All of lore.kernel.org
 help / color / mirror / Atom feed
* intel core2 full virtualized performance counters
@ 2009-06-26 12:58 Dietmar Hahn
  2009-06-29  6:43 ` Shan, Haitao
  0 siblings, 1 reply; 3+ messages in thread
From: Dietmar Hahn @ 2009-06-26 12:58 UTC (permalink / raw)
  To: xen-devel; +Cc: Haitao Shan

Hi list,

i played around with the full virtualized performance counters (vpmu stuff) on 
intel core2 processors in a HVM guest.
I got something to work on the processor family=6 and model=15. But with newer 
processors (model=26 -> E5520 and model=29 -> X7460) nothing happens.
I searched a little bit in the sources and found in vpmu_initialise() that only 
for model 15 and 23 the  vpmu->arch_vpmu_ops gets initialized.
Is this missing or do I something wrong? Maybe anybody is working on this?
Thanks in advance for all hints!

Dietmar.

^ permalink raw reply	[flat|nested] 3+ messages in thread

* RE: intel core2 full virtualized performance counters
  2009-06-26 12:58 intel core2 full virtualized performance counters Dietmar Hahn
@ 2009-06-29  6:43 ` Shan, Haitao
  2009-06-29 10:51   ` Dietmar Hahn
  0 siblings, 1 reply; 3+ messages in thread
From: Shan, Haitao @ 2009-06-29  6:43 UTC (permalink / raw)
  To: Dietmar Hahn, xen-devel

Dietmar Hahn wrote:
> Hi list,
> 
> i played around with the full virtualized performance counters (vpmu
> stuff) on intel core2 processors in a HVM guest.
> I got something to work on the processor family=6 and model=15. But
> with newer processors (model=26 -> E5520 and model=29 -> X7460)
> nothing happens. 
> I searched a little bit in the sources and found in vpmu_initialise()
> that only for model 15 and 23 the  vpmu->arch_vpmu_ops gets
> initialized. 
> Is this missing or do I something wrong? Maybe anybody is working on
> this? Thanks in advance for all hints!
> 
> Dietmar.

The performance monitor facilities differ across CPU models. At the time vpmu is developed, model 15 and model 23 are the only two  models that are supported. The vpmu is designed to work with model-specific hanlding code. But actually no one adds new support for newer generation of CPUs.

I do not know how much differences between Core 2 PMU and Core i7. If you only collect performance data using architectural performance counters and events, I think it would be easy for you to hack the code in vpmu_initialize a bit to get it work.

Best Regards
Shan Haitao

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: RE: intel core2 full virtualized performance counters
  2009-06-29  6:43 ` Shan, Haitao
@ 2009-06-29 10:51   ` Dietmar Hahn
  0 siblings, 0 replies; 3+ messages in thread
From: Dietmar Hahn @ 2009-06-29 10:51 UTC (permalink / raw)
  To: xen-devel; +Cc: Shan, Haitao


> Dietmar Hahn wrote:
> > Hi list,
> >
> > i played around with the full virtualized performance counters (vpmu
> > stuff) on intel core2 processors in a HVM guest.
> > I got something to work on the processor family=6 and model=15. But
> > with newer processors (model=26 -> E5520 and model=29 -> X7460)
> > nothing happens.
> > I searched a little bit in the sources and found in vpmu_initialise()
> > that only for model 15 and 23 the  vpmu->arch_vpmu_ops gets
> > initialized.
> > Is this missing or do I something wrong? Maybe anybody is working on
> > this? Thanks in advance for all hints!
> >
> > Dietmar.
>
> The performance monitor facilities differ across CPU models. At the time
> vpmu is developed, model 15 and model 23 are the only two  models that are
> supported. The vpmu is designed to work with model-specific hanlding code.
> But actually no one adds new support for newer generation of CPUs.

This is a little bit surprising for me because I thought Intel did this for 
using their performance analyzer software such as vtune in HVM guest.

>
> I do not know how much differences between Core 2 PMU and Core i7. If you
> only collect performance data using architectural performance counters and
> events, I think it would be easy for you to hack the code in
> vpmu_initialize a bit to get it work.

Yes I did this and it works. Thanks.

Dietmar.

>
> Best Regards
> Shan Haitao

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2009-06-29 10:51 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2009-06-26 12:58 intel core2 full virtualized performance counters Dietmar Hahn
2009-06-29  6:43 ` Shan, Haitao
2009-06-29 10:51   ` Dietmar Hahn

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.