Re: [PATCH 4 of 5] Add code to track the address of the VM generation id buffer across a

[Tim Deegan <tim@xen.org>]

At 13:26 +0000 on 28 Nov (1322486771), Paul Durrant wrote:
> > > +/* Address of VM generation id buffer */ #define
> > > +HVM_PARAM_VM_GENERATION_ID_ADDR 27
> > > +
> > > +#define HVM_NR_PARAMS          28
> > >
> > >  #endif /* __XEN_PUBLIC_HVM_PARAMS_H__ */
> > 
> > Since the hypervisor doesn't use this value, I don't think this is
> > the right place for it to live.  Tools-only VM metadata should
> > probably go in xenstore or in the toolstacks' own datastructures.
> > 
> 
> Yes, I was a little unsure about that. Is there any precedent for
> getting information from hvmloader out to the tools? Writing guest
> physical addresses into xenstore seems like the wrong thing to but
> apart from HVM params or the shared hvm info page, I couldn't think of
> another way apart from putting the VM generation id in a static
> well-known place. 

Presumably in this case the tools can find the address by walking the
ACPI tables, just like the guest OS would. :P  Not very pretty, though. 

I don't think there's anything wrong with putting a GPA into
xenstore.  All the other tools interactions are in guest-physical
addressing already.  IIRC HVMloader's xenbus client doesn't have a
xenstore_write() but it shouldn't be hard to add.  

I can't think of another way to pass data from hvmloader to the
tools right now -- I think the hvm-info page should probably be
replaced entirely with xenstore keys at some point. 

Tim.