I can confirm this on a different kernel and board revision, and I have
one data point that may be worth having: it stalls at essentially zero
TX load, not just under bulk transfer.
Setup
-----
3x Raspberry Pi 5 Model B Rev 1.1, 8 GB, NVMe via a PCIe HAT
Ubuntu 26.04 LTS, linux-raspi 7.0.0-1017.17 (7.0.0-1017-raspi)
macb, bus-info 1f00100000.ethernet, PHY Broadcom BCM54213PE
1000Mb/s, Full, autoneg on, flow control off
Switch: UniFi USW Enterprise 24 PoE
Offloads at default: tso on, sg on, gso on, gro on
Rings at default: RX 512 (max 8192), TX 512 (max 4096)
EEE: "enabled - inactive", link partner does not advertise EEE,
and tx_lpi_transitions stays 0 across a stall
The three nodes are a Docker Swarm control plane (all three managers).
The traffic is swarm gossip plus a handful of small containers.
Rate and load
-------------
24 stalls in ~344 node-hours (3 nodes, 2026-08-29 00:00 to 2026-09-02
18:48 UTC), so roughly one per 14 node-hours. Split: 10 / 7 / 7. The
window deliberately excludes earlier days when the cluster was being
physically worked on.
Measured TX over that period, from sysstat at 1-minute resolution:
node avg TX peak TX
------------------------------
swarm01 25.4 kB/s 95 kB/s
swarm02 30.2 kB/s 168 kB/s
swarm03 36.8 kB/s 151 kB/s
That is ~0.2-0.3 Mbit/s average and ~1.3 Mbit/s peak transmit, about
0.1% of link rate. RX peaks higher (~2.8 MB/s) during container image
pulls, but the stalls do not correlate with those.
I mention it because the reproducer in this thread is
"iperf -c <IP> -P 10 -t 3000 -w 4M", and the commit message says the
exact root cause is still unidentified. If the race needed sustained
bulk dataflow to be hit, this cluster should not be hitting it several
times a day. Whatever loses the TSTART write does not appear to need a
busy TX path.
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2133877
Title:
Complete network hang on Raspberry Pi 5 with kernel 6.17 under load -
possibly related to CPU frequency scaling
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux-raspi/+bug/2133877/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
