35 thoughts on - NIC Stability Problems Under Xen 4.4 / CentOS 6 / Linux 3.18

• Kevin Stange, It can be either kernel or update the NIC driver or firmware of the NIC
  card. Hope that helps!

  Xlord
  —–Original Message—

• Honeslty I would email Intel and see if they can help. This looks like the NIC decides something is wrong, throws off an PCIe error and then resets itself.

  It could also be an error in the Linux stack which would “eat” an interrupt when migrating interrupts (which was fixed upstream, see below). Are you running irqbalance? Could you try turning it off?

  Did you have these issues with an earlier kernel?

  The fix was ff1e22e7a638a0782f54f81a6c9cb139aca2da35
  Author: Boris Ostrovsky
  Date: Fri Mar 18 10:11:07 2016 -0400

  xen/events: Mask a moving irq

  and then there was a fix to this fix:
  commit f0f393877c71ad227d36705d61d1e4062bc29cf5
  Author: Ross Lagerwall
  Date: Tue May 10 16:11:00 2016 +0100

  xen/events: Don’t move disabled irqs

• This happens for several different NICs. Is there a good contact at Intel for this kind of thing, or should I just try to reach them through their web site?

  irqbalance is enabled on these servers. I’ll try disabling it.

  The last kernel these boxes ran was 2.6.18-412.el5xen under CentOS 5 and they were very stable, however the differences between 2.6.18 and 3.18
  are immense, especially with features like ASPM and other power management code. We’ve run into ASPM issues on systems before going from CentOS 5 to the CentOS 6 kernel 2.6.32, but not this particular hardware, which is why my first thought was to look at ASPM.

  They’ve all been upgraded to CentOS 6 and running the virt SIG kernel kernel-3.18.44-20.el6.x86_64. I haven’t run any previous versions 3.18
  or tried any other kernels.

  It surprises me that we would have all these issues if there isn’t a more widespread problem considering the hardware is fairly maintain and covers a lot of NIC chips.

• I had stopped irqbalance yesterday afternoon, but had a hypervisor’s NICs fail anyway in early morning this morning, so I’m pretty sure this is not the right tree to bark up.

• –LNgNxlHjaD7oMQXKqUuKo54FDnpM5bK5H
  Content-Type: text/plain; charset=windows-1252
  Content-Transfer-Encoding: quoted-printable

  Here is a set of drivers/fireware from Intel for those NICs:

  https://downloadcenter.intel.com/download/15817/Intel-Network-Adapter-Driver-for-PCI-E-Gigabit-Network-Connections-under-Linux-

  I will see if I can get a CentOS-6 build of the latest version of that from our older SRPM:

  http://vault.CentOS.org/6.7/xen4/Source/SPackages/e1000e-2.5.4-3.10.68.2.el6.CentOS.alt.src.rpm

  I am currently very busy with several c5, c6, c7 updates and the i686
  altarch c7 tree .. but I have this on my list. In the meantime, maybe someone else could also see if those drivers help you (or you could try to compile / install it).

  Do you have another machine that you can use to see if you can duplicate the issue NOT running the xen.gz hypervisor boot, but just the straight kernel?

  Thanks, Johnny Hughes

  –LNgNxlHjaD7oMQXKqUuKo54FDnpM5bK5H

• –BUWcnsL11lBJ37bs66R8eIeUuvwLqKoBq Content-Type: text/plain; charset=windows-1252
  Content-Transfer-Encoding: quoted-printable

  Actually .. I think this is the driver for you:

  https://downloadcenter.intel.com/download/13663

  And this explains how to make it work:

  http://www.intel.com/content/www/us/en/support/network-and-i-o/ethernet-products/000005767.html

  –BUWcnsL11lBJ37bs66R8eIeUuvwLqKoBq

• I can’t actually reproduce this problem reliably. It happens randomly when the servers are up and running anywhere between a few hours and a month or more, and I haven’t been able to isolate any specific way to cause it to happen. As a result I can’t really test different solutions on different servers to see what helps. I was hoping other people were seeing it so that I could get some direction. If I can reproduce it, it won’t take me very long to identify what the cause is. Right now if I
  do upgrade the drivers on the systems I won’t really know if it’s fixed until I don’t see another issue for several months.

