Fixes for setting event freq/periods
Subsystems (4)
Ian Rogers
reviewerv2 cover Fixes for setting event freq/periods
Some fixes that address issues for regular and pfm4 events with 2 additional perf_event_attr tests. Various authors, David Sharp isn't currently at Google.
v2. corrects the commit message following Athira Rajeev's suggestion.
David Sharp (1):
perf record: Set PERF_RECORD_PERIOD if attr->freq is set.
Ian Rogers (3):
perf test: Ensure sample_period is set libpfm4 events perf record: Don't clear event's period if set by a term perf test: Leader sampling shouldn't clear sample period
Stephane Eranian (1):
perf record: Prevent override of attr->sample_period for libpfm4
events
tools/perf/tests/attr/README | 2 ++ tools/perf/tests/attr/test-record-group2 | 29 ++++++++++++++++++++ tools/perf/tests/attr/test-record-pfm-period | 9 ++++++ tools/perf/util/evsel.c | 10 +++++-- tools/perf/util/record.c | 28 +++++++++++++------ 5 files changed, 67 insertions(+), 11 deletions(-) create mode 100644 tools/perf/tests/attr/test-record-group2 create mode 100644 tools/perf/tests/attr/test-record-pfm-period
Signature · 1 line
2.28.0.163.g6104cc2f0b6-goog
Ian Rogers
reviewerPATCH v2 1/5 perf record: Set PERF_RECORD_PERIOD if attr->freq is set.
From: David Sharp <[email protected]>
evsel__config() would only set PERF_RECORD_PERIOD if it set attr->freq from perf record options. When it is set by libpfm events, it would not get set. This changes evsel__config to see if attr->freq is set outside of whether or not it changes attr->freq itself.
--- tools/perf/util/evsel.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
tools/perf/util/evsel.c +6 -1
@@ -979,13 +979,18 @@ void evsel__config(struct evsel *evsel, struct record_opts *opts,if (!attr->sample_period || (opts->user_freq !=UINT_MAX || opts->user_interval !=ULLONG_MAX )) {if (opts->freq) { -evsel__set_sample_bit (evsel,PERIOD ); attr->freq =1 ; attr->sample_freq = opts->freq; }else { attr->sample_period = opts->default_interval; } } +/* +* If attr->freq was set (here or earlier), ask for period +* to be sampled. +*/ +if (attr->freq) +evsel__set_sample_bit (evsel,PERIOD );if (opts->no_samples) attr->sample_freq =0 ;
Signature · 1 line
2.28.0.163.g6104cc2f0b6-goog
Re: [PATCH v2 1/5] perf record: Set PERF_RECORD_PERIOD if attr->freq is set.
Quoted 9 lines · Ian Rogers
From: David Sharp <[email protected]>
evsel__config()would only setPERF_RECORD_PERIODif it setattr->freqfrom perf record options. When it is set by libpfm events, it would not get set. This changesevsel__configto see ifattr->freqis set outside of whether or not it changesattr->freqitself.Signed-off-by David Sharp <[email protected]>Signed-off-by Ian Rogers <[email protected]>
thanks,
jirka
Quoted 31 lines
--- tools/perf/util/evsel.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)tools/perf/util/evsel.c +6
-1@@ -979,13 +979,18 @@ void evsel__config(struct evsel *evsel, struct record_opts *opts,if (!attr->sample_period || (opts->user_freq !=UINT_MAX || opts->user_interval !=ULLONG_MAX )) {if (opts->freq) { -evsel__set_sample_bit (evsel,PERIOD ); attr->freq =1 ; attr->sample_freq = opts->freq; }else { attr->sample_period = opts->default_interval; } } +/* +* If attr->freq was set (here or earlier), ask for period +* to be sampled. +*/ +if (attr->freq) +evsel__set_sample_bit (evsel,PERIOD );if (opts->no_samples) attr->sample_freq =0 ;Signature · 1 line
2.28.0.163.g6104cc2f0b6-goog
Arnaldo Carvalho de Melo
maintainerRe: [PATCH v2 1/5] perf record: Set PERF_RECORD_PERIOD if attr->freq is set.
Em Tue, Jul 28, 2020 at 05:43:47PM +0200, Jiri Olsa escreveu:
Quoted 8 lines · Ian Rogers
Quoted 7 lines
From: David Sharp <[email protected]>
evsel__config()would only setPERF_RECORD_PERIODif it setattr->freqfrom perf record options. When it is set by libpfm events, it would not get set. This changesevsel__configto see ifattr->freqis set outside of whether or not it changesattr->freqitself.
Quoted 4 lines
Quoted 2 lines
Signed-off-by David Sharp <[email protected]>Signed-off-by Ian Rogers <[email protected]>Acked-by Jiri Olsa <[email protected]>
So, somebody else complained that its not PERF_RECORD_PERIOD (there is no such thing) that is being set, its PERF_SAMPLE_PERIOD.
Since you acked it I merged it now, with that correction,
- Arnaldo
Quoted 35 lines
thanks,
jirkaQuoted 31 lines
--- tools/perf/util/evsel.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)tools/perf/util/evsel.c +6
-1@@ -979,13 +979,18 @@ void evsel__config(struct evsel *evsel, struct record_opts *opts,if (!attr->sample_period || (opts->user_freq !=UINT_MAX || opts->user_interval !=ULLONG_MAX )) {if (opts->freq) { -evsel__set_sample_bit (evsel,PERIOD ); attr->freq =1 ; attr->sample_freq = opts->freq; }else { attr->sample_period = opts->default_interval; } } +/* +* If attr->freq was set (here or earlier), ask for period +* to be sampled. +*/ +if (attr->freq) +evsel__set_sample_bit (evsel,PERIOD );if (opts->no_samples) attr->sample_freq =0 ;Signature · 1 line
2.28.0.163.g6104cc2f0b6-goog
Signature · 1 line
- Arnaldo
Re: [PATCH v2 1/5] perf record: Set PERF_RECORD_PERIOD if attr->freq is set.
Quoted 13 lines · Arnaldo Carvalho de Melo
Em Tue, Jul 28, 2020 at 05:43:47PM +0200, Jiri Olsa escreveu:
Quoted 8 lines · Ian Rogers
Quoted 7 lines
From: David Sharp <[email protected]>
evsel__config()would only setPERF_RECORD_PERIODif it setattr->freqfrom perf record options. When it is set by libpfm events, it would not get set. This changesevsel__configto see ifattr->freqis set outside of whether or not it changesattr->freqitself.
Quoted 7 lines
Quoted 4 lines
Quoted 2 lines
Signed-off-by David Sharp <[email protected]>Signed-off-by Ian Rogers <[email protected]>Acked-by Jiri Olsa <[email protected]>So, somebody else complained that its not
PERF_RECORD_PERIOD(there is no such thing) that is being set, itsPERF_SAMPLE_PERIOD.
Hi Arnaldo
Thanks for adding in that correction.
Athira
Quoted 44 lines
Since you acked it I merged it now, with that correction,
- Arnaldo
Quoted 35 lines
thanks,
jirkaQuoted 31 lines
---
tools/perf/util/evsel.c| 7 ++++++-1 file changed, 6 insertions(+), 1 deletion(-)tools/perf/util/evsel.c +6
-1@@ -979,13 +979,18 @@ void evsel__config(struct evsel *evsel, struct record_opts *opts,if (!attr->sample_period || (opts->user_freq !=UINT_MAX || opts->user_interval !=ULLONG_MAX )) {if (opts->freq) { -evsel__set_sample_bit (evsel,PERIOD ); attr->freq =1 ; attr->sample_freq = opts->freq; }else { attr->sample_period = opts->default_interval; } } +/* +* If attr->freq was set (here or earlier), ask for period +* to be sampled. +*/ +if (attr->freq) +evsel__set_sample_bit (evsel,PERIOD );if (opts->no_samples) attr->sample_freq =0 ;Signature · 1 line
2.28.0.163.g6104cc2f0b6-googSignature · 1 line
- Arnaldo
Arnaldo Carvalho de Melo
maintainerRe: [PATCH v2 1/5] perf record: Set PERF_RECORD_PERIOD if attr->freq is set.
In reply to Ian RogersEm Tue, Jul 28, 2020 at 01:57:30AM -0700, Ian Rogers escreveu:
Quoted 3 lines
From: David Sharp <[email protected]>
evsel__config()would only setPERF_RECORD_PERIODif it setattr->freq
There is no such thing as 'PERF_RECORD_PERIOD', its PERF_SAMPLE_PERIOD, also...
Quoted 21 lines
from perf record options. When it is set by libpfm events, it would not get set. This changes
evsel__configto see ifattr->freqis set outside of whether or not it changesattr->freqitself.Signed-off-by David Sharp <[email protected]>Signed-off-by Ian Rogers <[email protected]>--- tools/perf/util/evsel.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)tools/perf/util/evsel.c +0
-1@@ -979,13 +979,18 @@ void evsel__config(struct evsel *evsel, struct record_opts *opts,if (!attr->sample_period || (opts->user_freq !=UINT_MAX || opts->user_interval !=ULLONG_MAX )) {if (opts->freq) { -evsel__set_sample_bit (evsel,PERIOD ); attr->freq =1 ; attr->sample_freq = opts->freq;
tools/perf/util/evsel.c void evsel__config(struct evsel *evsel, struct record_opts *opts, · line 993
Quoted 10 lines
tools/perf/util/evsel.c +6
-0}else { attr->sample_period = opts->default_interval; } } +/* +* If attr->freq was set (here or earlier), ask for period +* to be sampled. +*/ +if (attr->freq) +evsel__set_sample_bit (evsel,PERIOD );
Why can't the libpfm code set opts?
With this patch we will end up calling evsel__set_sample_bit(evsel, PERIOD) twice, which isn't a problem but looks strange.
- Arnaldo
Quoted 6 lines
tools/perf/util/evsel.c +0
-0if (opts->no_samples) attr->sample_freq =0 ;Signature · 1 line
2.28.0.163.g6104cc2f0b6-goog
Signature · 1 line
- Arnaldo
Ian Rogers
reviewerRe: [PATCH v2 1/5] perf record: Set PERF_RECORD_PERIOD if attr->freq is set.
Quoted 43 lines · Arnaldo Carvalho de Melo
Em Tue, Jul 28, 2020 at 01:57:30AM -0700, Ian Rogers escreveu:
Quoted 3 lines
From: David Sharp <[email protected]>
evsel__config()would only setPERF_RECORD_PERIODif it setattr->freqThere is no such thing as '
PERF_RECORD_PERIOD', itsPERF_SAMPLE_PERIOD, also...Quoted 21 lines
from perf record options. When it is set by libpfm events, it would not get set. This changes
evsel__configto see ifattr->freqis set outside of whether or not it changesattr->freqitself.Signed-off-by David Sharp <[email protected]>Signed-off-by Ian Rogers <[email protected]>--- tools/perf/util/evsel.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)tools/perf/util/evsel.c +0
-1@@ -979,13 +979,18 @@ void evsel__config(struct evsel *evsel, struct record_opts *opts,if (!attr->sample_period || (opts->user_freq !=UINT_MAX || opts->user_interval !=ULLONG_MAX )) {if (opts->freq) { -evsel__set_sample_bit (evsel,PERIOD ); attr->freq =1 ; attr->sample_freq = opts->freq;tools/perf/util/evsel.c void evsel__config(struct evsel *evsel, struct record_opts *opts, · line 993
Quoted 10 lines
tools/perf/util/evsel.c +6
-0}else { attr->sample_period = opts->default_interval; } } +/* +* If attr->freq was set (here or earlier), ask for period +* to be sampled. +*/ +if (attr->freq) +evsel__set_sample_bit (evsel,PERIOD );Why can't the libpfm code set opts?
Quoted 2 lines
With this patch we will end up calling
evsel__set_sample_bit(evsel, PERIOD)twice, which isn't a problem but looks strange.
Thanks Arnaldo! The case I was looking at was something like:
perf record --pfm-events cycles:freq=1000
For regular events this would be:
perf record -e cycles/freq=1000/
With libpfm4 events the perf_event_attr is set up (a public API in linux/perf_event.h) and then parse_events__add_event is used (an internal API) to make the evsel and this added to the evlist (parse_libpfm_events_option). This is similar to the parse_events function except rather than set up a perf_event_attr the regular parsing sets up config terms that are then applied to evsel and attr later in evsel__config, via evsel__apply_config_terms.
I think we can update this change so that in pfm.c after parse_events__add_event we do:
if (attr.freq)
evsel__set_sample_bit(evsel, PERIOD);
This code could also be part of parse_events__add_event. I think the intent in placing this code here was that it is close to the similar evsel__apply_config_terms and setting of sample bits in the evsel. The logic here is already dependent on reading the attr->sample_period.
I'm not sure I follow the double setting case - I think that is only possible with a config term or with period_set (-P).
Thanks,
Ian
Quoted 12 lines
- Arnaldo
Quoted 6 lines
tools/perf/util/evsel.c +0
-1if (opts->no_samples) attr->sample_freq =0 ; --2.28.0.163.g6104cc2f0b6-goog
--
- Arnaldo
Ian Rogers
reviewerRe: [PATCH v2 1/5] perf record: Set PERF_RECORD_PERIOD if attr->freq is set.
Quoted 73 lines · Ian Rogers
Quoted 43 lines · Arnaldo Carvalho de Melo
Em Tue, Jul 28, 2020 at 01:57:30AM -0700, Ian Rogers escreveu:
Quoted 3 lines
From: David Sharp <[email protected]>
evsel__config()would only setPERF_RECORD_PERIODif it setattr->freqThere is no such thing as '
PERF_RECORD_PERIOD', itsPERF_SAMPLE_PERIOD, also...Quoted 21 lines
from perf record options. When it is set by libpfm events, it would not get set. This changes
evsel__configto see ifattr->freqis set outside of whether or not it changesattr->freqitself.Signed-off-by David Sharp <[email protected]>Signed-off-by Ian Rogers <[email protected]>--- tools/perf/util/evsel.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)tools/perf/util/evsel.c +0
-1@@ -979,13 +979,18 @@ void evsel__config(struct evsel *evsel, struct record_opts *opts,if (!attr->sample_period || (opts->user_freq !=UINT_MAX || opts->user_interval !=ULLONG_MAX )) {if (opts->freq) { -evsel__set_sample_bit (evsel,PERIOD ); attr->freq =1 ; attr->sample_freq = opts->freq;tools/perf/util/evsel.c void evsel__config(struct evsel *evsel, struct record_opts *opts, · line 993
Quoted 10 lines
tools/perf/util/evsel.c +6
-0}else { attr->sample_period = opts->default_interval; } } +/* +* If attr->freq was set (here or earlier), ask for period +* to be sampled. +*/ +if (attr->freq) +evsel__set_sample_bit (evsel,PERIOD );Why can't the libpfm code set opts?
Quoted 2 lines
With this patch we will end up calling
evsel__set_sample_bit(evsel, PERIOD)twice, which isn't a problem but looks strange.Thanks Arnaldo! The case I was looking at was something like:
perf record --pfm-events cycles:freq=1000For regular events this would be:
perf record -e cycles/freq=1000/With libpfm4 events the
perf_event_attris set up (a public API inlinux/perf_event.h) and thenparse_events__add_eventis used (an internal API) to make the evsel and this added to the evlist (parse_libpfm_events_option). This is similar to theparse_eventsfunction except rather than set up aperf_event_attrthe regular parsing sets up config terms that are then applied to evsel and attr later inevsel__config, viaevsel__apply_config_terms.I think we can update this change so that in
pfm.cafterparse_events__add_eventwe do:
if (attr.freq)
evsel__set_sample_bit(evsel, PERIOD);This code could also be part of
parse_events__add_event. I think the intent in placing this code here was that it is close to the similarevsel__apply_config_termsand setting of sample bits in the evsel. The logic here is already dependent on reading theattr->sample_period.
Quoted 5 lines
I'm not sure I follow the double setting case - I think that is only possible with a config term or with
period_set(-P).Thanks,
Ian
Polite ping. Thanks,
Ian
Quoted 13 lines
Quoted 13 lines
- Arnaldo
Quoted 6 lines
tools/perf/util/evsel.c +0
-1if (opts->no_samples) attr->sample_freq =0 ; --2.28.0.163.g6104cc2f0b6-goog
--
- Arnaldo
Ian Rogers
reviewerPATCH v2 2/5 perf record: Prevent override of attr->sample_period for libpfm4 events
In reply to Ian RogersFrom: Stephane Eranian <[email protected]>
Before:
$ perf record -c 10000 --pfm-events=cycles:period=77777
Would yield a cycles event with period=10000, instead of 77777.
This was due to an ordering issue between libpfm4 parsing the event string and perf record initializing the event.
This patch fixes the problem by preventing override for events with attr->sample_period != 0 by the time perf_evsel__config() is invoked. This seems to have been the intent of the author.
--- tools/perf/util/evsel.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
tools/perf/util/evsel.c +1 -2
@@ -976,8 +976,7 @@ void evsel__config(struct evsel *evsel, struct record_opts *opts,* We default some events to have a default interval. But keep * it a weak assumption overridable by the user. */ -if (!attr->sample_period || (opts->user_freq !=UINT_MAX || - opts->user_interval !=ULLONG_MAX )) { +if (!attr->sample_period) {if (opts->freq) { attr->freq =1 ; attr->sample_freq = opts->freq;
Signature · 1 line
2.28.0.163.g6104cc2f0b6-goog
Re: [PATCH v2 2/5] perf record: Prevent override of attr->sample_period for libpfm4 events
Quoted 26 lines · Ian Rogers
From: Stephane Eranian <[email protected]>
Before:
$ perf record -c 10000 --pfm-events=cycles:period=77777Would yield a cycles event with period=10000, instead of 77777.
This was due to an ordering issue between libpfm4 parsing the event string and perf record initializing the event.
This patch fixes the problem by preventing override for events with
attr->sample_period!= 0 by the timeperf_evsel__config()is invoked. This seems to have been the intent of the author.Signed-off-by Stephane Eranian <[email protected]>Reviewed-by Ian Rogers <[email protected]>--- tools/perf/util/evsel.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)tools/perf/util/evsel.c +0
-0
tools/perf/util/evsel.c void evsel__config(struct evsel *evsel, struct record_opts *opts, · line 979
Quoted 6 lines
tools/perf/util/evsel.c +1
-2@@ -976,8 +976,7 @@ void evsel__config(struct evsel *evsel, struct record_opts *opts,* We default some events to have a default interval. But keep * it a weak assumption overridable by the user. */ -if (!attr->sample_period || (opts->user_freq !=UINT_MAX || - opts->user_interval !=ULLONG_MAX )) { +if (!attr->sample_period) {
I was wondering why this wouldn't break record/top but we take care of the via record_opts__config
as long as 'perf test attr' works it looks ok to me
thanks,
jirka
Re: [PATCH v2 2/5] perf record: Prevent override of attr->sample_period for libpfm4 events
Quoted 37 lines · Jiri Olsa
Quoted 26 lines · Ian Rogers
From: Stephane Eranian <[email protected]>
Before:
$ perf record -c 10000 --pfm-events=cycles:period=77777Would yield a cycles event with period=10000, instead of 77777.
This was due to an ordering issue between libpfm4 parsing the event string and perf record initializing the event.
This patch fixes the problem by preventing override for events with
attr->sample_period!= 0 by the timeperf_evsel__config()is invoked. This seems to have been the intent of the author.Signed-off-by Stephane Eranian <[email protected]>Reviewed-by Ian Rogers <[email protected]>--- tools/perf/util/evsel.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)tools/perf/util/evsel.c +0
-0tools/perf/util/evsel.c void evsel__config(struct evsel *evsel, struct record_opts *opts, · line 979
Quoted 6 lines
tools/perf/util/evsel.c +1
-2@@ -976,8 +976,7 @@ void evsel__config(struct evsel *evsel, struct record_opts *opts,* We default some events to have a default interval. But keep * it a weak assumption overridable by the user. */ -if (!attr->sample_period || (opts->user_freq !=UINT_MAX || - opts->user_interval !=ULLONG_MAX )) { +if (!attr->sample_period) {I was wondering why this wouldn't break record/top but we take care of the via
record_opts__config
Quoted 1 line
as long as 'perf test attr' works it looks ok to me
hum ;-)
[jolsa@krava perf]$ sudo ./perf test 17 -v
17: Setup struct perf_event_attr :
...
running './tests/attr/test-record-C0' expected sample_period=4000, got 3000
FAILED './tests/attr/test-record-C0' - match failure
jirka
Quoted 5 lines
Acked-by Jiri Olsa <[email protected]>thanks,
jirka
Ian Rogers
reviewerRe: [PATCH v2 2/5] perf record: Prevent override of attr->sample_period for libpfm4 events
Quoted 41 lines · Jiri Olsa
Quoted 37 lines · Jiri Olsa
Quoted 26 lines · Ian Rogers
From: Stephane Eranian <[email protected]>
Before:
$ perf record -c 10000 --pfm-events=cycles:period=77777Would yield a cycles event with period=10000, instead of 77777.
This was due to an ordering issue between libpfm4 parsing the event string and perf record initializing the event.
This patch fixes the problem by preventing override for events with
attr->sample_period!= 0 by the timeperf_evsel__config()is invoked. This seems to have been the intent of the author.Signed-off-by Stephane Eranian <[email protected]>Reviewed-by Ian Rogers <[email protected]>--- tools/perf/util/evsel.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)tools/perf/util/evsel.c +0
-0tools/perf/util/evsel.c void evsel__config(struct evsel *evsel, struct record_opts *opts, · line 979
Quoted 6 lines
tools/perf/util/evsel.c +1
-2@@ -976,8 +976,7 @@ void evsel__config(struct evsel *evsel, struct record_opts *opts,* We default some events to have a default interval. But keep * it a weak assumption overridable by the user. */ -if (!attr->sample_period || (opts->user_freq !=UINT_MAX || - opts->user_interval !=ULLONG_MAX )) { +if (!attr->sample_period) {I was wondering why this wouldn't break record/top but we take care of the via
record_opts__configQuoted 1 line
as long as 'perf test attr' works it looks ok to me
Quoted 8 lines
hum ;-)
[jolsa@krava perf]$ sudo ./perf test 17 -v
17: Setup structperf_event_attr:
...
running './tests/attr/test-record-C0' expected sample_period=4000, got 3000
FAILED './tests/attr/test-record-C0' - match failure
I'm not able to reproduce this. Do you have a build configuration or something else to look at? The test doesn't seem obviously connected with this patch.
Thanks,
Ian
Quoted 8 lines
jirka
Quoted 5 lines
Acked-by Jiri Olsa <[email protected]>thanks,
jirka
Ian Rogers
reviewerRe: [PATCH v2 2/5] perf record: Prevent override of attr->sample_period for libpfm4 events
Quoted 52 lines · Ian Rogers
Quoted 41 lines · Jiri Olsa
Quoted 37 lines · Jiri Olsa
Quoted 26 lines · Ian Rogers
From: Stephane Eranian <[email protected]>
Before:
$ perf record -c 10000 --pfm-events=cycles:period=77777Would yield a cycles event with period=10000, instead of 77777.
This was due to an ordering issue between libpfm4 parsing the event string and perf record initializing the event.
This patch fixes the problem by preventing override for events with
attr->sample_period!= 0 by the timeperf_evsel__config()is invoked. This seems to have been the intent of the author.Signed-off-by Stephane Eranian <[email protected]>Reviewed-by Ian Rogers <[email protected]>--- tools/perf/util/evsel.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)tools/perf/util/evsel.c +0
-0tools/perf/util/evsel.c void evsel__config(struct evsel *evsel, struct record_opts *opts, · line 979
Quoted 6 lines
tools/perf/util/evsel.c +1
-2@@ -976,8 +976,7 @@ void evsel__config(struct evsel *evsel, struct record_opts *opts,* We default some events to have a default interval. But keep * it a weak assumption overridable by the user. */ -if (!attr->sample_period || (opts->user_freq !=UINT_MAX || - opts->user_interval !=ULLONG_MAX )) { +if (!attr->sample_period) {I was wondering why this wouldn't break record/top but we take care of the via
record_opts__configQuoted 1 line
as long as 'perf test attr' works it looks ok to me
Quoted 8 lines
hum ;-)
[jolsa@krava perf]$ sudo ./perf test 17 -v
17: Setup structperf_event_attr:
...
running './tests/attr/test-record-C0' expected sample_period=4000, got 3000
FAILED './tests/attr/test-record-C0' - match failure
Quoted 6 lines
I'm not able to reproduce this. Do you have a build configuration or something else to look at? The test doesn't seem obviously connected with this patch.
Thanks,
Ian
Jiri, any update? Thanks,
Ian
Quoted 8 lines
Quoted 8 lines
jirka
Quoted 5 lines
Acked-by Jiri Olsa <[email protected]>thanks,
jirka
Re: [PATCH v2 2/5] perf record: Prevent override of attr->sample_period for libpfm4 events
Quoted 53 lines · Ian Rogers
Quoted 52 lines · Ian Rogers
Quoted 49 lines · Jiri Olsa
Quoted 37 lines · Jiri Olsa
Quoted 26 lines · Ian Rogers
From: Stephane Eranian <[email protected]>
Before:
$ perf record -c 10000 --pfm-events=cycles:period=77777Would yield a cycles event with period=10000, instead of 77777.
