[RFC/PATCH 1/2] perf script: Add --time-filter option

[RFC/PATCH 1/2] perf script: Add --time-filter option

[Namhyung Kim]

From: Namhyung Kim <namhyung.kim@lge.com>

The --time-filter option is for limiting samples within a range of
time.  A time range looks like <time1>-<time2> and at most one of them
can be omitted.  For instance:

  $ perf script --time-filter -2178446.12
  ...
         xchat  1772 [002] 2178446.070330: sched:sched_switch: prev_comm=xchat prev_pid=177
       swapper     0 [002] 2178446.070338: power:cpu_idle: state=4 cpu_id=2
       swapper     0 [001] 2178446.091952: sched:sched_wakeup: comm=synergys pid=1488 prio=
       swapper     0 [001] 2178446.091958: power:cpu_idle: state=4294967295 cpu_id=1
       swapper     0 [001] 2178446.091970: sched:sched_switch: prev_comm=swapper/1 prev_pid
      synergys  1488 [001] 2178446.091995: sched:sched_switch: prev_comm=synergys prev_pid=
       swapper     0 [001] 2178446.092003: power:cpu_idle: state=4 cpu_id=1
       swapper     0 [001] 2178446.116997: sched:sched_wakeup: comm=synergys pid=1488 prio=
       swapper     0 [001] 2178446.117004: power:cpu_idle: state=4294967295 cpu_id=1
       swapper     0 [001] 2178446.117016: sched:sched_switch: prev_comm=swapper/1 prev_pid
      synergys  1488 [001] 2178446.117040: sched:sched_switch: prev_comm=synergys prev_pid=
       swapper     0 [001] 2178446.117048: power:cpu_idle: state=4 cpu_id=1

Above example only displays samples which have a timestamp before
2178446.120000.

Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
 tools/perf/Documentation/perf-script.txt |  6 +++++
 tools/perf/builtin-script.c              | 46 ++++++++++++++++++++++++++++++++
 tools/perf/util/util.c                   | 25 +++++++++++++++++
 tools/perf/util/util.h                   |  2 ++
 4 files changed, 79 insertions(+)

diff --git a/tools/perf/Documentation/perf-script.txt b/tools/perf/Documentation/perf-script.txt
index e9cbfcddfa3f..e6d591c5c60c 100644
--- a/tools/perf/Documentation/perf-script.txt
+++ b/tools/perf/Documentation/perf-script.txt
@@ -203,6 +203,12 @@ OPTIONS
 --show-kernel-path::
 	Try to resolve the path of [kernel.kallsyms]
 
+--time-filter::
+	Display samples within a range of time only. A time range can be given
+	like 'time1-time2' and treated as a start time and end time
+        respectively. The time format is like "<sec>.<usec>". Either of time1
+	or time2 can be omitted.
+
 SEE ALSO
 --------
 linkperf:perf-record[1], linkperf:perf-script-perl[1],
diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
index 92d4658f56fb..fec624b9f8e3 100644
--- a/tools/perf/builtin-script.c
+++ b/tools/perf/builtin-script.c
@@ -28,6 +28,17 @@ static bool			system_wide;
 static const char		*cpu_list;
 static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
 
+#define TIME_FILTER_START 1
+#define TIME_FILTER_END   2
+
+struct time_range {
+	int filter;
+	u64 start;
+	u64 end;
+};
+
+static struct time_range trange;
+
 enum perf_output_field {
 	PERF_OUTPUT_COMM            = 1U << 0,
 	PERF_OUTPUT_TID             = 1U << 1,
@@ -510,6 +521,12 @@ static int process_sample_event(struct perf_tool *tool __maybe_unused,
 	if (cpu_list && !test_bit(sample->cpu, cpu_bitmap))
 		return 0;
 
+	if ((trange.filter & TIME_FILTER_START) && trange.start > sample->time)
+		return 0;
+
+	if ((trange.filter & TIME_FILTER_END) && trange.end < sample->time)
+		return 0;
+
 	scripting_ops->process_event(event, sample, evsel, machine, &al);
 
 	evsel->hists.stats.total_period += sample->period;
@@ -1236,6 +1253,33 @@ static int have_cmd(int argc, const char **argv)
 	return 0;
 }
 
+static int
+parse_time_filter(const struct option *opt, const char *str,
+		  int unset __maybe_unused)
+{
+	struct time_range *time_range = opt->value;
+	char *sep;
+
+	sep = strchr(str, '-');
+	if (sep == NULL || sep[1] == '\0') {
+		time_range->filter = TIME_FILTER_START;
+		time_range->start = parse_nsec_time(str);
+		return 0;
+	} else if (sep == str) {
+		time_range->filter = TIME_FILTER_END;
+		time_range->end = parse_nsec_time(++str);
+		return 0;
+	}
+
+	*sep++ = '\0';
+
+	time_range->filter = TIME_FILTER_START | TIME_FILTER_END;
+	time_range->start = parse_nsec_time(str);
+	time_range->end = parse_nsec_time(sep);
+
+	return 0;
+}
+
 int cmd_script(int argc, const char **argv, const char *prefix __maybe_unused)
 {
 	bool show_full_info = false;
@@ -1286,6 +1330,8 @@ int cmd_script(int argc, const char **argv, const char *prefix __maybe_unused)
 		    "display extended information from perf.data file"),
 	OPT_BOOLEAN('\0', "show-kernel-path", &symbol_conf.show_kernel_path,
 		    "Show the path of [kernel.kallsyms]"),
+	OPT_CALLBACK(0, "time-filter", &trange, "time[-time]",
+		    "Only display entries in the time range", parse_time_filter),
 	OPT_END()
 	};
 	const char * const script_usage[] = {
diff --git a/tools/perf/util/util.c b/tools/perf/util/util.c
index 59d868add275..4f68bae5f8b1 100644
--- a/tools/perf/util/util.c
+++ b/tools/perf/util/util.c
@@ -269,3 +269,28 @@ void perf_debugfs_set_path(const char *mntpt)
 	snprintf(debugfs_mountpoint, strlen(debugfs_mountpoint), "%s", mntpt);
 	set_tracing_events_path(mntpt);
 }
+
+/* XXX.YYYYYY -> nsec */
+u64 parse_nsec_time(const char *str)
+{
+	u64 time_sec, time_nsec;
+	char *end;
+
+	time_sec = strtol(str, &end, 10);
+	if (end && *end == '.') {
+		int i;
+		char nsec_buf[10];
+
+		strncpy(nsec_buf, end+1, 9);
+
+		/* make it nsec precision */
+		for (i = strlen(nsec_buf); i < 9; i++)
+			nsec_buf[i] = '0';
+		nsec_buf[9] = '\0';
+
+		time_nsec = strtol(nsec_buf, &end, 10);
+	} else
+		time_nsec = 0;
+
+	return time_sec * NSEC_PER_SEC + time_nsec;
+}
diff --git a/tools/perf/util/util.h b/tools/perf/util/util.h
index 7a484c97e500..85b24cef96d4 100644
--- a/tools/perf/util/util.h
+++ b/tools/perf/util/util.h
@@ -204,6 +204,8 @@ static inline int has_extension(const char *filename, const char *ext)
 #define NSEC_PER_MSEC	1000000L
 #endif
 
+u64 parse_nsec_time(const char *str);
+
 extern unsigned char sane_ctype[256];
 #define GIT_SPACE		0x01
 #define GIT_DIGIT		0x02
-- 
1.7.11.7

[RFC/PATCH 2/2] perf report: Add --time-filter option

[Namhyung Kim]

From: Namhyung Kim <namhyung.kim@lge.com>

The --time-filter option is for limiting samples within a range of
time.  A time range looks like <time1>-<time2> and at most one of them
can be omitted.  This can be useful when analyzing a part of a huge
data only.

Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
 tools/perf/Documentation/perf-report.txt |  6 ++++++
 tools/perf/builtin-report.c              | 33 ++++++++++++++++++++++++++++++++
 2 files changed, 39 insertions(+)

diff --git a/tools/perf/Documentation/perf-report.txt b/tools/perf/Documentation/perf-report.txt
index 7d5f4f38aa52..67544c2d8e69 100644
--- a/tools/perf/Documentation/perf-report.txt
+++ b/tools/perf/Documentation/perf-report.txt
@@ -210,6 +210,12 @@ OPTIONS
 	Demangle symbol names to human readable form. It's enabled by default,
 	disable with --no-demangle.
 
+--time-filter::
+	Report samples within a range of time only. A time range can be given
+	like 'time1-time2' and treated as a start time and end time
+	respectively. The time format is like "<sec>.<usec>". Either of time1
+	or time2 can be omitted.
+
 SEE ALSO
 --------
 linkperf:perf-stat[1], linkperf:perf-annotate[1]
diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
index d45bf9b0361d..2cd3f640cc4a 100644
--- a/tools/perf/builtin-report.c
+++ b/tools/perf/builtin-report.c
@@ -52,6 +52,7 @@ struct perf_report {
 	symbol_filter_t		annotate_init;
 	const char		*cpu_list;
 	const char		*symbol_filter_str;
+	u64			time_start, time_end;
 	DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
 };
 
@@ -311,6 +312,12 @@ static int process_sample_event(struct perf_tool *tool,
 	if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
 		return 0;
 
+	if (rep->time_start && rep->time_start > sample->time)
+		return 0;
+
+	if (rep->time_end && rep->time_end < sample->time)
+		return 0;
+
 	if (sort__mode == SORT_MODE__BRANCH) {
 		if (perf_report__add_branch_hist_entry(tool, &al, sample,
 						       evsel, machine)) {
@@ -700,6 +707,30 @@ parse_branch_mode(const struct option *opt __maybe_unused,
 	return 0;
 }
 
+static int
+parse_time_filter(const struct option *opt, const char *str,
+		  int unset __maybe_unused)
+{
+	struct perf_report *rep = opt->value;
+	char *sep;
+
+	sep = strchr(str, '-');
+	if (sep == NULL || sep[1] == '\0') {
+		rep->time_start = parse_nsec_time(str);
+		return 0;
+	} else if (sep == str) {
+		rep->time_end = parse_nsec_time(++str);
+		return 0;
+	}
+
+	*sep++ = '\0';
+
+	rep->time_start = parse_nsec_time(str);
+	rep->time_end = parse_nsec_time(sep);
+
+	return 0;
+}
+
 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
 {
 	struct perf_session *session;
@@ -809,6 +840,8 @@ int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
 	OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
 		    "Disable symbol demangling"),
 	OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
+	OPT_CALLBACK(0, "time-filter", &report, "time[-time]",
+		    "Only display entries in the time range", parse_time_filter),
 	OPT_END()
 	};
 
-- 
1.7.11.7

Re: [RFC/PATCH 2/2] perf report: Add --time-filter option

[Stephane Eranian]

Hi,

I think what could also be useful would be an option to consider only
the samples when the rate as reported by the timestamp exceed a
user provided rate per sec. That would eliminate samples which are
coming from low activity periods.


