lkml.org 
[lkml]   [2020]   [Jun]   [2]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [PATCHv3] perf stat: Ensure group is defined on top of the same cpu mask
Em Tue, Jun 02, 2020 at 12:17:36PM +0200, Jiri Olsa escreveu:
> Jin Yao reported the issue (and posted first versions of this change)
> with groups being defined over events with different cpu mask.

> This causes assert aborts in get_group_fd, like:

> # perf stat -M "C2_Pkg_Residency" -a -- sleep 1
> perf: util/evsel.c:1464: get_group_fd: Assertion `!(fd == -1)' failed.
> Aborted

> All the events in the group have to be defined over the same
> cpus so the group_fd can be found for every leader/member pair.

> Adding check to ensure this condition is met and removing the
> group (with warning) if we detect mixed cpus, like:

> $ sudo perf stat -e '{power/energy-cores/,cycles},{instructions,power/energy-cores/}'
> WARNING: event cpu maps do not match, disabling group:
> anon group { power/energy-cores/, cycles }
> anon group { instructions, power/energy-cores/ }

So it doesn't disable the 'group', it disables the 'grouping' of those
events, right? I.e. reading the WARNING, I thought that it would count
nothing, since it lists both groups as being disabled, but when I tested
I noticed that:

[root@seventh ~]# perf stat -e '{power/energy-cores/,cycles},{instructions,power/energy-cores/}'
WARNING: grouped events cpus do not match, disabling group:
anon group { power/energy-cores/, cycles }
anon group { instructions, power/energy-cores/ }
^C
Performance counter stats for 'system wide':

12.62 Joules power/energy-cores/
106,920,637 cycles
80,228,899 instructions # 0.75 insn per cycle
12.62 Joules power/energy-cores/

14.514476987 seconds time elapsed


[root@seventh ~]#

I.e. it counted the events, ungrouped, or am I missing something?

If I do:

[root@seventh ~]# perf stat -e '{power/energy-cores/,power/energy-ram/},{instructions,cycles}' -a sleep 2

Performance counter stats for 'system wide':

1.73 Joules power/energy-cores/
0.92 Joules power/energy-ram/
12,191,658 instructions # 0.67 insn per cycle
18,275,233 cycles

2.001272492 seconds time elapsed

[root@seventh ~]#

It works, grouped. One observation, shouldn't we somehow show in the
output that the first two were indeed grouped, ditto for the second two?

Also, this needs improvement:

[root@seventh ~]# perf stat -e '{power/energy-cores/,power/energy-ram/},{instructions,cycles}' sleep 2
Error:
The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (power/energy-cores/).
/bin/dmesg | grep -i perf may provide additional information.

[root@seventh ~]#

Probably stating that the power/ events can only be done on a system
wide mode or per-cpu?

I'm applying the patch now, with the above observations as committer
notes, we can improve this in follow on patch,

- Arnaldo

> Ian asked also for cpu maps details, it's displayed in verbose mode:
>
> $ sudo perf stat -e '{cycles,power/energy-cores/}' -v
> WARNING: group events cpu maps do not match, disabling group:
> anon group { power/energy-cores/, cycles }
> power/energy-cores/: 0
> cycles: 0-7
> anon group { instructions, power/energy-cores/ }
> instructions: 0-7
> power/energy-cores/: 0
>
> Fixes: 6a4bb04caacc8 ("perf tools: Enable grouping logic for parsed events")
> Co-developed-by: Jin Yao <yao.jin@linux.intel.com>
> Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
> Signed-off-by: Jiri Olsa <jolsa@kernel.org>
> ---
> tools/perf/builtin-stat.c | 55 +++++++++++++++++++++++++++++++++++++++
> 1 file changed, 55 insertions(+)
>
> v3 changes:
> - reword the warning with Ian's suggestion
>
> diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
> index b2b79aa161dd..9be020e0098a 100644
> --- a/tools/perf/builtin-stat.c
> +++ b/tools/perf/builtin-stat.c
> @@ -190,6 +190,59 @@ static struct perf_stat_config stat_config = {
> .big_num = true,
> };
>
> +static bool cpus_map_matched(struct evsel *a, struct evsel *b)
> +{
> + if (!a->core.cpus && !b->core.cpus)
> + return true;
> +
> + if (!a->core.cpus || !b->core.cpus)
> + return false;
> +
> + if (a->core.cpus->nr != b->core.cpus->nr)
> + return false;
> +
> + for (int i = 0; i < a->core.cpus->nr; i++) {
> + if (a->core.cpus->map[i] != b->core.cpus->map[i])
> + return false;
> + }
> +
> + return true;
> +}
> +
> +static void evlist__check_cpu_maps(struct evlist *evlist)
> +{
> + struct evsel *evsel, *pos, *leader;
> + char buf[1024];
> +
> + evlist__for_each_entry(evlist, evsel) {
> + leader = evsel->leader;
> +
> + /* Check that leader matches cpus with each member. */
> + if (leader == evsel)
> + continue;
> + if (cpus_map_matched(leader, evsel))
> + continue;
> +
> + /* If there's mismatch disable the group and warn user. */
> + WARN_ONCE(1, "WARNING: grouped events cpus do not match, disabling group:\n");
> + evsel__group_desc(leader, buf, sizeof(buf));
> + pr_warning(" %s\n", buf);
> +
> + if (verbose) {
> + cpu_map__snprint(leader->core.cpus, buf, sizeof(buf));
> + pr_warning(" %s: %s\n", leader->name, buf);
> + cpu_map__snprint(evsel->core.cpus, buf, sizeof(buf));
> + pr_warning(" %s: %s\n", evsel->name, buf);
> + }
> +
> + for_each_group_evsel(pos, leader) {
> + pos->leader = pos;
> + pos->core.nr_members = 0;
> + }
> + evsel->leader->core.nr_members = 0;
> + }
> +}
> +
> static inline void diff_timespec(struct timespec *r, struct timespec *a,
> struct timespec *b)
> {
> @@ -2113,6 +2166,8 @@ int cmd_stat(int argc, const char **argv)
> goto out;
> }
>
> + evlist__check_cpu_maps(evsel_list);
> +
> /*
> * Initialize thread_map with comm names,
> * so we could print it out on output.
> --
> 2.25.4
>

--

- Arnaldo

\
 
 \ /
  Last update: 2020-06-02 15:43    [W:0.088 / U:0.240 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site