Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1587496 > unrolled thread

[PATCH] perf tools: Force uncore events to system wide monitoring

Started byJiri Olsa <jolsa@kernel.org>
First post2017-02-24 11:20 +0100
Last post2017-02-27 12:50 +0100
Articles 5 — 4 participants

Back to article view | Back to linux.kernel

Contents

  [PATCH] perf tools: Force uncore events to system wide monitoring Jiri Olsa <jolsa@kernel.org> - 2017-02-24 11:20 +0100
    Re: [PATCH] perf tools: Force uncore events to system wide monitoring Arnaldo Carvalho de Melo <acme@kernel.org> - 2017-02-24 14:40 +0100
      Re: [PATCH] perf tools: Force uncore events to system wide monitoring Jiri Olsa <jolsa@redhat.com> - 2017-02-24 15:10 +0100
        Re: [PATCH] perf tools: Force uncore events to system wide monitoring Jiri Olsa <jolsa@redhat.com> - 2017-02-27 10:40 +0100
          Re: [PATCH] perf tools: Force uncore events to system wide monitoring Borislav Petkov <bp@alien8.de> - 2017-02-27 12:50 +0100

#1587496 — [PATCH] perf tools: Force uncore events to system wide monitoring

FromJiri Olsa <jolsa@kernel.org>
Date2017-02-24 11:20 +0100
Subject[PATCH] perf tools: Force uncore events to system wide monitoring
Message-ID<tel8J-79A-11@gated-at.bofh.it>
Adding system_wide flag to uncore pmu objects and passing
this flag along to the their events.

Making system wide (-a) the default option if no target
was specified and one of following conditions is met:

  - there's no workload specified (current behaviour)
  - there is workload specified but all requested
    events are system wide events

Mixed events core/uncore with workload:
  $ perf stat -e 'uncore_cbox_0/clockticks/,cycles' sleep 1

   Performance counter stats for 'sleep 1':

     <not supported>      uncore_cbox_0/clockticks/
             980,489      cycles

         1.000897406 seconds time elapsed

Uncore event with workload:
  $ perf stat -e 'uncore_cbox_0/clockticks/' sleep 1

   Performance counter stats for 'system wide':

  281,473,897,192,670      uncore_cbox_0/clockticks/

         1.000833784 seconds time elapsed

Suggested-by: Borislav Petkov <bp@alien8.de>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@alien8.de>
Link: http://lkml.kernel.org/n/tip-rh8kybg6xdadfrw6x6uj37i2@git.kernel.org
---
 tools/perf/arch/x86/util/pmu.c |  6 +++++-
 tools/perf/builtin-stat.c      | 32 +++++++++++++++++++++++++++++---
 tools/perf/util/parse-events.c |  1 +
 tools/perf/util/pmu.h          |  1 +
 4 files changed, 36 insertions(+), 4 deletions(-)

diff --git a/tools/perf/arch/x86/util/pmu.c b/tools/perf/arch/x86/util/pmu.c
index 79fe07158d00..654290f87a19 100644
--- a/tools/perf/arch/x86/util/pmu.c
+++ b/tools/perf/arch/x86/util/pmu.c
@@ -11,8 +11,12 @@ struct perf_event_attr *perf_pmu__get_default_config(struct perf_pmu *pmu __mayb
 #ifdef HAVE_AUXTRACE_SUPPORT
 	if (!strcmp(pmu->name, INTEL_PT_PMU_NAME))
 		return intel_pt_pmu_default_config(pmu);
-	if (!strcmp(pmu->name, INTEL_BTS_PMU_NAME))
+	if (!strcmp(pmu->name, INTEL_BTS_PMU_NAME)) {
 		pmu->selectable = true;
+		return NULL;
+	}
 #endif
+	/* All uncore PMUs are monitored system wide. */
+	pmu->system_wide = !strncmp(pmu->name, "uncore", 6);
 	return NULL;
 }
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 13b54999ad79..c32c7fa6f092 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -2339,6 +2339,34 @@ static int __cmd_report(int argc, const char **argv)
 	return 0;
 }
 
+static void setup_system_wide(int forks)
+{
+	/*
+	 * Make system wide (-a) the default target if
+	 * no target was specified and one of following
+	 * conditions is met:
+	 *
+	 *   - there's no workload specified
+	 *   - there is workload specified but all requested
+	 *     events are system wide events
+	 */
+	if (!target__none(&target))
+		return;
+
+	if (!forks)
+		target.system_wide = true;
+	else {
+		struct perf_evsel *counter;
+
+		evlist__for_each_entry(evsel_list, counter) {
+			if (!counter->system_wide)
+				return;
+		}
+
+		target.system_wide = true;
+	}
+}
+
 int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
 {
 	const char * const stat_usage[] = {
@@ -2445,9 +2473,7 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
 	} else if (big_num_opt == 0) /* User passed --no-big-num */
 		big_num = false;
 
-	/* Make system wide (-a) the default target. */
-	if (!argc && target__none(&target))
-		target.system_wide = true;
+	setup_system_wide(argc);
 
 	if (run_count < 0) {
 		pr_err("Run count must be a positive number\n");
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index 67a8aebc67ab..8b06b21f1bbc 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -1254,6 +1254,7 @@ int parse_events_add_pmu(struct parse_events_evlist *data,
 		evsel->scale = info.scale;
 		evsel->per_pkg = info.per_pkg;
 		evsel->snapshot = info.snapshot;
+		evsel->system_wide = pmu->system_wide;
 	}
 
 	return evsel ? 0 : -ENOMEM;
diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h
index 00852ddc7741..0ce3ffacbf92 100644
--- a/tools/perf/util/pmu.h
+++ b/tools/perf/util/pmu.h
@@ -21,6 +21,7 @@ struct perf_pmu {
 	char *name;
 	__u32 type;
 	bool selectable;
+	bool system_wide;
 	struct perf_event_attr *default_config;
 	struct cpu_map *cpus;
 	struct list_head format;  /* HEAD struct perf_pmu_format -> list */
-- 
2.7.4

[toc] | [next] | [standalone]

#1587670

FromArnaldo Carvalho de Melo <acme@kernel.org>
Date2017-02-24 14:40 +0100
Message-ID<teogi-MF-15@gated-at.bofh.it>
In reply to#1587496
Em Fri, Feb 24, 2017 at 11:17:14AM +0100, Jiri Olsa escreveu:
> Adding system_wide flag to uncore pmu objects and passing
> this flag along to the their events.

Boris, ack? Tested-by?

- Arnaldo
 
> Making system wide (-a) the default option if no target
> was specified and one of following conditions is met:
> 
>   - there's no workload specified (current behaviour)
>   - there is workload specified but all requested
>     events are system wide events
> 
> Mixed events core/uncore with workload:
>   $ perf stat -e 'uncore_cbox_0/clockticks/,cycles' sleep 1
> 
>    Performance counter stats for 'sleep 1':
> 
>      <not supported>      uncore_cbox_0/clockticks/
>              980,489      cycles
> 
>          1.000897406 seconds time elapsed
> 
> Uncore event with workload:
>   $ perf stat -e 'uncore_cbox_0/clockticks/' sleep 1
> 
>    Performance counter stats for 'system wide':
> 
>   281,473,897,192,670      uncore_cbox_0/clockticks/
> 
>          1.000833784 seconds time elapsed
> 
> Suggested-by: Borislav Petkov <bp@alien8.de>
> Signed-off-by: Jiri Olsa <jolsa@kernel.org>
> Cc: Namhyung Kim <namhyung@kernel.org>
> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
> Cc: Adrian Hunter <adrian.hunter@intel.com>
> Cc: Borislav Petkov <bp@alien8.de>
> Link: http://lkml.kernel.org/n/tip-rh8kybg6xdadfrw6x6uj37i2@git.kernel.org
> ---
>  tools/perf/arch/x86/util/pmu.c |  6 +++++-
>  tools/perf/builtin-stat.c      | 32 +++++++++++++++++++++++++++++---
>  tools/perf/util/parse-events.c |  1 +
>  tools/perf/util/pmu.h          |  1 +
>  4 files changed, 36 insertions(+), 4 deletions(-)
> 
> diff --git a/tools/perf/arch/x86/util/pmu.c b/tools/perf/arch/x86/util/pmu.c
> index 79fe07158d00..654290f87a19 100644
> --- a/tools/perf/arch/x86/util/pmu.c
> +++ b/tools/perf/arch/x86/util/pmu.c
> @@ -11,8 +11,12 @@ struct perf_event_attr *perf_pmu__get_default_config(struct perf_pmu *pmu __mayb
>  #ifdef HAVE_AUXTRACE_SUPPORT
>  	if (!strcmp(pmu->name, INTEL_PT_PMU_NAME))
>  		return intel_pt_pmu_default_config(pmu);
> -	if (!strcmp(pmu->name, INTEL_BTS_PMU_NAME))
> +	if (!strcmp(pmu->name, INTEL_BTS_PMU_NAME)) {
>  		pmu->selectable = true;
> +		return NULL;
> +	}
>  #endif
> +	/* All uncore PMUs are monitored system wide. */
> +	pmu->system_wide = !strncmp(pmu->name, "uncore", 6);
>  	return NULL;
>  }
> diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
> index 13b54999ad79..c32c7fa6f092 100644
> --- a/tools/perf/builtin-stat.c
> +++ b/tools/perf/builtin-stat.c
> @@ -2339,6 +2339,34 @@ static int __cmd_report(int argc, const char **argv)
>  	return 0;
>  }
>  
> +static void setup_system_wide(int forks)
> +{
> +	/*
> +	 * Make system wide (-a) the default target if
> +	 * no target was specified and one of following
> +	 * conditions is met:
> +	 *
> +	 *   - there's no workload specified
> +	 *   - there is workload specified but all requested
> +	 *     events are system wide events
> +	 */
> +	if (!target__none(&target))
> +		return;
> +
> +	if (!forks)
> +		target.system_wide = true;
> +	else {
> +		struct perf_evsel *counter;
> +
> +		evlist__for_each_entry(evsel_list, counter) {
> +			if (!counter->system_wide)
> +				return;
> +		}
> +
> +		target.system_wide = true;
> +	}
> +}
> +
>  int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
>  {
>  	const char * const stat_usage[] = {
> @@ -2445,9 +2473,7 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
>  	} else if (big_num_opt == 0) /* User passed --no-big-num */
>  		big_num = false;
>  
> -	/* Make system wide (-a) the default target. */
> -	if (!argc && target__none(&target))
> -		target.system_wide = true;
> +	setup_system_wide(argc);
>  
>  	if (run_count < 0) {
>  		pr_err("Run count must be a positive number\n");
> diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
> index 67a8aebc67ab..8b06b21f1bbc 100644
> --- a/tools/perf/util/parse-events.c
> +++ b/tools/perf/util/parse-events.c
> @@ -1254,6 +1254,7 @@ int parse_events_add_pmu(struct parse_events_evlist *data,
>  		evsel->scale = info.scale;
>  		evsel->per_pkg = info.per_pkg;
>  		evsel->snapshot = info.snapshot;
> +		evsel->system_wide = pmu->system_wide;
>  	}
>  
>  	return evsel ? 0 : -ENOMEM;
> diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h
> index 00852ddc7741..0ce3ffacbf92 100644
> --- a/tools/perf/util/pmu.h
> +++ b/tools/perf/util/pmu.h
> @@ -21,6 +21,7 @@ struct perf_pmu {
>  	char *name;
>  	__u32 type;
>  	bool selectable;
> +	bool system_wide;
>  	struct perf_event_attr *default_config;
>  	struct cpu_map *cpus;
>  	struct list_head format;  /* HEAD struct perf_pmu_format -> list */
> -- 
> 2.7.4

[toc] | [prev] | [next] | [standalone]

#1587683

FromJiri Olsa <jolsa@redhat.com>
Date2017-02-24 15:10 +0100
Message-ID<teoJk-1cU-9@gated-at.bofh.it>
In reply to#1587670
On Fri, Feb 24, 2017 at 10:32:22AM -0300, Arnaldo Carvalho de Melo wrote:
> Em Fri, Feb 24, 2017 at 11:17:14AM +0100, Jiri Olsa escreveu:
> > Adding system_wide flag to uncore pmu objects and passing
> > this flag along to the their events.
> 
> Boris, ack? Tested-by?

actualy we discussed that on irc and I need to send new version ;-)

jirka

[toc] | [prev] | [next] | [standalone]

#1588563

FromJiri Olsa <jolsa@redhat.com>
Date2017-02-27 10:40 +0100
Message-ID<tfpWF-3Dy-11@gated-at.bofh.it>
In reply to#1587683
On Fri, Feb 24, 2017 at 03:00:03PM +0100, Jiri Olsa wrote:
> On Fri, Feb 24, 2017 at 10:32:22AM -0300, Arnaldo Carvalho de Melo wrote:
> > Em Fri, Feb 24, 2017 at 11:17:14AM +0100, Jiri Olsa escreveu:
> > > Adding system_wide flag to uncore pmu objects and passing
> > > this flag along to the their events.
> > 
> > Boris, ack? Tested-by?
> 
> actualy we discussed that on irc and I need to send new version ;-)

Boris,
would you mind testing new version?

thanks,
jirka


---
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index f4f555a67e9b..1320352cda04 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -2350,6 +2350,34 @@ static int __cmd_report(int argc, const char **argv)
 	return 0;
 }
 
+static void setup_system_wide(int forks)
+{
+	/*
+	 * Make system wide (-a) the default target if
+	 * no target was specified and one of following
+	 * conditions is met:
+	 *
+	 *   - there's no workload specified
+	 *   - there is workload specified but all requested
+	 *     events are system wide events
+	 */
+	if (!target__none(&target))
+		return;
+
+	if (!forks)
+		target.system_wide = true;
+	else {
+		struct perf_evsel *counter;
+
+		evlist__for_each_entry(evsel_list, counter) {
+			if (!counter->system_wide)
+				return;
+		}
+
+		target.system_wide = true;
+	}
+}
+
 int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
 {
 	const char * const stat_usage[] = {
@@ -2456,9 +2484,7 @@ int cmd_stat(int argc, const char **argv, const char *prefix __maybe_unused)
 	} else if (big_num_opt == 0) /* User passed --no-big-num */
 		big_num = false;
 
-	/* Make system wide (-a) the default target. */
-	if (!argc && target__none(&target))
-		target.system_wide = true;
+	setup_system_wide(argc);
 
 	if (run_count < 0) {
 		pr_err("Run count must be a positive number\n");
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index 67a8aebc67ab..54355d3caf09 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -316,8 +316,9 @@ __add_event(struct list_head *list, int *idx,
 		return NULL;
 
 	(*idx)++;
-	evsel->cpus     = cpu_map__get(cpus);
-	evsel->own_cpus = cpu_map__get(cpus);
+	evsel->cpus        = cpu_map__get(cpus);
+	evsel->own_cpus    = cpu_map__get(cpus);
+	evsel->system_wide = !!cpus;
 
 	if (name)
 		evsel->name = strdup(name);

[toc] | [prev] | [next] | [standalone]

#1588628

FromBorislav Petkov <bp@alien8.de>
Date2017-02-27 12:50 +0100
Message-ID<tfqg2-3L5-19@gated-at.bofh.it>
In reply to#1588563
On Mon, Feb 27, 2017 at 10:28:59AM +0100, Jiri Olsa wrote:
> would you mind testing new version?

