On 10/12/2020 18:27, John Garry wrote:
>> Its unpublished, I'll send it to the tmp.perf/core branch now.
> 
> I use cross-compile for arm64 to build, and it's ok.
> 
> I notice that the failures are for architectures which don't have an 
> entry under pmu-events/arch, so maybe we're missing some 'weak' 
> definition of pmu_sys_event_tables.
> 
> I'll check now.
> 

Hi Arnaldo,

Can you try this on top:

----8<-----

 From 201aa2cb7bc8723765afd84a5d3972248af0f0a1 Mon Sep 17 00:00:00 2001
From: John Garry <john.garry@huawei.com>
Date: Thu, 10 Dec 2020 19:45:14 +0000
Subject: [PATCH] perf jevents: Add system events table for empty mapping

For architectures which have no PMU event lists - like arm32 - an empty
mapping table is printed. This is how the "pmu_events_map" symbol -
referenced in util/pmu.c::perf_pmu__find_map() - is created for those
architectures.

Since pmu-events.c now includes a new table - "pmu_sys_event_tables" -
which is also referenced from util/pmu.c, also add this for the empty
mappings.

Signed-off-by: John Garry <john.garry@huawei.com>

diff --git a/tools/perf/pmu-events/jevents.c 
b/tools/perf/pmu-events/jevents.c
index e930096ad713..28e20d9ec0f5 100644
--- a/tools/perf/pmu-events/jevents.c
+++ b/tools/perf/pmu-events/jevents.c
@@ -816,19 +816,30 @@ static void print_mapping_test_table(FILE *outfp)
  	fprintf(outfp, "},\n");
  }

+static void print_system_event_mapping_table_prefix(FILE *outfp)
+{
+	fprintf(outfp, "\nstruct pmu_sys_events pmu_sys_event_tables[] = {");
+}
+
+static void print_system_event_mapping_table_suffix(FILE *outfp)
+{
+	fprintf(outfp, "\n\t{\n\t\t.table = 0\n\t},");
+
+	fprintf(outfp, "\n};\n");
+}
+
  static int process_system_event_tables(FILE *outfp)
  {
  	struct sys_event_table *sys_event_table;

-	fprintf(outfp, "\nstruct pmu_sys_events pmu_sys_event_tables[] = {");
+	print_system_event_mapping_table_prefix(outfp);

  	list_for_each_entry(sys_event_table, &sys_event_tables, list) {
  		fprintf(outfp, "\n\t{\n\t\t.table = %s,\n\t},",
  			sys_event_table->soc_id);
  	}
-	fprintf(outfp, "\n\t{\n\t\t.table = 0\n\t},");

-	fprintf(outfp, "\n};\n");
+	print_system_event_mapping_table_suffix(outfp);

  	return 0;
  }
@@ -938,6 +949,9 @@ static void create_empty_mapping(const char 
*output_file)
  	fprintf(outfp, "#include \"pmu-events/pmu-events.h\"\n");
  	print_mapping_table_prefix(outfp);
  	print_mapping_table_suffix(outfp);
+	print_system_event_mapping_table_prefix(outfp);
+	print_system_event_mapping_table_suffix(outfp);
+
  	fclose(outfp);
  }

---->8----

Obviously I never tested building for one of test architectures which 
does not use PMU events - sorry!

I'll review this more tomorrow.

Thanks!


> 
>>
>> More results from testing:
>>
>>    59    13.57 ubuntu:16.04-x-arm            : FAIL 
>> arm-linux-gnueabihf-gcc (Ubuntu/Linaro 5.4.0-6ubuntu1~16.04.9) 5.4.0 
>> 20160609
>>
>> [perfbuilder@five ~]$ tail -20 dm.log/ubuntu\:16.04-x-arm
>>    CC       /tmp/build/perf/util/expr.o
>>    LD       /tmp/build/perf/util/intel-pt-decoder/perf-in.o
>>    LD       /tmp/build/perf/util/perf-in.o
>>    LD       /tmp/build/perf/perf-in.o
>>    LINK     /tmp/build/perf/perf
>> /tmp/build/perf/perf-in.o: In function `pmu_for_each_sys_event':
>> /git/linux/tools/perf/util/pmu.c:816: undefined reference to 
>> `pmu_sys_event_tables'
>> /git/linux/tools/perf/util/pmu.c:816: undefined reference to 
>> `pmu_sys_event_tables'
>> /tmp/build/perf/perf-in.o: In function `pmu_add_sys_aliases':
>> /git/linux/tools/perf/util/pmu.c:886: undefined reference to 
>> `pmu_sys_event_tables'
>> /git/linux/tools/perf/util/pmu.c:886: undefined reference to 
>> `pmu_sys_event_tables'
>> collect2: error: ld returned 1 exit status
>> Makefile.perf:659: recipe for target '/tmp/build/perf/perf' failed
>> make[2]: *** [/tmp/build/perf/perf] Error 1
>> Makefile.perf:232: recipe for target 'sub-make' failed
>> make[1]: *** [sub-make] Error 2
>> Makefile:69: recipe for target 'all' failed
>> make: *** [all] Error 2
>> make: Leaving directory '/git/linux/tools/perf'
>> + exit 1