This was due to an ordering issue between libpfm4 parsing the event string and perf record initializing the event.
This patch fixes the problem by preventing override for events with
attr->sample_period!= 0 by the timeperf_evsel__config()is invoked. This seems to have been the intent of the author.Signed-off-by Stephane Eranian <[email protected]>Reviewed-by Ian Rogers <[email protected]>--- tools/perf/util/evsel.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)tools/perf/util/evsel.c +0
-0tools/perf/util/evsel.c void evsel__config(struct evsel *evsel, struct record_opts *opts, · line 979
Quoted 6 lines
tools/perf/util/evsel.c +1
-2@@ -976,8 +976,7 @@ void evsel__config(struct evsel *evsel, struct record_opts *opts,* We default some events to have a default interval. But keep * it a weak assumption overridable by the user. */ -if (!attr->sample_period || (opts->user_freq !=UINT_MAX || - opts->user_interval !=ULLONG_MAX )) { +if (!attr->sample_period) {I was wondering why this wouldn't break record/top but we take care of the via
record_opts__configQuoted 1 line
as long as 'perf test attr' works it looks ok to me
hum ;-)
[jolsa@krava perf]$ sudo ./perf test 17 -v
17: Setup structperf_event_attr:
...
running './tests/attr/test-record-C0' expected sample_period=4000, got 3000
FAILED './tests/attr/test-record-C0' - match failure
Quoted 8 lines
Quoted 6 lines
I'm not able to reproduce this. Do you have a build configuration or something else to look at? The test doesn't seem obviously connected with this patch.
Thanks,
IanJiri, any update? Thanks,
sorry, I rebased and ran it again and it passes for me now, so it got fixed along the way
jirka
Ian Rogers
reviewerRe: [PATCH v2 2/5] perf record: Prevent override of attr->sample_period for libpfm4 events
Quoted 59 lines · Jiri Olsa
Quoted 57 lines · Ian Rogers
Quoted 56 lines · Ian Rogers
Quoted 41 lines · Jiri Olsa
Quoted 37 lines · Jiri Olsa
Quoted 26 lines · Ian Rogers
From: Stephane Eranian <[email protected]>
Before:
$ perf record -c 10000 --pfm-events=cycles:period=77777Would yield a cycles event with period=10000, instead of 77777.
This was due to an ordering issue between libpfm4 parsing the event string and perf record initializing the event.
This patch fixes the problem by preventing override for events with
attr->sample_period!= 0 by the timeperf_evsel__config()is invoked. This seems to have been the intent of the author.Signed-off-by Stephane Eranian <[email protected]>Reviewed-by Ian Rogers <[email protected]>--- tools/perf/util/evsel.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)tools/perf/util/evsel.c +0
-0tools/perf/util/evsel.c void evsel__config(struct evsel *evsel, struct record_opts *opts, · line 979
Quoted 6 lines
tools/perf/util/evsel.c +1
-2@@ -976,8 +976,7 @@ void evsel__config(struct evsel *evsel, struct record_opts *opts,* We default some events to have a default interval. But keep * it a weak assumption overridable by the user. */ -if (!attr->sample_period || (opts->user_freq !=UINT_MAX || - opts->user_interval !=ULLONG_MAX )) { +if (!attr->sample_period) {I was wondering why this wouldn't break record/top but we take care of the via
record_opts__configQuoted 1 line
as long as 'perf test attr' works it looks ok to me
Quoted 8 lines
hum ;-)
[jolsa@krava perf]$ sudo ./perf test 17 -v
17: Setup structperf_event_attr:
...
running './tests/attr/test-record-C0' expected sample_period=4000, got 3000
FAILED './tests/attr/test-record-C0' - match failureI'm not able to reproduce this. Do you have a build configuration or something else to look at? The test doesn't seem obviously connected with this patch.
Quoted 7 lines
Quoted 4 lines
Quoted 2 lines
Thanks,
IanJiri, any update? Thanks,
sorry, I rebased and ran it again and it passes for me now, so it got fixed along the way
No worries, thanks for the update! It'd be nice to land this and the other libpfm fixes.
Ian
Quoted 2 lines
jirka
Arnaldo Carvalho de Melo
maintainerRe: [PATCH v2 2/5] perf record: Prevent override of attr->sample_period for libpfm4 events
Em Fri, Sep 04, 2020 at 09:22:10AM -0700, Ian Rogers escreveu:
Quoted 17 lines · Jiri Olsa
Quoted 15 lines · Ian Rogers
Quoted 14 lines · Ian Rogers
Quoted 13 lines · Jiri Olsa
Quoted 8 lines · Jiri Olsa
Quoted 1 line · Ian Rogers
[jolsa@krava perf]$ sudo ./perf test 17 -v
17: Setup structperf_event_attr:running './tests/attr/test-record-C0' expected sample_period=4000, got 3000
FAILED './tests/attr/test-record-C0' - match failureI'm not able to reproduce this. Do you have a build configuration or something else to look at? The test doesn't seem obviously connected with this patch.
Quoted 7 lines
Quoted 4 lines
Quoted 1 line
Jiri, any update? Thanks,
sorry, I rebased and ran it again and it passes for me now, so it got fixed along the way
No worries, thanks for the update! It'd be nice to land this and the other libpfm fixes.
I applied it and it generated this regression:
FAILED '/home/acme/libexec/perf-core/tests/attr/test-record-pfm-period' - match failure
I'll look at the other patches that are pending in this regard to see what needs to be squashed so that we don't break bisect.
- Arnaldo
Arnaldo Carvalho de Melo
maintainerRe: [PATCH v2 2/5] perf record: Prevent override of attr->sample_period for libpfm4 events
Em Fri, Sep 04, 2020 at 03:48:03PM -0300, Arnaldo Carvalho de Melo escreveu:
Quoted 26 lines
Em Fri, Sep 04, 2020 at 09:22:10AM -0700, Ian Rogers escreveu:
Quoted 17 lines · Jiri Olsa
Quoted 15 lines · Ian Rogers
Quoted 14 lines · Ian Rogers
Quoted 13 lines · Jiri Olsa
Quoted 8 lines · Jiri Olsa
Quoted 1 line · Ian Rogers
[jolsa@krava perf]$ sudo ./perf test 17 -v
17: Setup structperf_event_attr:running './tests/attr/test-record-C0' expected sample_period=4000, got 3000
FAILED './tests/attr/test-record-C0' - match failureI'm not able to reproduce this. Do you have a build configuration or something else to look at? The test doesn't seem obviously connected with this patch.
Quoted 7 lines
Quoted 4 lines
Quoted 1 line
Jiri, any update? Thanks,
sorry, I rebased and ran it again and it passes for me now, so it got fixed along the way
No worries, thanks for the update! It'd be nice to land this and the other libpfm fixes.
Quoted 6 lines
I applied it and it generated this regression:
FAILED '
/home/acme/libexec/perf-core/tests/attr/test-record-pfm-period' - match failureI'll look at the other patches that are pending in this regard to see what needs to be squashed so that we don't break bisect.
So, more context:
running '/home/acme/libexec/perf-core/tests/attr/test-record-pfm-period' expected exclude_hv=0, got 1
FAILED '/home/acme/libexec/perf-core/tests/attr/test-record-pfm-period' - match failure test child finished with -1
---- end ----
Setup struct perf_event_attr: FAILED!
[root@five ~]#
Ian, can you take a look at this?
- Arnaldo
Arnaldo Carvalho de Melo
maintainerRe: [PATCH v2 2/5] perf record: Prevent override of attr->sample_period for libpfm4 events
Em Fri, Sep 04, 2020 at 03:50:13PM -0300, Arnaldo Carvalho de Melo escreveu:
Quoted 44 lines
Em Fri, Sep 04, 2020 at 03:48:03PM -0300, Arnaldo Carvalho de Melo escreveu:
Quoted 26 lines
Em Fri, Sep 04, 2020 at 09:22:10AM -0700, Ian Rogers escreveu:
Quoted 17 lines · Jiri Olsa
Quoted 15 lines · Ian Rogers
Quoted 14 lines · Ian Rogers
Quoted 13 lines · Jiri Olsa
Quoted 8 lines · Jiri Olsa
Quoted 1 line · Ian Rogers
[jolsa@krava perf]$ sudo ./perf test 17 -v
17: Setup structperf_event_attr:running './tests/attr/test-record-C0' expected sample_period=4000, got 3000
FAILED './tests/attr/test-record-C0' - match failureI'm not able to reproduce this. Do you have a build configuration or something else to look at? The test doesn't seem obviously connected with this patch.
Quoted 7 lines
Quoted 4 lines
Quoted 1 line
Jiri, any update? Thanks,
sorry, I rebased and ran it again and it passes for me now, so it got fixed along the way
No worries, thanks for the update! It'd be nice to land this and the other libpfm fixes.
Quoted 6 lines
I applied it and it generated this regression:
FAILED '
/home/acme/libexec/perf-core/tests/attr/test-record-pfm-period' - match failureI'll look at the other patches that are pending in this regard to see what needs to be squashed so that we don't break bisect.
So, more context:
running '
/home/acme/libexec/perf-core/tests/attr/test-record-pfm-period' expected exclude_hv=0, got 1
FAILED '/home/acme/libexec/perf-core/tests/attr/test-record-pfm-period' - match failure test child finished with -1
---- end ----
Setup structperf_event_attr: FAILED!
[root@five ~]#
Quoted 1 line
Ian, can you take a look at this?
Further tests I've performed:
Committer testing:
Not linking with libpfm:
# ldd ~/bin/perf | grep libpfm
#
Before:
# perf record -c 10000 -e cycles/period=12345/,instructions sleep 0.0001
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.052 MB perf.data (258 samples) ]
# perf evlist -v
cycles/period=12345/: size: 120, { sample_period, sample_freq }: 12345, sample_type: IP|TID|TIME|ID, read_format: ID, disabled: 1, inherit: 1, mmap: 1, comm: 1, enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1, ksymbol: 1, bpf_event: 1
instructions: size: 120, config: 0x1, { sample_period, sample_freq }: 10000, sample_type: IP|TID|TIME|ID, read_format: ID, disabled: 1, inherit: 1, enable_on_exec: 1, sample_id_all: 1, exclude_guest: 1
#
After:
#
# perf record -c 10000 -e cycles/period=12345/,instructions sleep 0.0001
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.053 MB perf.data (284 samples) ]
# perf evlist -v
cycles/period=12345/: size: 120, { sample_period, sample_freq }: 12345, sample_type: IP|TID|TIME|ID, read_format: ID, disabled: 1, inherit: 1, mmap: 1, comm: 1, enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1, ksymbol: 1, bpf_event: 1
instructions: size: 120, config: 0x1, { sample_period, sample_freq }: 10000, sample_type: IP|TID|TIME|ID, read_format: ID, disabled: 1, inherit: 1, enable_on_exec: 1, sample_id_all: 1, exclude_guest: 1
#
Linking with libpfm:
# ldd ~/bin/perf | grep libpfm
libpfm.so.4 => /lib64/libpfm.so.4 (0x00007f54c7d75000)
#
# perf record -c 10000 --pfm-events=cycles:period=77777 sleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.043 MB perf.data (141 samples) ]
# perf evlist -v
cycles:period=77777: size: 120, { sample_period, sample_freq }: 10000, sample_type: IP|TID|TIME, read_format: ID, disabled: 1, inherit: 1, exclude_hv: 1, mmap: 1, comm: 1, enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1, ksymbol: 1, bpf_event: 1
#
After:
# perf record -c 10000 --pfm-events=cycles:period=77777 sleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.039 MB perf.data (19 samples) ]
# perf evlist -v
cycles:period=77777: size: 120, { sample_period, sample_freq }: 77777, sample_type: IP|TID|TIME, read_format: ID, disabled: 1, inherit: 1, exclude_hv: 1, mmap: 1, comm: 1, enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1, ksymbol: 1, bpf_event: 1
#Ian Rogers
reviewerRe: [PATCH v2 2/5] perf record: Prevent override of attr->sample_period for libpfm4 events
Quoted 91 lines · Arnaldo Carvalho de Melo
Em Fri, Sep 04, 2020 at 03:50:13PM -0300, Arnaldo Carvalho de Melo escreveu:
Quoted 44 lines
Em Fri, Sep 04, 2020 at 03:48:03PM -0300, Arnaldo Carvalho de Melo escreveu:
Quoted 26 lines
Em Fri, Sep 04, 2020 at 09:22:10AM -0700, Ian Rogers escreveu:
Quoted 17 lines · Jiri Olsa
Quoted 15 lines · Ian Rogers
Quoted 14 lines · Ian Rogers
Quoted 13 lines · Jiri Olsa
Quoted 8 lines · Jiri Olsa
Quoted 1 line · Ian Rogers
[jolsa@krava perf]$ sudo ./perf test 17 -v
17: Setup structperf_event_attr:running './tests/attr/test-record-C0' expected sample_period=4000, got 3000
FAILED './tests/attr/test-record-C0' - match failureI'm not able to reproduce this. Do you have a build configuration or something else to look at? The test doesn't seem obviously connected with this patch.
Quoted 7 lines
Quoted 4 lines
Quoted 1 line
Jiri, any update? Thanks,
sorry, I rebased and ran it again and it passes for me now, so it got fixed along the way
No worries, thanks for the update! It'd be nice to land this and the other libpfm fixes.
Quoted 6 lines
I applied it and it generated this regression:
FAILED '
/home/acme/libexec/perf-core/tests/attr/test-record-pfm-period' - match failureI'll look at the other patches that are pending in this regard to see what needs to be squashed so that we don't break bisect.
So, more context:
running '
/home/acme/libexec/perf-core/tests/attr/test-record-pfm-period' expected exclude_hv=0, got 1
FAILED '/home/acme/libexec/perf-core/tests/attr/test-record-pfm-period' - match failure test child finished with -1
---- end ----
Setup structperf_event_attr: FAILED!
[root@five ~]#Quoted 1 line
Ian, can you take a look at this?
Further tests I've performed:
Committer testing: Not linking with libpfm: # ldd ~/bin/perf | grep libpfm # Before: # perf record -c 10000 -e cycles/period=12345/,instructions sleep 0.0001 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.052 MB perf.data (258 samples) ] # perf evlist -v cycles/period=12345/: size: 120, { sample_period, sample_freq }: 12345, sample_type: IP|TID|TIME|ID, read_format: ID, disabled: 1, inherit: 1, mmap: 1, comm: 1, enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1, ksymbol: 1, bpf_event: 1 instructions: size: 120, config: 0x1, { sample_period, sample_freq }: 10000, sample_type: IP|TID|TIME|ID, read_format: ID, disabled: 1, inherit: 1, enable_on_exec: 1, sample_id_all: 1, exclude_guest: 1 # After: # # perf record -c 10000 -e cycles/period=12345/,instructions sleep 0.0001 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.053 MB perf.data (284 samples) ] # perf evlist -v cycles/period=12345/: size: 120, { sample_period, sample_freq }: 12345, sample_type: IP|TID|TIME|ID, read_format: ID, disabled: 1, inherit: 1, mmap: 1, comm: 1, enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1, ksymbol: 1, bpf_event: 1 instructions: size: 120, config: 0x1, { sample_period, sample_freq }: 10000, sample_type: IP|TID|TIME|ID, read_format: ID, disabled: 1, inherit: 1, enable_on_exec: 1, sample_id_all: 1, exclude_guest: 1 # Linking with libpfm: # ldd ~/bin/perf | grep libpfm libpfm.so.4 => /lib64/libpfm.so.4 (0x00007f54c7d75000) # # perf record -c 10000 --pfm-events=cycles:period=77777 sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.043 MB perf.data (141 samples) ] # perf evlist -v cycles:period=77777: size: 120, { sample_period, sample_freq }: 10000, sample_type: IP|TID|TIME, read_format: ID, disabled: 1, inherit: 1, exclude_hv: 1, mmap: 1, comm: 1, enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1, ksymbol: 1, bpf_event: 1 #
Quoted 9 lines
After:
# perf record -c 10000 --pfm-events=cycles:period=77777 sleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.039 MB perf.data (19 samples) ]
# perf evlist -v
cycles:period=77777: size: 120, { sample_period, sample_freq }: 77777, sample_type: IP|TID|TIME, read_format: ID, disabled: 1, inherit: 1, exclude_hv: 1, mmap: 1, comm: 1, enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1, ksymbol: 1, bpf_event: 1
#Hi Arnaldo,
I've been trying to reproduce the test failure you mention and I've not been able to. This follow up e-mail seems to show things working as intended. Did the issue resolve itself?
Thanks,
Ian
Ian Rogers
reviewerRe: [PATCH v2 2/5] perf record: Prevent override of attr->sample_period for libpfm4 events
Quoted 103 lines · Ian Rogers
Quoted 91 lines · Arnaldo Carvalho de Melo
Em Fri, Sep 04, 2020 at 03:50:13PM -0300, Arnaldo Carvalho de Melo escreveu:
Quoted 44 lines
Em Fri, Sep 04, 2020 at 03:48:03PM -0300, Arnaldo Carvalho de Melo escreveu:
Quoted 26 lines
Em Fri, Sep 04, 2020 at 09:22:10AM -0700, Ian Rogers escreveu:
Quoted 17 lines · Jiri Olsa
Quoted 15 lines · Ian Rogers
Quoted 14 lines · Ian Rogers
Quoted 13 lines · Jiri Olsa
Quoted 8 lines · Jiri Olsa
Quoted 1 line · Ian Rogers
[jolsa@krava perf]$ sudo ./perf test 17 -v
17: Setup structperf_event_attr:running './tests/attr/test-record-C0' expected sample_period=4000, got 3000
FAILED './tests/attr/test-record-C0' - match failureI'm not able to reproduce this. Do you have a build configuration or something else to look at? The test doesn't seem obviously connected with this patch.
Quoted 7 lines
Quoted 4 lines
Quoted 1 line
Jiri, any update? Thanks,
sorry, I rebased and ran it again and it passes for me now, so it got fixed along the way
No worries, thanks for the update! It'd be nice to land this and the other libpfm fixes.
Quoted 6 lines
I applied it and it generated this regression:
FAILED '
/home/acme/libexec/perf-core/tests/attr/test-record-pfm-period' - match failureI'll look at the other patches that are pending in this regard to see what needs to be squashed so that we don't break bisect.
So, more context:
running '
/home/acme/libexec/perf-core/tests/attr/test-record-pfm-period' expected exclude_hv=0, got 1
FAILED '/home/acme/libexec/perf-core/tests/attr/test-record-pfm-period' - match failure test child finished with -1
---- end ----
Setup structperf_event_attr: FAILED!
[root@five ~]#Quoted 1 line
Ian, can you take a look at this?
Further tests I've performed:
Committer testing: Not linking with libpfm: # ldd ~/bin/perf | grep libpfm # Before: # perf record -c 10000 -e cycles/period=12345/,instructions sleep 0.0001 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.052 MB perf.data (258 samples) ] # perf evlist -v cycles/period=12345/: size: 120, { sample_period, sample_freq }: 12345, sample_type: IP|TID|TIME|ID, read_format: ID, disabled: 1, inherit: 1, mmap: 1, comm: 1, enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1, ksymbol: 1, bpf_event: 1 instructions: size: 120, config: 0x1, { sample_period, sample_freq }: 10000, sample_type: IP|TID|TIME|ID, read_format: ID, disabled: 1, inherit: 1, enable_on_exec: 1, sample_id_all: 1, exclude_guest: 1 # After: # # perf record -c 10000 -e cycles/period=12345/,instructions sleep 0.0001 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.053 MB perf.data (284 samples) ] # perf evlist -v cycles/period=12345/: size: 120, { sample_period, sample_freq }: 12345, sample_type: IP|TID|TIME|ID, read_format: ID, disabled: 1, inherit: 1, mmap: 1, comm: 1, enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1, ksymbol: 1, bpf_event: 1 instructions: size: 120, config: 0x1, { sample_period, sample_freq }: 10000, sample_type: IP|TID|TIME|ID, read_format: ID, disabled: 1, inherit: 1, enable_on_exec: 1, sample_id_all: 1, exclude_guest: 1 # Linking with libpfm: # ldd ~/bin/perf | grep libpfm libpfm.so.4 => /lib64/libpfm.so.4 (0x00007f54c7d75000) # # perf record -c 10000 --pfm-events=cycles:period=77777 sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.043 MB perf.data (141 samples) ] # perf evlist -v cycles:period=77777: size: 120, { sample_period, sample_freq }: 10000, sample_type: IP|TID|TIME, read_format: ID, disabled: 1, inherit: 1, exclude_hv: 1, mmap: 1, comm: 1, enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1, ksymbol: 1, bpf_event: 1 #Quoted 9 lines
After: # perf record -c 10000 --pfm-events=cycles:period=77777 sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.039 MB perf.data (19 samples) ] # perf evlist -v cycles:period=77777: size: 120, { sample_period, sample_freq }: 77777, sample_type: IP|TID|TIME, read_format: ID, disabled: 1, inherit: 1, exclude_hv: 1, mmap: 1, comm: 1, enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1, ksymbol: 1, bpf_event: 1 #
Quoted 5 lines
Hi Arnaldo,
I've been trying to reproduce the test failure you mention and I've not been able to. This follow up e-mail seems to show things working as intended. Did the issue resolve itself?
It looks like the test to ensure the period for pfm events worked:
https://lore.kernel.org/lkml/[email protected]/
has been applied without the fixes in patches 1 and 2. I've resent the patches, hopefully addressing all feedback.
https://lore.kernel.org/lkml/[email protected]/
Thanks,
Ian
Arnaldo Carvalho de Melo
maintainerRe: [PATCH v2 2/5] perf record: Prevent override of attr->sample_period for libpfm4 events
In reply to Ian RogersEm Tue, Jul 28, 2020 at 01:57:31AM -0700, Ian Rogers escreveu:
Quoted 6 lines
From: Stephane Eranian <[email protected]>
Before:
$ perf record -c 10000 --pfm-events=cycles:period=77777Would yield a cycles event with period=10000, instead of 77777.
I tried the equivalent without libpfm and it works:
$ perf record -c 10000 -e cycles/period=20000/ sleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.024 MB perf.data (23 samples) ]
$ perf evlist -v
cycles/period=20000/u: size: 120, { sample_period, sample_freq }: 20000, sample_type: IP|TID|TIME, read_format: ID, disabled: 1, inherit: 1, exclude_kernel: 1, exclude_hv: 1, mmap: 1, comm: 1, enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1, ksymbol: 1, bpf_event: 1
$Quoted 31 lines
This was due to an ordering issue between libpfm4 parsing the event string and perf record initializing the event.
This patch fixes the problem by preventing override for events with
attr->sample_period!= 0 by the timeperf_evsel__config()is invoked. This seems to have been the intent of the author.Signed-off-by Stephane Eranian <[email protected]>Reviewed-by Ian Rogers <[email protected]>--- tools/perf/util/evsel.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)tools/perf/util/evsel.c +1
-2@@ -976,8 +976,7 @@ void evsel__config(struct evsel *evsel, struct record_opts *opts,* We default some events to have a default interval. But keep * it a weak assumption overridable by the user. */ -if (!attr->sample_period || (opts->user_freq !=UINT_MAX || - opts->user_interval !=ULLONG_MAX )) { +if (!attr->sample_period) {if (opts->freq) { attr->freq =1 ; attr->sample_freq = opts->freq;Signature · 1 line
2.28.0.163.g6104cc2f0b6-goog
Signature · 1 line
- Arnaldo
Ian Rogers
reviewerPATCH v2 3/5 perf test: Ensure sample_period is set libpfm4 events
In reply to Ian RogersTest that a command line option doesn't override the period set on a libpfm4 event.
Without libpfm4 test passes as unsupported.
--- tools/perf/tests/attr/README | 1 + tools/perf/tests/attr/test-record-pfm-period | 9 +++++++++ 2 files changed, 10 insertions(+) create mode 100644 tools/perf/tests/attr/test-record-pfm-period
tools/perf/tests/attr/README +1 -0
@@ -53,6 +53,7 @@ Following tests are defined (with perf commands): perf record -i kill (test-record-no-inherit) perf record -n kill (test-record-no-samples) perf record -c 100 -P kill (test-record-period) + perf record -c 1 --pfm-events=cycles:period=2 (test-record-pfm-period) perf record -R kill (test-record-raw) perf stat -e cycles kill (test-stat-basic) perf stat kill (test-stat-default)
tools/perf/tests/attr/test-record-pfm-period new file +9 -0
@@ -0,0 +1,9 @@ +[config] +command = record +args = --no-bpf-event -c 10000 --pfm-events=cycles:period=77777 kill >/dev/null 2>&1 +ret = 1 + +[event:base-record] +sample_period=77777 +sample_type=7 +freq=0
Signature · 1 line
2.28.0.163.g6104cc2f0b6-goog
Re: [PATCH v2 3/5] perf test: Ensure sample_period is set libpfm4 events
Em Tue, Jul 28, 2020 at 01:57:32AM -0700, Ian Rogers escreveu:
Quoted 3 lines
Test that a command line option doesn't override the period set on a libpfm4 event.