On Wed, May 15, 2013 at 11:23 AM, Namhyung Kim <namhyung@kernel.org> wrote:
> From: Namhyung Kim <namhyung.kim@lge.com>
>
> The --time-filter option is for limiting samples within a range of
> time.  A time range looks like <time1>-<time2> and at most one of them
> can be omitted.  This can be useful when analyzing a part of a huge
> data only.
>
> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
> ---
>  tools/perf/Documentation/perf-report.txt |  6 ++++++
>  tools/perf/builtin-report.c              | 33 ++++++++++++++++++++++++++++++++
>  2 files changed, 39 insertions(+)
>
> diff --git a/tools/perf/Documentation/perf-report.txt b/tools/perf/Documentation/perf-report.txt
> index 7d5f4f38aa52..67544c2d8e69 100644
> --- a/tools/perf/Documentation/perf-report.txt
> +++ b/tools/perf/Documentation/perf-report.txt
> @@ -210,6 +210,12 @@ OPTIONS
>         Demangle symbol names to human readable form. It's enabled by default,
>         disable with --no-demangle.
>
> +--time-filter::
> +       Report samples within a range of time only. A time range can be given
> +       like 'time1-time2' and treated as a start time and end time
> +       respectively. The time format is like "<sec>.<usec>". Either of time1
> +       or time2 can be omitted.
> +
>  SEE ALSO
>  --------
>  linkperf:perf-stat[1], linkperf:perf-annotate[1]
> diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
> index d45bf9b0361d..2cd3f640cc4a 100644
> --- a/tools/perf/builtin-report.c
> +++ b/tools/perf/builtin-report.c
> @@ -52,6 +52,7 @@ struct perf_report {
>         symbol_filter_t         annotate_init;
>         const char              *cpu_list;
>         const char              *symbol_filter_str;
> +       u64                     time_start, time_end;
>         DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
>  };
>
> @@ -311,6 +312,12 @@ static int process_sample_event(struct perf_tool *tool,
>         if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
>                 return 0;
>
> +       if (rep->time_start && rep->time_start > sample->time)
> +               return 0;
> +
> +       if (rep->time_end && rep->time_end < sample->time)
> +               return 0;
> +
>         if (sort__mode == SORT_MODE__BRANCH) {
>                 if (perf_report__add_branch_hist_entry(tool, &al, sample,
>                                                        evsel, machine)) {
> @@ -700,6 +707,30 @@ parse_branch_mode(const struct option *opt __maybe_unused,
>         return 0;
>  }
>
> +static int
> +parse_time_filter(const struct option *opt, const char *str,
> +                 int unset __maybe_unused)
> +{
> +       struct perf_report *rep = opt->value;
> +       char *sep;
> +
> +       sep = strchr(str, '-');
> +       if (sep == NULL || sep[1] == '\0') {
> +               rep->time_start = parse_nsec_time(str);
> +               return 0;
> +       } else if (sep == str) {
> +               rep->time_end = parse_nsec_time(++str);
> +               return 0;
> +       }
> +
> +       *sep++ = '\0';
> +
> +       rep->time_start = parse_nsec_time(str);
> +       rep->time_end = parse_nsec_time(sep);
> +
> +       return 0;
> +}
> +
>  int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
>  {
>         struct perf_session *session;
> @@ -809,6 +840,8 @@ int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
>         OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
>                     "Disable symbol demangling"),
>         OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
> +       OPT_CALLBACK(0, "time-filter", &report, "time[-time]",
> +                   "Only display entries in the time range", parse_time_filter),
>         OPT_END()
>         };
>
> --
> 1.7.11.7
>

Re: [RFC/PATCH 1/2] perf script: Add --time-filter option

[Andi Kleen]

On Wed, May 15, 2013 at 06:23:58PM +0900, Namhyung Kim wrote:
> From: Namhyung Kim <namhyung.kim@lge.com>
> 
> The --time-filter option is for limiting samples within a range of
> time.  A time range looks like <time1>-<time2> and at most one of them
> can be omitted.  For instance:

It would be more useful for report or inject than for script.
After all with a script it's trivial to do an own filter.
Not so easy with report.

-Andi

Re: [RFC/PATCH 1/2] perf script: Add --time-filter option

[David Ahern]

