From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S934352Ab1KJM6M (ORCPT ); Thu, 10 Nov 2011 07:58:12 -0500 Received: from mx1.redhat.com ([209.132.183.28]:39457 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1757658Ab1KJM6J (ORCPT ); Thu, 10 Nov 2011 07:58:09 -0500 From: Gleb Natapov To: kvm@vger.kernel.org Cc: linux-kernel@vger.kernel.org, avi@redhat.com, mtosatti@redhat.com, mingo@elte.hu, a.p.zijlstra@chello.nl, acme@ghostprotocols.net Subject: [PATCHv3 06/10] x86, perf: disable non available architectural events. Date: Thu, 10 Nov 2011 14:57:26 +0200 Message-Id: <1320929850-10480-7-git-send-email-gleb@redhat.com> In-Reply-To: <1320929850-10480-1-git-send-email-gleb@redhat.com> References: <1320929850-10480-1-git-send-email-gleb@redhat.com> Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Intel CPUs report non-available architectural events in cpuid leaf 0AH.EBX. Use it to disable events that are not available according to CPU. Signed-off-by: Gleb Natapov --- arch/x86/include/asm/perf_event.h | 14 ++++++++++++++ arch/x86/kernel/cpu/perf_event.h | 5 +++++ arch/x86/kernel/cpu/perf_event_intel.c | 29 ++++++++++++++++++++++++----- 3 files changed, 43 insertions(+), 5 deletions(-) diff --git a/arch/x86/include/asm/perf_event.h b/arch/x86/include/asm/perf_event.h index f61c62f..c6998bc 100644 --- a/arch/x86/include/asm/perf_event.h +++ b/arch/x86/include/asm/perf_event.h @@ -57,6 +57,7 @@ (1 << (ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX)) #define ARCH_PERFMON_BRANCH_MISSES_RETIRED 6 +#define ARCH_PERFMON_EVENTS_COUNT 7 /* * Intel "Architectural Performance Monitoring" CPUID @@ -72,6 +73,19 @@ union cpuid10_eax { unsigned int full; }; +union cpuid10_ebx { + struct { + unsigned int no_unhalted_core_cycles:1; + unsigned int no_instructions_retired:1; + unsigned int no_unhalted_reference_cycles:1; + unsigned int no_llc_reference:1; + unsigned int no_llc_misses:1; + unsigned int no_branch_instruction_retired:1; + unsigned int no_branch_misses_retired:1; + } split; + unsigned int full; +}; + union cpuid10_edx { struct { unsigned int num_counters_fixed:5; diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h index b9698d4..cd0ebcd 100644 --- a/arch/x86/kernel/cpu/perf_event.h +++ b/arch/x86/kernel/cpu/perf_event.h @@ -259,6 +259,11 @@ struct x86_pmu { int num_counters_fixed; int cntval_bits; u64 cntval_mask; + union { + unsigned long events_maskl; + unsigned long events_mask[BITS_TO_LONGS(ARCH_PERFMON_EVENTS_COUNT)]; + }; + int events_mask_len; int apic; u64 max_period; struct event_constraint * diff --git a/arch/x86/kernel/cpu/perf_event_intel.c b/arch/x86/kernel/cpu/perf_event_intel.c index e09ca20..301369a 100644 --- a/arch/x86/kernel/cpu/perf_event_intel.c +++ b/arch/x86/kernel/cpu/perf_event_intel.c @@ -1544,13 +1544,23 @@ static void intel_clovertown_quirks(void) x86_pmu.pebs_constraints = NULL; } +static int intel_event_id_to_hw_id[] = { + PERF_COUNT_HW_CPU_CYCLES, + PERF_COUNT_HW_INSTRUCTIONS, + PERF_COUNT_HW_BUS_CYCLES, + PERF_COUNT_HW_CACHE_REFERENCES, + PERF_COUNT_HW_CACHE_MISSES, + PERF_COUNT_HW_BRANCH_INSTRUCTIONS, + PERF_COUNT_HW_BRANCH_MISSES, +}; + __init int intel_pmu_init(void) { union cpuid10_edx edx; union cpuid10_eax eax; + union cpuid10_ebx ebx; unsigned int unused; - unsigned int ebx; - int version; + int version, bit; if (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) { switch (boot_cpu_data.x86) { @@ -1566,8 +1576,8 @@ __init int intel_pmu_init(void) * Check whether the Architectural PerfMon supports * Branch Misses Retired hw_event or not. */ - cpuid(10, &eax.full, &ebx, &unused, &edx.full); - if (eax.split.mask_length <= ARCH_PERFMON_BRANCH_MISSES_RETIRED) + cpuid(10, &eax.full, &ebx.full, &unused, &edx.full); + if (eax.split.mask_length < ARCH_PERFMON_EVENTS_COUNT) return -ENODEV; version = eax.split.version_id; @@ -1643,7 +1653,7 @@ __init int intel_pmu_init(void) /* UOPS_EXECUTED.CORE_ACTIVE_CYCLES,c=1,i=1 */ intel_perfmon_event_map[PERF_COUNT_HW_STALLED_CYCLES_BACKEND] = 0x1803fb1; - if (ebx & 0x40) { + if (ebx.split.no_branch_misses_retired) { /* * Erratum AAJ80 detected, we work it around by using * the BR_MISP_EXEC.ANY event. This will over-count @@ -1651,6 +1661,7 @@ __init int intel_pmu_init(void) * architectural event which is often completely bogus: */ intel_perfmon_event_map[PERF_COUNT_HW_BRANCH_MISSES] = 0x7f89; + ebx.split.no_branch_misses_retired = 0; pr_cont("erratum AAJ80 worked around, "); } @@ -1729,5 +1740,13 @@ __init int intel_pmu_init(void) break; } } + x86_pmu.events_maskl = ebx.full; + x86_pmu.events_mask_len = eax.split.mask_length; + + /* disable event that reported as not presend by cpuid */ + for_each_set_bit(bit, x86_pmu.events_mask, + min(x86_pmu.events_mask_len, x86_pmu.max_events)) + intel_perfmon_event_map[intel_event_id_to_hw_id[bit]] = 0; + return 0; } -- 1.7.7.1