Without libpfm4 test passes as unsupported.
Thanks, applied.
- Arnaldo
Ian Rogers
reviewerPATCH v2 4/5 perf record: Don't clear event's period if set by a term
In reply to Ian RogersIf events in a group explicitly set a frequency or period with leader sampling, don't disable the samples on those events.
Prior to 5.8:
perf record -e '{cycles/period=12345000/,instructions/period=6789000/}:S' would clear the attributes then apply the config terms. In commit 5f34278867b7 leader sampling configuration was moved to after applying the config terms, in the example, making the instructions' event have its period cleared.
This change makes it so that sampling is only disabled if configuration terms aren't present.
--- tools/perf/util/record.c | 28 ++++++++++++++++++++-------- 1 file changed, 20 insertions(+), 8 deletions(-)
tools/perf/util/record.c +20 -8
@@ -2,6 +2,7 @@#include "debug.h" #include "evlist.h" #include "evsel.h" +#include "evsel_config.h" #include "parse-events.h" #include <errno.h> #include <limits.h>
@@ -38,6 +39,9 @@ static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *evstruct perf_event_attr *attr = &evsel->core.attr;struct evsel *leader = evsel->leader;struct evsel *read_sampler; +struct evsel_config_term *term; +struct list_head *config_terms = &evsel->config_terms; +int term_types, freq_mask;if (!leader->sample_read)return ;
@@ -47,16 +51,24 @@ static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *evif (evsel == read_sampler)return ; +/* Determine the evsel's config term types. */ + term_types =0 ; +list_for_each_entry (term, config_terms, list) { + term_types |=1 << term->type; + }/* -* Disable sampling for all group members other than the leader in -* case the leader 'leads' the sampling, except when the leader is an -* AUX area event, in which case the 2nd event in the group is the one -* that 'leads' the sampling. +* Disable sampling for all group members except those with explicit +* config terms or the leader. In the case of an AUX area event, the 2nd +* event in the group is the one that 'leads' the sampling. */ - attr->freq =0 ; - attr->sample_freq =0 ; - attr->sample_period =0 ; - attr->write_backward =0 ; + freq_mask = (1 <<EVSEL__CONFIG_TERM_FREQ ) | (1 <<EVSEL__CONFIG_TERM_PERIOD ); +if ((term_types & freq_mask) ==0 ) { + attr->freq =0 ; + attr->sample_freq =0 ; + attr->sample_period =0 ; + } +if ((term_types & (1 <<EVSEL__CONFIG_TERM_OVERWRITE )) ==0 ) + attr->write_backward =0 ;/* * We don't get a sample for slave events, we make them when delivering
Signature · 1 line
2.28.0.163.g6104cc2f0b6-goog
Arnaldo Carvalho de Melo
maintainerRe: [PATCH v2 4/5] perf record: Don't clear event's period if set by a term
Em Tue, Jul 28, 2020 at 01:57:33AM -0700, Ian Rogers escreveu:
Quoted 11 lines
If events in a group explicitly set a frequency or period with leader sampling, don't disable the samples on those events.
Prior to 5.8:
perf record -e '{cycles/period=12345000/,instructions/period=6789000/}:S' would clear the attributes then apply the config terms. In commit5f34278867b7leader sampling configuration was moved to after applying the config terms, in the example, making the instructions' event have its period cleared.
This change makes it so that sampling is only disabled if configuration terms aren't present.
Adrian, can you take a look at this one?
- Arnaldo
Quoted 64 lines
Fixes 5f34278867b7 ("perf evlist: Move leader-sampling configuration")Signed-off-by Ian Rogers <[email protected]>--- tools/perf/util/record.c | 28 ++++++++++++++++++++-------- 1 file changed, 20 insertions(+), 8 deletions(-)tools/perf/util/record.c +20
-8@@ -2,6 +2,7 @@#include "debug.h" #include "evlist.h" #include "evsel.h" +#include "evsel_config.h" #include "parse-events.h" #include <errno.h> #include <limits.h> @@ -38,6 +39,9 @@ static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *evstruct perf_event_attr *attr = &evsel->core.attr;struct evsel *leader = evsel->leader;struct evsel *read_sampler; +struct evsel_config_term *term; +struct list_head *config_terms = &evsel->config_terms; +int term_types, freq_mask;if (!leader->sample_read)return ;@@ -47,16 +51,24 @@ static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *evif (evsel == read_sampler)return ; +/* Determine the evsel's config term types. */ + term_types =0 ; +list_for_each_entry (term, config_terms, list) { + term_types |=1 << term->type; + }/* -* Disable sampling for all group members other than the leader in -* case the leader 'leads' the sampling, except when the leader is an -* AUX area event, in which case the 2nd event in the group is the one -* that 'leads' the sampling. +* Disable sampling for all group members except those with explicit +* config terms or the leader. In the case of an AUX area event, the 2nd +* event in the group is the one that 'leads' the sampling. */ - attr->freq =0 ; - attr->sample_freq =0 ; - attr->sample_period =0 ; - attr->write_backward =0 ; + freq_mask = (1 <<EVSEL__CONFIG_TERM_FREQ ) | (1 <<EVSEL__CONFIG_TERM_PERIOD ); +if ((term_types & freq_mask) ==0 ) { + attr->freq =0 ; + attr->sample_freq =0 ; + attr->sample_period =0 ; + } +if ((term_types & (1 <<EVSEL__CONFIG_TERM_OVERWRITE )) ==0 ) + attr->write_backward =0 ;/* * We don't get a sample for slave events, we make them when delivering Signature · 1 line
2.28.0.163.g6104cc2f0b6-goog
Signature · 1 line
- Arnaldo
Adrian Hunter
reviewerRe: [PATCH v2 4/5] perf record: Don't clear event's period if set by a term
In reply to Ian RogersQuoted 5 lines · Ian Rogers
If events in a group explicitly set a frequency or period with leader sampling, don't disable the samples on those events.
Prior to 5.8:
perf record -e '{cycles/period=12345000/,instructions/period=6789000/}:S'
Might be worth explaining this use-case some more. Perhaps add it to the leader sampling documentation for perf-list.
Quoted 49 lines
would clear the attributes then apply the config terms. In commit
5f34278867b7leader sampling configuration was moved to after applying the config terms, in the example, making the instructions' event have its period cleared.
This change makes it so that sampling is only disabled if configuration terms aren't present.Fixes 5f34278867b7 ("perf evlist: Move leader-sampling configuration")Signed-off-by Ian Rogers <[email protected]>--- tools/perf/util/record.c | 28 ++++++++++++++++++++-------- 1 file changed, 20 insertions(+), 8 deletions(-)tools/perf/util/record.c +9
-4@@ -2,6 +2,7 @@#include "debug.h" #include "evlist.h" #include "evsel.h" +#include "evsel_config.h" #include "parse-events.h" #include <errno.h> #include <limits.h> @@ -38,6 +39,9 @@ static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *evstruct perf_event_attr *attr = &evsel->core.attr;struct evsel *leader = evsel->leader;struct evsel *read_sampler; +struct evsel_config_term *term; +struct list_head *config_terms = &evsel->config_terms; +int term_types, freq_mask;if (!leader->sample_read)return ;@@ -47,16 +51,24 @@ static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *evif (evsel == read_sampler)return ; +/* Determine the evsel's config term types. */ + term_types =0 ; +list_for_each_entry (term, config_terms, list) { + term_types |=1 << term->type; + }/* -* Disable sampling for all group members other than the leader in -* case the leader 'leads' the sampling, except when the leader is an -* AUX area event, in which case the 2nd event in the group is the one -* that 'leads' the sampling.
tools/perf/util/record.c static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *ev · line 65
Quoted 10 lines
tools/perf/util/record.c +5
-4+* Disable sampling for all group members except those with explicit +* config terms or the leader. In the case of an AUX area event, the 2nd +* event in the group is the one that 'leads' the sampling. */ - attr->freq =0 ; - attr->sample_freq =0 ; - attr->sample_period =0 ; - attr->write_backward =0 ; + freq_mask = (1 <<EVSEL__CONFIG_TERM_FREQ ) | (1 <<EVSEL__CONFIG_TERM_PERIOD ); +if ((term_types & freq_mask) ==0 ) {
It would be nicer to have a helper e.g.
