yahoo-eng-team team mailing list archive

[bryan <2067757@xxxxxxxxxxxxxxxxxx>]

Public bug reported:

>From Linux kernel v4.19 onwards, the nested KVM parameter is enabled by
default for Intel and AMD. (Though your Linux distribution might
override this default, here is the official documentation of this:
https://www.kernel.org/doc/html/v5.7/virt/kvm/running-nested-guests.html

We are using OpenStack Zed on CentOS 9 and the VM is running on AMD
compute nodes, and the kernel version is: 5.14.0-386.el9.x86_64.

When we created an instance on AMD server and set the "cpu_mode" to
"none", we found that the "svm" feature is passed to the instance XML on
libvirt, but when we execu "lscpu" inside the VM, we can not see the
"svm" feature, so we could not create a L2 instance inside the VM.

However, when we set the "cpu_mode" to "host-passthrough" and hard
reboot the VM, the "svm" is set correctly within the VM.

For intel servers, we can create nested instances by default, and the
"cpu_mode" is also set to "none", and everything works well.

We guess it might because of some CPU feature dependencies which cause
this issue. Can you help us to take a look? Thanks

** Affects: nova
     Importance: Undecided
         Status: New


** Tags: amd

** Description changed:

  From Linux kernel v4.19 onwards, the nested KVM parameter is enabled by
  default for Intel and AMD. (Though your Linux distribution might
  override this default, here is the official documentation of this:
  https://www.kernel.org/doc/html/v5.7/virt/kvm/running-nested-guests.html
  
  We are using OpenStack Zed on CentOS 9 and the VM is running on AMD
  compute nodes, and the kernel version is: 5.14.0-386.el9.x86_64.
  
  When we created an instance on AMD server and set the "cpu_mode" to
  "none", we found that the "svm" feature is passed to the instance XML on
  libvirt, but when we execu "lscpu" inside the VM, we can not see the
  "svm" feature, so we could not create a L2 instance inside the VM.
  
  However, when we set the "cpu_mode" to "host-passthrough" and hard
  reboot the VM, the "svm" is set correctly within the VM.
  
+ For intel servers, we can create nested instances by default, and the
+ "cpu_mode" is also set to "none", and everything works well.
+ 
  We guess it might because of some CPU feature dependencies which cause
  this issue. Can you help us to take a look? Thanks

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to OpenStack Compute (nova).
https://bugs.launchpad.net/bugs/2067757

Title:
  AMD server do not support nested virtualization

Status in OpenStack Compute (nova):
  New

Bug description:
  From Linux kernel v4.19 onwards, the nested KVM parameter is enabled
  by default for Intel and AMD. (Though your Linux distribution might
  override this default, here is the official documentation of this:
  https://www.kernel.org/doc/html/v5.7/virt/kvm/running-nested-
  guests.html

  We are using OpenStack Zed on CentOS 9 and the VM is running on AMD
  compute nodes, and the kernel version is: 5.14.0-386.el9.x86_64.

  When we created an instance on AMD server and set the "cpu_mode" to
  "none", we found that the "svm" feature is passed to the instance XML
  on libvirt, but when we execu "lscpu" inside the VM, we can not see
  the "svm" feature, so we could not create a L2 instance inside the VM.

  However, when we set the "cpu_mode" to "host-passthrough" and hard
  reboot the VM, the "svm" is set correctly within the VM.

  For intel servers, we can create nested instances by default, and the
  "cpu_mode" is also set to "none", and everything works well.

  We guess it might because of some CPU feature dependencies which cause
  this issue. Can you help us to take a look? Thanks

To manage notifications about this bug go to:
https://bugs.launchpad.net/nova/+bug/2067757/+subscriptions