On 5/15/13 3:23 AM, Namhyung Kim wrote:
> From: Namhyung Kim <namhyung.kim@lge.com>
>
> The --time-filter option is for limiting samples within a range of
> time.  A time range looks like <time1>-<time2> and at most one of them
> can be omitted.  For instance:
>
>    $ perf script --time-filter -2178446.12
>    ...
>           xchat  1772 [002] 2178446.070330: sched:sched_switch: prev_comm=xchat prev_pid=177
>         swapper     0 [002] 2178446.070338: power:cpu_idle: state=4 cpu_id=2
>         swapper     0 [001] 2178446.091952: sched:sched_wakeup: comm=synergys pid=1488 prio=
>         swapper     0 [001] 2178446.091958: power:cpu_idle: state=4294967295 cpu_id=1
>         swapper     0 [001] 2178446.091970: sched:sched_switch: prev_comm=swapper/1 prev_pid
>        synergys  1488 [001] 2178446.091995: sched:sched_switch: prev_comm=synergys prev_pid=
>         swapper     0 [001] 2178446.092003: power:cpu_idle: state=4 cpu_id=1
>         swapper     0 [001] 2178446.116997: sched:sched_wakeup: comm=synergys pid=1488 prio=
>         swapper     0 [001] 2178446.117004: power:cpu_idle: state=4294967295 cpu_id=1
>         swapper     0 [001] 2178446.117016: sched:sched_switch: prev_comm=swapper/1 prev_pid
>        synergys  1488 [001] 2178446.117040: sched:sched_switch: prev_comm=synergys prev_pid=
>         swapper     0 [001] 2178446.117048: power:cpu_idle: state=4 cpu_id=1
>
> Above example only displays samples which have a timestamp before
> 2178446.120000.
>
> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
> Signed-off-by: Namhyung Kim <namhyung@kernel.org>
> ---
>   tools/perf/Documentation/perf-script.txt |  6 +++++
>   tools/perf/builtin-script.c              | 46 ++++++++++++++++++++++++++++++++
>   tools/perf/util/util.c                   | 25 +++++++++++++++++
>   tools/perf/util/util.h                   |  2 ++
>   4 files changed, 79 insertions(+)
>
> diff --git a/tools/perf/Documentation/perf-script.txt b/tools/perf/Documentation/perf-script.txt
> index e9cbfcddfa3f..e6d591c5c60c 100644
> --- a/tools/perf/Documentation/perf-script.txt
> +++ b/tools/perf/Documentation/perf-script.txt
> @@ -203,6 +203,12 @@ OPTIONS
>   --show-kernel-path::
>   	Try to resolve the path of [kernel.kallsyms]
>
> +--time-filter::
> +	Display samples within a range of time only. A time range can be given
> +	like 'time1-time2' and treated as a start time and end time
> +        respectively. The time format is like "<sec>.<usec>". Either of time1
> +	or time2 can be omitted.

I have this option internally on all analysis commands for while now -- 
on report, script and my timehist command. Very useful feature.

How about just --time? less typing.

> +
>   SEE ALSO
>   --------
>   linkperf:perf-record[1], linkperf:perf-script-perl[1],
> diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
> index 92d4658f56fb..fec624b9f8e3 100644
> --- a/tools/perf/builtin-script.c
> +++ b/tools/perf/builtin-script.c
> @@ -28,6 +28,17 @@ static bool			system_wide;
>   static const char		*cpu_list;
>   static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
>
> +#define TIME_FILTER_START 1
> +#define TIME_FILTER_END   2
> +
> +struct time_range {
> +	int filter;
> +	u64 start;
> +	u64 end;
> +};

The FILTER parts should not be needed.

> +
> +static struct time_range trange;
> +
>   enum perf_output_field {
>   	PERF_OUTPUT_COMM            = 1U << 0,
>   	PERF_OUTPUT_TID             = 1U << 1,
> @@ -510,6 +521,12 @@ static int process_sample_event(struct perf_tool *tool __maybe_unused,
>   	if (cpu_list && !test_bit(sample->cpu, cpu_bitmap))
>   		return 0;
>
> +	if ((trange.filter & TIME_FILTER_START) && trange.start > sample->time)
> +		return 0;

How about just:
if (trange.start && trange.start < sample->time)
     return 0;

> +
> +	if ((trange.filter & TIME_FILTER_END) && trange.end < sample->time)
> +		return 0;

and similarly:
if (trange.end && trange.end > sample->time)
     return 0;

> +
>   	scripting_ops->process_event(event, sample, evsel, machine, &al);
>
>   	evsel->hists.stats.total_period += sample->period;
> @@ -1236,6 +1253,33 @@ static int have_cmd(int argc, const char **argv)
>   	return 0;
>   }
>
> +static int
> +parse_time_filter(const struct option *opt, const char *str,
> +		  int unset __maybe_unused)
> +{
> +	struct time_range *time_range = opt->value;
> +	char *sep;
> +
> +	sep = strchr(str, '-');
> +	if (sep == NULL || sep[1] == '\0') {
> +		time_range->filter = TIME_FILTER_START;
> +		time_range->start = parse_nsec_time(str);
> +		return 0;
> +	} else if (sep == str) {
> +		time_range->filter = TIME_FILTER_END;
> +		time_range->end = parse_nsec_time(++str);
> +		return 0;
> +	}
> +
> +	*sep++ = '\0';
> +
> +	time_range->filter = TIME_FILTER_START | TIME_FILTER_END;
> +	time_range->start = parse_nsec_time(str);
> +	time_range->end = parse_nsec_time(sep);

I would expect parse_nsec_time to fail. e.g., a time string like 123455.a

David


> +
> +	return 0;
> +}
> +
>   int cmd_script(int argc, const char **argv, const char *prefix __maybe_unused)
>   {
>   	bool show_full_info = false;
> @@ -1286,6 +1330,8 @@ int cmd_script(int argc, const char **argv, const char *prefix __maybe_unused)
>   		    "display extended information from perf.data file"),
>   	OPT_BOOLEAN('\0', "show-kernel-path", &symbol_conf.show_kernel_path,
>   		    "Show the path of [kernel.kallsyms]"),
> +	OPT_CALLBACK(0, "time-filter", &trange, "time[-time]",
> +		    "Only display entries in the time range", parse_time_filter),
>   	OPT_END()
>   	};
>   	const char * const script_usage[] = {
> diff --git a/tools/perf/util/util.c b/tools/perf/util/util.c
> index 59d868add275..4f68bae5f8b1 100644
> --- a/tools/perf/util/util.c
> +++ b/tools/perf/util/util.c
> @@ -269,3 +269,28 @@ void perf_debugfs_set_path(const char *mntpt)
>   	snprintf(debugfs_mountpoint, strlen(debugfs_mountpoint), "%s", mntpt);
>   	set_tracing_events_path(mntpt);
>   }
> +
> +/* XXX.YYYYYY -> nsec */
> +u64 parse_nsec_time(const char *str)
> +{
> +	u64 time_sec, time_nsec;
> +	char *end;
> +
> +	time_sec = strtol(str, &end, 10);
> +	if (end && *end == '.') {
> +		int i;
> +		char nsec_buf[10];
> +
> +		strncpy(nsec_buf, end+1, 9);
> +
> +		/* make it nsec precision */
> +		for (i = strlen(nsec_buf); i < 9; i++)
> +			nsec_buf[i] = '0';
> +		nsec_buf[9] = '\0';
> +
> +		time_nsec = strtol(nsec_buf, &end, 10);
> +	} else
> +		time_nsec = 0;
> +
> +	return time_sec * NSEC_PER_SEC + time_nsec;
> +}
> diff --git a/tools/perf/util/util.h b/tools/perf/util/util.h
> index 7a484c97e500..85b24cef96d4 100644
> --- a/tools/perf/util/util.h
> +++ b/tools/perf/util/util.h
> @@ -204,6 +204,8 @@ static inline int has_extension(const char *filename, const char *ext)
>   #define NSEC_PER_MSEC	1000000L
>   #endif
>
> +u64 parse_nsec_time(const char *str);
> +
>   extern unsigned char sane_ctype[256];
>   #define GIT_SPACE		0x01
>   #define GIT_DIGIT		0x02
>

Re: [RFC/PATCH 2/2] perf report: Add --time-filter option

[Namhyung Kim]

Hi Stephane,

On Wed, 15 May 2013 13:51:20 +0200, Stephane Eranian wrote:
> Hi,
>
> I think what could also be useful would be an option to consider only
> the samples when the rate as reported by the timestamp exceed a
> user provided rate per sec. That would eliminate samples which are
> coming from low activity periods.

Sounds good.  I'll try to play with it next time.

Thanks,
Namhyung

Re: [RFC/PATCH 1/2] perf script: Add --time-filter option

[Namhyung Kim]

Hi Andi,

On Wed, 15 May 2013 15:56:08 +0200, Andi Kleen wrote:
> On Wed, May 15, 2013 at 06:23:58PM +0900, Namhyung Kim wrote:
>> From: Namhyung Kim <namhyung.kim@lge.com>
>> 
>> The --time-filter option is for limiting samples within a range of
>> time.  A time range looks like <time1>-<time2> and at most one of them
>> can be omitted.  For instance:
>
> It would be more useful for report or inject than for script.
> After all with a script it's trivial to do an own filter.
> Not so easy with report.

