perf: add a dummy software event to keep tracking
Adrian Hunter
reviewercover perf: add a dummy software event to keep tracking
Hi
This is an alternative to the 'keep tracking' flag patch which is here:
http://marc.info/?l=linux-kernel&m=137242545521246&w=2
perf tools is updated and a test added to demonstrate the new event.
Adrian Hunter (3):
perf: add a dummy software event to keep tracking perf tools: add support for PERF_COUNT_SW_DUMMY perf tools: add 'keep tracking' test
include/uapi/linux/perf_event.h | 1 + tools/perf/Makefile | 1 + tools/perf/tests/builtin-test.c | 4 ++ tools/perf/tests/keep-tracking.c | 150 +++++++++++++++++++++++++++++++++++++++ tools/perf/tests/tests.h | 1 + tools/perf/util/evlist.c | 42 ++++++++++- tools/perf/util/evlist.h | 5 ++ tools/perf/util/parse-events.c | 4 ++ tools/perf/util/parse-events.l | 1 + tools/perf/util/python.c | 1 + 10 files changed, 208 insertions(+), 2 deletions(-) create mode 100644 tools/perf/tests/keep-tracking.c
Regards
Adrian
Adrian Hunter
reviewerPATCH 1/3 perf: add a dummy software event to keep tracking
When an event is disabled the "tracking" events selected by the 'mmap', 'comm' and 'task' bits of struct perf_event_attr, are also disabled. However, the information those events provide is necessary to resolve symbols for when the main event is re-enabled.
The "tracking" events can be kept enabled by putting them on another event, but that requires an event that otherwise does nothing. A new software event PERF_COUNT_SW_DUMMY is added for that purpose.
--- include/uapi/linux/perf_event.h | 1 + 1 file changed, 1 insertion(+)
include/uapi/linux/perf_event.h +1 -0
@@ -109,6 +109,7 @@ enum perf_sw_ids {PERF_COUNT_SW_PAGE_FAULTS_MAJ =6 ,PERF_COUNT_SW_ALIGNMENT_FAULTS =7 ,PERF_COUNT_SW_EMULATION_FAULTS =8 , +PERF_COUNT_SW_DUMMY =9 , PERF_COUNT_SW_MAX, /* non-ABI */ };
Signature · 1 line
1.7.11.7
Adrian Hunter
reviewerQuoted 6 lines · Adrian Hunter
When an event is disabled the "tracking" events selected by the 'mmap', 'comm' and 'task' bits of struct
perf_event_attr, are also disabled. However, the information those events provide is necessary to resolve symbols for when the main event is re-enabled.
Quoted 6 lines
The "tracking" events can be kept enabled by putting them on another event, but that requires an event that otherwise does nothing. A new software event
PERF_COUNT_SW_DUMMYis added for that purpose.Signed-off-by Adrian Hunter <[email protected]>
Any comments?
Quoted 17 lines
--- include/uapi/linux/perf_event.h | 1 + 1 file changed, 1 insertion(+)include/uapi/linux/perf_event.h +1
-0@@ -109,6 +109,7 @@ enum perf_sw_ids {PERF_COUNT_SW_PAGE_FAULTS_MAJ =6 ,PERF_COUNT_SW_ALIGNMENT_FAULTS =7 ,PERF_COUNT_SW_EMULATION_FAULTS =8 , +PERF_COUNT_SW_DUMMY =9 , PERF_COUNT_SW_MAX, /* non-ABI */ };
Em Mon, Aug 26, 2013 at 04:15:34PM +0300, Adrian Hunter escreveu:
Quoted 7 lines · Adrian Hunter
Quoted 6 lines
When an event is disabled the "tracking" events selected by the 'mmap', 'comm' and 'task' bits of struct
perf_event_attr, are also disabled. However, the information those events provide is necessary to resolve symbols for when the main event is re-enabled.
Quoted 8 lines
Quoted 6 lines
The "tracking" events can be kept enabled by putting them on another event, but that requires an event that otherwise does nothing. A new software event
PERF_COUNT_SW_DUMMYis added for that purpose.Signed-off-by Adrian Hunter <[email protected]>Any comments?
Ok with me, Peter?