  The different combinations of NICs overlap both the e1000e and igb drivers, but the most egregious issues have been with the igb ones. I’ll try to give this a shot and report back if I still see issues with a server after doing so, but it might be a week or two before I find out.

• So the NICs giving issues in most cases were igb drivers. I’ve tried replacing the drivers on some HVs with the version you suggested, but it doesn’t seem to have helped with stability. Any other ideas?

• Have you tried to eliminate all power management features all over?

  Are the devices connected to the same network infrastructure?

  There has to be something common.

  I’ve been using Intel NICs with Xen/CentOS for ages with no issues.

  Karel

• I’ve been trying to find and disable all power management features but having relatively little luck with that solving the problems. Stabbing the the dark I’ve tried different ACPI settings, including completely disabling it, disabling CPU frequency scaling, and setting pcie_aspm=off on the kernel command line. Are there other kernel options that might be useful to try?

  There are two onboard NICs and two NICs on a dual-port card in each server. All devices connect to a cisco switch pair in VSS and the links are paired in LACP.

  The NICs having issues are running a native VLAN, a tagged VLAN, iSCSI
  and NFS traffic, as well as some basic management stuff over SSH, and they are configured with an MTU of 9000 on the native VLAN. It’s a lot of features, but I can’t really turn them off and then actually have enough load on the NICs to reproduce the issue. Several of these servers were installed and being burned in for 3 months without ever having an issue, but suddenly collapsed when I tried to bring 20 or so real-world VMs up on them.

  The other NICs in the system that are connected don’t exhibit issues and run only VM network interfaces. They are also in LACP and running VLAN
  tags, but normal 1500 MTU.

  So far it seems to correlate with NICs on the expansion cards, but it’s a coincidence that these cards are the ones with the storage and management traffic. I’m trying to swap some of this load to the onboard NICs to see if the issues migrate over with it, or if they stay with the expansion cards.

  If the issue exists on both NIC types, then it rules out the specific NIC chipset as the culprit. It could point to the driver, but upgrading it to a newer version did not help and actually appeared to make everything worse. This issue might actually be more to do with the PCIe bridge than the NICs, but these are still different motherboards with different PCIe bridges (5520 vs C600) experiencing the same issues.

  I figured that must be so. Everyone uses Intel NICs. If this was a common issue, it would probably be causing a lot of people a lot of trouble.

• pci=nomsi

  https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1521173/comments/13

• Incidentally, already found that one and I’m trying it currently on one of the boxes. So far there’s been no issues, but it’s only been since Friday.

  Also, I found this:

  https://xen.crc.id.au/support/guides/install/

  There’s a 4.4 kernel here built for Xen Dom0, which I’m giving a whirl to see how stable it is, also only since Friday. I’m not using anything else he’s packaged from his repo.

  On a related note, does the SIG have plans to replace the 3.18 kernel which is marked as projected EOL of January 2017
  (https://www.kernel.org/category/releases.html)?

• –A6WpREJ02fvBT7CI am currently working on a 4.4 kernel as a replacement for the 3.18
  kernel. I have it working well no el7, but not yet working well on el6. I hope to have something to release in the first 2 weeks of Feb. for testing.

  –A6WpREJ02fvBT7COn34TX2mb3redkj5Je

• What kind of issues are you having with 4.4? Since I’m testing that
  “Xen Made Easy” build of 4.4, are there any things I should watch out for? Might be worth looking at what he did for his builds to see if that helps get yours working better.

  http://au1.mirror.crc.id.au/repo/el6/SRPM/

• May I chip in here? In our environment we’re randomly seeing:

  Jan 17 23:40:14 xen01 kernel: ixgbe 0000:04:00.1 eth6: Detected Tx Unit Hang Jan 17 23:40:14 xen01 kernel: Tx Queue <0>
  Jan 17 23:40:14 xen01 kernel: TDH, TDT <9a>, <127>
  Jan 17 23:40:14 xen01 kernel: next_to_use <127>
  Jan 17 23:40:14 xen01 kernel: next_to_clean <98>
  Jan 17 23:40:14 xen01 kernel: ixgbe 0000:04:00.1 eth6:
  tx_buffer_info[next_to_clean]
  Jan 17 23:40:14 xen01 kernel: time_stamp <218443db3>
  Jan 17 23:40:14 xen01 kernel: jiffies <218445368>
  Jan 17 23:40:14 xen01 kernel: ixgbe 0000:04:00.1 eth6: tx hang 1
  detected on queue 0, resetting adapter Jan 17 23:40:14 xen01 kernel: ixgbe 0000:04:00.1 eth6: Reset adapter Jan 17 23:40:15 xen01 kernel: ixgbe 0000:04:00.1 eth6: PCIe transaction pending bit also did not clear. Jan 17 23:40:15 xen01 kernel: ixgbe 0000:04:00.1: master disable timed out Jan 17 23:40:15 xen01 kernel: bonding: bond1: link status down for interface eth6, disabling it in 200 ms. Jan 17 23:40:15 xen01 kernel: bonding: bond1: link status definitely down for interface eth6, disabling it
  […] repeated every second or so.

  We’ve been experienced ixgbe stability issues on CentOS 6.x with various
  3.x kernels for years with different ixgbe driver versions and, to date, the only way to completely get rid of the issue was to switch from Intel to Broadcom. Just like in your case, the problem pops up randomly and the only reliable temporary fix is to reboot the affected Xen node. Another temporary fix that worked several times but not always was to migrate / shutdown the domUs, deactivate the volume groups, log out of all the iSCSI targets, “ifdown bond1” and “modprobe -r ixgbe” followed by “ifup bond1”.

  The set up is:
  – Intel Dual 10Gb Ethernet – either X520-T2 or X540-T2
  – Tried Xen kernels from both xen.crc.id.au and CentOS 6 Xen repos
  – LACP bonding to connect to the NFS & iSCSI storage using Brocade VDX6740T fabric. MTU

• Welcome. It’s a relief to know someone else has been having a similar nightmare! Perhaps that’s not encouraging…

  You said 3.x kernels specifically. The kernel on Xen Made Easy now is a
  4.4 kernel. Any chance you have tested with that one?

  Did you ever try without MTU

• Not yet, however the future Xen nodes we’ll deploy will run CentOS 7 and Xen with kernel 4.4.

  Yes, also with all sorts of configuration combinations like LACP rate slow/fast, “options ixgbe LRO=0,0” and so on. No improvement.

  This is interesting. We’ve never had any problems with the 1Gb NICs, but we’re only using 10Gb for the storage network. Could it be a common problem with either the adapters, or the drivers which only replicate running the Xen enabled kernel?

  We’ve purchased a number of servers with Broadcom BCM957810A1008G, sold by Dell as QLogic 57810 dual 10Gb Base-T adapters, none of them going up
  & down like a yo-yo so far.

  It’d be more like 6-9 months for us, making it terrible to debug it :-/

  Adi Pircalabu

• I’ll keep you (and others here) posted on my own experiences with that
  4.4 build over the next few weeks to report on any issues. I’m hoping something happened between 3.18 and 4.4 that fixed underlying problems.

  Alright, I’ll assume that probably won’t help then. I tried it on one box which hasn’t had the issue again yet, but that doesn’t guarantee anything.

  Since I’ve never run the 3.18 kernel on a box of this type without running in a dom0 and since I can’t reproduce this kind of issue without a fair amount of NIC load over a tremendous period of time, it’s impossible to test if it’s tied to Xen.

  However, I know this hardware works well under 2.6.32-*.el6 and
  3.10.0-*.el7 kernels without stability problems, as it did with
  2.6.18-*.el5xen (Xen 3.4.4).

