perf/x86/intel: Fix PEBS data source interpretation on Nehalem/Westmere

Jiri reported some time ago that some entries in the PEBS data source table
in perf do not agree with the SDM. We investigated and the bits
changed for Sandy Bridge, but the SDM was not updated.

perf already implements the bits correctly for Sandy Bridge
and later. This patch patches it up for Nehalem and Westmere.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: <stable@vger.kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: jolsa@kernel.org
Link: http://lkml.kernel.org/r/1456871124-15985-1-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>

diff --git a/arch/x86/events/intel/core.c b/arch/x86/events/intel/core.c
index edac81c4295a6b5bc0d0e6d755848f0c3feffdef..68fa55b4d42e672ff466eecaed9782ff4e404795 100644 (file)
--- a/arch/x86/events/intel/core.c
+++ b/arch/x86/events/intel/core.c
@@ -3417,6 +3417,7 @@ __init int intel_pmu_init(void)
                intel_perfmon_event_map[PERF_COUNT_HW_STALLED_CYCLES_BACKEND] =
                        X86_CONFIG(.event=0xb1, .umask=0x3f, .inv=1, .cmask=1);
 
+               intel_pmu_pebs_data_source_nhm();
                x86_add_quirk(intel_nehalem_quirk);
 
                pr_cont("Nehalem events, ");
@@ -3480,6 +3481,7 @@ __init int intel_pmu_init(void)
                intel_perfmon_event_map[PERF_COUNT_HW_STALLED_CYCLES_BACKEND] =
                        X86_CONFIG(.event=0xb1, .umask=0x3f, .inv=1, .cmask=1);
 
+               intel_pmu_pebs_data_source_nhm();
                pr_cont("Westmere events, ");
                break;
 

diff --git a/arch/x86/events/intel/ds.c b/arch/x86/events/intel/ds.c
index a99a8cbac729d7558bb47d88ece7d1d69436cddb..ce7211a07c0b809767c88481d41127cd6f331b6d 100644 (file)
--- a/arch/x86/events/intel/ds.c
+++ b/arch/x86/events/intel/ds.c
@@ -51,7 +51,8 @@ union intel_x86_pebs_dse {
 #define OP_LH (P(OP, LOAD) | P(LVL, HIT))
 #define SNOOP_NONE_MISS (P(SNOOP, NONE) | P(SNOOP, MISS))
 
-static const u64 pebs_data_source[] = {
+/* Version for Sandy Bridge and later */
+static u64 pebs_data_source[] = {
        P(OP, LOAD) | P(LVL, MISS) | P(LVL, L3) | P(SNOOP, NA),/* 0x00:ukn L3 */
        OP_LH | P(LVL, L1)  | P(SNOOP, NONE),   /* 0x01: L1 local */
        OP_LH | P(LVL, LFB) | P(SNOOP, NONE),   /* 0x02: LFB hit */
@@ -70,6 +71,14 @@ static const u64 pebs_data_source[] = {
        OP_LH | P(LVL, UNC) | P(SNOOP, NONE), /* 0x0f: uncached */
 };
 
+/* Patch up minor differences in the bits */
+void __init intel_pmu_pebs_data_source_nhm(void)
+{
+       pebs_data_source[0x05] = OP_LH | P(LVL, L3)  | P(SNOOP, HIT);
+       pebs_data_source[0x06] = OP_LH | P(LVL, L3)  | P(SNOOP, HITM);
+       pebs_data_source[0x07] = OP_LH | P(LVL, L3)  | P(SNOOP, HITM);
+}
+
 static u64 precise_store_data(u64 status)
 {
        union intel_x86_pebs_dse dse;

diff --git a/arch/x86/events/perf_event.h b/arch/x86/events/perf_event.h
index 24e259eb0b3ada4103db142c1d2a969defafa0fe..68155cafa8a13d88907493a4bf32f0e083189b73 100644 (file)
--- a/arch/x86/events/perf_event.h
+++ b/arch/x86/events/perf_event.h
@@ -907,6 +907,8 @@ void intel_pmu_lbr_init_skl(void);
 
 void intel_pmu_lbr_init_knl(void);
 
+void intel_pmu_pebs_data_source_nhm(void);
+
 int intel_pmu_setup_lbr_filter(struct perf_event *event);
 
 void intel_pt_interrupt(void);