Looks ok to me.

Thanks.

[boris@pd: ~/kernel/linux/tools/perf> ./perf stat -v -e amd_nb/event=0xe0,umask=0x1f/ sleep 1
Using CPUID AuthenticAMD-21-2
amd_nb/event=0xe0,umask=0x1f/: 977574 1002133816 1002133816

 Performance counter stats for 'system wide':

           977,574      amd_nb/event=0xe0,umask=0x1f/                                   

       1.002162083 seconds time elapsed

[boris@pd: ~/kernel/linux/tools/perf> ./perf stat -v -e amd_nb/event=0xe0,umask=0x1f/,cycles sleep 1
Using CPUID AuthenticAMD-21-2
Warning:
amd_nb/event=0xe0,umask=0x1f/ event is not supported by the kernel.
failed to read counter amd_nb/event=0xe0,umask=0x1f/
amd_nb/event=0xe0,umask=0x1f/: 0 0 0
cycles: 1132814 754123 754123

 Performance counter stats for 'sleep 1':

   <not supported>      amd_nb/event=0xe0,umask=0x1f/                                   
         1,132,814      cycles                                                      

       1.001673257 seconds time elapsed

[boris@pd: ~/kernel/linux/tools/perf> ./perf stat -v -e amd_nb/event=0xe0,umask=0x1f/,amd_nb/event=0xe1,umask=0x3/ sleep 1
Using CPUID AuthenticAMD-21-2
amd_nb/event=0xe0,umask=0x1f/: 1002677 1001767669 1001767669
amd_nb/event=0xe1,umask=0x3/: 0 1001764647 1001764647

 Performance counter stats for 'system wide':

         1,002,677      amd_nb/event=0xe0,umask=0x1f/                                   
                 0      amd_nb/event=0xe1,umask=0x3/                                   

       1.001811216 seconds time elapsed

[boris@pd: ~/kernel/linux/tools/perf> ./perf stat -v -e amd_nb/event=0xe0,umask=0x1f/,amd_nb/event=0xe1,umask=0x3/,amd_nb/event=0xe2,umask=0x3/ sleep 1
Using CPUID AuthenticAMD-21-2
amd_nb/event=0xe0,umask=0x1f/: 216137 1001633975 1001633975
amd_nb/event=0xe1,umask=0x3/: 0 1001635564 1001635564
amd_nb/event=0xe2,umask=0x3/: 1084188 1001637368 1001637368

 Performance counter stats for 'system wide':

           216,137      amd_nb/event=0xe0,umask=0x1f/                                   
                 0      amd_nb/event=0xe1,umask=0x3/                                   
         1,084,188      amd_nb/event=0xe2,umask=0x3/                                   

       1.001687549 seconds time elapsed

[boris@pd: ~/kernel/linux/tools/perf> ./perf stat -v -a -e amd_nb/event=0xe0,umask=0x1f/,amd_nb/event=0xe1,umask=0x3/,amd_nb/event=0xe2,umask=0x3/,cycles sleep 1
Using CPUID AuthenticAMD-21-2
amd_nb/event=0xe0,umask=0x1f/: 226857 1001760122 1001760122
amd_nb/event=0xe1,umask=0x3/: 0 1001757534 1001757534
amd_nb/event=0xe2,umask=0x3/: 1117142 1001753602 1001753602
cycles: 109689123 8014364579 8014364579

 Performance counter stats for 'system wide':

           226,857      amd_nb/event=0xe0,umask=0x1f/                                   
                 0      amd_nb/event=0xe1,umask=0x3/                                   
         1,117,142      amd_nb/event=0xe2,umask=0x3/                                   
       109,689,123      cycles                                                      

       1.001970476 seconds time elapsed

-- 
Regards/Gruss,
    Boris.

Good mailing practices for 400: avoid top-posting and trim the reply.

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web