How do you disable hpet in a kvm?

[Fred Brier]

I installed PBX In A Flash (PIAF) as a VM on a Fedora 11 64 bit host.  
PIAF is a 32 bit CentOS 5.2 distro for running Asterisk.  I, like many 
others, am getting the error message "rtc: lost some interrupts at 
1024hz" repeating over and over on the console.  Some people have 
apparently gotten rid of this error message by disabling hpet in the 
BIOS.  However, this is a VM (kvm).

I tried adding nohpet as a kernel parameter in /boot/grub/menu.lst, but 
that effects the linux OS, not the BIOS.  It did not fix it.

I found references to a -no-hpet qemu parameter.  However, I cannot find 
how to set it in a VM created using virt-manager and an ISO image.  In 
the /etc/libvirt of the host I found XML configuration files describing 
the VMs.  These appear to be the same as those created using the virsh 
dumpxml command.  There does not appear to be an XML element as there is 
for acpi and apic.  The man page describing the XML format does not have 
an HPET related option.  I could not find an example or sample of it 
being disabled in a libvirt XML file.  Nor does there appear to be a 
schema (xsd or dtd) for the libvirt file format, which might have had an 
hpet element or attribute.

Ideally, is there a way to disable HPET in the BIOS for a specific VM, 
and not all VMs.  Thank you.

Fred
//

[Laine Stump]

On 06/08/2010 02:15 PM, Frederick N. Brier wrote:
>  I installed PBX In A Flash (PIAF) as a VM on a Fedora 11 64 bit host.
>  
>  PIAF is a 32 bit CentOS 5.2 distro for running Asterisk.  I, like many 
>  others, am getting the error message "rtc: lost some interrupts at 
>  1024hz" repeating over and over on the console.  Some people have 
>  apparently gotten rid of this error message by disabling hpet in the 
>  BIOS.  However, this is a VM (kvm).
>
>  I tried adding nohpet as a kernel parameter in /boot/grub/menu.lst, 
>  but that effects the linux OS, not the BIOS.  It did not fix it.
>
>  I found references to a -no-hpet qemu parameter.  However, I cannot 
>  find how to set it in a VM created using virt-manager and an ISO 
>  image.  In the /etc/libvirt of the host I found XML configuration 
>  files describing the VMs.  These appear to be the same as those 
>  created using the virsh dumpxml command.  There does not appear to be 
>  an XML element as there is for acpi and apic.  The man page describing 
>  the XML format does not have an HPET related option.  I could not find 
>  an example or sample of it being disabled in a libvirt XML file.  Nor 
>  does there appear to be a schema (xsd or dtd) for the libvirt file 
>  format, which might have had an hpet element or attribute.
>
>  Ideally, is there a way to disable HPET in the BIOS for a specific VM, 
>  and not all VMs.  Thank you. 
If you are running libvirt 0.8.0 or newer, and if your qemu-kvm supports 
the -no-hpet option (check the output of "qemu-kvm --help"), you can add 
the -no-hpet option for qemu-kvm for a particular domain by editing the 
domain's XML (with "virsh edit domainname") and adding

<timer name='hpet' present='no'/>

to the "clock" section. It will end up looking something like this:


<clock offset='utc'>
<timer name='hpet' present='no'/>
</clock>

[Fred Brier]

On 06/08/2010 04:37 PM, Laine Stump wrote:
>  On 06/08/2010 02:15 PM, Frederick N. Brier wrote:
> > I installed PBX In A Flash (PIAF) as a VM on a Fedora 11 64 bit 
> > host.  PIAF is a 32 bit CentOS 5.2 distro for running Asterisk.  I, 
> > like many others, am getting the error message "rtc: lost some 
> > interrupts at 1024hz" repeating over and over on the console.  Some 
> > people have apparently gotten rid of this error message by disabling 
> > hpet in the BIOS.  However, this is a VM (kvm).
> >
> > I tried adding nohpet as a kernel parameter in /boot/grub/menu.lst, 
> > but that effects the linux OS, not the BIOS.  It did not fix it.
> >
> > I found references to a -no-hpet qemu parameter.  However, I cannot 
> > find how to set it in a VM created using virt-manager and an ISO 
> > image.  In the /etc/libvirt of the host I found XML configuration 
> > files describing the VMs.  These appear to be the same as those 
> > created using the virsh dumpxml command.  There does not appear to be 
> > an XML element as there is for acpi and apic.  The man page 
> > describing the XML format does not have an HPET related option.  I 
> > could not find an example or sample of it being disabled in a libvirt 
> > XML file.  Nor does there appear to be a schema (xsd or dtd) for the 
> > libvirt file format, which might have had an hpet element or attribute.
> >
> > Ideally, is there a way to disable HPET in the BIOS for a specific 
> > VM, and not all VMs?  Thank you.
>
>  If you are running libvirt 0.8.0 or newer, and if your qemu-kvm 
>  supports the -no-hpet option (check the output of "qemu-kvm --help"), 
>  you can add the -no-hpet option for qemu-kvm for a particular domain 
>  by editing the domain's XML (with "virsh edit domainname") and adding
>
>  <timer name='hpet' present='no'/>
>
>  to the "clock" section. It will end up looking something like this:
>
>
>  <clock offset='utc'>
>  <timer name='hpet' present='no'/>
>  </clock> 
Brilliant answer.  This is exactly the feature I need, but my Fedora
11 
hosts are at libvirt 0.6.2 and Fedora 13 appears to only be up to 
0.7.7.  Is it possible (and stable) to upgrade a Fedora 11 or 13 to 
libvirt 0.8.0+?  Could you please post a link to documentation on how to 
upgrade libvirt?  Is there a specific 0.8.0 version you would upgrade 
to?  Thank you.

Fred

[Ján ONDREJ (SAL)]

>  >If you are running libvirt 0.8.0 or newer, and if your qemu-kvm
>  >supports the -no-hpet option (check the output of "qemu-kvm
>  >--help"), you can add the -no-hpet option for qemu-kvm for a
>  >particular domain by editing the domain's XML (with "virsh edit
>  >domainname") and adding
>  >
>  ><timer name='hpet' present='no'/>
>  >
>  >to the "clock" section. It will end up looking something like this:
>  >
>  >
>  ><clock offset='utc'>
>  ><timer name='hpet' present='no'/>
>  ></clock>
>  Brilliant answer.  This is exactly the feature I need, but my Fedora
>  11 hosts are at libvirt 0.6.2 and Fedora 13 appears to only be up to
>  0.7.7.  Is it possible (and stable) to upgrade a Fedora 11 or 13 to
>  libvirt 0.8.0+?  Could you please post a link to documentation on
>  how to upgrade libvirt?  Is there a specific 0.8.0 version you would
>  upgrade to?  Thank you. 
I vote for F13 update to libvirt 0.8.0 too. :-)
Works well for me on one testing machine.

Another solution is to use F13 + virt-preview repository. This contains
needed libvirt.

			SAL

[Laine Stump]

On 06/09/2010 02:18 AM, Ján ONDREJ (SAL) wrote:
> > Is it possible (and stable) to upgrade a Fedora 11 or 13 to
> > libvirt 0.8.0+?  Could you please post a link to documentation on
> > how to upgrade libvirt?  Is there a specific 0.8.0 version you would
> > upgrade to?  Thank you.
> >      
>  I vote for F13 update to libvirt 0.8.0 too. :-)
>  Works well for me on one testing machine.
>
>  Another solution is to use F13 + virt-preview repository. This contains
>  needed libvirt.
>      
Actually there are a *lot* of very good bugfixes in 0.8.1. If you're 
upgrading, you might as well go all the way.

I don't know that there are pre-built packages for F11 anywhere; the 
quickest route to upgrading may be to get the source and build it (there 
are a few dependencies you'll need to install). (I also haven't tried 
putting libvirt 0.8.1 on F11, as I don't have any machiens left running 
it; it's possible there is a dependency in the new libvirt that isn't 
available with yum on F11, and you would have to build *that* package 
from source as well. If that happens, you'll just need to make a 
decision about when it becomes less useful to continue ;-)

[Fred Brier]

On 06/09/2010 09:48 AM, Laine Stump wrote:
>  On 06/09/2010 02:18 AM, Ján ONDREJ (SAL) wrote:
> >> Is it possible (and stable) to upgrade a Fedora 11 or 13 to
> >> libvirt 0.8.0+?  Could you please post a link to documentation on
> >> how to upgrade libvirt?  Is there a specific 0.8.0 version you would
> >> upgrade to?  Thank you.
> > I vote for F13 update to libvirt 0.8.0 too. :-)
> > Works well for me on one testing machine.
> >
> > Another solution is to use F13 + virt-preview repository. This contains
> > needed libvirt.
>  Actually there are a *lot* of very good bugfixes in 0.8.1. If you're 
>  upgrading, you might as well go all the way.
>
>  I don't know that there are pre-built packages for F11 anywhere; the 
>  quickest route to upgrading may be to get the source and build it 
>  (there are a few dependencies you'll need to install). (I also haven't 
>  tried putting libvirt 0.8.1 on F11, as I don't have any machiens left 
>  running it; it's possible there is a dependency in the new libvirt 
>  that isn't available with yum on F11, and you would have to build 
>  *that* package from source as well. If that happens, you'll just need 
>  to make a decision about when it becomes less useful to continue ;-) 
I have followed all your suggestions, and it appears to have worked.  No 
more rtc lost interrupt messages.  I upgraded the host to F13 using 
preupgrade.  This was the second time I have done this, and it has 
worked cleanly both times.  I then added the virt-preview repo, updated 
libvirt and the related packages, rebooted, added the <timer> element to 
the guest VM, and started the VM.  No annoying console messages.  Life 
is good.  Now back to finishing configuring Asterisk.  Thank you VERY 
much for all the help.

Fred