[libvirt] About the vfio error when using SR-IOV

[Alex Williamson]

On Mon, 22 Feb 2016 10:23:16 -0500
Laine Stump <laine at laine.org> wrote:

> On 02/22/2016 04:29 AM, Xiao Ma (xima2) wrote:
> > Hi, All
> >
> > I want to use the SR-IOV of intel 82576 NIC.
> >
> > I enabled IOMMU and VT-d and SR-IOV in BIOS.
> > And enabled VT-d in kernel.
> > The OS information is bellow:
> > [root at host3 nova]# cat /etc/redhat-release
> > CentOS Linux release 7.1.1503 (Core)
> > [root at host3 nova]# uname -an
> > Linux host3.localdomain 3.10.0-229.el7.x86_64 #1 SMP Fri Mar 6 11:36:42 UTC 2015 x86_64 x86_64 x86_64 GNU/Linux
> > [root at host3 nova]# rpm -qa|grep qemu
> > libvirt-daemon-driver-qemu-1.2.8-16.el7.x86_64
> > qemu-kvm-1.5.3-86.el7.x86_64
> > qemu-kvm-common-1.5.3-86.el7.x86_64
> > ipxe-roms-qemu-20130517-6.gitc4bce43.el7.noarch
> > qemu-img-1.5.3-86.el7.x86_64
> > [root at host3 nova]# rpm -qa|grep libvirt
> > libvirt-daemon-1.2.8-16.el7.x86_64
> > libvirt-daemon-driver-nwfilter-1.2.8-16.el7.x86_64
> > libvirt-python-1.2.8-7.el7.x86_64
> > libvirt-daemon-driver-storage-1.2.8-16.el7.x86_64
> > libvirt-daemon-driver-qemu-1.2.8-16.el7.x86_64
> > libvirt-daemon-config-nwfilter-1.2.8-16.el7.x86_64
> > libvirt-daemon-driver-secret-1.2.8-16.el7.x86_64
> > libvirt-daemon-driver-interface-1.2.8-16.el7.x86_64
> > libvirt-client-1.2.8-16.el7.x86_64
> > libvirt-glib-0.1.7-3.el7.x86_64
> > libvirt-daemon-driver-network-1.2.8-16.el7.x86_64
> > libvirt-daemon-driver-nodedev-1.2.8-16.el7.x86_64
> > libvirt-daemon-kvm-1.2.8-16.el7.x86_64
> >
> > And I can see the vf of the NIC after ‘ echo '7' > /sys/class/net/ens1f1/device/sriov_numvfs '
> >
> > [root at host3 VTS2.1-demo]# lspci |grep -i ethernet
> > 08:00.0 Ethernet controller: QLogic Corp. 10GbE Converged Network Adapter (TCP/IP Networking) (rev 02)
> > 08:00.1 Ethernet controller: QLogic Corp. 10GbE Converged Network Adapter (TCP/IP Networking) (rev 02)
> > 0f:00.0 Ethernet controller: Intel Corporation 82576 Gigabit Network Connection (rev 01)
> > 0f:00.1 Ethernet controller: Intel Corporation 82576 Gigabit Network Connection (rev 01)
> > 10:10.1 Ethernet controller: Intel Corporation 82576 Virtual Function (rev 01)
> > 10:10.3 Ethernet controller: Intel Corporation 82576 Virtual Function (rev 01)
> > 10:10.5 Ethernet controller: Intel Corporation 82576 Virtual Function (rev 01)
> > 10:10.7 Ethernet controller: Intel Corporation 82576 Virtual Function (rev 01)
> > 10:11.1 Ethernet controller: Intel Corporation 82576 Virtual Function (rev 01)
> > 10:11.3 Ethernet controller: Intel Corporation 82576 Virtual Function (rev 01)
> > 10:11.5 Ethernet controller: Intel Corporation 82576 Virtual Function (rev 01)
> >
> >
> > I configured the interface as bellow in XML:
> >
> >         <interface type="hostdev" managed="yes">
> >         <mac address="fa:16:3e:f7:57:5f"/>
> >         <source>
> >             <address type="pci" domain="0x0000" bus="0x10" slot="0x10" function="0x3"/>
> >         </source>
> >         <vlan>
> >             <tag id="1000"/>
> >         </vlan>
> >         </interface>
> >
> >
> > But the error output when I boot one vm:
> >
> > [root at host3 VTS2.1-demo]# virsh create vtc.demo.xml
> > error: Failed to create domain from vtc.demo.xml
> > error: internal error: early end of file from monitor: possible problem:
> > 2016-02-22T07:38:42.169035Z qemu-kvm: -device vfio-pci,host=10:10.3,id=hostdev0,bus=pci.0,addr=0x3: vfio: error, group 17 is not viable, please ensure all devices within the iommu_group are bound to their vfio bus driver.  
> 
> One possible meaning is that vfio sees multiple devices in the same 
> iommu group as the VF at 10:10.3 (in case there is some other possible 
> cause, I'm Cc'ing Alex Williamson, the vfio author). You can check this 
> by looking at the output of "virsh nodedev-dumpxml pci_0000_10_10_3" and 
> look at the "iommuGroup" section - if there are multiple addresses 
> listed there, then there are multiple devices in the same iommu group.
> 
> It could be that your particular chipset needs a "quirk" in the kernel 
> to be told that the VFs really can be in separate iommu groups; without 
> that quirk, all 14 VFs show up in the same iommu group, so the only way 
> to assign one to a guest is to assign *all* of them to the same guest 
> (or at least detach all of them from the VF driver and attach to 
> vfio-pci, then only assign one of them to a guest while the others sit 
> unused).
> 
> Since you're still running a 7.1 kernel, you may want to try updating to 
> the latest available and see if that solves your problem.
> 
> If you have further questions, please include the output of "virsh 
> nodedev-dumpxxml pci_0000_10_10_3" and full lspci output (among other 
> things, that should tell us which chipset your machine uses).

It's almost certain this is due to the 82576 card being installed into
a root port that does not have native ACS support.  Updating to a newer
kernel may or may not solve that problem, it really depends on what the
upstream port is from the device.  In addition to the libvirt commands
above, /sys/kernel/iommu_groups/ will show you all of the isolated
groups on the system.  For further details, see:

http://vfio.blogspot.com/2014/08/iommu-groups-inside-and-out.html

Thanks,
Alex