[PVE-User] Compiling newer kvm-kmod on PVE 2.1

[David Black]

Hi Dietmar,

I found vanilla Linux kernels - the only ones kvm-kmod-3.5 would compile against, don't have the proper volume label/LVM support to mount /dev/mapper/pve-root at boot time.  As you mention, with the PVE kernel based on RHEL6, I wasn't surprised kvm-kmod also would not compile against its source.  What did work is a stock Fedora 3.5.3 kernel with the extra-modules package, the latter to pick up the dlm module for PVE clustering.  Live migrating a CentOS 6 VM back and forth worked with another node running stock PVE 2.1 - something I expected would break, but didn't.

The ESXi 5.0u1 installer at first appears to boot normally, then very slowly goes through initializing the chipset and scheduler.  It eventually purple screens while initializing timing.  XenServer 6.0.2 installs normally, and I haven't yet tried running a VM.  BTW, on that same 3.5.3 host kernel and kvm-intel loaded with nested=1, XenServer does see the CPU's vmx flag but warns no hardware virtualization is available.  I also downloaded Hyper-V 2012, and out of curiosity plan to shortly try that.  The current kvm-intel nested hypervisor code may only work with kvm on kvm.  In this respect, ESXi 5 is far ahead.

Anyway, after finding the current OpenVZ patch set is only available for 2.6.32, it seems to me that for a stable kernel supporting the root filesystem on LVM, KVM and OpenVZ, PVE 2.1 may already have the optimal combination.  :-)

Dave

----- Original Message -----
> > It appears substantial backporting may have been done to the
> >  2.6.32-14-pve
> > kernel because function prototypes from vanilla kernels as new as
> > 3.0.0 are
> > present
> 
> PVE is based on RHEL6 kernel, so many feature are already backported.
> 
> - Dietmar
>