Severe size-dependent packet loss on CB1 onboard Ethernet (Manta M8P V2.0 + CB1)
- Dominant language
- No language data
- Stars
- 496
- Forks
- 86
- PR merge metrics
- No merged PRs in 30d
Description
Symptom: Fluidd UI hangs for seconds, webcam stream collapses after a few seconds, SSH is laggy. Bulk TCP (apt-get, iperf3) hits line rate, which masks the issue.
Hardware/OS: BTT Manta M8P V2.0 + CB1, 100 Mb/s Full link. Reproduced on CB1 Debian 12 (k6.6) and Debian 13 minimal/Klipper (k7.0).
Key data:
iperf3: 95 Mbit/s both directions
ping -M do from CB1 → router: 56B 6–8%, 500B 16–24%, 1472B 30–37% loss
ping laptop → CB1: 32B ~2%, 1472B ~40% loss
ip -s -s link: 0 CRC, 0 FIFO, 0 overrun errors despite 30%+ loss → packets dropped below the MAC
dmesg clean, no link flaps or PHY resets
Ruled out: cable (2 known-good), switch (bypassed), router port (laptop fine on same port), CPU/RAM/disk load, PSU undervoltage, USB peripherals, link speed (10/100 both lossy), NIC offloads (disabling collapses sunxi-gmac throughput).
Reconciliation: TCP at 30% loss / 1 ms RTT can't exceed ~20 Mbit/s, yet iperf3 hits 95 Mbit/s → loss is intermittent; PHY enters/exits a degraded state, matching the "works for a few seconds, then collapses" UX.
Contributor guide
No contributing guide indexed for this repository
Research direction
Reproduce the loss using ping -M do and iperf3 on the listed CB1 hardware and Debian versions, then inspect ip -s -s link and dmesg alongside the sunxi-gmac behavior. Done means identifying the cause of the intermittent, size-dependent Ethernet loss and documenting or implementing a verified resolution.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- debian, linux
- Domain
- networking, operating-systems
- Issue type
- Bug
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Quiet
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100