| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <1343120493-23059-1-git-send-email-namhyung@kernel.org>
Date: Tue, 24 Jul 2012 18:01:21 +0900
From: Namhyung Kim <namhyung@...nel.org>
To: Arnaldo Carvalho de Melo <acme@...stprotocols.net>
Cc: Peter Zijlstra <a.p.zijlstra@...llo.nl>,
Paul Mackerras <paulus@...ba.org>,
Ingo Molnar <mingo@...nel.org>,
LKML <linux-kernel@...r.kernel.org>,
Stephane Eranian <eranian@...gle.com>,
Jiri Olsa <jolsa@...hat.com>,
Ulrich Drepper <drepper@...il.com>,
Andi Kleen <andi@...stfloor.org>
Subject: [RFC/PATCHSET 00/12] perf report: Add support to event group viewing (v1)
Hi all,
This is a patchset to support event grouping on perf report.
It depends on other patches like refactoring hist print [1],
processing file header feature [2] and (obviously) Jiri's event group
management [3]. All of this need to be reviewed though. ;)
The basic idea is move group member's hist entries to a leader, and
sort/collapse them on the leader's tree. The leader will have all of
group members' stat in it. The output is sorted by the leader's period
and in turn first child and so on.
To use it, 'perf record' should group events when recording. And then
perf report parses the saved command line and reconstruct the group
relation. Currently only the '-e { event1,event2 }' syntax is supported
(i.e. --group option is *NOT* supported) to make things easy. But it'd
not be that hard to support --group also.
But I think re-using event parsing routine (at least, in its current
form) has some problems especially if perf report will not run on the
same machine that runs perf record. I cannot find a better way than
extending/changing the perf file format to let perf record know about
the group relationship. Any thought?
Here is an example:
$ ./perf record -e cycles:u -e '{cache-references,cache-misses}:u' noploop 1
Without --group:
$ ./perf report --stdio
...
# Samples: 4K of event 'cycles:u'
# Event count (approx.): 3692183905
#
# Overhead Command Shared Object Symbol
# ........ ....... .................. .......................
#
99.98% noploop noploop [.] main
0.01% noploop ld-2.15.so [.] _dl_relocate_object
0.00% noploop [kernel.kallsyms] [k] page_fault
0.00% noploop libc-2.15.so [.] __execvpe
0.00% noploop libpthread-2.15.so [.] __read_nocancel
0.00% noploop ld-2.15.so [.] _start
# Samples: 26 of event 'cache-references:u'
# Event count (approx.): 4229
#
# Overhead Command Shared Object Symbol
# ........ ....... .................. ..................................
#
55.85% noploop ld-2.15.so [.] do_lookup_x
19.18% noploop ld-2.15.so [.] strlen
14.09% noploop libc-2.15.so [.] getenv
3.17% noploop ld-2.15.so [.] _dl_fini
2.65% noploop noploop [.] main
1.44% noploop perf [.] perf_evlist__prepare_workload
1.44% noploop [kernel.kallsyms] [k] page_fault
0.99% noploop [kernel.kallsyms] [k] apic_timer_interrupt
0.33% noploop noploop [.] handler
0.33% noploop libc-2.15.so [.] __run_exit_handlers
0.33% noploop [kernel.kallsyms] [k] call_function_single_interrupt
0.12% noploop ld-2.15.so [.] _dl_start
0.05% noploop libpthread-2.15.so [.] __read_nocancel
0.02% noploop ld-2.15.so [.] _start
# Samples: 23 of event 'cache-misses:u'
# Event count (approx.): 4312
#
# Overhead Command Shared Object Symbol
# ........ ....... .................. ....................................
#
46.94% noploop ld-2.15.so [.] do_lookup_x
21.78% noploop libc-2.15.so [.] getenv
18.65% noploop ld-2.15.so [.] calloc
6.05% noploop ld-2.15.so [.] rtld_lock_default_lock_recursive
2.92% noploop noploop [.] main
1.41% noploop [kernel.kallsyms] [k] page_fault
1.32% noploop libc-2.15.so [.] execvp
0.32% noploop libc-2.15.so [.] __run_exit_handlers
0.32% noploop ld-2.15.so [.] _dl_fini
0.12% noploop perf [.] perf_evlist__prepare_workload
0.12% noploop ld-2.15.so [.] _dl_start
0.02% noploop libpthread-2.15.so [.] __read_nocancel
0.02% noploop ld-2.15.so [.] _start
With --group:
$ ./perf report --group --stdio
...
# Samples: 4K of event 'cycles:u'
# Event count (approx.): 7384367810
#
# Overhead Command Shared Object Symbol
# ........ ....... .................. .......................
#
99.98% noploop noploop [.] main
0.01% noploop ld-2.15.so [.] _dl_relocate_object
0.00% noploop [kernel.kallsyms] [k] page_fault
0.00% noploop libc-2.15.so [.] __execvpe
0.00% noploop libpthread-2.15.so [.] __read_nocancel
0.00% noploop ld-2.15.so [.] _start
# Samples: 49 of event 'anon_group { cache-references:u, cache-misses:u }'
# Event count (approx.): 12770
#
# Overhead Command Shared Object Symbol
# ................ ....... .................. ....................................
#
55.85% 46.94% noploop ld-2.15.so [.] do_lookup_x
19.18% 0.00% noploop ld-2.15.so [.] strlen
14.09% 21.78% noploop libc-2.15.so [.] getenv
3.17% 0.32% noploop ld-2.15.so [.] _dl_fini
2.65% 2.92% noploop noploop [.] main
1.44% 1.41% noploop [kernel.kallsyms] [k] page_fault
1.44% 0.12% noploop perf [.] perf_evlist__prepare_workload
0.99% 0.00% noploop [kernel.kallsyms] [k] apic_timer_interrupt
0.33% 0.32% noploop libc-2.15.so [.] __run_exit_handlers
0.33% 0.00% noploop noploop [.] handler
0.33% 0.00% noploop [kernel.kallsyms] [k] call_function_single_interrupt
0.12% 0.12% noploop ld-2.15.so [.] _dl_start
0.05% 0.02% noploop libpthread-2.15.so [.] __read_nocancel
0.02% 0.02% noploop ld-2.15.so [.] _start
0.00% 18.65% noploop ld-2.15.so [.] calloc
0.00% 6.05% noploop ld-2.15.so [.] rtld_lock_default_lock_recursive
0.00% 1.32% noploop libc-2.15.so [.] execvp
(Hmm.. looks like there's a bug in group event counting in the header
area. But I believe the period value itself is intact.)
You can access it via my tree as well (if you want to test it).
git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git perf/report-group-v1
Any comments are welcome, thanks.
Namhyung
[1] https://lkml.org/lkml/2012/7/19/263
[2] https://lkml.org/lkml/2012/7/21/75
[3] https://lkml.org/lkml/2012/7/19/419
Namhyung Kim (12):
perf tools: Add a couple of helper routines to handle groups
perf hist: Convert to struct he_stat
perf hist: Collapse group hist_entries to a leader
perf hist: Maintain total periods of group members in the leader
perf report: Make another loop for output resorting
perf header: Reconstruct group relationship by parsing cmdline
perf ui/hist: Add support to group viewing
perf ui/browser: Add support to group viewing
perf ui/gtk: Add support to group viewing
perf report: Show leader events only when event group is enabled
perf report: Show group description when event group is enabled
perf report: Add --group option
tools/perf/builtin-report.c | 29 +++++++
tools/perf/ui/browsers/hists.c | 105 +++++++++++++++++++++---
tools/perf/ui/gtk/browser.c | 69 +++++++++++++---
tools/perf/ui/hist.c | 158 +++++++++++++++++++++++++++++------
tools/perf/util/evsel.c | 25 ++++++
tools/perf/util/evsel.h | 21 +++++
tools/perf/util/header.c | 110 +++++++++++++++++++++++++
tools/perf/util/hist.c | 177 ++++++++++++++++++++++++++++++++++------
tools/perf/util/hist.h | 1 +
tools/perf/util/parse-events.c | 6 +-
tools/perf/util/sort.h | 17 ++--
tools/perf/util/symbol.c | 4 +
tools/perf/util/symbol.h | 3 +-
13 files changed, 641 insertions(+), 84 deletions(-)
--
1.7.10.4
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists