Re: [tip:perf/core] perf/x86: Fix USER/KERNEL tagging of samples

[Ingo Molnar <mingo@kernel.org>]

* Peter Zijlstra <peterz@infradead.org> wrote:

> On Tue, 2012-07-10 at 11:02 +0200, Peter Zijlstra wrote:
> > Ingo, do you want me to do a version where I simply bail on everything
> > if regs->{cs,ss} != {__USER_CS, __USER32_CS} || regs->flags & VM ?
> 
> Here's a variant that does that.. 

>  arch/x86/include/asm/perf_event.h         |   11 +++++--
>  arch/x86/kernel/cpu/perf_event.c          |   46 +++++++++++++++++++++++++-----
>  arch/x86/kernel/cpu/perf_event.h          |   20 +++++++++++++
>  arch/x86/kernel/cpu/perf_event_amd_ibs.c  |    4 +-
>  arch/x86/kernel/cpu/perf_event_intel_ds.c |    7 ++--
>  5 files changed, 74 insertions(+), 14 deletions(-)

this is the full thing:

>  arch/x86/include/asm/perf_event.h         |   11 ++-
>  arch/x86/kernel/cpu/perf_event.c          |   89 ++++++++++++++++++++++++++----
>  arch/x86/kernel/cpu/perf_event.h          |   20 ++++++
>  arch/x86/kernel/cpu/perf_event_amd_ibs.c  |    4 +
>  arch/x86/kernel/cpu/perf_event_intel_ds.c |    7 +-
>  5 files changed, 114 insertions(+), 17 deletions(-)

so that's 40 LOC difference.

Hm, I expected there to be more of a difference, so let me 
change my mind again in view of the evidence: now I tend to
lean Linus's way, we might as well apply those extra 40 lines
now that you've written them :-)

Even if it is not enough to do proper segmented profiling, 
should anyone be interested in such a profiling mode they'll 
have a much easier job making it work, the rest looks mostly a 
user space side job. Your larger patch looks safe enough at the 
boundaries.

Thanks,

	Ingo