Re: [PATCH 14/14] perf record: Add --dir option to store data in directory

[Alexey Budankov <alexey.budankov@linux.intel.com>]

On 03.02.2019 18:30, Jiri Olsa wrote:
> Adding --dir option to store data in directory. It's next
> step for multiple threads in record. It's not possible
> to make directory data via --dir option, like:
> 
>   $ perf record --dir perf bench sched messaging
>   $ ls -l perf.data
>   total 344
>   -rw-------. 1 jolsa jolsa 43864 Jan 20 22:26 data.0
>   -rw-------. 1 jolsa jolsa 30464 Jan 20 22:26 data.1
>   -rw-------. 1 jolsa jolsa 53816 Jan 20 22:26 data.2
>   -rw-------. 1 jolsa jolsa 30368 Jan 20 22:26 data.3
>   -rw-------. 1 jolsa jolsa 40088 Jan 20 22:26 data.4
>   -rw-------. 1 jolsa jolsa 42592 Jan 20 22:26 data.5
>   -rw-------. 1 jolsa jolsa 56136 Jan 20 22:26 data.6
>   -rw-------. 1 jolsa jolsa 25992 Jan 20 22:26 data.7
>   -rw-------. 1 jolsa jolsa  8832 Jan 20 22:26 header
> 
> There's a data file created for every cpu and it's storing
> data for those cpu maps.
> 
> It's possible to transform directory data into standard
> perf.data file via following inject command:
> 
>   $ perf inject -o perf.data.file -i perf.data
> 
> Link: http://lkml.kernel.org/n/tip-0kjm8wpglzu2tm18tpagfm4d@git.kernel.org
> Signed-off-by: Jiri Olsa <jolsa@kernel.org>
> ---
>  tools/perf/Documentation/perf-record.txt |  3 ++
>  tools/perf/builtin-record.c              | 59 ++++++++++++++++++++++--
>  tools/perf/util/mmap.h                   | 23 ++++-----
>  3 files changed, 70 insertions(+), 15 deletions(-)
> 
> diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt
> index d232b13ea713..8dcdc8cabcad 100644
> --- a/tools/perf/Documentation/perf-record.txt
> +++ b/tools/perf/Documentation/perf-record.txt
> @@ -505,6 +505,9 @@ config terms. For example: 'cycles/overwrite/' and 'instructions/no-overwrite/'.
>  
>  Implies --tail-synthesize.
>  
> +--dir::
> +Store data into directory with one data file for cpu.
> +

Makes sense to mention compatibility with -o option and per-thread buffer mapping.

>  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 cd02ab3ec4ff..87e39b9cc7bd 100644
> --- a/tools/perf/builtin-record.c
> +++ b/tools/perf/builtin-record.c
> @@ -111,17 +111,21 @@ static bool switch_output_time(struct record *rec)
>  	       trigger_is_ready(&switch_output_trigger);
>  }
>  
> -static int record__write(struct record *rec, struct perf_mmap *map __maybe_unused,
> +static int record__write(struct record *rec, struct perf_mmap *map,
>  			 void *bf, size_t size)
>  {
> -	struct perf_data_file *file = &rec->session->data->file;
> +	struct perf_data_file *file = &rec->data.file;
> +
> +	if (map && map->file)
> +		file = map->file;

For AIO mode per-cpu streaming could be done in parallel because Posix 
AIO API uses a separate thread for every open data.# or header fd.

>  
>  	if (perf_data_file__write(file, bf, size) < 0) {
>  		pr_err("failed to write perf data, error: %m\n");
>  		return -1;
>  	}
>  
> -	rec->bytes_written += size;
> +	if (file == &rec->data.file)
> +		rec->bytes_written += size;

switch-output logic now tracks header file size only? 
If so, it probably needs to be corrected or simply disabled 
for --output_dir mode.

