Proxmox 8.1.3 breaks Jumbo Frames on Intel i225-V ?

[phastier]

Hi all, as posted on another thread (https://forum.proxmox.com/threads/weird-bond-issue-since-8-1-i225-v-bug.137162/), I'm experiencing issues since upgrading from 8.0 to 8.1.3, sorry to open another thread.

I did a lot of investigations, but what I definitely know by now is that despite the NIC apparently showing a 9000 MTU, no packers above 1990 or 1970 bytes go through the interface (strange values by the way).

I managed to update the NIC firmware, I do not see any change. None of the other member of the cluster show this issue and still continue to use a 9000 MTU, only difference is they are not i225-V NICs.

I would really appreciate any help, this drives me crazy.

THANK YOU !

(please note that it was partially an issue in 8.0. I had to force autonegotiation to off. This is also not possible anymore in 8.1.3, autonegotiation stays on matter what ethtool option I use)

 

[spirit]

if it's a bug, it's a kernel bug. Maybe try to rollback to 6.2 kernel to compare.

#proxmox-boot-tool kernel pin 6.2.16-19-pve
#reboot

 

[phastier]

@spirit YES!!!!!! It did fix the issue ! You're a genius.

I still had to do "ethtool -s enpxs0 speed 1000 duplex full autoneg off" on all 4 bond members, but at least it works again.
Jumbo frames now work properly on this host, and Ceph is back to life.

So... how do we properly report a kernel bug like this ? I'm no kernel expert in that manner...

 

[phastier]

Bugzilla : https://bugzilla.proxmox.com/show_bug.cgi?id=5100

 

[phastier]

OK... so now kernel 6.2.16-20 also broke it. I tried 6.5.11-7, still a failure....

 

[phastier]

OK... so now kernel 6.2.16-20 also broke it. I tried 6.5.11-7, still a failure....

Going back., once more, to 6.2.16-19, reenables Jumbo frames.

I updated Bugzilla https://bugzilla.proxmox.com/show_bug.cgi?id=5100.

 

[jduekGuy]

I see also weird SSL errors on my VM's, could this be related?

*the internet traffic of these VM's are running via a virtualized pfsense which is running on a hypervisor with a Intel i225 network card.

Pending: write EPROTO 10E08095FFFF0000:error:0A000119:SSL routines:ssl3_get_record:decryption failed or bad record mac:../deps/openssl/openssl/ssl/record/ssl3_record.c:623

 

[phastier]

Hi all !

Just a quick update...

It still does not work with the latest 6.8.4-3 kernel...