  I suspect the above errors are actually due to something PCIe related, and I have a subset of boxes which are actually being impacted by two distinct problems with equivalent impact, which increases the likelihood that the boxes will die. Another set of boxes only ever sees the unit hangs which seem unrecoverable even unloading/reloading the driver. A
  third set has random recoverable unit hangs only. With so much diversity, it’s even harder to pin any specific causes to the problems.

  The fact we’re both pushing NFS and iSCSI traffic over these links makes me wonder if there’s something about that kind of traffic that increases the chances of causing these issues. When I put VM network traffic over the same NICs, they seem a lot less prone to failures, but also end up pushing less traffic in general.

  I had a bunch of these on relatively light VM load for 3 months for
  “burn in” with no issues but they’ve been pretty aggressively failing since I started to try to put real loads on them. Still, it’s odd because some of the boxes with identical hardware and similar VM loads have not yet blown up after 3 or more weeks, and maybe they won’t for several months.

• Someone in this thread: https://sourceforge.net/p/e1000/bugs/530/#2855
  reported that *”With these kernels I was only able to work around the issue by disabling tx-checksumming offload with ethtool.”*

  However, that was reported for Kernels 4.2.6 / 4.2.8 / 4.4.8 and 4.4.10. I
  just thought it could be something you could rule out and hence mentioned it:

  ethtool –offload eth6 rx off tx off

  Another thing to rule out in case its a regression with Intel NICs and TSO:

  # tso => tcp-segmentation-offload
  # gso => generic-segmentation-offload
  # gro => generic-receive-offload
  # sg => scatter-gather
  # ufo => udp-fragmentation-offload (Cannot change)
  # lro => large-receive-offload (Cannot change)

  ethtool -K eth6 tso off gso off gro off sg off

• Nice, useful information. I’ve just disabled tx & rx checksumming on all the 10Gb interfaces on the affected servers, see how it goes. But as I
  said yesterday, in our environment it takes months to replicate.

  Thanks,

  Adi Pircalabu

• I was able to discover something new, which might not conclusively prove anything, but it at least seems to rule out the pci=nomsi kernel option from being effective.

  I had one server booted with that option as well as MTU 1500. It was stable for quite a long time, so I decided to try turning the MTU back to 9000 and within 12 hours, the interface on the expansion NIC with the jumbo MTU failed.

  The other NIC in the LACP bundle is onboard and didn’t fail. The other NIC on the dual-port expansion card also didn’t fail. This leads me to believe that ONE of the bugs I’m experiencing is related to 82575EB +
  jumbo frames.

  I still think I’m also having a PCI-e issue that is separate and additional on top of that, and which has not reared its head recently, making it difficult for me to gather any new data.

  One of the things I’ve done that seemed to help a lot with stability was balance the LACP so that one NIC from onboard and one NIC from expansion card is in each LAG. Previously we just had the first LAG onboard and the second on the expansion card. This way, at least, given the expansion NIC’s propensity toward failing first, I don’t have to crash the server and all running VMs to recover.

  I’ve seen absolutely no issues yet with the 4.4 kernel either, but I am not willing to call that a win because of the quiet from even the servers on which no tweaks have been applied yet.

  I will continue the story as I have more material! :)

• Kevin Stange, Good attempt, jumbo frame is extremely important if hosting for IaaS, not to mention other provider who need feather for network specific application.

  Xlord
  —–Original Message—

• Thanks for the heads-up Kevin, appreciated. One thing I need to clarify, though: what kernel was this machine running at the time?

  Adi Pircalabu

• Kernel running at the time was the Virt SIG’s 3.18.44-20 kernel.

  As a further note, within an additional 24 hours, the onboard Intel
  82576 that was switched to enable jumbo frames also failed and we had to reboot the server. The expansion and onboard ports without jumbo frames did not fail. Since reboot, it’s on the 4.4.47 kernel from Xen Made Easy now with jumbo frames and has not exhibited issues since Friday.

• Kevin Stange Sound interesting.

