Skip to contentSkip to conversation

perf script: Add --time-filter option

PATCH In review Namhyung Kim 5 participants 12 messages May 15, 2013 – May 16, 2013 perf events

lore
Share & tools
lore

Namhyung Kim

author maintainer
#0 lore raw

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

From: Namhyung Kim <[email protected]>

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.

---
 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(+)
tools/perf/Documentation/perf-script.txt +6 -0
index e9cbfcddfa3f..e6d591c5c60c 100644
@@ -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],
tools/perf/builtin-script.c +46 -0

70 lines not on this page · raw message

tools/perf/util/util.c +25 -0
index 59d868add275..4f68bae5f8b1 100644
@@ -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;
+}
tools/perf/util/util.h +2 -0
index 7a484c97e500..85b24cef96d4 100644
@@ -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
Signature · 1 line
1.7.11.7
author maintainer
#1 lore raw

PATCH 2/2 perf report: Add --time-filter option

From: Namhyung Kim <[email protected]>

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.

---
 tools/perf/Documentation/perf-report.txt |  6 ++++++
 tools/perf/builtin-report.c              | 33 ++++++++++++++++++++++++++++++++
 2 files changed, 39 insertions(+)
tools/perf/Documentation/perf-report.txt +6 -0
index 7d5f4f38aa52..67544c2d8e69 100644
@@ -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]
tools/perf/builtin-report.c +33 -0
index d45bf9b0361d..2cd3f640cc4a 100644
@@ -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()
 	};
 
Signature · 1 line
1.7.11.7
#2 lore raw

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

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.

Quoted 99 lines · Namhyung Kim

From: Namhyung Kim <[email protected]>

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.

---
 tools/perf/Documentation/perf-report.txt |  6 ++++++
 tools/perf/builtin-report.c              | 33 ++++++++++++++++++++++++++++++++
 2 files changed, 39 insertions(+)
tools/perf/Documentation/perf-report.txt +6 -0
index 7d5f4f38aa52..67544c2d8e69 100644
@@ -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]
tools/perf/builtin-report.c +33 -1
index d45bf9b0361d..2cd3f640cc4a 100644
@@ -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

author maintainer
#3 lore raw

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

Hi Stephane,

Quoted 6 lines · 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.

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

Thanks,
Namhyung

#4 lore raw
In reply to Namhyung Kim
Quoted 5 lines · Namhyung Kim

From: Namhyung Kim <[email protected]>

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

author maintainer
#5 lore raw

Hi Andi,

Quoted 10 lines · Andi Kleen
Quoted 5 lines · Namhyung Kim

From: Namhyung Kim <[email protected]>

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

#6 lore raw
In reply to Andi Kleen
Quoted 8 lines · Andi Kleen
Quoted 5 lines · Namhyung Kim

From: Namhyung Kim <[email protected]>

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 ...

Quoted 1 line

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

#7 lore raw
In reply to Namhyung Kim
Quoted 38 lines · Namhyung Kim

From: Namhyung Kim <[email protected]>

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.

---
  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(+)
tools/perf/Documentation/perf-script.txt +0 -0
index e9cbfcddfa3f..e6d591c5c60c 100644

tools/perf/Documentation/perf-script.txt OPTIONS · line 210

Quoted 8 lines
tools/perf/Documentation/perf-script.txt +5 -0
@@ -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.

Quoted 10 lines
tools/perf/Documentation/perf-script.txt +1 -0
+
  SEE ALSO
  --------
  linkperf:perf-record[1], linkperf:perf-script-perl[1],
tools/perf/builtin-script.c +0 -0
index 92d4658f56fb..fec624b9f8e3 100644
@@ -28,6 +28,17 @@ static bool			system_wide;
  static const char		*cpu_list;

tools/perf/builtin-script.c static bool system_wide; · line 38

Quoted 10 lines
tools/perf/builtin-script.c +8 -0
  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.

tools/perf/builtin-script.c static int process_sample_event(struct perf_tool *tool __maybe_unused, · line 525

Quoted 12 lines
tools/perf/builtin-script.c +5 -0
+
+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;

tools/perf/builtin-script.c static int process_sample_event(struct perf_tool *tool __maybe_unused, · line 528

Quoted 3 lines
tools/perf/builtin-script.c +3 -0
+
+	if ((trange.filter & TIME_FILTER_END) && trange.end < sample->time)
+		return 0;

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

    return 0;
Quoted 21 lines
tools/perf/builtin-script.c +14 -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) {

tools/perf/builtin-script.c static int have_cmd(int argc, const char **argv) · line 1278

Quoted 10 lines
tools/perf/builtin-script.c +10 -0
+		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

Quoted 63 lines
tools/perf/builtin-script.c +6 -0
+
+	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[] = {
tools/perf/util/util.c +25 -0
index 59d868add275..4f68bae5f8b1 100644
@@ -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;
+}
tools/perf/util/util.h +2 -0
index 7a484c97e500..85b24cef96d4 100644
@@ -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
author maintainer
#8 lore raw

Hi David,

Quoted 1 line · David Ahern

On 5/15/13 3:23 AM, Namhyung Kim wrote:

[SNIP]

Quoted 10 lines

tools/perf/Documentation/perf-script.txt OPTIONS · line 210

Quoted 5 lines

+--time-filter::

tools/perf/Documentation/perf-script.txt +4 -0
+	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? ;)

Quoted 13 lines
Quoted 13 lines
tools/perf/Documentation/perf-script.txt +1 -0
+
  SEE ALSO
  --------
  linkperf:perf-record[1], linkperf:perf-script-perl[1],
tools/perf/builtin-script.c +0 -0
index 92d4658f56fb..fec624b9f8e3 100644
@@ -28,6 +28,17 @@ static bool			system_wide;
  static const char		*cpu_list;
  static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
 
Quoted 10 lines

tools/perf/builtin-script.c static bool system_wide; · line 38

Quoted 8 lines
tools/perf/builtin-script.c +8 -0
+#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.

Quoted 18 lines

tools/perf/builtin-script.c static int process_sample_event(struct perf_tool *tool __maybe_unused, · line 525

Quoted 13 lines
tools/perf/builtin-script.c +5 -0
+
+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;
Quoted 7 lines

tools/perf/builtin-script.c static int process_sample_event(struct perf_tool *tool __maybe_unused, · line 528

Quoted 3 lines
tools/perf/builtin-script.c +3 -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.

Quoted 26 lines
Quoted 26 lines
tools/perf/builtin-script.c +18 -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;
+	}
Quoted 8 lines

tools/perf/builtin-script.c static int have_cmd(int argc, const char **argv) · line 1278

Quoted 6 lines
tools/perf/builtin-script.c +6 -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

#9 lore raw
Quoted 3 lines · Namhyung Kim
Quoted 1 line

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?)?

Quoted 1 line

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.

Quoted 5 lines
Quoted 1 line

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

author maintainer
#10 lore raw

Hi David,

Quoted 7 lines · David Ahern
Quoted 3 lines · Namhyung Kim
Quoted 1 line

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?

Quoted 7 lines
Quoted 2 lines

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.

Quoted 9 lines
Quoted 6 lines
Quoted 2 lines

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

#11 lore raw
Quoted 10 lines · Namhyung Kim
Quoted 7 lines · David Ahern
Quoted 3 lines · Namhyung Kim
Quoted 1 line

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