perf events parser: Add missing Intel CPU events to parser
authorAdrian Hunter <adrian.hunter@intel.com>
Thu, 26 Mar 2020 08:01:47 +0000 (10:01 +0200)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Fri, 3 Apr 2020 12:37:56 +0000 (09:37 -0300)
perf list expects CPU events to be parseable by name, e.g.

    # perf list | grep el-capacity-read
      el-capacity-read OR cpu/el-capacity-read/          [Kernel PMU event]

But the event parser does not recognize them that way, e.g.

    # perf test -v "Parse event"
    <SNIP>
    running test 54 'cycles//u'
    running test 55 'cycles:k'
    running test 0 'cpu/config=10,config1,config2=3,period=1000/u'
    running test 1 'cpu/config=1,name=krava/u,cpu/config=2/u'
    running test 2 'cpu/config=1,call-graph=fp,time,period=100000/,cpu/config=2,call-graph=no,time=0,period=2000/'
    running test 3 'cpu/name='COMPLEX_CYCLES_NAME:orig=cycles,desc=chip-clock-ticks',period=0x1,event=0x2/ukp'
    -> cpu/event=0,umask=0x11/
    -> cpu/event=0,umask=0x13/
    -> cpu/event=0x54,umask=0x1/
    failed to parse event 'el-capacity-read:u,cpu/event=el-capacity-read/u', err 1, str 'parser error'
    event syntax error: 'el-capacity-read:u,cpu/event=el-capacity-read/u'
                           \___ parser error test child finished with 1
    ---- end ----
    Parse event definition strings: FAILED!

This happens because the parser splits names by '-' in order to deal
with cache events. For example 'L1-dcache' is a token in
parse-events.l which is matched to 'L1-dcache-load-miss' by the
following rule:

    PE_NAME_CACHE_TYPE '-' PE_NAME_CACHE_OP_RESULT '-' PE_NAME_CACHE_OP_RESULT opt_event_config

And so there is special handling for 2-part PMU names i.e.

    PE_PMU_EVENT_PRE '-' PE_PMU_EVENT_SUF sep_dc

but no handling for 3-part names, which are instead added as tokens e.g.

    topdown-[a-z-]+

While it would be possible to add a rule for 3-part names, that would
not work if the first parts were also a valid PMU name e.g.
'el-capacity-read' would be matched to 'el-capacity' before the parser
reached the 3rd part.

The parser would need significant change to rationalize all this, so
instead fix for now by adding missing Intel CPU events with 3-part names
to the event parser as tokens.

Missing events were found by using:

    grep -r EVENT_ATTR_STR arch/x86/events/intel/core.c

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Link: http://lore.kernel.org/lkml/90c7ae07-c568-b6d3-f9c4-d0c1528a0610@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/util/parse-events.l

index 7b1c8ee..baa48f2 100644 (file)
@@ -342,11 +342,13 @@ bpf-output                                        { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_BPF_OUT
         * Because the prefix cycles is mixed up with cpu-cycles.
         * loads and stores are mixed up with cache event
         */
-cycles-ct                                      { return str(yyscanner, PE_KERNEL_PMU_EVENT); }
-cycles-t                                       { return str(yyscanner, PE_KERNEL_PMU_EVENT); }
-mem-loads                                      { return str(yyscanner, PE_KERNEL_PMU_EVENT); }
-mem-stores                                     { return str(yyscanner, PE_KERNEL_PMU_EVENT); }
-topdown-[a-z-]+                                        { return str(yyscanner, PE_KERNEL_PMU_EVENT); }
+cycles-ct                              |
+cycles-t                               |
+mem-loads                              |
+mem-stores                             |
+topdown-[a-z-]+                                |
+tx-capacity-[a-z-]+                    |
+el-capacity-[a-z-]+                    { return str(yyscanner, PE_KERNEL_PMU_EVENT); }
 
 L1-dcache|l1-d|l1d|L1-data             |
 L1-icache|l1-i|l1i|L1-instruction      |