approach for heavily hw-oriented values in smbios for guests

approach for heavily hw-oriented values in smbios for guests

[Andrew D. Ball]

When I was implementing smbios, I occasionally put in mock values for
things like the data width of dimms, much to my chagrin.

What do y'all think the best approach to such situations would be?  If
specific migration groups were enforced, such that a guest will not be
migrated to a different type of cpu/memory/etc., putting accuracte data
there makes sense to me.  Otherwise, I'd prefer to put no values.
However, even though the SMBIOS 2.4 standard specifies lots of places
where a value of 'unknown' is syntactically valid, the comformance
guidelines require a ton of very hardware oriented things to be fully
specified.

Thanks for your help!
Andrew

Re: approach for heavily hw-oriented values in smbios for guests

[Keir Fraser]

On 16/11/06 15:54, "Andrew D. Ball" <aball@linux.vnet.ibm.com> wrote:

> When I was implementing smbios, I occasionally put in mock values for
> things like the data width of dimms, much to my chagrin.
> 
> What do y'all think the best approach to such situations would be?  If
> specific migration groups were enforced, such that a guest will not be
> migrated to a different type of cpu/memory/etc., putting accuracte data
> there makes sense to me.  Otherwise, I'd prefer to put no values.
> However, even though the SMBIOS 2.4 standard specifies lots of places
> where a value of 'unknown' is syntactically valid, the comformance
> guidelines require a ton of very hardware oriented things to be fully
> specified.

I'm pretty sure that consistent lying is the best approach for most such
fields.

 -- Keir

Re: approach for heavily hw-oriented values in smbios for guests

[Alan]

On Thu, 16 Nov 2006 10:54:36 -0500
"Andrew D. Ball" <aball@linux.vnet.ibm.com> wrote:

> However, even though the SMBIOS 2.4 standard specifies lots of places
> where a value of 'unknown' is syntactically valid, the comformance
> guidelines require a ton of very hardware oriented things to be fully
> specified.

Xen is just very hot-pluggable. I guess you either lie or change it to
match the hardware each time. Which is more correct unfortunately appears
to depend upon the desired answer but I'd favour the real hardware - for
diagnostics.

Alan

Re: approach for heavily hw-oriented values in smbios for guests

[Andrew D. Ball]

As far as I know, SMBIOS data reflects the state of the system at boot
time only.  No updates after any sort of hot-plugging.  However, I
distantly remember some possibility of updating something, maybe a power
state.

However, the format of the data is so fixed, that dynamic updates, like
adding vCPUs are infeasible.

Peace.
Andrew


On Thu, 2006-11-16 at 18:32 +0000, Alan wrote:
> On Thu, 16 Nov 2006 10:54:36 -0500
> "Andrew D. Ball" <aball@linux.vnet.ibm.com> wrote:
> 
> > However, even though the SMBIOS 2.4 standard specifies lots of places
> > where a value of 'unknown' is syntactically valid, the comformance
> > guidelines require a ton of very hardware oriented things to be fully
> > specified.
> 
> Xen is just very hot-pluggable. I guess you either lie or change it to
> match the hardware each time. Which is more correct unfortunately appears
> to depend upon the desired answer but I'd favour the real hardware - for
> diagnostics.
> 
> Alan
>