- Arnaldo
Quoted 17 lines
Quoted 17 lines
--- include/uapi/linux/perf_event.h | 1 + 1 file changed, 1 insertion(+)include/uapi/linux/perf_event.h +1
-0@@ -109,6 +109,7 @@ enum perf_sw_ids {PERF_COUNT_SW_PAGE_FAULTS_MAJ =6 ,PERF_COUNT_SW_ALIGNMENT_FAULTS =7 ,PERF_COUNT_SW_EMULATION_FAULTS =8 , +PERF_COUNT_SW_DUMMY =9 , PERF_COUNT_SW_MAX, /* non-ABI */ };
Quoted 13 lines · Arnaldo Carvalho de Melo
Em Mon, Aug 26, 2013 at 04:15:34PM +0300, Adrian Hunter escreveu:
Quoted 12 lines · Adrian Hunter
Quoted 11 lines
When an event is disabled the "tracking" events selected by the 'mmap', 'comm' and 'task' bits of struct
perf_event_attr, are also disabled. However, the information those events provide is necessary to resolve symbols for when the main event is re-enabled.The "tracking" events can be kept enabled by putting them on another event, but that requires an event that otherwise does nothing. A new software event
PERF_COUNT_SW_DUMMYis added for that purpose.
Quoted 5 lines
Quoted 3 lines
Quoted 1 line
Signed-off-by Adrian Hunter <[email protected]>Any comments?
Ok with me, Peter?
Yeah I suppose so.. I don't really like it and I've made a start at looking at the problem twice now but never got anywhere, so sure lets go with this.
Adrian Hunter
reviewerPATCH 2/3 perf tools: add support for PERF_COUNT_SW_DUMMY
In reply to Adrian HunterAdd support for the new dummy software eventPERF_COUNT_SW_DUMMY.
--- tools/perf/util/parse-events.c | 4 ++++ tools/perf/util/parse-events.l | 1 + tools/perf/util/python.c | 1 + 3 files changed, 6 insertions(+)
tools/perf/util/parse-events.c +4 -0
@@ -108,6 +108,10 @@ static struct event_symbol event_symbols_sw[PERF_COUNT_SW_MAX] = { .symbol = "emulation-faults", .alias = "", }, + [PERF_COUNT_SW_DUMMY] = { + .symbol = "dummy", + .alias = "", + }, };#define __PERF_EVENT_FIELD (config, name) \
tools/perf/util/parse-events.l +1 -0
@@ -144,6 +144,7 @@ context-switches|cs { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW cpu-migrations|migrations { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_CPU_MIGRATIONS); } alignment-faults { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_ALIGNMENT_FAULTS); } emulation-faults { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_EMULATION_FAULTS); } +dummy { return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_DUMMY); } L1-dcache|l1-d|l1d|L1-data | L1-icache|l1-i|l1i|L1-instruction |
tools/perf/util/python.c +1 -0
@@ -967,6 +967,7 @@ static struct { { "COUNT_SW_PAGE_FAULTS_MAJ", PERF_COUNT_SW_PAGE_FAULTS_MAJ }, { "COUNT_SW_ALIGNMENT_FAULTS", PERF_COUNT_SW_ALIGNMENT_FAULTS }, { "COUNT_SW_EMULATION_FAULTS", PERF_COUNT_SW_EMULATION_FAULTS }, + { "COUNT_SW_DUMMY", PERF_COUNT_SW_DUMMY }, { "SAMPLE_IP", PERF_SAMPLE_IP }, { "SAMPLE_TID", PERF_SAMPLE_TID },
Signature · 1 line
1.7.11.7
Adrian Hunter
reviewerPATCH 3/3 perf tools: add 'keep tracking' test
In reply to Adrian HunterAdd a test for the newly added PERF_COUNT_SW_DUMMY event. The test checks that tracking events continue when an event is disabled but a dummy software event is not disabled.
