Re: [PATCH 0/7] perf stat: Change event enable code

[Adrian Hunter <adrian.hunter@intel.com>]

On 09/12/15 15:44, Adrian Hunter wrote:
> On 08/12/15 15:53, Arnaldo Carvalho de Melo wrote:
>> Em Tue, Dec 08, 2015 at 09:29:51AM +0200, Adrian Hunter escreveu:
>>> On 07/12/15 23:09, Arnaldo Carvalho de Melo wrote:
>>>> Em Thu, Dec 03, 2015 at 10:06:39AM +0100, Jiri Olsa escreveu:
>>>>> while testing ftrace:function event I noticed we create
>>>>> stat counters as enabled (except for enable_on_exec couters).
>>>>>
>>>>> This way we count also filter setup and other config code
>>>>> which might be crucial for some events.
>>>>>
>>>>> Posponing the events enable once everything is ready.
>>>>>
>>>>> The last patch is RFC as I wasn't sure there's some hidden
>>>>> catch about perf_evlist__(enable|disable)_event functions
>>>>> I missed.. Adrian?
>>
>>>> They look the same, Adrian?
>>
>>>> Applied the first 6, will give some more time to Adrian to chime in.
>>  
>>> Looks like there might already be a problem using evsel->threads instead of
>>> evlist->threads with the logic relating to evsel->system_wide getting lost -
>>> but that happened already in "perf evlist: Factor
>>> perf_evlist__(enable|disable) functions".  Probably the threads should not
>>> be propagated in that case, but it needs more investigation.  I will try to
>>> look at it today.
>>
>> Thanks! Is that covered by any 'perf test' entry? Do you think having
>> some sort of Intel PT test to run on capable machines would be feasible?
> 
> There is "Test tracking with sched_switch".  There was an issue where 'perf
> record' was working differently to the tests.  I will try to find where the
> gaps are.  Seems I have run out of time again today though.

I was wrong about there being any problem using evsel->threads.  While the
patch "perf evlist: Factor perf_evlist__(enable|disable) function" changes
the number of threads (from perf_evlist__nr_threads() to thread_map__nr()),
the system_wide check is still done in perf_evsel__run_ioctl(), so
everything is fine.

WRT "[RFC 7/7] perf tools: Remove perf_evlist__(enable|disable)_event
functions" it might be worth putting the evsel->fd checks that
perf_evlist__[enable|disable]_event() have into perf_evsel__[enable|disable]().
But otherwise it looks fine.

The gap in testing that I was thinking of is below:

From: Adrian Hunter <adrian.hunter@intel.com>
Date: Fri, 11 Dec 2015 11:05:11 +0200
Subject: [PATCH] perf tools: Make perf_evlist__open() open evsels with their
 cpus and threads (like perf record does)

'perf record' uses perf_evsel__open() to open events and passes the evsel->cpus
and evsel->threads.  Many tests and some tools instead use perf_evlist__open()
which passes instead evlist->cpus and evlist->threads.

Make perf_evlist__open() follow the 'perf record' behaviour so that a consistent
approach is taken.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
---
 tools/perf/util/evlist.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index d1b6c206bb93..306dacb33d8e 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -1470,7 +1470,7 @@ int perf_evlist__open(struct perf_evlist *evlist)
 	perf_evlist__update_id_pos(evlist);
 
 	evlist__for_each(evlist, evsel) {
-		err = perf_evsel__open(evsel, evlist->cpus, evlist->threads);
+		err = perf_evsel__open(evsel, evsel->cpus, evsel->threads);
 		if (err < 0)
 			goto out_err;
 	}
-- 
1.9.1