Alexey

>  
>  	if (switch_output_size(rec))
>  		trigger_hit(&switch_output_trigger);
> @@ -563,6 +567,25 @@ static int record__mmap_evlist(struct record *rec,
>  	return 0;
>  }
>  
> +static int record__mmap_dir_data(struct record *rec)
> +{
> +	struct perf_evlist *evlist = rec->evlist;
> +	struct perf_data *data = &rec->data;
> +	int i, ret, nr = evlist->nr_mmaps;
> +
> +	ret = perf_data__create_dir(data, nr);
> +	if (ret)
> +		return ret;
> +
> +	for (i = 0; i < nr; i++) {
> +		struct perf_mmap *map = &evlist->mmap[i];
> +
> +		map->file = &data->dir.files[i];
> +	}
> +
> +	return 0;
> +}
> +
>  static int record__mmap(struct record *rec)
>  {
>  	return record__mmap_evlist(rec, rec->evlist);
> @@ -792,8 +815,12 @@ static int record__mmap_read_evlist(struct record *rec, struct perf_evlist *evli
>  	/*
>  	 * Mark the round finished in case we wrote
>  	 * at least one event.
> +	 *
> +	 * No need for round events in directory mode,
> +	 * because per-cpu files/maps have sorted data
> +	 * from kernel.
>  	 */
> -	if (bytes_written != rec->bytes_written)
> +	if (!perf_data__is_dir(&rec->data) && bytes_written != rec->bytes_written)
>  		rc = record__write(rec, NULL, &finished_round_event, sizeof(finished_round_event));
>  
>  	if (overwrite)
> @@ -851,6 +878,9 @@ record__finish_output(struct record *rec)
>  	rec->session->header.data_size += rec->bytes_written;
>  	data->file.size = lseek(perf_data__fd(data), 0, SEEK_CUR);
>  
> +	if (perf_data__is_dir(data))
> +		perf_data__update_dir(data);
> +
>  	if (!rec->no_buildid) {
>  		process_buildids(rec);
>  
> @@ -1170,11 +1200,23 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
>  	if (data->is_pipe && rec->evlist->nr_entries == 1)
>  		rec->opts.sample_id = true;
>  
> +	if (data->is_pipe && perf_data__is_dir(data)) {
> +		pr_err("Directory output is not allowed for pipe output\n");
> +		err = -1;
> +		goto out_child;
> +	}
> +
>  	if (record__open(rec) != 0) {
>  		err = -1;
>  		goto out_child;
>  	}
>  
> +	if (perf_data__is_dir(data)) {
> +		err = record__mmap_dir_data(rec);
> +		if (err)
> +			goto out_child;
> +	}
> +
>  	err = bpf__apply_obj_config();
>  	if (err) {
>  		char errbuf[BUFSIZ];
> @@ -1962,6 +2004,8 @@ static struct option __record_options[] = {
>  		     &nr_cblocks_default, "n", "Use <n> control blocks in asynchronous trace writing mode (default: 1, max: 4)",
>  		     record__aio_parse),
>  #endif
> +	OPT_BOOLEAN(0, "dir", &record.data.is_dir,
> +		    "Store data into directory perf.data"),
>  	OPT_END()
>  };
>  
> @@ -2113,6 +2157,13 @@ int cmd_record(int argc, const char **argv)
>  		goto out;
>  	}
>  
> +	if (perf_data__is_dir(&rec->data)) {
> +		if (!rec->opts.sample_time) {
> +			pr_err("Sample timestamp is required for indexing\n");
> +			goto out;
> +		}
> +	}
> +
>  	if (rec->opts.target.tid && !rec->opts.no_inherit_set)
>  		rec->opts.no_inherit = true;
>  
> diff --git a/tools/perf/util/mmap.h b/tools/perf/util/mmap.h
> index e566c19b242b..3e8595a8d6ce 100644
> --- a/tools/perf/util/mmap.h
> +++ b/tools/perf/util/mmap.h
> @@ -19,17 +19,18 @@ struct aiocb;
>   * @refcnt - e.g. code using PERF_EVENT_IOC_SET_OUTPUT to share this
>   */
>  struct perf_mmap {
> -	void		 *base;
> -	int		 mask;
> -	int		 fd;
> -	int		 cpu;
> -	refcount_t	 refcnt;
> -	u64		 prev;
> -	u64		 start;
> -	u64		 end;
> -	bool		 overwrite;
> -	struct auxtrace_mmap auxtrace_mmap;
> -	char		 event_copy[PERF_SAMPLE_MAX_SIZE] __aligned(8);
> +	void			 *base;
> +	int			 mask;
> +	int			 fd;
> +	int			 cpu;
> +	refcount_t		 refcnt;
> +	u64			 prev;
> +	u64			 start;
> +	u64			 end;
> +	bool			 overwrite;
> +	struct auxtrace_mmap	 auxtrace_mmap;
> +	struct perf_data_file	*file;
> +	char			 event_copy[PERF_SAMPLE_MAX_SIZE] __aligned(8);
>  #ifdef HAVE_AIO_SUPPORT
>  	struct {
>  		void		 **data;
>