From Fedora Project Wiki

(Mention /dev/vfio/vfio permissions bug)
(Drop explicit VFIO bits, it's the default now)
Line 7: Line 7:
 
* Functioning F19+ host and F19+ VM.
 
* Functioning F19+ host and F19+ VM.
 
* PCI device you can afford to devote entirely to a VM
 
* PCI device you can afford to devote entirely to a VM
 +
* Verify your host as necessary hardware support for VFIO
 +
** If on an Intel CPU, this command should show at least 2 matches: <code>dmesg <nowiki>|</nowiki> grep -e DMAR -e IOMMU</code>
 +
** If on an AMD CPU, this command should show some output: <code>dmesg <nowiki>|</nowiki> grep AMD-Vi</code>
  
 
|actions=
 
|actions=
Line 17: Line 20:
 
# Shut down the guest and check the device is functional within the host
 
# Shut down the guest and check the device is functional within the host
 
# Select the device in virt-manager and ''Remove'' it
 
# Select the device in virt-manager and ''Remove'' it
 
== Device assignment with VFIO ==
 
 
[http://www.linux-kvm.org/wiki/images/d/d1/2011-forum-VFIO.pdf VFIO] is the new and improved PCI assignment method for kvm. It will likely become the default through libvirt in the future.
 
 
# Verify your host as necessary hardware support for VFIO
 
#* If on an Intel CPU, this command should show at least 2 matches: <code>dmesg <nowiki>|</nowiki> grep -e DMAR -e IOMMU</code>
 
#* If on an AMD CPU, this command should show some output: <code>dmesg <nowiki>|</nowiki> grep AMD-Vi</code>
 
# On the host, do <code>sudo chmod 666 /dev/vfio/vfio</code>  ([https://bugzilla.redhat.com/show_bug.cgi?id=967230 this is an F19 bug])
 
# Follow the regular PCI assignment steps up to step 4.
 
# Change the host device to use the VFIO method:
 
#* <code>sudo virsh edit test-day-vm</code>
 
#* Add <code><driver name="vfio"/></code> to the PCI <code><hostdev></code> block
 
#* Save and exit
 
# Continue on at step 4 in the above steps.
 
  
 
|results=
 
|results=

Revision as of 17:04, 16 September 2014

Description

Assign a physical PCI device to a KVM guest, and verify that it works as expected.

Setup

  • Functioning F19+ host and F19+ VM.
  • PCI device you can afford to devote entirely to a VM
  • Verify your host as necessary hardware support for VFIO
    • If on an Intel CPU, this command should show at least 2 matches: dmesg | grep -e DMAR -e IOMMU
    • If on an AMD CPU, this command should show some output: dmesg | grep AMD-Vi

How to test

  1. Run virt-manager, open an existing shutoff guest and go to the details page.
  2. Add Hardware->PCI Host Device
  3. Choose the device you want to assign from the list, like: 00:19.0 Interface eth0 (82566DM-2 Gigabit Network Connection)
  4. Click Finish
  5. Start the guest and check the device is functional within the guest
  6. Shut down the guest and check the device is functional within the host
  7. Select the device in virt-manager and Remove it

Expected Results

No obvious errors occur. PCI device assignment can be quirky and does not always work with every PCI device or host PCI layout. If you get an error, ask in #fedora-test-day IRC first and we can likely provide some direction.