--- tools/perf/Makefile | 1 + tools/perf/tests/builtin-test.c | 4 ++ tools/perf/tests/keep-tracking.c | 150 +++++++++++++++++++++++++++++++++++++++ tools/perf/tests/tests.h | 1 + tools/perf/util/evlist.c | 42 ++++++++++- tools/perf/util/evlist.h | 5 ++ 6 files changed, 201 insertions(+), 2 deletions(-) create mode 100644 tools/perf/tests/keep-tracking.c
tools/perf/Makefile +1 -0
@@ -392,6 +392,7 @@ LIB_OBJS += $(OUTPUT)tests/sw-clock.o ifeq ($(ARCH),x86) LIB_OBJS += $(OUTPUT)tests/perf-time-to-tsc.o endif +LIB_OBJS += $(OUTPUT)tests/keep-tracking.o BUILTIN_OBJS += $(OUTPUT)builtin-annotate.o BUILTIN_OBJS += $(OUTPUT)builtin-bench.o
tools/perf/tests/builtin-test.c +4 -0
@@ -100,6 +100,10 @@ static struct test { },#endif { + .desc = "Test using a dummy software event to keep tracking", + .func = test__keep_tracking, + }, + { .func = NULL, }, };
tools/perf/tests/keep-tracking.c new file +150 -0
150 lines not on this page · raw message
tools/perf/tests/tests.h +1 -0
@@ -36,5 +36,6 @@ int test__bp_signal_overflow(void);int test__task_exit (void );int test__sw_clock_freq (void );int test__perf_time_to_tsc (void ); +int test__keep_tracking (void );#endif /* TESTS_H */
tools/perf/util/evlist.c +40 -2
60 lines not on this page · raw message
tools/perf/util/evlist.h +5 -0
@@ -101,6 +101,11 @@ void perf_evlist__munmap(struct perf_evlist *evlist);void perf_evlist__disable (struct perf_evlist *evlist);void perf_evlist__enable (struct perf_evlist *evlist); +int perf_evlist__disable_event (struct perf_evlist *evlist, +struct perf_evsel *evsel); +int perf_evlist__enable_event (struct perf_evlist *evlist, +struct perf_evsel *evsel); +void perf_evlist__set_selected (struct perf_evlist *evlist,struct perf_evsel *evsel);
Signature · 1 line
1.7.11.7
Namhyung Kim
maintainerRe: [PATCH 3/3] perf tools: add 'keep tracking' test
Quoted 4 lines · Adrian Hunter
Add a test for the newly added
PERF_COUNT_SW_DUMMYevent. The test checks that tracking events continue when an event is disabled but a dummy software event is not disabled.
[SNIP]
Quoted 79 lines
tools/perf/tests/keep-tracking.c new file +79
-0+int test__keep_tracking (void ) +{ +struct perf_record_opts opts = { + .mmap_pages =UINT_MAX , + .user_freq =UINT_MAX , + .user_interval =ULLONG_MAX , + .freq =4000 , + .target = { + .uses_mmap = true, + }, + }; +struct thread_map *threads =NULL ; +struct cpu_map *cpus =NULL ; +struct perf_evlist *evlist =NULL ; +struct perf_evsel *evsel =NULL ; +int found, err =-1 ; +const char *comm; + + threads =thread_map__new (-1 ,getpid (),UINT_MAX ); +CHECK_NOT_NULL__ (threads); + + cpus =cpu_map__new (NULL ); +CHECK_NOT_NULL__ (cpus); + + evlist =perf_evlist__new (); +CHECK_NOT_NULL__ (evlist); + +perf_evlist__set_maps (evlist, cpus, threads); + +CHECK__ (parse_events (evlist,"dummy:u" )); +CHECK__ (parse_events (evlist,"cycles:u" )); + +perf_evlist__config (evlist, &opts); + + evsel =perf_evlist__first (evlist); + + evsel->attr.comm =1 ; + evsel->attr.disabled =1 ; + evsel->attr.enable_on_exec =0 ; + +CHECK__ (perf_evlist__open (evlist)); + +CHECK__ (perf_evlist__mmap (evlist,UINT_MAX , false)); + +/* +* First, test that a 'comm' event can be found when the event is +* enabled. +*/ + +perf_evlist__enable (evlist); + + comm ="Test COMM 1" ; +CHECK__ (prctl (PR_SET_NAME , (unsigned long )comm,0 ,0 ,0 )); + +perf_evlist__disable (evlist); + + found =find_comm (evlist, comm); +if (found !=1 ) { +pr_debug ("First time, failed to find tracking event.\n" ); + goto out_err; + } + +/* +* Secondly, test that a 'comm' event can be found when the event is +* disabled with the dummy event still enabled. +*/ + +perf_evlist__enable (evlist); + + evsel =perf_evlist__last (evlist); + +CHECK__ (perf_evlist__disable_event (evlist, evsel)); + + comm ="Test COMM 2" ; +CHECK__ (prctl (PR_SET_NAME , (unsigned long )comm,0 ,0 ,0 )); + +perf_evlist__disable (evlist); + + found =find_comm (evlist, comm);
tools/perf/tests/keep-tracking.c line 139
Quoted 10 lines
tools/perf/tests/keep-tracking.c new file +10
-0+if (found !=1 ) { +pr_debug ("Seconf time, failed to find tracking event.\n" ); + goto out_err; + } + + err =0 ; + +out_err: +if (evlist) { +perf_evlist__disable (evlist);
It seems the both of goto out_err already called perf_evlist__disable().
