nvsp_rndis_pkt_complete error status: 2 on HyperV

[mikekuzak]

Description

Hi,
We are running flatcar on Hyper-V.
Dmesg shows some a lot of logs like this:

[Tue May 20 14:29:58 2025] hv_netvsc 04793a47-94db-4aeb-a9ec-04de09a581e3 eth0: nvsp_rndis_pkt_complete error status: 2
[Tue May 20 14:29:58 2025] hv_netvsc 04793a47-94db-4aeb-a9ec-04de09a581e3 eth0: nvsp_rndis_pkt_complete error status: 2
[Tue May 20 14:30:04 2025] net_ratelimit: 120 callbacks suppressed
[Tue May 20 14:30:04 2025] hv_netvsc 04793a47-94db-4aeb-a9ec-04de09a581e3 eth0: nvsp_rndis_pkt_complete error status: 2
[Tue May 20 14:30:04 2025] hv_netvsc 04793a47-94db-4aeb-a9ec-04de09a581e3 eth0: nvsp_rndis_pkt_complete error status: 2
[Tue May 20 14:30:04 2025] hv_netvsc 04793a47-94db-4aeb-a9ec-04de09a581e3 eth0: nvsp_rndis_pkt_complete error status: 2
[Tue May 20 14:30:04 2025] hv_netvsc 04793a47-94db-4aeb-a9ec-04de09a581e3 eth0: nvsp_rndis_pkt_complete error status: 2
[Tue May 20 14:30:04 2025] hv_netvsc 04793a47-94db-4aeb-a9ec-04de09a581e3 eth0: nvsp_rndis_pkt_complete error status: 2
[Tue May 20 14:30:04 2025] hv_netvsc 04793a47-94db-4aeb-a9ec-04de09a581e3 eth0: nvsp_rndis_pkt_complete error status: 2
[Tue May 20 14:30:04 2025] hv_netvsc 04793a47-94db-4aeb-a9ec-04de09a581e3 eth0: nvsp_rndis_pkt_complete error status: 2
[Tue May 20 14:30:04 2025] hv_netvsc 04793a47-94db-4aeb-a9ec-04de09a581e3 eth0: nvsp_rndis_pkt_complete error status: 2
[Tue May 20 14:30:04 2025] hv_netvsc 04793a47-94db-4aeb-a9ec-04de09a581e3 eth0: nvsp_rndis_pkt_complete error status: 2
[Tue May 20 14:30:04 2025] hv_netvsc 04793a47-94db-4aeb-a9ec-04de09a581e3 eth0: nvsp_rndis_pkt_complete error status: 2

6.6.83-flatcar 1 SMP PREEMPT_DYNAMIC Mon Mar 17 16:07:40 -00 2025 x86_64 Intel(R) Xeon(R) Gold 6254 CPU @ 3.10GHz GenuineIntel GNU/Linux

Hyper-V Version: 10.0.17763.7009

Impact

Package lose. Potentially linked to:
https://bugzilla.kernel.org/show_bug.cgi?id=217503

Does anyone else see this behavior ?

[tormath1]

Hi @mikekuzak thanks for the report.

From the upstream issue, I read:

The fix for this problem has been backported to these kernel stable releases that were made available today:

• 6.14.8
• 6.12.30
• 6.6.92
• 6.1.140

Those updates will be ingested soon and should be available in the set of releases. In the meantime, are you available to test a nightly image to see if it fixes your issue ?

[mikekuzak]

@tormath1
Sure ..

[tormath1]

Hey @mikekuzak, so you have the choice: you can try with current Stable nightly: https://bincache.flatcar-linux.net/images/amd64/4152.2.3+nightly-20250527-2100/ (kernel 6.6.y)

Or with the latest official Alpha (kernel 6.12.y): https://www.flatcar.org/releases#release-4344.0.0

[mikekuzak]

Running now Flatcar stable 4230.2.0 with Kernel 6.6.94. Issue seems gone. Thanks @tormath1

[tgelter]

In case it's helpful for anyone else, my team has observed significant issues with etcd (specifically during upgrades) related to this issue. On Kubernetes clusters which have high object counts (large etcd database) and high usage (high number of watchers/other K8s API server requests), we were unable to upgrade a single etcd VM due to the packet loss related to this bug. In order to work around it, we had to patch the etcd server binary to adjust the publish-timeout from 7s to 10m(!) and re-attempt node replacement multiple times before successfully moving on. Upon moving to the latest stable version (4230.2.1), we were able to successfully upgrade an example cluster.