perf vendor events intel: Add NO_NMI_WATCHDOG metric constraint
authorKan Liang <kan.liang@linux.intel.com>
Mon, 24 Feb 2020 21:59:24 +0000 (13:59 -0800)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Tue, 10 Mar 2020 17:56:46 +0000 (14:56 -0300)
commitb95fcd2c1c25bd14f55d5d6ab268b3ab00b8a774
tree7fc0f5e60193d252b91c5757d66c3ff7b00262ae
parentab483d8bc8acb83f0103bc38ef8f2c27d98ffd1b
perf vendor events intel: Add NO_NMI_WATCHDOG metric constraint

Add NO_NMI_WATCHDOG metric constraint to Page_Walks_Utilization for Sky Lake
and Cascade Lake.

Committer testing:

On a Lenovo T480S, Intel(R) Core(TM) i7-8650U Kaby Lake, that looking at x86's
mapfile.csv file is a:

  $ grep -w skylake tools/perf/pmu-events/arch/x86/mapfile.csv
  GenuineIntel-6-[4589]E,v24,skylake,core
  $

So uses the constraint added in this patch in this file:

  tools/perf/pmu-events/arch/x86/skylake/skl-metrics.json

Before:

  # perf stat -a -M Page_Walks_Utilization sleep 2

   Performance counter stats for 'system wide':

       <not counted>      itlb_misses.walk_pending                                      (0.00%)
       <not counted>      dtlb_load_misses.walk_pending                                     (0.00%)
       <not counted>      dtlb_store_misses.walk_pending                                     (0.00%)
       <not counted>      ept.walk_pending                                              (0.00%)
       <not counted>      cycles                                                        (0.00%)

         2.001750514 seconds time elapsed

  Some events weren't counted. Try disabling the NMI watchdog:
   echo 0 > /proc/sys/kernel/nmi_watchdog
   perf stat ...
   echo 1 > /proc/sys/kernel/nmi_watchdog
  The events in group usually have to be from the same PMU. Try reorganizing the group.
  #

After:

  # perf stat -a -M Page_Walks_Utilization sleep 2
  Splitting metric group Page_Walks_Utilization into standalone metrics.
  Try disabling the NMI watchdog to comply NO_NMI_WATCHDOG metric constraint:
      echo 0 > /proc/sys/kernel/nmi_watchdog
      perf stat ...
      echo 1 > /proc/sys/kernel/nmi_watchdog
  ,
   Performance counter stats for 'system wide':

          36,883,102      itlb_misses.walk_pending  #      0.1 Page_Walks_Utilization   (79.99%)
         123,104,146      dtlb_load_misses.walk_pending                                     (80.02%)
          13,720,795      dtlb_store_misses.walk_pending                                     (79.99%)
                   0      ept.walk_pending                                              (79.99%)
       1,519,948,400      cycles                                                        (80.01%)

         2.002170780 seconds time elapsed

  #

Before and after, if we disable the nmi_watchdog we get:

  # echo 0 > /proc/sys/kernel/nmi_watchdog
  # perf stat -a -M Page_Walks_Utilization sleep 2

   Performance counter stats for 'system wide':

          33,721,658      itlb_misses.walk_pending  #      0.1 Page_Walks_Utilization
          84,070,996      dtlb_load_misses.walk_pending
           9,816,071      dtlb_store_misses.walk_pending
                   0      ept.walk_pending
         704,920,899      cycles

         2.002331670 seconds time elapsed

  #

  More information about the metric expressions:

  # perf stat -v -a -M Page_Walks_Utilization sleep 2
  Using CPUID GenuineIntel-6-8E-A
  metric expr ( itlb_misses.walk_pending + dtlb_load_misses.walk_pending + dtlb_store_misses.walk_pending + ept.walk_pending ) / ( 2 * cycles ) for Page_Walks_Utilization
  found event itlb_misses.walk_pending
  found event dtlb_load_misses.walk_pending
  found event dtlb_store_misses.walk_pending
  found event ept.walk_pending
  found event cycles
  adding {itlb_misses.walk_pending,dtlb_load_misses.walk_pending,dtlb_store_misses.walk_pending,ept.walk_pending,cycles}:W
   -> cpu/umask=0x10,(null)=0x186a3,event=0x85/
   -> cpu/umask=0x10,(null)=0x1e8483,event=0x8/
   -> cpu/umask=0x10,(null)=0x1e8483,event=0x49/
   -> cpu/umask=0x10,(null)=0x1e8483,event=0x4f/
  itlb_misses.walk_pending: 8085772 16010162799 16010162799
  dtlb_load_misses.walk_pending: 28134579 16010162799 16010162799
  dtlb_store_misses.walk_pending: 7276535 16010162799 16010162799
  ept.walk_pending: 2 16010162799 16010162799
  cycles: 315140605 16010162799 16010162799

   Performance counter stats for 'system wide':

           8,085,772      itlb_misses.walk_pending  #      0.1 Page_Walks_Utilization
          28,134,579      dtlb_load_misses.walk_pending
           7,276,535      dtlb_store_misses.walk_pending
                   2      ept.walk_pending
         315,140,605      cycles

         2.002333181 seconds time elapsed

  #

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Link: http://lore.kernel.org/lkml/1582581564-184429-6-git-send-email-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/pmu-events/arch/x86/cascadelakex/clx-metrics.json
tools/perf/pmu-events/arch/x86/skylake/skl-metrics.json
tools/perf/pmu-events/arch/x86/skylakex/skx-metrics.json