Re: [Xen-devel] [PATCH RFC 1/2] linux/vnuma: vnuma support for pv guest

[Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>]

On Thu, Aug 29, 2013 at 02:41:23PM +0100, David Vrabel wrote:
> On 29/08/13 01:11, Matt Wilson wrote:
> > 
> > I think it wouldn't be too hard to construct the static ACPI tables
> > and provide them in acpi_os_get_root_pointer().
> 
> Elena's approach looks fine to me from a Linux point of view.

From an maintaince perspective the question is:

 - Why not re-use existing code? As in, if both PV and HVM can use
   SRAT, why not do it that way? This way you are exercising the same
   code path in both guests and it means less bugs to chase.

   Incidententaly this also means that the mechanism to fetch the
   NUMA information from the hypervisor and construct the SRAT tables
   from can be done the same in both hvmloader and Linux kernel.
   Which is nice - b/c if you find bugs in one, there is a chance
   they are the same in the other.

   Also this means one hypercall for both PV and HVM which is neat as
   well.

 - Lastly, the question is - a new 'interface' in the generic code is
   added to deal exclusively with Xen. Why special case it when one can
   use the SRAT for example? Is there a compeling technical argument?
   (Is SRAT inferior to vNUMA?)

Of course the counter argument is:

 - It will require more work as now Linux has to use the hypercall to
   construct an ACPI SRAT table. And we need to create a "fake" SRAT
   table and as well an RS.. something structure so that ACPI code
   can find it. Or perhaps this is something libxl can do? Which would
   mean the code to construct this and for HVM is both in Xen code base.
   And the only thing Linux needs to do is turn NUMA on for PV guests
   if it finds an ACPI table (perhaps by scanning the E820?).

 - Will it work? As in will an PV domU be able to run with just one
   ACPI table - just the SRAT and nothing else? Or will it fall flat
   on its face?
> 
> It's still not clear how presenting the information via ACPI would be an
> improvement.  I also think it is preferrable for the hypercall ABI for
> PV guests to be independent of architecture specifics where possible.

I concur on the hypercall ABI - but I think the hypercall should be
available on both PV and HVM so that it can be used in both.
> 
> If you posted Xen and Linux series producing/using the ACPI SRAT and
> SLIT tables for dom0 and domU PV guests we could better evaluate that
> option but until then I'm inclined to consider Elena's approach as the
> better one.

I think short term Elena's option is better one as it gets it going and
we can experiement with it. Long term I think stashing the data in ACPI
SRAT/SLIT is right. But Elena might not get to it in the next couple of
months - which means somebody else will have to sign up for that.

P.S.
I did for fun enable ACPI support in the PV domU guests and it boots fine.
It does complain about missing tables, but that was it.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel