perf list: Fix checking for supported events on older kernels
authorVince Weaver <vincent.weaver@maine.edu>
Mon, 30 Dec 2013 20:39:45 +0000 (15:39 -0500)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Mon, 10 Feb 2014 14:34:31 +0000 (11:34 -0300)
"perf list" listing of hardware events doesn't work on older ARM devices.
The change enabling event detection:

 commit b41f1cec91c37eeea6fdb15effbfa24ea0a5536b
 Author: Namhyung Kim <namhyung.kim@lge.com>
 Date:   Tue Aug 27 11:41:53 2013 +0900

     perf list: Skip unsupported events

uses the following code in tools/perf/util/parse-events.c:

        struct perf_event_attr attr = {
                .type = type,
                .config = config,
                .disabled = 1,
                .exclude_kernel = 1,
        };

On ARM machines pre-dating the Cortex-A15 this doesn't work, as these
machines don't support .exclude_kernel.  So starting with 3.12 "perf
list" does not report any hardware events at all on older machines (seen
on Rasp-Pi, Pandaboard, Beagleboard, etc).

This version of the patch makes changes suggested by Namhyung Kim to
check for EACCESS and retry (instead of just dropping the
exclude_kernel) so we can properly handle machines where
/proc/sys/kernel/perf_event_paranoid is set to 2.

Reported-by: Chad Paradis <chad.paradis@umit.maine.edu>
Signed-off-by: Vince Weaver <vincent.weaver@maine.edu>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Chad Paradis <chad.paradis@umit.maine.edu>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/alpine.DEB.2.10.1312301536150.28814@vincent-weaver-1.um.maine.edu
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/util/parse-events.c

index d248fca6d7ed7302c77d5500a79c853d3073ea07..1e15df10a88c2c1ff7b62dbfbcc05f39ffbe4cd1 100644 (file)
@@ -1091,12 +1091,12 @@ int is_valid_tracepoint(const char *event_string)
 static bool is_event_supported(u8 type, unsigned config)
 {
        bool ret = true;
+       int open_return;
        struct perf_evsel *evsel;
        struct perf_event_attr attr = {
                .type = type,
                .config = config,
                .disabled = 1,
-               .exclude_kernel = 1,
        };
        struct {
                struct thread_map map;
@@ -1108,7 +1108,20 @@ static bool is_event_supported(u8 type, unsigned config)
 
        evsel = perf_evsel__new(&attr);
        if (evsel) {
-               ret = perf_evsel__open(evsel, NULL, &tmap.map) >= 0;
+               open_return = perf_evsel__open(evsel, NULL, &tmap.map);
+               ret = open_return >= 0;
+
+               if (open_return == -EACCES) {
+                       /*
+                        * This happens if the paranoid value
+                        * /proc/sys/kernel/perf_event_paranoid is set to 2
+                        * Re-run with exclude_kernel set; we don't do that
+                        * by default as some ARM machines do not support it.
+                        *
+                        */
+                       evsel->attr.exclude_kernel = 1;
+                       ret = perf_evsel__open(evsel, NULL, &tmap.map) >= 0;
+               }
                perf_evsel__delete(evsel);
        }