at least since late August, across different ports and link partners): - The 2.5G link retrains constantly: `NIC Link is Down` → ~3 s later `NIC Link is Up 2500 Mbps Full Duplex`. Up to ~70 events/day, in bursts, around the clock (04:00 included). Every few days a 25–30 s outage that comes back with a _different_ flow-control result (RX vs RX/TX) — autoneg landing on different outcomes run after run. - The ugly variant: it downshifts and **latches**. Link lands at 1000 or 100 Mbps and then sits there for days — rock stable, zero errors — until `ethtool -r` reclaims it. Two ports have latched at 100M; the 2.5G port latches at 1000. - Sometimes the MAC gives up entirely: full `Register Dump` + `Reset adapter` in the kernel log, plus the odd `exceed max 2 second`. **Ruled out:** - EEE — disabled, LPI counters zero - Cable/copper — exactly 1 CRC error in 37 days; the behavior follows the _port_, not the cable; three ports, three partners, different cables - The router — same disease against a GL.iNet AX1800, a MikroTik hAP be³, and an older uplink; one evening a second port on a different link flapped within seconds of the first - Driver version — identical signature under two kernel builds (7.0.14-12 and -15); it predates the kernel upgrade - Offloads — TSO on/off makes no difference So the evidence points at the I226 PHYs themselves (or board-level signal integrity/power). I understand a driver only writes the autoneg advertisement and retry logic — it can't make a marginal PHY _hold_ link. But drivers have surprised me before, and these boards' i225-V predecessors had a well-earned reputation for exactly this kind of nonsense — I thought the I226 was supposed to be the fixed one. **What's the fix?** 1. Has anyone made i226 2.5G stable via a specific kernel/igc version? Which one? 2. Has anyone patched `igc` — or written their own driver/PHY handling for these chips? 3. Has anyone flashed newer NVM/firmware onto these boards? Intel's tools look OEM/UEFI-only. Really hoping there's a fix out there.
document fetched 2026-11-01 00:28 · canonical https://reddit.com/r/selfhosted/comments/1xkelym/chronic_25gbe_link_flapping_autoneg