  Xlord

  —–Original Message—

• […]

  Fingers crossed for 4.4.47 kernel, thanks and keep us posted :)
  Cheers,

  –

• –I5AeMcbrlu8U3tjN7JnU8D4P6Waa9Dkxi Content-Type: text/plain; charset=windows-1252
  Content-Transfer-Encoding: quoted-printable

  Kevin,

  Please try the 4.9.11-22 kernel that I just released for CentOS-6 (along with the newer linux-firmare packages and xfsprogs).

  If you enable the xen-testing repository in your CentOS-Xen.repo file
  (assuming it is pointing to xen-44 and not xen-46) then a ‘yum upgrade’
  should replace all the needed packages.

  The actual path is here for the packages:

  https://buildlogs.CentOS.org/CentOS/6/virt/x86_64/xen-44/

  Hopefully this helps.

  Thanks, Johnny Hughes

  –I5AeMcbrlu8U3tjN7JnU8D4P6Waa9Dkxi

• –c2C5EnaJpnEXWGafhgteSlVcec0Ji0BAL
  Content-Type: text/plain; charset=windows-1252
  Content-Transfer-Encoding: quoted-printable

  I should have said .. ‘just releaed for testing’ :)

  I have been using this for 4 or 5 days with no issues in production, but it needs testing before final release :)

  –c2C5EnaJpnEXWGafhgteSlVcec0Ji0BAL

• Currently I’ve moved most of my servers onto the 4.4 kernel from xen made easy and they’ve been stable. I have some indications of an issue with one of my 3.18 servers right now which required it to be rebooted, so I’m going to bring the 4.9 kernel up on that server to see how it does. It may take a few weeks or more to draw any conclusions.

• Currently running 4.9.11 on a few servers and they’ve been working fine. No new issues have come up so far, anyway.

  I still can’t rest assured the NIC issue is fixed, but no 4.4 or 4.9
  server has yet had a NIC issue, with some being up almost a full month. It looks promising! (I’m knocking on all the wood everywhere, though.)

• I’m ready to call this conclusive. The problems I was having across the board seemed to be caused by something seriously broken in 3.18. Most of my servers are now on 4.9.13 or newer and everything has been working very well.

  I’m not going to post any further updates unless something breaks. Thanks to everyone that provided tips and suggestions along the way.

• Do you mind sharing what hardware have you been running the 4.9 kernel on other than “Supermicro X9DRT, Dual Xeon E5-2650, 2x I350, 2x 82571EB” and
  “Supermicro X9DRD-iF/LF, Dual Xeon E5-2630, 2x I350, 2x 82575EB” if any? Are you using any SATA/SAS controllers?

  Thanks, Sarah

• We have no expansion cards installed except for the dual-port gigabit NICs. We’re using the onboard SATA controller for only the local Dom0
  OS, and iSCSI and NFS for managing storage for VMs and images.

• We’ve got some other motherboards as well, I think this list is exhaustive:

  Supermicro X8DT3
  Supermicro X8DT6
  Supermicro X9DRD-iF/LF
  Supermicro X9DRT
  Supermicro X9SCL/X9SCM

  These are -F variants which means they include a BMC chip with a separate NIC. A few of the X8DT3 are the LN4 variant, which has 4
  onboard NICs and therefore we did not use an expansion NIC.

• FYI, Here’s one of our machines, which just crapped itself earlier today without being subjected to any significant load. Before the crash it was running kernel-3.18.44-20.el6.x86_64, now it’s on kernel-4.9.13-22.el6.x86_64:
  – Dell PowerEdge R620
  – 2 x Intel(R) Xeon(R) CPU E5-2630 v2 @ 2.60GHz, 6 cores each
  – Dual Intel 10-Gigabit X540-AT2 (rev 01)
  Before the crash both em interfaces member of bond1, which connects to the storage network, had tx & rx checksumming turned off. xen_commandline: dom0_mem36M,max:2048M dom0_max_vcpus=1
  dom0_vcpus_pin cpuinfo com15200,8n1 console=com1,tty loglvl=all guest_loglvl=all

  –