Right.  I actually sent the patch 2/2 for report side and it's the main
aim for this patch.  This was included almost for verification. :)

Anyway, I didn't think about the inject command, I will apply it to the
inject too in next spin.

Thanks,
Namhyung

Re: [RFC/PATCH 1/2] perf script: Add --time-filter option

[Namhyung Kim]

Hi David,

On Wed, 15 May 2013 09:16:55 -0600, David Ahern wrote:
> On 5/15/13 3:23 AM, Namhyung Kim wrote:
[SNIP]
>> +--time-filter::
>> +	Display samples within a range of time only. A time range can be given
>> +	like 'time1-time2' and treated as a start time and end time
>> +        respectively. The time format is like "<sec>.<usec>". Either of time1
>> +	or time2 can be omitted.
>
> I have this option internally on all analysis commands for while now -- 
> on report, script and my timehist command. Very useful feature.
>
> How about just --time? less typing.

Thanks, I'm fine with '--time' too but '--time-filter' looks more
obvious.  What does the timehist command do, btw? ;)

>
>> +
>>   SEE ALSO
>>   --------
>>   linkperf:perf-record[1], linkperf:perf-script-perl[1],
>> diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
>> index 92d4658f56fb..fec624b9f8e3 100644
>> --- a/tools/perf/builtin-script.c
>> +++ b/tools/perf/builtin-script.c
>> @@ -28,6 +28,17 @@ static bool			system_wide;
>>   static const char		*cpu_list;
>>   static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
>>
>> +#define TIME_FILTER_START 1
>> +#define TIME_FILTER_END   2
>> +
>> +struct time_range {
>> +	int filter;
>> +	u64 start;
>> +	u64 end;
>> +};
>
> The FILTER parts should not be needed.

Right.  I'll remove it.

>
>> +
>> +static struct time_range trange;
>> +
>>   enum perf_output_field {
>>   	PERF_OUTPUT_COMM            = 1U << 0,
>>   	PERF_OUTPUT_TID             = 1U << 1,
>> @@ -510,6 +521,12 @@ static int process_sample_event(struct perf_tool *tool __maybe_unused,
>>   	if (cpu_list && !test_bit(sample->cpu, cpu_bitmap))
>>   		return 0;
>>
>> +	if ((trange.filter & TIME_FILTER_START) && trange.start > sample->time)
>> +		return 0;
>
> How about just:
> if (trange.start && trange.start < sample->time)
>     return 0;
>
>> +
>> +	if ((trange.filter & TIME_FILTER_END) && trange.end < sample->time)
>> +		return 0;
>
> and similarly:
> if (trange.end && trange.end > sample->time)
>     return 0;

Okay.

>
>> +
>>   	scripting_ops->process_event(event, sample, evsel, machine, &al);
>>
>>   	evsel->hists.stats.total_period += sample->period;
>> @@ -1236,6 +1253,33 @@ static int have_cmd(int argc, const char **argv)
>>   	return 0;
>>   }
>>
>> +static int
>> +parse_time_filter(const struct option *opt, const char *str,
>> +		  int unset __maybe_unused)
>> +{
>> +	struct time_range *time_range = opt->value;
>> +	char *sep;
>> +
>> +	sep = strchr(str, '-');
>> +	if (sep == NULL || sep[1] == '\0') {
>> +		time_range->filter = TIME_FILTER_START;
>> +		time_range->start = parse_nsec_time(str);
>> +		return 0;
>> +	} else if (sep == str) {
>> +		time_range->filter = TIME_FILTER_END;
>> +		time_range->end = parse_nsec_time(++str);
>> +		return 0;
>> +	}
>> +
>> +	*sep++ = '\0';
>> +
>> +	time_range->filter = TIME_FILTER_START | TIME_FILTER_END;
>> +	time_range->start = parse_nsec_time(str);
>> +	time_range->end = parse_nsec_time(sep);
>
> I would expect parse_nsec_time to fail. e.g., a time string like 123455.a

It looks like current strtol() returns 0 when failed to parse like
above.  Hmm.. do I have to check whether the return value is 0 or just
ignore invalid inputs?

