Re: [RFC][PATCH 0/3] perf support for amd guest/host-only bits

[Ingo Molnar <mingo@elte.hu>]

* Roedel, Joerg <Joerg.Roedel@amd.com> wrote:

> On Tue, May 10, 2011 at 11:21:32AM -0400, Ingo Molnar wrote:
> > 
> > * Joerg Roedel <joerg.roedel@amd.com> wrote:
> > 
> > > So any feedback is greatly appreciated :-)
> > 
> > it's nice to see this hw feature utilized :)
> 
> Yeah, I planned to do this for some time now and finally found the time
> to do it :)
> 
> > 
> > > Diffstat:
> > > 
> > >  arch/x86/include/asm/perf_event.h    |    3 +++
> > >  arch/x86/kernel/cpu/perf_event_amd.c |    6 ++++++
> > >  include/linux/perf_event.h           |    5 ++++-
> > >  kernel/perf_event.c                  |    4 ++++
> > >  tools/perf/util/parse-events.c       |   10 +++++++++-
> > >  5 files changed, 26 insertions(+), 2 deletions(-)
> > 
> > I see you have not touched tools/perf/builtin-kvm.c - how does this feature 
> > integrate with the features of 'perf kvm'? (are you using perf kvm by any 
> > chance?)
> 
> builtin-kvm looks like a wrapper for other perf-cmds which just sets up
> some additional state to tell the sub-cmds that guest-symbols need to be
> resolved and so on.
>
> Implementing this into the place where the other exclude-bits get set (which 
> is in generic code) seemed like a good idea to me. Touching builtin-kvm was 
> unnecessary for this.

ok, then let me put it in another way:

'perf kvm' is the tool that is making Linux instrumentation friendlier to KVM 
developers. Can you think of ways to utilize this new feature there?

For example a new 'perf kvm stat' feature could pass the exclusion bits to perf 
stat automatically. 'perf kvm record' could record on the guest only by 
default.

Things like that.

Thanks,

	Ingo