Re: [BUG] perf: intel_pt won't display kernel function

[Song Liu <songliubraving@fb.com>]

> On Apr 3, 2019, at 11:50 AM, Arnaldo Carvalho de Melo <arnaldo.melo@gmail.com> wrote:
> 
> Em Wed, Apr 03, 2019 at 04:27:56PM +0000, Song Liu escreveu:
>> 
>> 
>>> On Apr 3, 2019, at 8:15 AM, Arnaldo Carvalho de Melo <arnaldo.melo@gmail.com> wrote:
>>> 
>>> Em Wed, Apr 03, 2019 at 11:53:53AM -0300, Arnaldo Carvalho de Melo escreveu:
>>>> Em Wed, Apr 03, 2019 at 04:37:38PM +0200, Jiri Olsa escreveu:
>>>>> hi,
>>>>> perf script --call-trace stop working for me recently,
>>>>> and displays only user space functions
>>>>> 
>>>>> I bisected that to:
>>>>> 7b612e291a5a perf tools: Synthesize PERF_RECORD_* for loaded BPF programs
>>>>> 
>>>>> data from following comands will display user space functions only:
>>>>> # perf-with-kcore record pt -e intel_pt// -- ls
>>>>> # perf-with-kcore script pt --call-trace
>>>>> 
>>>>> when I disable the bpf synthesizing (patch below), kernel functions are back
>>>>> 
>>>>> I guess the new events mess up with intel_pt decoder somehow
>>>> 
>>>> I.e. I'm adding the patch below to my perf/urgent branch.
>>> 
>>> Song, that is what I have, can I have your Acked-by and please consider
>>> taking a look at the bug Jiri reported,
>>> 
>>> Thanks,
>>> 
>>> - Arnaldo
>> 
>> Current logic with --no-bpf-event is to generate PERF_RECORD_KSYMBOL, but not
>> PERF_RECORD_BPF_EVENT:
> 
> I see... The opts is done later, after querying the kernel for existing
> BPF programs so that at least the KSYMBOL ones can be generated. So I'll
> keep it as is, no need for this patch.
> 
>> https://urldefense.proofpoint.com/v2/url?u=https-3A__elixir.bootlin.com_linux_v5.1-2Drc3_source_tools_perf_util_bpf-2Devent.c-23L254&d=DwIBAg&c=5VD0RTtNlTh3ycd41b3MUw&r=dR8692q0_uaizy0jkrBJQM5k2hfm4CiFxYT8KaysFrg&m=8ITYgIrqFBIy1uQFQPHrsX5kBX5RlmhwcknKYJa2eGs&s=MdB2zA9JstwC93qKEonsfTQeQYiw-yeOLbPxm3QMRo8&e=
>> 
>> I will look into the intel_pt problem. 
> 
> Ok.
> 
>> In the meanwhile, let's fix it for now. 
> 
> I'll wait a bit.

I guess something like this might be a better fix? (Sorry for missing 
commit message):

diff --git i/tools/perf/util/map.c w/tools/perf/util/map.c
index e32628cd20a7..741430a35dca 100644
--- i/tools/perf/util/map.c
+++ w/tools/perf/util/map.c
@@ -261,6 +261,12 @@ bool __map__is_extra_kernel_map(const struct map *map)
        return kmap && kmap->name[0];
 }

+bool __map__is_bpf_prog(const struct map *map)
+{
+       const char *name = map->dso->short_name;
+       return name && (strstr(name, "bpf_prog_") == name);
+}
+
 bool map__has_symbols(const struct map *map)
 {
        return dso__has_symbols(map->dso);
diff --git i/tools/perf/util/map.h w/tools/perf/util/map.h
index 0e20749f2c55..01079f1f4375 100644
--- i/tools/perf/util/map.h
+++ w/tools/perf/util/map.h
@@ -159,10 +159,12 @@ int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name,

 bool __map__is_kernel(const struct map *map);
 bool __map__is_extra_kernel_map(const struct map *map);
+bool __map__is_bpf_prog(const struct map *map);

 static inline bool __map__is_kmodule(const struct map *map)
 {
-       return !__map__is_kernel(map) && !__map__is_extra_kernel_map(map);
+       return !__map__is_kernel(map) && !__map__is_extra_kernel_map(map) &&
+               !__map__is_bpf_prog(map);
 }

 bool map__has_symbols(const struct map *map);

Thanks,
Song

> 
>> Acked-by: Song Liu <songliubraving@fb.com>
>> 
>> Thanks, 
>> Song
>> 
>> 
>>> commit 011318ccc2024ba03e96c32a06f74ca5d6ab5503
>>> Author: Arnaldo Carvalho de Melo <acme@redhat.com>
>>> Date:   Wed Apr 3 12:05:15 2019 -0300
>>> 
>>>   perf record: Do not synthesize BPF records when --no-bpf-event is used
>>> 
>>>   By default we synthesize and ask the kernel for BPF events, having a
>>>   --no-bpf-event option to disable that, which can be useful, for
>>>   instance, if there are still bugs in that code, which seems to be the
>>>   case as reported by Jiri Olsa in:
>>> 
>>>     "[BUG] perf: intel_pt won't display kernel function"
>>>     https://lore.kernel.org/lkml/20190403143738.GB32001@krava
>>> 
>>>   So add the check for record_opts.no_bpf_event when considering
>>>   synthesizing BPF events for pre-existing BPF programs in 'perf record'.
>>> 
>>>   The reported bug needs further analysis and is a separate matter.
>>> 
>>>   Cc: Adrian Hunter <adrian.hunter@intel.com>
>>>   Cc: Alexei Starovoitov <ast@kernel.org>
>>>   Cc: Andrii Nakryiko <andrii.nakryiko@gmail.com>
>>>   Cc: Daniel Borkmann <daniel@iogearbox.net>
>>>   Cc: Jiri Olsa <jolsa@kernel.org>
>>>   Cc: Martin KaFai Lau <kafai@fb.com>
>>>   Cc: Namhyung Kim <namhyung@kernel.org>
>>>   Cc: Peter Zijlstra <peterz@infradead.org>
>>>   Cc: Song Liu <songliubraving@fb.com>
>>>   Cc: Yonghong Song <yhs@fb.com>
>>>   Fixes: 7b612e291a5a ("perf tools: Synthesize PERF_RECORD_* for loaded BPF programs")
>>>   Link: https://lore.kernel.org/lkml/20190403145353.GE32553@kernel.org
>>>   Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
>>> 
>>> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
>>> index 4e2d953d4bc5..17d772f192ad 100644
>>> --- a/tools/perf/builtin-record.c
>>> +++ b/tools/perf/builtin-record.c
>>> @@ -1114,10 +1114,11 @@ static int record__synthesize(struct record *rec, bool tail)
>>> 		return err;
>>> 	}
>>> 
>>> -	err = perf_event__synthesize_bpf_events(session, process_synthesized_event,
>>> -						machine, opts);
>>> -	if (err < 0)
>>> -		pr_warning("Couldn't synthesize bpf events.\n");
>>> +	if (!opts->no_bpf_event) {
>>> +		err = perf_event__synthesize_bpf_events(session, process_synthesized_event, machine, opts);
>>> +		if (err < 0)
>>> +			pr_warning("Couldn't synthesize bpf events.\n");
>>> +	}
>>> 
>>> 	err = __machine__synthesize_threads(machine, tool, &opts->target, rec->evlist->threads,
>>> 					    process_synthesized_event, opts->sample_address,
> 
> -- 
> 
> - Arnaldo