Thanks,
Namhyung
Quoted 11 lines
tools/perf/tests/keep-tracking.c new file +11
-0+perf_evlist__munmap (evlist); +perf_evlist__close (evlist); +perf_evlist__delete (evlist); + } +if (cpus) +cpu_map__delete (cpus); +if (threads) +thread_map__delete (threads); + +return err; +}
Adrian Hunter
reviewerRe: [PATCH 3/3] perf tools: add 'keep tracking' test
Quoted 91 lines · Namhyung Kim
Quoted 4 lines · Adrian Hunter
Add a test for the newly added
PERF_COUNT_SW_DUMMYevent. The test checks that tracking events continue when an event is disabled but a dummy software event is not disabled.[SNIP]
Quoted 83 lines
tools/perf/tests/keep-tracking.c new file +83
-0+int test__keep_tracking (void ) +{ +struct perf_record_opts opts = { + .mmap_pages =UINT_MAX , + .user_freq =UINT_MAX , + .user_interval =ULLONG_MAX , + .freq =4000 , + .target = { + .uses_mmap = true, + }, + }; +struct thread_map *threads =NULL ; +struct cpu_map *cpus =NULL ; +struct perf_evlist *evlist =NULL ; +struct perf_evsel *evsel =NULL ; +int found, err =-1 ; +const char *comm; + + threads =thread_map__new (-1 ,getpid (),UINT_MAX ); +CHECK_NOT_NULL__ (threads); + + cpus =cpu_map__new (NULL ); +CHECK_NOT_NULL__ (cpus); + + evlist =perf_evlist__new (); +CHECK_NOT_NULL__ (evlist); + +perf_evlist__set_maps (evlist, cpus, threads); + +CHECK__ (parse_events (evlist,"dummy:u" )); +CHECK__ (parse_events (evlist,"cycles:u" )); + +perf_evlist__config (evlist, &opts); + + evsel =perf_evlist__first (evlist); + + evsel->attr.comm =1 ; + evsel->attr.disabled =1 ; + evsel->attr.enable_on_exec =0 ; + +CHECK__ (perf_evlist__open (evlist)); + +CHECK__ (perf_evlist__mmap (evlist,UINT_MAX , false)); + +/* +* First, test that a 'comm' event can be found when the event is +* enabled. +*/ + +perf_evlist__enable (evlist); + + comm ="Test COMM 1" ; +CHECK__ (prctl (PR_SET_NAME , (unsigned long )comm,0 ,0 ,0 )); + +perf_evlist__disable (evlist); + + found =find_comm (evlist, comm); +if (found !=1 ) { +pr_debug ("First time, failed to find tracking event.\n" ); + goto out_err; + } + +/* +* Secondly, test that a 'comm' event can be found when the event is +* disabled with the dummy event still enabled. +*/ + +perf_evlist__enable (evlist); + + evsel =perf_evlist__last (evlist); + +CHECK__ (perf_evlist__disable_event (evlist, evsel)); + + comm ="Test COMM 2" ; +CHECK__ (prctl (PR_SET_NAME , (unsigned long )comm,0 ,0 ,0 )); + +perf_evlist__disable (evlist); + + found =find_comm (evlist, comm); +if (found !=1 ) { +pr_debug ("Seconf time, failed to find tracking event.\n" ); + goto out_err; + }
Quoted 8 lines
tools/perf/tests/keep-tracking.c line 139
Quoted 6 lines
tools/perf/tests/keep-tracking.c new file +6
-0+ + err =0 ; + +out_err: +if (evlist) { +perf_evlist__disable (evlist);It seems the both of goto
out_erralready calledperf_evlist__disable().
There are more "goto out_err"s in the CHECK__ macro
Quoted 18 lines
Thanks,
NamhyungQuoted 11 lines
tools/perf/tests/keep-tracking.c new file +11
-0+perf_evlist__munmap (evlist); +perf_evlist__close (evlist); +perf_evlist__delete (evlist); + } +if (cpus) +cpu_map__delete (cpus); +if (threads) +thread_map__delete (threads); + +return err; +}
Namhyung Kim
maintainerHi Adrian,
Quoted 9 lines · Adrian Hunter
Hi
This is an alternative to the 'keep tracking' flag patch which is here:
http://marc.info/?l=linux-kernel&m=137242545521246&w=2perf tools is updated and a test added to demonstrate the new event.
It looks like a perfect fit for tracking fork/mmap events on perf-ftrace. :)
Thanks,
Namhyung