Flint 3 BE9300: Anyone seeing packet loss on eth0 WAN @ 2.5GbE? eth1 works but stuck at 1G fallback

fredu · July 7, 2025, 1:46pm

Hi all, (apologies for the ChatGPT-ish format.. Its quite good at summarising stuff. I am not!)

Curious if anyone else has run into this — I’d love to compare notes before pushing GL.iNet for a final fix.

Setup:

GL-BE9300 (Flint 3)
Virgin Media Hub 5 in modem mode (UK Gig1)
WAN port should negotiate 2.5GbE
Router mode with standard NAT, DHCP, MSS clamping etc.

What’s happening

eth0 (default WAN port) links up perfectly at 2.5GbE, but once the router gets a public IP and starts NAT/routing, I get severe packet loss (~70–90% loss). Makes WAN basically unusable.
Same port works perfectly in AP mode as a LAN bridge at 2.5GbE → no packet loss. So the PHY is fine — issue shows up only in WAN mode.
eth1 works great as a fallback WAN port in router mode — pings clean, no loss — but it’s a fixed 10G port with no NBase-T fallback. My modem only offers 2.5GbE so the link drops to 1G fallback, capping my real throughput at ~940Mbps (should be ~1.1Gbps).
Flow offload toggles don’t help. VLAN config is simple. No weird MTU or mangle issues.

Root cause guess

To me, this screams firmware/driver bug:

DSA switch config, flow offload, or PHY register tuning probably needs work for multi-gig WAN.
Hardware seems physically fine — it’s the WAN path that chokes.

Has anyone else seen this?

Would love to hear:

Anyone else using a Flint 3/BE9300 with a 2.5GbE modem/ONT?
Does your eth0 WAN path stay clean under real routing load?
Any tricks to get eth1 to negotiate 2.5GbE with NBase-T fallback?

Workaround for now:

Using eth1 as WAN keeps me online — but stuck at 1G because it won’t negotiate properly with the modem.

If you’ve found a fix, or hit the same problem, let me know! I’m ready to share logs if GL.iNet wants to pick this up.

Cheers.

phantasm22 · July 7, 2025, 2:49pm

Have you tried enabling "Use broadcast flag" in Luci?

is this IPv4 or v6?

You may also need to tweak the MTU to match your provider

fredu · July 7, 2025, 5:03pm

Hi, No, I havn't tried that. I have an existing GL-MT6000 which has worked flawlessly. The settings are the same on both routers (no broadcast flag, 1500MTU)

Will give it a go, but suspect that if it does anything constructive, its going to just be a sticking plaster
thanks

phantasm22 · July 7, 2025, 6:05pm

You can also try disabling network acceleration to see if that has any effect.

Just trying to understand your statement around eth1 being fixed 10G port. It's actually a 2.5GbE port and should negotiate the same as eth0 (WAN port). It could be you are seeing some sensitivity with your ethernet cable as well. Try a short distance cat6+ cable as a pure test.

fredu · July 7, 2025, 6:30pm

yup, that's one of the reasons I purchased the upgraded device (2.5G Ports throughout). Its very odd behaviour. I use 0.5Mtr shielded cat6 throughout.

WAN coming from VM modem passthrough.
Settings for eth0 [ethtool eth0] (WAN Connected to eth0: MAC address cloned to match existing MT6000 router):

        Supported ports: [ TP MII ]
        Supported link modes:   10baseT/Half 10baseT/Full
                                100baseT/Half 100baseT/Full
                                1000baseT/Full
                                2500baseT/Full
        Supported pause frame use: Symmetric Receive-only
        Supports auto-negotiation: Yes
        Supported FEC modes: Not reported
        Advertised link modes:  10baseT/Half 10baseT/Full
                                100baseT/Half 100baseT/Full
                                1000baseT/Full
                                2500baseT/Full
        Advertised pause frame use: Symmetric Receive-only
        Advertised auto-negotiation: Yes
        Advertised FEC modes: Not reported
        Link partner advertised link modes:  100baseT/Half 100baseT/Full
                                             1000baseT/Half 1000baseT/Full
                                             2500baseT/Full
        Link partner advertised pause frame use: Symmetric Receive-only
        Link partner advertised auto-negotiation: Yes
        Link partner advertised FEC modes: Not reported
        Speed: 2500Mb/s
        Duplex: Full
        Port: MII
        PHYAD: 1
        Transceiver: external
        Auto-negotiation: on
        Link detected: yes

Settings for eth1 [ethtool eth1] (Set to WAN - WAN Connected to eth1: MAC address cloned to match existing MT6000 router):
This one is weird, as in the GUI, it tels you its negotiated at 2.5G, but in reality, its actually 1G.

Only 10G is listed — no 1G, no 2.5G, no 5G fallback.
Auto-negotiation is OFF and not supported.

