Re: [PATCH 2/2] perf, tools: Add --delay support to perf record

[Namhyung Kim <namhyung@kernel.org>]

Hi Andi,

On Tue,  7 Jan 2014 14:14:07 -0800, Andi Kleen wrote:
> From: Andi Kleen <ak@linux.intel.com>
>
> perf stat has a --delay option to delay measuring the workload.
> This is useful to skip measuring the startup phase of the program, which
> is often very different from the main workload.
>
> The same is useful for perf record when sampling. Add --delay to perf record too.
> -D was already taken for record, so there is only a long option.
>
> Signed-off-by: Andi Kleen <ak@linux.intel.com>
> ---
>  tools/perf/Documentation/perf-record.txt | 4 ++++
>  tools/perf/builtin-record.c              | 9 ++++++++-
>  tools/perf/perf.h                        | 1 +
>  tools/perf/util/evsel.c                  | 5 +++--
>  4 files changed, 16 insertions(+), 3 deletions(-)
>
> diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt
> index 43b42c4..e349151 100644
> --- a/tools/perf/Documentation/perf-record.txt
> +++ b/tools/perf/Documentation/perf-record.txt
> @@ -207,6 +207,10 @@ Force the use of per-cpu mmaps.  By default, when tasks are specified (i.e. -p,
>  forces per-cpu mmaps.  A side-effect of that is that inheritance is
>  automatically enabled.  Add the -i option also to disable inheritance.
>  
> +--delay msecs::
> +After starting the program, wait msecs before measuring. This is useful to
> +filter out the startup phase of the program, which is often very different.
> +
>  SEE ALSO
>  --------
>  linkperf:perf-stat[1], linkperf:perf-list[1]
> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
> index 7c8020a..110158e 100644
> --- a/tools/perf/builtin-record.c
> +++ b/tools/perf/builtin-record.c
> @@ -506,7 +506,7 @@ static int __cmd_record(struct perf_record *rec, int argc, const char **argv)
>  	 * (apart from group members) have enable_on_exec=1 set,
>  	 * so don't spoil it by prematurely enabling them.
>  	 */
> -	if (!target__none(&opts->target))
> +	if (!target__none(&opts->target) && !opts->initial_delay)
>  		perf_evlist__enable(evsel_list);
>  
>  	/*
> @@ -515,6 +515,11 @@ static int __cmd_record(struct perf_record *rec, int argc, const char **argv)
>  	if (forks)
>  		perf_evlist__start_workload(evsel_list);
>  
> +	if (opts->initial_delay) {
> +		usleep(opts->initial_delay * 1000);
> +		perf_evlist__enable(evsel_list);
> +	}
> +
>  	for (;;) {
>  		int hits = rec->samples;
>  
> @@ -890,6 +895,8 @@ const struct option record_options[] = {
>  		    "sample transaction flags (special events only)"),
>  	OPT_BOOLEAN(0, "force-per-cpu", &record.opts.target.force_per_cpu,
>  		    "force the use of per-cpu mmaps"),
> +	OPT_UINTEGER(0, "delay", &record.opts.initial_delay,
> +		  "ms to wait before starting measurement after program start"),
>  	OPT_END()
>  };
>  
> diff --git a/tools/perf/perf.h b/tools/perf/perf.h
> index b079304..d052f1d 100644
> --- a/tools/perf/perf.h
> +++ b/tools/perf/perf.h
> @@ -268,6 +268,7 @@ struct perf_record_opts {
>  	u64	     user_interval;
>  	u16	     stack_dump_size;
>  	bool	     sample_transaction;
> +	unsigned     initial_delay;
>  };
>  
>  #endif
> diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
> index 46dd4c2..4a70656 100644
> --- a/tools/perf/util/evsel.c
> +++ b/tools/perf/util/evsel.c
> @@ -689,14 +689,15 @@ void perf_evsel__config(struct perf_evsel *evsel,
>  	 * Disabling only independent events or group leaders,
>  	 * keeping group members enabled.
>  	 */
> -	if (perf_evsel__is_group_leader(evsel))
> +	if (perf_evsel__is_group_leader(evsel) || opts->initial_delay)
>  		attr->disabled = 1;

It seems this hunk is not needed since it'll disable group member events
which is not what we want.  Note that the perf_evlist__enable/disable()
only work for the leaders.

Thanks,
Namhyung

>  
>  	/*
>  	 * Setting enable_on_exec for independent events and
>  	 * group leaders for traced executed by perf.
>  	 */
> -	if (target__none(&opts->target) && perf_evsel__is_group_leader(evsel))
> +	if (target__none(&opts->target) && perf_evsel__is_group_leader(evsel) &&
> +		!opts->initial_delay)
>  		attr->enable_on_exec = 1;
>  }