Message-ID: <2aed81ff-2d28-3af5-0657-16ee69705c03@linux.intel.com>
Date:   Fri, 3 Mar 2023 11:40:08 -0500
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:102.0) Gecko/20100101
 Thunderbird/102.8.0
Subject: Re: [PATCH v2 03/10] perf record: Early auxtrace initialization
 before event parsing
Content-Language: en-US
To:     Ian Rogers <irogers@google.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>,
        Arnaldo Carvalho de Melo <acme@kernel.org>,
        Mark Rutland <mark.rutland@arm.com>,
        Alexander Shishkin <alexander.shishkin@linux.intel.com>,
        Jiri Olsa <jolsa@kernel.org>,
        Namhyung Kim <namhyung@kernel.org>,
        Zhengjun Xing <zhengjun.xing@linux.intel.com>,
        Ravi Bangoria <ravi.bangoria@amd.com>,
        Adrian Hunter <adrian.hunter@intel.com>,
        "Steinar H. Gunderson" <sesse@google.com>,
        Kim Phillips <kim.phillips@amd.com>,
        Florian Fischer <florian.fischer@muhq.space>,
        James Clark <james.clark@arm.com>,
        Suzuki Poulouse <suzuki.poulose@arm.com>,
        Sean Christopherson <seanjc@google.com>,
        Leo Yan <leo.yan@linaro.org>,
        John Garry <john.g.garry@oracle.com>,
        Kajol Jain <kjain@linux.ibm.com>,
        linux-perf-users@vger.kernel.org, linux-kernel@vger.kernel.org
Cc:     Stephane Eranian <eranian@google.com>
References: <20230302212531.1043318-1-irogers@google.com>
 <20230302212531.1043318-4-irogers@google.com>
From:   "Liang, Kan" <kan.liang@linux.intel.com>
In-Reply-To: <20230302212531.1043318-4-irogers@google.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk


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.
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);