Re: [PATCH V2 4/5] ara virt interface of perf to support kvm guest os statistics collection in guest os

["Zhang, Yanmin" <yanmin_zhang@linux.intel.com>]

On Tue, 2010-06-22 at 10:24 +0200, Jes Sorensen wrote:
> On 06/21/10 11:31, Zhang, Yanmin wrote:
> > @@ -583,10 +584,20 @@ static void x86_pmu_disable_all(void)
> >  	}
> >  }
> >  
> > +#ifdef CONFIG_KVM_PERF
> > +static int kvm_hw_perf_enable(void);
> > +static int kvm_hw_perf_disable(void);
> > +#endif
> 
> Please put these prototypes into a header ... and create dummy stubs for
> them when CONFIG_KVM_PERF is not set.
Ok. I just didn't want to touch too much generic codes of perf.

> 
> >  void hw_perf_disable(void)
> >  {
> >  	struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
> >  
> > +#ifdef CONFIG_KVM_PERF
> > +	if (!kvm_hw_perf_disable())
> > +		return;
> > +#endif
> 
> If you stub them out we can avoid all the ugly #ifdefs
Ok.

> 
> > @@ -810,6 +821,11 @@ void hw_perf_enable(void)
> >  	struct hw_perf_event *hwc;
> >  	int i, added = cpuc->n_added;
> >  
> > +#ifdef CONFIG_KVM_PERF
> > +	if (!kvm_hw_perf_enable())
> > +		return;
> > +#endif
> 
> and here....
Ok.

> 
> > @@ -1317,6 +1334,11 @@ void __init init_hw_perf_events(void)
> >  
> >  	pr_info("Performance Events: ");
> >  
> > +#ifdef CONFIG_KVM_PERF
> > +	if (!kvm_init_hw_perf_events())
> > +		return;
> > +#endif
> 
> and again here :)
Ok. Peter is working out a couple of patches to support multiple PMU. His patches
change pmu difition and we might move some into the callbacks. That will become
much clearer.

Yanmin