Thanks,
Namhyung

Re: [RFC/PATCH 1/2] perf script: Add --time-filter option

[David Ahern]

On 5/15/13 7:56 PM, Namhyung Kim wrote:
>> How about just --time? less typing.
>
> Thanks, I'm fine with '--time' too but '--time-filter' looks more

yes, I just have really long command lines now. Is there a consistent 
single letter (X?)?

> obvious.  What does the timehist command do, btw? ;)

task scheduling time history including run time and time between 
sched-in. It needs to be updated to use tracepoints and perhaps fold 
into perf-sched (e.g., perf sched history). Too many features; too 
little time. I hope to get to it in the next month or so.


>> I would expect parse_nsec_time to fail. e.g., a time string like 123455.a
>
> It looks like current strtol() returns 0 when failed to parse like
> above.  Hmm.. do I have to check whether the return value is 0 or just
> ignore invalid inputs?

end will point to the next character not converted so make sure it is as 
expected (*end == '\0').

David

Re: [RFC/PATCH 1/2] perf script: Add --time-filter option

[Namhyung Kim]

Hi David,

On Wed, 15 May 2013 20:42:30 -0600, David Ahern wrote:
> On 5/15/13 7:56 PM, Namhyung Kim wrote:
>>> How about just --time? less typing.
>>
>> Thanks, I'm fine with '--time' too but '--time-filter' looks more
>
> yes, I just have really long command lines now. Is there a consistent
> single letter (X?)?

Do you mean a single-letter option "-X" for this?

>
>> obvious.  What does the timehist command do, btw? ;)
>
> task scheduling time history including run time and time between
> sched-in. It needs to be updated to use tracepoints and perhaps fold
> into perf-sched (e.g., perf sched history). Too many features; too
> little time. I hope to get to it in the next month or so.

Looks useful.

>
>>> I would expect parse_nsec_time to fail. e.g., a time string like 123455.a
>>
>> It looks like current strtol() returns 0 when failed to parse like
>> above.  Hmm.. do I have to check whether the return value is 0 or just
>> ignore invalid inputs?
>
> end will point to the next character not converted so make sure it is
> as expected (*end == '\0').

Arh, okay.  So I'll change it to check the 'end' and fails if an
unexpected input comes.

Thanks,
Namhyung

Re: [RFC/PATCH 1/2] perf script: Add --time-filter option

[Ingo Molnar]

* Andi Kleen <andi@firstfloor.org> wrote:

> On Wed, May 15, 2013 at 06:23:58PM +0900, Namhyung Kim wrote:
> > From: Namhyung Kim <namhyung.kim@lge.com>
> > 
> > The --time-filter option is for limiting samples within a range of
> > time.  A time range looks like <time1>-<time2> and at most one of them
> > can be omitted.  For instance:
> 
> It would be more useful for report or inject than for script.

The 2/2 patch does it for report ...

> After all with a script it's trivial to do an own filter.

Utility features/shortcuts are almost always useful, especially if they 
span multiple tools.

Thanks,

	Ingo

Re: [RFC/PATCH 1/2] perf script: Add --time-filter option

[David Ahern]

On 5/16/13 2:50 AM, Namhyung Kim wrote:
> On Wed, 15 May 2013 20:42:30 -0600, David Ahern wrote:
>> On 5/15/13 7:56 PM, Namhyung Kim wrote:
>>>> How about just --time? less typing.
>>>
>>> Thanks, I'm fine with '--time' too but '--time-filter' looks more
>>
>> yes, I just have really long command lines now. Is there a consistent
>> single letter (X?)?
>
> Do you mean a single-letter option "-X" for this?

Yes, by consistent I meant a letter not used by any of the current 
analysis commands (report, script, diff). We are to the point where 
options are getting added only as --some-name now. Be nice to have 
single letter shortcuts.

David