[libvirt] [RFC] exporting KVM host power saving capabilities through libvirt

[Dave Allan]

On Tue, Jun 28, 2011 at 08:55:31PM +0530, Vaidyanathan Srinivasan wrote:
> Hi,
> 
> Linux host systems running KVM support various power management
> capabilities.  Most of the features like DVFS and sleep states can be
> independently exploited by the host system itself based on system
> utilisation subject to policies set by the administrator.
> 
> However, system-wide low power states like S3 and S4 would require
> external communication and interaction with the systems management
> stack in order to be used.  The first steps in this direction would be
> to allow systems management stack to discover host power saving
> capabilities like S3 and S4 along with various other host CPU
> capabilities.
> 
> Libvirt seems to be the main glue layer between the platform and the
> systems-management stack.  Adding host power savings capabilities as
> part of libvirt host discovery mechanism seems to be one possible
> approach without addition of any new APIs or agents.

Can you provide the use cases you're looking to address with this
work?

> libvirt has virConnectGetCapabilities() that would export an XML file
> describing the capabilities of the host platform and guest features.
> 
> KVM hypervisor's capability to support S3 can be exported as a host
> feature in the XML as follows:
> 
> <host>
>     <uuid>94a3492f-2635-2491-8c87-8de976fad119</uuid>
>     <cpu>
>       <arch>x86_64</arch>
>       <features>        <<<=== New host feature fields
>         <S3/>
>         <S4/>
>       </features>

Just my $.02, but calling it <features> seems to be confusingly close
to the existing <feature> tag.  Maybe <power_management> ?

>       <model>Westmere</model>
>       <vendor>Intel</vendor>
>       <topology sockets='1' cores='2' threads='2'/>
>       <feature name='rdtscp'/>
>       <feature name='xtpr'/>
>       <feature name='tm2'/>
>       <feature name='est'/>
>       <feature name='vmx'/>     <<<= These are host CPU features
> 
> System management software that works through libvirt, already queries
> this capabilities XML file and hence no new API is needed.
> 
> As simple discovery mechanism can be added to libvirt as follows:
> 
> Index: libvirt-0.9.0/src/qemu/qemu_capabilities.c
> ===================================================================
> --- libvirt-0.9.0.orig/src/qemu/qemu_capabilities.c
> +++ libvirt-0.9.0/src/qemu/qemu_capabilities.c
> @@ -738,6 +738,14 @@ virCapsPtr qemuCapsInit(virCapsPtr old_c
>      virCapabilitiesAddHostMigrateTransport(caps,
>                                             "tcp");
> 
> +    /* Add host energy management host capabilities */
> +
> +    /* if "pm-is-supported --suspend" == 0 */
> +    virCapabilitiesAddHostFeature(caps, "S3");
> +
> +    /* if "pm-is-supported --hibernate" == 0 */
> +    virCapabilitiesAddHostFeature(caps, "S4");
> +
>      /* First the pure HVM guests */
>      for (i = 0 ; i < ARRAY_CARDINALITY(arch_info_hvm) ; i++)
>          if (qemuCapsInitGuest(caps, old_caps,
> 
> 
> Please let me know your comments, I will code a working prototype
> shortly and post for review/discussion.

Dave