        Supported ports: [ ]
        Supported link modes:   10000baseT/Full
        Supported pause frame use: No
        Supports auto-negotiation: No
        Supported FEC modes: Not reported
        Advertised link modes:  10000baseT/Full
        Advertised pause frame use: No
        Advertised auto-negotiation: No
        Advertised FEC modes: Not reported
        Speed: 10000Mb/s
        Duplex: Full
        Port: MII
        PHYAD: 0
        Transceiver: internal
        Auto-negotiation: off
        Link detected: yes

Correspondingly, on my existing MT6000 that works:
eth0 (WAN) is:

Settings for eth0:
        Supported ports: [ MII ]
        Supported link modes:   2500baseT/Full
        Supported pause frame use: Symmetric Receive-only
        Supports auto-negotiation: No
        Supported FEC modes: Not reported
        Advertised link modes:  2500baseT/Full
        Advertised pause frame use: Symmetric Receive-only
        Advertised auto-negotiation: No
        Advertised FEC modes: Not reported
        Speed: 2500Mb/s
        Duplex: Full
        Port: MII
        PHYAD: 0
        Transceiver: internal
        Auto-negotiation: on
        Current message level: 0x000000ff (255)
                               drv probe link timer ifdown ifup rx_err tx_err
        Link detected: yes

eth1 (LAN) is:

        Supported ports: [ TP MII ]
        Supported link modes:   10baseT/Half 10baseT/Full
                                100baseT/Half 100baseT/Full
                                1000baseT/Full
                                2500baseT/Full
        Supported pause frame use: Symmetric Receive-only
        Supports auto-negotiation: Yes
        Supported FEC modes: Not reported
        Advertised link modes:  10baseT/Half 10baseT/Full
                                100baseT/Half 100baseT/Full
                                1000baseT/Full
                                2500baseT/Full
        Advertised pause frame use: Symmetric Receive-only
        Advertised auto-negotiation: Yes
        Advertised FEC modes: Not reported
        Link partner advertised link modes:  100baseT/Half 100baseT/Full
                                             1000baseT/Half 1000baseT/Full
                                             2500baseT/Full
        Link partner advertised pause frame use: Symmetric Receive-only
        Link partner advertised auto-negotiation: Yes
        Link partner advertised FEC modes: Not reported
        Speed: 2500Mb/s
        Duplex: Full
        Port: MII
        PHYAD: 1
        Transceiver: external
        Auto-negotiation: on
        Current message level: 0x000000ff (255)
                               drv probe link timer ifdown ifup rx_err tx_err
        Link detected: yes

Chatting to a friend, and he remembers there were similar issues on the Flint 2 with the open source firmware. They reverted to the MediaTek SDK a few weeks after release.

phantasm22 · July 7, 2025, 7:44pm

New to me but eth1 is actually an internal Qualcomm switch, which is abstracted via VLANs and switch port mappings.

If you cat your /etc/config/network you'll see the following:

config switch_vlan 'vlan_lan'
  option device 'switch1'
  option vlan '1'
  option ports '4 5 6 7 3t'

So eth1.1 is part of the br-lan bridge, and it’s tagged with VLAN 1 and is the internal uplink to the LAN switch that serves ports LAN1–LAN4

If you run swconfig dev switch1 show you'll see port 3 uplink with 10Gbit tagged as 3t and whatever port (7, 6, 5, or 4) your LAN is plugged into with the link negotiated speed (in my case 2.5Gbit). Note that port 7 = LAN 1, port 6 = LAN 2, port 5 = LAN 3, and port 4 = LAN 4

root@GL-BE9300:~# swconfig dev switch1 show

Global attributes:

Port 0:

disable: ???

pvid: 1

link: ???

Port 1:

disable: ???

pvid: 1

link: ???

Port 2:

disable: ???

pvid: 1

link: ???

Port 3:

disable: 0

pvid: 1

link: port:3 link:up speed:10000baseT full-duplex

Port 4:

disable: 0

pvid: 1

link: port:4 link:down

Port 5:

disable: 0

pvid: 1

link: port:5 link:down

Port 6:

disable: 0

pvid: 1

link: port:6 link:down

Port 7:

disable: 0

pvid: 1

link: port:7 link:up speed:2500baseT full-duplex txflow rxflow

Port 8:

disable: 0

pvid: 1

link: port:8 link:down

VLAN 1:

ports: 3t 4 5 6 7

root@GL-BE9300:~#

xplode · July 9, 2025, 7:23am

I wonder if 10G can be mapped to one port ... ill experiment eith it when my Flint 3 arrive.

phantasm22 · July 9, 2025, 2:55pm

While you could redirect the 10GbE to a specific port, you'll still most likely get a maximum speed of 2.5GbE due to the electrical/physical nature of the port. The most you would likely get is unrestricted access to the backplane at 10GbE

fredu · July 14, 2025, 10:51am

Quick update on the WAN instability some of us (? - or is it just me?!) hit:

BE-9300 + Virgin-Media Hub 5 — quick WAN update (Jul 2025)

Bug Primary 2.5 Gb WAN jack (eth0, Realtek RTL8221) links at 2.5 G but drops 70-90 % packets when cabled straight to a Hub 5 in modem-mode → unusable.
Fallback Plug the Hub 5 into the second RJ-45 (eth1, Aquantia AQR113).
Works rock-solid at 1 Gb (≈ 940 Mb/s), even though the driver falsely says “10 Gb”.
Cause (per GL.iNet) Wrong PHY-tuning file loaded for the RTL8221; AQR113 speed reporting also needs a fix.
Status I believe Engineers are building a beta firmware to restore full 2.5 Gb on eth0 and correct the speed read-outs.

TL;DR Use eth1 for WAN until the patch drops; I’ll post again when a test build is out.

bruce · July 16, 2025, 7:33am

Hi,

It seems that encountered the 2.5G network port speed negotiation compatibility problem in the first network port.
After the network port speed is reduced to 1G, the network is normal. Others Virgin users feedback this method as well.

a bit strange, we're trying to find a Virgin Hub to check this issue locally.

The physical network port 2/3/4/5 is derived from the switch chip, so ethtool eth1 will display 10000Mbps/10Gbps. I'm not sure if this is an issue, will let the R&D to check it out.