hvm hibernate

hvm hibernate

[Ky Srinivasan]

With the support of S4 sleep state (hibernate) for  HVM guests,  we now have a situation where the guest is resuming from hibernate while the hypervisor assumes that it is spinning up a new guest VM. And so, if the hypervisor maintains any guest specific state that needs to persist across a hibernate, there is no way to save and restore this state. I am running into this problem on win2k8 guests where the Hyperv shim has to maintain guest specific state that needs to persist across a hibernate.
Given that we already support save/restore which is functionally equivalent to hibernate, do we need to support S4 sleep state? 

Regards,

K. Y

RE: hvm hibernate

[Ian Pratt]

> With the support of S4 sleep state (hibernate) for  HVM guests,  we now
> have a situation where the guest is resuming from hibernate while the
> hypervisor assumes that it is spinning up a new guest VM. And so, if
> the hypervisor maintains any guest specific state that needs to persist
> across a hibernate, there is no way to save and restore this state. I
> am running into this problem on win2k8 guests where the Hyperv shim has
> to maintain guest specific state that needs to persist across a
> hibernate.
> Given that we already support save/restore which is functionally
> equivalent to hibernate, do we need to support S4 sleep state?

It's needed for Microsoft WHQL logo tests.

Have you switched over to using the hyperv code in xen-unstable or are you using the original Novell code?

Ian 

RE: hvm hibernate

[Ky Srinivasan]

>>> On 4/1/2009 at  4:01 PM, in message
<4FA716B1526C7C4DB0375C6DADBC4EA34172EC1B83@LONPMAILBOX01.citrite.net>, Ian
Pratt <Ian.Pratt@eu.citrix.com> wrote: 
>>  With the support of S4 sleep state (hibernate) for  HVM guests,  we now
>> have a situation where the guest is resuming from hibernate while the
>> hypervisor assumes that it is spinning up a new guest VM. And so, if
>> the hypervisor maintains any guest specific state that needs to persist
>> across a hibernate, there is no way to save and restore this state. I
>> am running into this problem on win2k8 guests where the Hyperv shim has
>> to maintain guest specific state that needs to persist across a
>> hibernate.
>> Given that we already support save/restore which is functionally
>> equivalent to hibernate, do we need to support S4 sleep state?
> 
> It's needed for Microsoft WHQL logo tests.
> 
> Have you switched over to using the hyperv code in xen-unstable or are you 
> using the original Novell code?

We have shipped the Novell shim in  sles11. This issue came about while running SVVP (MSFT's virtualization validation suite). If the firmware supports s4 sleep state then and only then sleep related tests are run. As I noted in my email, if the hypervisor needs to manage state that persists across hibernate, you will have a problem. I suspect, the hyperv code in xen-unstable will run into the same problem - for instance guest_os_id is written by the guest via an MSR write and that is maintained by the hypervisor. If this guest were to hibernate and come back, this state will not be correctly set.

Regards,

K. Y
> 
> Ian 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel