Re: [PATCH v2 03/10] perf record: Early auxtrace initialization before event parsing

["Liang, Kan" <kan.liang@linux.intel.com>]

On 2023-03-06 4:31 a.m., Adrian Hunter wrote:
> On 5/03/23 10:32, Adrian Hunter wrote:
>> On 3/03/23 18:40, Liang, Kan wrote:
>>>
>>>
>>> On 2023-03-02 4:25 p.m., Ian Rogers wrote:
>>>> This allows event parsing to use the evsel__is_aux_event function,
>>>> which is important when determining event grouping.
>>>>
>>>> Signed-off-by: Ian Rogers <irogers@google.com>
>>>> ---
>>>>  tools/perf/arch/x86/util/auxtrace.c | 17 +++++++++++++----
>>>>  tools/perf/builtin-record.c         |  6 ++++++
>>>>  tools/perf/util/auxtrace.h          |  2 ++
>>>>  3 files changed, 21 insertions(+), 4 deletions(-)
>>>>
>>>> diff --git a/tools/perf/arch/x86/util/auxtrace.c b/tools/perf/arch/x86/util/auxtrace.c
>>>> index 3da506e13f49..de1e4842ea2e 100644
>>>> --- a/tools/perf/arch/x86/util/auxtrace.c
>>>> +++ b/tools/perf/arch/x86/util/auxtrace.c
>>>> @@ -15,6 +15,19 @@
>>>>  #include "../../../util/intel-bts.h"
>>>>  #include "../../../util/evlist.h"
>>>>  
>>>> +void auxtrace__early_init(void)
>>>> +{
>>>> +	struct perf_pmu *intel_pt_pmu;
>>>> +	struct perf_pmu *intel_bts_pmu;
>>>> +
>>>> +	intel_pt_pmu = perf_pmu__find(INTEL_PT_PMU_NAME);
>>>> +	if (intel_pt_pmu)
>>>> +		intel_pt_pmu->auxtrace = true;
>>>> +	intel_bts_pmu = perf_pmu__find(INTEL_BTS_PMU_NAME);
>>>> +	if (intel_bts_pmu)
>>>> +		intel_bts_pmu->auxtrace = true;
>>>> +}
>>>> +
>>>>  static
>>>>  struct auxtrace_record *auxtrace_record__init_intel(struct evlist *evlist,
>>>>  						    int *err)
>>>> @@ -26,11 +39,7 @@ struct auxtrace_record *auxtrace_record__init_intel(struct evlist *evlist,
>>>>  	bool found_bts = false;
>>>>  
>>>>  	intel_pt_pmu = perf_pmu__find(INTEL_PT_PMU_NAME);
>>>> -	if (intel_pt_pmu)
>>>> -		intel_pt_pmu->auxtrace = true;
>>>>  	intel_bts_pmu = perf_pmu__find(INTEL_BTS_PMU_NAME);
>>>> -	if (intel_bts_pmu)
>>>> -		intel_bts_pmu->auxtrace = true;
>>>>  
>>>>  	evlist__for_each_entry(evlist, evsel) {
>>>>  		if (intel_pt_pmu && evsel->core.attr.type == intel_pt_pmu->type)
>>>> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
>>>> index 8374117e66f6..a0870c076dc0 100644
>>>> --- a/tools/perf/builtin-record.c
>>>> +++ b/tools/perf/builtin-record.c
>>>> @@ -3940,6 +3940,10 @@ static int record__init_thread_masks(struct record *rec)
>>>>  	return ret;
>>>>  }
>>>>  
>>>> +__weak void auxtrace__early_init(void)
>>>> +{
>>>> +}
>>>> +
>>>>  int cmd_record(int argc, const char **argv)
>>>>  {
>>>>  	int err;
>>>> @@ -3985,6 +3989,8 @@ int cmd_record(int argc, const char **argv)
>>>>  	if (err)
>>>>  		return err;
>>>>  
>>>> +	auxtrace__early_init();
>>>
>>> So the auxtrace__early_init() will be unconditionally invoked even there
>>> is no PT or BTS events, right?
>>>
>>> Maybe we should move the auxtrace__early_init() to evsel__is_aux_event()
>>> and cache the value. The initialization will only be invoked when it's
>>> required.
>>
>> Although perf_pmu__find() will be called unconditionally via
>> record__auxtrace_init() anyway.
> 
> However auxtrace__early_init() is before parsing 'verbose' so
> debug prints don't work anymore.
> 
> How about this instead:

Yes, I think it should be a better place to initialize them.

Thanks,
Kan
> 
> diff --git a/tools/perf/arch/x86/util/auxtrace.c
> b/tools/perf/arch/x86/util/auxtrace.c
> index 3da506e13f49d..330d03216b0e6 100644
> --- a/tools/perf/arch/x86/util/auxtrace.c
> +++ b/tools/perf/arch/x86/util/auxtrace.c
> @@ -26,11 +26,7 @@ struct auxtrace_record
> *auxtrace_record__init_intel(struct evlist *evlist,
>  	bool found_bts = false;
> 
>  	intel_pt_pmu = perf_pmu__find(INTEL_PT_PMU_NAME);
> -	if (intel_pt_pmu)
> -		intel_pt_pmu->auxtrace = true;
>  	intel_bts_pmu = perf_pmu__find(INTEL_BTS_PMU_NAME);
> -	if (intel_bts_pmu)
> -		intel_bts_pmu->auxtrace = true;
> 
>  	evlist__for_each_entry(evlist, evsel) {
>  		if (intel_pt_pmu && evsel->core.attr.type == intel_pt_pmu->type)
> diff --git a/tools/perf/arch/x86/util/pmu.c b/tools/perf/arch/x86/util/pmu.c
> index 358340b342431..f73b80dcd8bdc 100644
> --- a/tools/perf/arch/x86/util/pmu.c
> +++ b/tools/perf/arch/x86/util/pmu.c
> @@ -27,10 +27,14 @@ static bool cached_list;
>  struct perf_event_attr *perf_pmu__get_default_config(struct perf_pmu
> *pmu __maybe_unused)
>  {
>  #ifdef HAVE_AUXTRACE_SUPPORT
> -	if (!strcmp(pmu->name, INTEL_PT_PMU_NAME))
> +	if (!strcmp(pmu->name, INTEL_PT_PMU_NAME)) {
> +		pmu->auxtrace = true;
>  		return intel_pt_pmu_default_config(pmu);
> -	if (!strcmp(pmu->name, INTEL_BTS_PMU_NAME))
> +	}
> +	if (!strcmp(pmu->name, INTEL_BTS_PMU_NAME)) {
> +		pmu->auxtrace = true;
>  		pmu->selectable = true;
> +	}
>  #endif
>  	return NULL;
>  }
> 
> 
> 
>>
>>> Something as below (not tested.)
>>>
>>> +void auxtrace__init(void)
>>> +{
>>> +	struct perf_pmu *intel_pt_pmu;
>>> +	struct perf_pmu *intel_bts_pmu;
>>> +	static bool cached;
>>> +
>>> +	if (cached)
>>> +		return;
>>> +	intel_pt_pmu = perf_pmu__find(INTEL_PT_PMU_NAME);
>>> +	if (intel_pt_pmu)
>>> +		intel_pt_pmu->auxtrace = true;
>>> +	intel_bts_pmu = perf_pmu__find(INTEL_BTS_PMU_NAME);
>>> +	if (intel_bts_pmu)
>>> +		intel_bts_pmu->auxtrace = true;
>>> +}
>>>
>>> bool evsel__is_aux_event(struct evsel *evsel)
>>> {
>>> 	struct perf_pmu *pmu = evsel__find_pmu(evsel);
>>> +	auxtrace__init();
>>> 	return pmu && pmu->auxtrace;
>>> }
>>>
>>>
>>>
>>> Thanks,
>>> Kan
>>>
>>>> +
>>>>  	argc = parse_options(argc, argv, record_options, record_usage,
>>>>  			    PARSE_OPT_STOP_AT_NON_OPTION);
>>>>  	if (quiet)
>>>> diff --git a/tools/perf/util/auxtrace.h b/tools/perf/util/auxtrace.h
>>>> index 29eb82dff574..49a86aa6ac94 100644
>>>> --- a/tools/perf/util/auxtrace.h
>>>> +++ b/tools/perf/util/auxtrace.h
>>>> @@ -457,6 +457,8 @@ struct addr_filters {
>>>>  
>>>>  struct auxtrace_cache;
>>>>  
>>>> +void auxtrace__early_init(void);
>>>> +
>>>>  #ifdef HAVE_AUXTRACE_SUPPORT
>>>>  
>>>>  u64 compat_auxtrace_mmap__read_head(struct auxtrace_mmap *mm);
>>
>