perf/x86: Use INST_RETIRED.TOTAL_CYCLES_PS for cycles:pp for Skylake
authorAndi Kleen <ak@linux.intel.com>
Fri, 4 Dec 2015 11:50:32 +0000 (03:50 -0800)
committerIngo Molnar <mingo@kernel.org>
Wed, 6 Jan 2016 10:15:32 +0000 (11:15 +0100)
I added UOPS_RETIRED.ALL by mistake to the Skylake PEBS event list for
cycles:pp. But the event is not documented for Skylake, and has some
issues.

The recommended replacement for cycles:pp is to use
INST_RETIRED.ANY+pebs as a base, similar to what CPUs before Sandy
Bridge did. This new event is called INST_RETIRED.TOTAL_CYCLES_PS. The
event is not really new, but has been already used by perf before
Sandy Bridge for the original cycles:p

Note the SDM doesn't document that event either, but it's being
documented in the latest version of the event list on:

  https://download.01.org/perfmon/SKL

This patch does:

 - Remove UOPS_RETIRED.ALL from the Skylake PEBS event list

 - Add INST_RETIRED.ANY to the Skylake PEBS event list, and an table entry to
   allow cmask=16,inv=1 for cycles:pp

 - We don't need an extra entry for the base INST_RETIRED event,
   because it is already covered by the catch-all PEBS table entry.

 - Switch Skylake to use the Core2 PEBS alias (which is
   INST_RETIRED.TOTAL_CYCLES_PS)

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: hpa@zytor.com
Link: http://lkml.kernel.org/r/1448929689-13771-1-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
arch/x86/kernel/cpu/perf_event_intel.c
arch/x86/kernel/cpu/perf_event_intel_ds.c

index 33b4b677bf6eda02eb232a63c0817d7960adaf6b..5ed6e0d8ab95ac61c7fba5168e810ab10df30bc2 100644 (file)
@@ -3521,7 +3521,7 @@ __init int intel_pmu_init(void)
                x86_pmu.event_constraints = intel_skl_event_constraints;
                x86_pmu.pebs_constraints = intel_skl_pebs_event_constraints;
                x86_pmu.extra_regs = intel_skl_extra_regs;
-               x86_pmu.pebs_aliases = intel_pebs_aliases_snb;
+               x86_pmu.pebs_aliases = intel_pebs_aliases_core2;
                /* all extra regs are per-cpu when HT is on */
                x86_pmu.flags |= PMU_FL_HAS_RSP_1;
                x86_pmu.flags |= PMU_FL_NO_HT_SHARING;
index cd1993e46d190fb86e15e3e805b781ff19132d62..56b501532623f41f026a5a9c424c1caedf2a956a 100644 (file)
@@ -718,9 +718,8 @@ struct event_constraint intel_hsw_pebs_event_constraints[] = {
 
 struct event_constraint intel_skl_pebs_event_constraints[] = {
        INTEL_FLAGS_UEVENT_CONSTRAINT(0x1c0, 0x2),      /* INST_RETIRED.PREC_DIST */
-       INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_NA(0x01c2, 0xf), /* UOPS_RETIRED.ALL */
-       /* UOPS_RETIRED.ALL, inv=1, cmask=16 (cycles:p). */
-       INTEL_FLAGS_EVENT_CONSTRAINT(0x108001c2, 0xf),
+       /* INST_RETIRED.TOTAL_CYCLES_PS (inv=1, cmask=16) (cycles:p). */
+       INTEL_FLAGS_EVENT_CONSTRAINT(0x108000c0, 0x0f),
        INTEL_PLD_CONSTRAINT(0x1cd, 0xf),                     /* MEM_TRANS_RETIRED.* */
        INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_LD(0x11d0, 0xf), /* MEM_INST_RETIRED.STLB_MISS_LOADS */
        INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_ST(0x12d0, 0xf), /* MEM_INST_RETIRED.STLB_MISS_STORES */