Re: [PATCH 1/2] perf, x86: Add basic Ivy Bridge support

[Stephane Eranian]

On Fri, Jun 8, 2012 at 12:29 AM, Andi Kleen <andi@xxxxxxxxxxxxxx> wrote:
> From: Andi Kleen <ak@xxxxxxxxxxxxxxx>
>
> Very similar to Sandy Bridge, but there is no PEBS problem.
>
> Signed-off-by: Andi Kleen <ak@xxxxxxxxxxxxxxx>
> ---
> Âarch/x86/kernel/cpu/perf_event_intel.c | Â Â9 ++++++++-
> Â1 files changed, 8 insertions(+), 1 deletions(-)
>
> diff --git a/arch/x86/kernel/cpu/perf_event_intel.c b/arch/x86/kernel/cpu/perf_event_intel.c
> index 166546e..1249c56 100644
> --- a/arch/x86/kernel/cpu/perf_event_intel.c
> +++ b/arch/x86/kernel/cpu/perf_event_intel.c
> @@ -1698,6 +1698,7 @@ __init int intel_pmu_init(void)
> Â Â Â Âunion cpuid10_ebx ebx;
> Â Â Â Âunsigned int unused;
> Â Â Â Âint version;
> + Â Â Â char *name;
>
> Â Â Â Âif (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) {
> Â Â Â Â Â Â Â Âswitch (boot_cpu_data.x86) {
> @@ -1839,9 +1840,15 @@ __init int intel_pmu_init(void)
> Â Â Â Â Â Â Â Âpr_cont("Westmere events, ");
> Â Â Â Â Â Â Â Âbreak;
>
> + Â Â Â case 58: /* IvyBridge */
> + Â Â Â Â Â Â Â name = "Ivy";
> + Â Â Â Â Â Â Â goto snb_ivb_common;
> +
> Â Â Â Âcase 42: /* SandyBridge */
> Â Â Â Â Â Â Â Âx86_add_quirk(intel_sandybridge_quirk);
> Â Â Â Âcase 45: /* SandyBridge, "Romely-EP" */
> + Â Â Â Â Â Â Â name = "Sandy";
> + Â Â Â snb_ivb_common:
> Â Â Â Â Â Â Â Âmemcpy(hw_cache_event_ids, snb_hw_cache_event_ids,
> Â Â Â Â Â Â Â Â Â Â Â sizeof(hw_cache_event_ids));
>
> @@ -1861,7 +1868,7 @@ __init int intel_pmu_init(void)
> Â Â Â Â Â Â Â Âintel_perfmon_event_map[PERF_COUNT_HW_STALLED_CYCLES_BACKEND] =
> Â Â Â Â Â Â Â Â Â Â Â ÂX86_CONFIG(.event=0xb1, .umask=0x01, .inv=1, .cmask=1);
>

But as far as I know for the other generic stall event:
               /* UOPS_DISPATCHED.THREAD,c=1,i=1 to count stall cycles*/
                intel_perfmon_event_map[PERF_COUNT_HW_STALLED_CYCLES_BACKEND] =
                        X86_CONFIG(.event=0xb1, .umask=0x01, .inv=1, .cmask=1);

Event 0xb1, umask 0x1 does not exist on IVB. Only the
UOPS_DISPATCHED.CORE version does.
Given I don't have a definition for STALLED_CYCLES_BACKEND, I don't
know if measuring across
both HT thread would fit the bill. I would advise you keep this one
undefined on IVB otherwise this
may lead to confusion when comparing with SNB.


> - Â Â Â Â Â Â Â pr_cont("SandyBridge events, ");
> + Â Â Â Â Â Â Â pr_cont("%sBridge events, ", name);
> Â Â Â Â Â Â Â Âbreak;
>
> Â Â Â Âdefault:
> --
> 1.7.7.6
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/