2.5.2 to 2.6.0: network adapter no longer working

MatiasR

Our physical server is Intel + Windows Server 2012 and is running pfSense on Hyper V. It has 2 network adapters onboard and a PCIe card with 2 more network adapters.

With version 2.5.2 everything was running fine, with 3 of the 4 network adapters being used by pfSense (2 WANs, LAN, fail over).

After updating to 2.6.0 it asked me to reassign network adapters, as only the 2 onboard Broadcom based network adapters remained, and the Intel 82576 based card network adapter vanished. I had to reconfigure the assignments to keep the main WAN and the LAN working, and my fail over is gone.

Boot log shows the following error:

hn2: <Hyper-V Network Interface> on vmbus0
hn2: gpadl_conn(chan14) failed: 3221225540
hn2: rxbuf gpadl conn failed: 5
device_attach: hn2 attach returned 5

I tried disabling virtual machine queue and IPsec task offloading but still the hn2 is failing.

What can I do to restore my network adapter?

Thanks in advance!

MatiasR

I created another VM and installed a new 2.5.2 instance, restored my configs and everything is working fine again. So I will keep the 2.5.2 running in production until the VM with 2.6.x fixes the issues with the NICs.

bmeeks

Just an FYI based on my observations here over the years. pfSense seems to run with the least issues on virtualization platforms like VMware's ESXi. The majority of issues posters place in this group seem to happen in Hyper-V and Proxmox environments.

dehsgr

Same issue here. 2.5.2 is running flawlessly and after upgrade I got the same error:

Seems to be an issue loading/initializing the last interface...

dehsgr

Any ideas how we'll get update process working fine?

dehsgr

Strange. I cannot imagine that only two people have this issue.

netchr

Tried to update from 2.5.2 to 2.6.0 today and got the same error:

hn7: <Hyper-V Network Interface> on vmbus0
hn7: gpadl_conn(chan21) failed: 3221225540
hn7: rxbuf gpadl conn failed: 5
device_attach: hn7 attach retuned 5

Additionally, the previously loaded interfaces show the following info:

hnX: # of RX rings (4) does not match # of RSS buckets (8)

hn7 is the last out of 8 interfaces (hn0 ... hn7). Two of them got a VLAN config (hn3 and hn5). pfSense is running as VM on Hyper-V 2016.

Physical NICs on the Hyper-V host are Intel I210.

Are there any similarities to the other affected systems, e.g. Hyper-V or Intel I210 NIC?

xpxp2002

@netchr said in 2.5.2 to 2.6.0: network adapter no longer working:

hnX: # of RX rings (4) does not match # of RSS buckets (8)

Did you ever figure out the resolution for this? I have the same message for each of my VM network adapters on mine, too. It has a relationship to the number of vCPUs assigned to the VM, and probably consequently, the number of vRSS queues exposed to the VM.

I do not enable multiple threads for the pfSense VM, as most of what I've read recommends against using Hyper-Threading with a latency-sensitive VM like a router/firewall. It seems like there are always double the number of RSS buckets available compared to the RX rings, though. Even if I temporarily enable HT for the VM, it then shows 8 RX rings/16 RSS buckets.

netchr

@xpxp2002
Unfortunately, I couldn't figure out a solution. As a workaround, I switched from 8 virtual interfaces down to 7 by using further VLANs. Afterwards, the update went through without issues.

dehsgr

I waited for 2.7.0 but the issue remains there. That's very, very bad. Seems to be an issue with underlying FreeBSD as this is recognized there too, also in OPNSense. Irritating is for me the fact that there are different amounts of LAN interfaces causing this issue. Clean booting a newer pfSense image hits the same issue...