Xen devel - Sep 2011 - Architectural question: How to put BTS and PEBS in Xen

Hi list,

I''m on the way to coding the virtualisation of BTS (Branch Trace Store)
and
later PEBS support for Intel CPU''s in Xen.

As some aspects overlap with the PMU (Performance Management Unit) I used the
existing infrastructure of the VPMU stuff to do some coding and tests.
But currently this infrastructure is only switched on with a
''vpmu'' boot flag
in the  hypervisor.
As reminder this flag was introduced because of an inexplicable behaviour of
the Nehalem CPUs while using the performance counters which could lead to an
endless interrupt loop (see check_pmc_quirk()).

As currently no one else than we seem to use/need this feature for simplicity
I would move the BTS code complete to the VPMU environment and therfore it
would only be usable when using the ''vpmu'' flag on boot.

Any advices would be welcome!

Thanks.
Dietmar.


-- 
Company details: http://ts.fujitsu.com/imprint.html

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 27/09/2011 02:51, "Dietmar Hahn"
<dietmar.hahn@ts.fujitsu.com> wrote:
> Hi list,
> 
> I''m on the way to coding the virtualisation of BTS (Branch Trace
Store) and
> later PEBS support for Intel CPU''s in Xen.
> 
> As some aspects overlap with the PMU (Performance Management Unit) I used
the
> existing infrastructure of the VPMU stuff to do some coding and tests.
> But currently this infrastructure is only switched on with a
''vpmu'' boot flag
> in the  hypervisor.
> As reminder this flag was introduced because of an inexplicable behaviour
of
> the Nehalem CPUs while using the performance counters which could lead to
an
> endless interrupt loop (see check_pmc_quirk()).
> 
> As currently no one else than we seem to use/need this feature for
simplicity
> I would move the BTS code complete to the VPMU environment and therfore it
> would only be usable when using the ''vpmu'' flag on boot.
It sounds quite sensible to me.

 -- Keir
> Any advices would be welcome!
> 
> Thanks.
> Dietmar.
> 


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel