Solved (Workaround) FreeBSD 10.1, sudden network down

I have the exact same problem with my Intel NICs. I've seen it with both 82571 and 82574 based NICs.

With tso, rxcsum and txcsum enabled the 82574 NICs will stop responding within an hour, but the 82571 NICs usually keep working a couple of days. When tso, rxcsum and txcsum are disabled the 82574 NICs keep working for about 48 hours. There is no difference for the 82571 NICs. When they stop responding I have to power the server off completely or the NICs will stop responding again within minutes of coming online.

Loading bhyve with kldload vmm seems to accelerate the process and the NICs stop working after just a few hours.

I'm running FreeBSD 10.1-p5.
 
No news regarding this?

82579LM receives watchdog error:
Code:
Feb 21 04:23:12 Feb 21 04:23:06 kernel: em0: Watchdog timeout -- resetting
Feb 21 04:23:12 Feb 21 04:23:06 kernel: em0: Queue(0) tdh = 685, hw tdt = 340
Feb 21 04:23:12 Feb 21 04:23:06 kernel: em0: TX(0) desc avail = 331,Next TX to Clean = 671
Feb 21 04:23:12 Feb 21 04:23:06 kernel: em0: link state changed to DOWN
Feb 21 04:23:12 Feb 21 04:23:10 kernel: em0: link state changed to UP
and 82574L just stops working, no traffic in or out, but the interface is still up.

I couldn't find a PR for this, does anyone know?
 
I have similar issues after upgrading to FreeBSD 10.1 on my server with ZFS, Samba and native iSCSI Target (ctld). The server under load (~600Mbps / 45k pps) suddenly stops responding on console, but network connections remains available. SSH sessions becomes very lagging and opening a folder via SMB requires more than 6 seconds. If I use ctld iSCSI connections being dropped. HDD's aren't busy during this time and in the top output everything is good (cpu, memory). Copying huge files (~6 Tb) between ZFS pools doesn't freeze the system. I've tried disabling TSO4, applying different sysctl tweaks but nothing happens. Everything was fine on FreeBSD 10.0 . My configuration: Supermicro X9SCL-F, Intel Xeon E3-1220, 8Gb RAM.
I've also noticed an interesting degradation of network performance on the monitoring graph after the console freezes (~21:35):
2015-03-03_2.55.15.png
 
I've tried today with
Code:
-tso4 -tso -txcsum -rxcsum -vlanhwtso
on each NIC but the server suddenly hangs after 3 hours of 400 Mbps /45k pps load. Nothing helps.
 
I am sorry to hear this, I can't help either, this workaround helped me and I don't have issues ever since then, it could be something else?
Anyway you can always submit a bug report here https://bugs.freebsd.org/bugzilla/
If your server is on production I would change the network adapter with a different one till this issue is sorted or downgrade if possible.

Take a look here for downgrade instructions.
 
Back
Top