Re: KVM Guest Detection

[Chris J Arges <chris.j.arges@canonical.com>]

On 02/09/2015 07:26 AM, Paolo Bonzini wrote:
> 
> 
> On 06/02/2015 21:08, Chris J Arges wrote:
>> Is there a architecture and machine type independent way to detect that
>> one is running inside a KVM guest? I've noticed the following systemd
>> code which does this detection and it seems to be very architecture
>> dependent for KVM:
>> https://github.com/systemd/systemd/blob/master/src/shared/virt.c
> 
> No, there is no other way.  In fact on some architectures (PPC and s390
> for example) KVM implements the same paravirtualized architecture that
> is used by proprietary hypervisors, and should be indistiguishable from
> them.
> 
>> In addition one could grep for strings in the kernel log or CPU types if
>> using QEMU CPU model.
> 
> KVM can also be used without QEMU though.  Also, not all architectures
> support custom model names the way x86 has them in /proc/cpuinfo.
> 

Yes, I understand.

>> Given the many ways to do this, would it make sense to create a sysfs
>> entry (similar to how Xen does this with /sys/hypervisor/type), so that
>> one can easily tell they are running in a KVM guest?
> 
> Why do you need that?  If you are disabling something if you are on a
> virtualized platform, then that is most of the time (and I am not saying
> always only because of things like microcode.service) wrong.
> 

A use-case is disabling KSM if running inside an L1 guest. This would be
changed to ensure better defaults rather than work around issues. The
qemu-kvm packaging in Ubuntu enables KSM by default if you install the
package. If we are attempting to do things like nested guests, then it
may make sense to disable KSM when installing qemu-kvm in the L1 guest.
Another solution to this issue is to not enable KSM in the package and
let the user configure this when necessary.

> In order of likelihood:
> 
> 1) you are working around a bug in KVM, and the developers won't know;
> 
> 2) you are sweeping under the carpet a bug in your program (e.g. in the
> service that you're writing a unit file for);
> 
> 3) your assumptions are flawed.  For example the (now removed) readahead
> daemon in systemd was disabled under virtualization... except that most
> of the time production virtual machines are run with O_DIRECT so the
> benefit from readahead is the same as on bare metal.
> 
> Paolo
> 

Thanks for the response. My hope was that there could be a more uniform
way of exposing the fact that we are running on a hypervisor for
userspace programs.

--chris

>> I can work on a patch, but initial feedback would be helpful.
>>
>> Thanks,
>> --chris j arges
>> --
>> To unsubscribe from this list: send the line "unsubscribe kvm" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>
>