KVM: Find already attached/consumed VFs by a VM?

kp^ · 09-04-2018, 02:37 PM

Using virt-manager in RHEL/SLES...

Is there a way to find out if a VF (virtual function) is already attached/consumed by a VM without powering it on?

Does the driver and or kernel report this somewhere?

The only mechanism I found was to dump and parse the xml of each VM. I'm still in the process of writing this code. This is a pita and imho think there might be a better way...

Code:

 virsh dumpxml rhel7.4

or for a quick view I can dump the xml and use xmllint to check:

Code:

virsh dumpxml rhel7.4 | xmllint --xpath '//hostdev' -

If you do a 'lspci -ks 0000:6d:01.1' for example the 'Kernel driver in use' is the driver that is installed on the host. The second you power on the VM this get switched to 'vfio-pci'. This tells me this VF is in use, but I dont think I should be forced to power on my VM until I'm ready.

Context: I created a bash script to add X number of VFs to a VM (vfs must already be spawned). The issue is that the same VF gets added to multiple VM's. I don't know there is an issue until I power on the VM.

Thanks!

*Not sure if this should be here or in the programming section, please forgive.

berndbausch · 09-04-2018, 07:26 PM

Quote:

Originally Posted by kp^

Using virt-manager in RHEL/SLES...

Is there a way to find out if a VF (virtual function) is already attached/consumed by a VM without powering it on?
Does the driver and or kernel report this somewhere?

Even if virt-manager told you whether an interface is based on a VF, how would you do this check in a shell script? The XML definition of your VM is an infinitely better solution. XML can be parsed, GUIs can't.

AFAIK libvirt, and all the tools that build on libvirt including virsh and the virtual machine manager GUI, don't know and don't care what kind of network device they are connected to anyway. You give them a device and that's it. It can be a physical or a virtual function, or a bridge or something else.

For example:

Code:

<interface type='hostdev' managed='yes'>
  <source>
    <address type='pci' domain='0x0' bus='4' slot='9' function='6' />
  </source>
  <mac address='aa:be:ef:23:23:01' />
</interface>

So, it depends on how exactly you specify interfaces. If it is via PCI address like in the above example (I don't know if there are other ways to map a VM interface to its host's VF), you just use lspci to check whether 4:9.6 is a virtual function or not. I don't think this is very hard to do in a shell script, with some awk and sed magic, or an XML parser.

kp^ · 09-06-2018, 01:47 PM

Thanks for the feedback berndbausch!

I do alot of testing/validation on network devices so I have various scripts that show me this type of info. I know if an interface is a PF or vf based on the driver. I also know the domain:bus:slot.function of each device etc... Dig into "/sys/class/net/<interface>" for example....

Yes, so the VF is added by a PCI address like you've shown below. Below is an example of two vfs that are attached to a VM..

I will prolly have to create a thread in the programming section!

Code:

<hostdev mode='subsystem' type='pci' managed='yes'>
      <source>
        <address domain='0x0000' bus='0x44' slot='0x01' function='0x4'/>
      </source>
      <address type='pci' domain='0x0000' bus='0x00' slot='0x08' function='0x0'/>
    </hostdev>
    <hostdev mode='subsystem' type='pci' managed='yes'>
      <source>
        <address domain='0x0000' bus='0x44' slot='0x01' function='0x0'/>
      </source>
      <address type='pci' domain='0x0000' bus='0x00' slot='0x1a' function='0x0'/>
    </hostdev>