Re: [tip:perf/urgent] perf evlist: Open event on evsel cpus and threads

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

On 3/09/2015 6:27 p.m., Arnaldo Carvalho de Melo wrote:
> Em Thu, Sep 03, 2015 at 04:34:24PM +0300, Adrian Hunter escreveu:
>> On 01/09/15 11:31, tip-bot for Kan Liang wrote:
>>> Commit-ID:  d988d5ee647861706bc7a391ddbc29429b50f00e
>>> Gitweb:     http://git.kernel.org/tip/d988d5ee647861706bc7a391ddbc29429b50f00e
>>> Author:     Kan Liang <kan.liang@intel.com>
>>> AuthorDate: Fri, 21 Aug 2015 02:23:14 -0400
>>> Committer:  Arnaldo Carvalho de Melo <acme@redhat.com>
>>> CommitDate: Mon, 31 Aug 2015 17:28:01 -0300
>>>
>>> perf evlist: Open event on evsel cpus and threads
>>>
>>> An evsel may have different cpus and threads than the evlist it is in.
>>>
>>> Use it's own cpus and threads, when opening the evsel in 'perf record'.
>>>
>>> Signed-off-by: Kan Liang <kan.liang@intel.com>
>>> Cc: Jiri Olsa <jolsa@kernel.org>
>>> Link: http://lkml.kernel.org/r/1440138194-17001-1-git-send-email-kan.liang@intel.com
>>> Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
>>
>> Just noticed this breaks Intel PT.  Will have to investigate further.
>
> What kind of breakage?

Can't open the sched_switch event

>
> It all should be equivalent to before, its just that it uses
> evsel->{threads,cpus} while before it was using evlist->{threads,cpus},
> but that should point to the same thing if that
> perf_evlist__propagate_maps() method was called, so I assume this is
> some segfault?

I think maybe it doesn't consider evsels added later

>
> Something we could catch in a 'test' entry? Even if that required Intel
> PT hardware that would be something important to have, all this stuff is
> growing in complexity, we need those tests...

There is "Test tracking with sched_switch" but you need to expose it
to the same issue i.e.

diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index d51a520..6aeffdd 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -1433,7 +1433,7 @@ 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;
  	}



>
> - Arnaldo
>
>>> ---
>>>   tools/perf/builtin-record.c | 2 +-
>>>   tools/perf/util/evlist.c    | 4 ++++
>>>   2 files changed, 5 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
>>> index a660022..1d14f38 100644
>>> --- a/tools/perf/builtin-record.c
>>> +++ b/tools/perf/builtin-record.c
>>> @@ -279,7 +279,7 @@ static int record__open(struct record *rec)
>>>
>>>   	evlist__for_each(evlist, pos) {
>>>   try_again:
>>> -		if (perf_evsel__open(pos, evlist->cpus, evlist->threads) < 0) {
>>> +		if (perf_evsel__open(pos, pos->cpus, pos->threads) < 0) {
>>>   			if (perf_evsel__fallback(pos, errno, msg, sizeof(msg))) {
>>>   				if (verbose)
>>>   					ui__warning("%s\n", msg);
>>> diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
>>> index 8d00039..d51a520 100644
>>> --- a/tools/perf/util/evlist.c
>>> +++ b/tools/perf/util/evlist.c
>>> @@ -1181,6 +1181,10 @@ int perf_evlist__apply_filters(struct perf_evlist *evlist, struct perf_evsel **e
>>>   		if (evsel->filter == NULL)
>>>   			continue;
>>>
>>> +		/*
>>> +		 * filters only work for tracepoint event, which doesn't have cpu limit.
>>> +		 * So evlist and evsel should always be same.
>>> +		 */
>>>   		err = perf_evsel__apply_filter(evsel, ncpus, nthreads, evsel->filter);
>>>   		if (err) {
>>>   			*err_evsel = evsel;
>>>