Proxmox 6.8.12-9-pve kernel has introduced a problem with e1000e Driver and network connection lost after some hours
- Thread starter leandrosgf
- Start date
The scripts are now updated. it should ask you if you want to leave the offloading features enabled or not
Using Kernel 6.8.12-11 and this script: https://community-scripts.github.io/ProxmoxVE/scripts?id=post-pve-install.
So far, on the I219-LM (Rev 10) it seems to be stable.
@Twitch6004 i applied your Script like one week ago.
Amazing. No more network looses since then. Thanks a lot
Amazing. No more network looses since then. Thanks a lot
For those who have recently applied the script with hardware offloading turned off, it would be appreciated if you could report back on how it affected your CPU load. Since disabling offloading shifts packet segmentation and checksum calculations to the CPU, this could have a noticeable impact—especially on systems handling significant traffic. What kind of difference have you seen on your systems?
Got the same issue on:
HW: ThinkCenter M720q
SW: Proxmox v8.4.0
Network adapter:
Network card loses internet in a few hours, which is pretty critical. I'm wondering, whether these things are tested at Proxmox, given the scale and how many subscribers are impacted?
HW: ThinkCenter M720q
SW: Proxmox v8.4.0
Network adapter:
Code:
modinfo e1000e | grep version
srcversion: F65CFC0DF4BFD42B230512D
vermagic: 6.8.12-9-pve SMP preempt mod_unload modversionsNetwork card loses internet in a few hours, which is pretty critical. I'm wondering, whether these things are tested at Proxmox, given the scale and how many subscribers are impacted?
As per this thread alone, you may come to realize, that it seems to only affect specific cards and revisions. So it would be impossible for proxmox to have ALL the different revisions of the i217/i219-V/-LM cards available. It was mentioned, that such problems seem to occur not only in this specific card but occasionally on Realtech, Marvell or Broadcom as well (seen similar hickups on Windows machines as well....)
So I'd guess, there is more to blame at Intel than at Proxmox or Linux....
As you were asking: Still everything up and running since the last post.
Code:
Linux pveG4 6.8.12-13-pve #1 SMP PREEMPT_DYNAMIC PMX 6.8.12-13 (2025-07-22T10:00Z) x86_64 GNU/Linux
Code:
journalctl -u network-fix.service | grep -v 'Network connectivity OK'New install of proxmox 9 on a cluster of 3 lenovo p330 tiny PCs. Before I had Proxmox 7 and 8, rarely updated though, but running strong 24/7 for 2 years with no problem. With the new proxmox 9 within 4 hours had a network card hang on me.
The reason I don't want to disable any offloading is that for my use-case at least performance is much more important than security, being an isolated lab.
Issued these commands to downgrade the kernel on proxmox 9:
I recommend deleting the file /etc/apt/sources.list.d/pve8.list after this.
The reason I don't want to disable any offloading is that for my use-case at least performance is much more important than security, being an isolated lab.
Issued these commands to downgrade the kernel on proxmox 9:
Bash:
echo "deb http://download.proxmox.com/debian/pve bookworm pve-no-subscription" > /etc/apt/sources.list.d/pve8.list
apt update
apt install proxmox-kernel-6.8.12-8-pve
proxmox-boot-tool kernel pin 6.8.12-8-pveI recommend deleting the file /etc/apt/sources.list.d/pve8.list after this.
Last edited:
Upgraded to ProxMox 9 last week and now experiencing loss of connectivity to host after network load. Was able to recreate this just transferring files via FTP from a VM. Applied your script a few hours ago and haven't had issue. On previous version of ProxMox the VM would freeze under load and resume after a minute or so, but never affect the host's connectivity.
any feedback from people using the script? I wonder if turning off hardware offloading will have significant impact on modern CPU (I'm talking intel 9th gen or above)
It will impact the processor, no doubt about it. The impact will be proportional to the amount and type of traffic, as the processor would now have to segment and calculate checksums of each packet, which is a very inefficient task for a CPU. The hardware embedded in NICs is highly specialised in this task and can do this much more efficiently.
Ethernet controller: Intel Corporation Ethernet Connection (7) I219-LM (rev 10)
Is this stable on the latest kernel?
I'm using a USB ethernet adapter with SNAT and it works but being USB it's a bit slow.
The offloading fix doesnt always work for me.
Is this stable on the latest kernel?
I'm using a USB ethernet adapter with SNAT and it works but being USB it's a bit slow.
The offloading fix doesnt always work for me.
anyone know if this is on their radar for upcoming fix? As older kernel doesn't seem to have this problem.
Last edited:
Proxmox 9.0.5 Debian Trixie (6.14.8-2) intel i219-LM the same problems with the network card freezing. And in Windows there were never such problems at all, only on Linux something happens. In general, the problem was solved with minimal edits to the file interfaces:
iface eno1 inet manual
offload-tso off
offload-gso off
I really didn't want to disable all the functionality of the network card due to problems in the Linux kernels and found this minimal set of changes that works
iface eno1 inet manual
offload-tso off
offload-gso off
I really didn't want to disable all the functionality of the network card due to problems in the Linux kernels and found this minimal set of changes that works