Frequent Kernel Panics - References to MACVLAN and NVIDIA

[kl0wn]

Hey All,

 

I have issues with Kernel Panics relating to macvlan and in some cases the nvidia driver but there doesn't seem to be any rhyme or reason. Example below and Diagnostics attached

 

Aug 26 08:27:25 NAS kernel: <TASK>
Aug 26 08:27:25 NAS kernel: netif_rx_ni+0x53/0x85
Aug 26 08:27:25 NAS kernel: macvlan_broadcast+0x116/0x144 [macvlan]
Aug 26 08:27:25 NAS kernel: macvlan_process_broadcast+0xc7/0x110 [macvlan]
Aug 26 08:27:25 NAS kernel: process_one_work+0x198/0x27a
Aug 26 08:27:25 NAS kernel: worker_thread+0x19c/0x240
Aug 26 08:27:25 NAS kernel: ? rescuer_thread+0x28b/0x28b
Aug 26 08:27:25 NAS kernel: kthread+0xde/0xe3
Aug 26 08:27:25 NAS kernel: ? set_kthread_struct+0x32/0x32
Aug 26 08:27:25 NAS kernel: ret_from_fork+0x22/0x30
Aug 26 08:27:25 NAS kernel: </TASK>

nas-diagnostics-20220826-1931.zip

[JorgeB]

For the mcvlan errors change docker network to ipvlan.

[kl0wn]

I tried that but I'm having trouble understanding how exactly to implement IPVLAN in practice. If all of the containers are using the same MAC w/ different IP addresses...how am I then able to effectively route traffic across the network? For reference my docker containers are on a VLAN as well w/ no DHCP.

 

Thanks!

[JorgeB]

I don't use macvlan/ipvlan but from my understanding they work similarity, i.e., you just need to change the custom network type and dockers with a custom IP address should remain working.

[kl0wn]

Copy - the problem is IPVLAN uses the concept of 1 MAC Address to many IP's; which is a bit of a nightmare for me personally. I've read a lot of posts and there seems to be a common theme of switch to IPVLAN while also Disabling "Host Access to Customer Networks". I went thru a variety of macvlan/ipvlan iterations, with/without: VLAN's, dedicated NIC w/ PVID, DHCP on/off...

 

Right now I'm using macvlan w/ "Host Access to Custom Networks" disabled; so far so good for ~6hours. I'll report back in 24 or so.

 

NOTE: Bridge and Host containers will lose the abiltiy to communicate with containers that both have a static IP and live on the Host.

[kl0wn]

I can confidently mark this thread as resolved. If anyone else is having an issue with kernel panics after version 6.9+ disabling "Host Access to Custom Networks" has resulted in no panics for over 24hr. Hopefully this helps someone else out.

 

@JorgeB thank you for your guidance and assistance as well!