if (!evsel__have_config_term(evsel, FREQ) &&
!evsel__have_config_term(evsel, PERIOD)) {tools/perf/util/record.c static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *ev · line 68
Quoted 3 lines
tools/perf/util/record.c +3
-0+ attr->freq =0 ; + attr->sample_freq =0 ; + attr->sample_period =0 ;
If we are not sampling, then maybe we should also put here:
attr->write_backward = 0;
tools/perf/util/record.c static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *ev · line 69
Quoted 1 line
tools/perf/util/record.c +1
-0+ }
Then, if we are sampling this evsel shouldn't the backward setting match the leader? e.g.
if (attr->sample_freq) attr->write_backward = leader->core.attr.write_backward;
Quoted 6 lines
tools/perf/util/record.c +2
-0+if ((term_types & (1 <<EVSEL__CONFIG_TERM_OVERWRITE )) ==0 ) + attr->write_backward =0 ;/* * We don't get a sample for slave events, we make them when delivering
Ian Rogers
reviewerRe: [PATCH v2 4/5] perf record: Don't clear event's period if set by a term
Quoted 73 lines · Adrian Hunter
Quoted 5 lines · Ian Rogers
If events in a group explicitly set a frequency or period with leader sampling, don't disable the samples on those events.
Prior to 5.8:
perf record -e '{cycles/period=12345000/,instructions/period=6789000/}:S'Might be worth explaining this use-case some more. Perhaps add it to the leader sampling documentation for perf-list.
Quoted 49 lines
would clear the attributes then apply the config terms. In commit
5f34278867b7leader sampling configuration was moved to after applying the config terms, in the example, making the instructions' event have its period cleared.
This change makes it so that sampling is only disabled if configuration terms aren't present.Fixes 5f34278867b7 ("perf evlist: Move leader-sampling configuration")Signed-off-by Ian Rogers <[email protected]>--- tools/perf/util/record.c | 28 ++++++++++++++++++++-------- 1 file changed, 20 insertions(+), 8 deletions(-)tools/perf/util/record.c +9
-4@@ -2,6 +2,7 @@#include "debug.h" #include "evlist.h" #include "evsel.h" +#include "evsel_config.h" #include "parse-events.h" #include <errno.h> #include <limits.h> @@ -38,6 +39,9 @@ static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *evstruct perf_event_attr *attr = &evsel->core.attr;struct evsel *leader = evsel->leader;struct evsel *read_sampler; +struct evsel_config_term *term; +struct list_head *config_terms = &evsel->config_terms; +int term_types, freq_mask;if (!leader->sample_read)return ;@@ -47,16 +51,24 @@ static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *evif (evsel == read_sampler)return ; +/* Determine the evsel's config term types. */ + term_types =0 ; +list_for_each_entry (term, config_terms, list) { + term_types |=1 << term->type; + }/* -* Disable sampling for all group members other than the leader in -* case the leader 'leads' the sampling, except when the leader is an -* AUX area event, in which case the 2nd event in the group is the one -* that 'leads' the sampling. tools/perf/util/record.c static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *ev · line 65
Quoted 10 lines
tools/perf/util/record.c +5
-4+* Disable sampling for all group members except those with explicit +* config terms or the leader. In the case of an AUX area event, the 2nd +* event in the group is the one that 'leads' the sampling. */ - attr->freq =0 ; - attr->sample_freq =0 ; - attr->sample_period =0 ; - attr->write_backward =0 ; + freq_mask = (1 <<EVSEL__CONFIG_TERM_FREQ ) | (1 <<EVSEL__CONFIG_TERM_PERIOD ); +if ((term_types & freq_mask) ==0 ) {It would be nicer to have a helper e.g.
Quoted 2 lines
if (!evsel__have_config_term(evsel, FREQ) &&
!evsel__have_config_term(evsel, PERIOD)) {Sure. The point of doing it this way was to avoid repeatedly iterating over the config term list.
Quoted 8 lines
tools/perf/util/record.c static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *ev · line 68
Quoted 3 lines
tools/perf/util/record.c +3
-0+ attr->freq =0 ; + attr->sample_freq =0 ; + attr->sample_period =0 ;If we are not sampling, then maybe we should also put here:
attr->write_backward = 0;
Quoted 7 lines
tools/perf/util/record.c static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *ev · line 69
Quoted 1 line
tools/perf/util/record.c +1
-0+ }Then, if we are sampling this evsel shouldn't the backward setting match the leader? e.g.
if (attr->sample_freq) attr->write_backward = leader->core.attr.write_backward;
Perhaps that should be a follow up change? This change is trying to make the behavior match the previous behavior.
Thanks,
Ian
Quoted 7 lines
Quoted 6 lines
tools/perf/util/record.c +2
-0+if ((term_types & (1 <<EVSEL__CONFIG_TERM_OVERWRITE )) ==0 ) + attr->write_backward =0 ;/* * We don't get a sample for slave events, we make them when delivering
Adrian Hunter
reviewerRe: [PATCH v2 4/5] perf record: Don't clear event's period if set by a term
Quoted 74 lines · Ian Rogers
Quoted 73 lines · Adrian Hunter
Quoted 5 lines · Ian Rogers
If events in a group explicitly set a frequency or period with leader sampling, don't disable the samples on those events.
Prior to 5.8:
perf record -e '{cycles/period=12345000/,instructions/period=6789000/}:S'Might be worth explaining this use-case some more. Perhaps add it to the leader sampling documentation for perf-list.
Quoted 49 lines
would clear the attributes then apply the config terms. In commit
5f34278867b7leader sampling configuration was moved to after applying the config terms, in the example, making the instructions' event have its period cleared.
This change makes it so that sampling is only disabled if configuration terms aren't present.Fixes 5f34278867b7 ("perf evlist: Move leader-sampling configuration")Signed-off-by Ian Rogers <[email protected]>--- tools/perf/util/record.c | 28 ++++++++++++++++++++-------- 1 file changed, 20 insertions(+), 8 deletions(-)tools/perf/util/record.c +9
-4@@ -2,6 +2,7 @@#include "debug.h" #include "evlist.h" #include "evsel.h" +#include "evsel_config.h" #include "parse-events.h" #include <errno.h> #include <limits.h> @@ -38,6 +39,9 @@ static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *evstruct perf_event_attr *attr = &evsel->core.attr;struct evsel *leader = evsel->leader;struct evsel *read_sampler; +struct evsel_config_term *term; +struct list_head *config_terms = &evsel->config_terms; +int term_types, freq_mask;if (!leader->sample_read)return ;@@ -47,16 +51,24 @@ static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *evif (evsel == read_sampler)return ; +/* Determine the evsel's config term types. */ + term_types =0 ; +list_for_each_entry (term, config_terms, list) { + term_types |=1 << term->type; + }/* -* Disable sampling for all group members other than the leader in -* case the leader 'leads' the sampling, except when the leader is an -* AUX area event, in which case the 2nd event in the group is the one -* that 'leads' the sampling. tools/perf/util/record.c static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *ev · line 65
Quoted 10 lines
tools/perf/util/record.c +5
-4+* Disable sampling for all group members except those with explicit +* config terms or the leader. In the case of an AUX area event, the 2nd +* event in the group is the one that 'leads' the sampling. */ - attr->freq =0 ; - attr->sample_freq =0 ; - attr->sample_period =0 ; - attr->write_backward =0 ; + freq_mask = (1 <<EVSEL__CONFIG_TERM_FREQ ) | (1 <<EVSEL__CONFIG_TERM_PERIOD ); +if ((term_types & freq_mask) ==0 ) {It would be nicer to have a helper e.g.
Quoted 5 lines
Quoted 2 lines
if (!evsel__have_config_term(evsel, FREQ) && !evsel__have_config_term(evsel, PERIOD)) {Sure. The point of doing it this way was to avoid repeatedly iterating over the config term list.
But perhaps it is premature optimization
Quoted 11 lines
Quoted 11 lines
tools/perf/util/record.c static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *ev · line 68
Quoted 4 lines
tools/perf/util/record.c +3
-0+ attr->freq =0 ; + attr->sample_freq =0 ; + attr->sample_period =0 ;If we are not sampling, then maybe we should also put here:
attr->write_backward = 0;Quoted 1 line
tools/perf/util/record.c +1
-0+ }
Quoted 8 lines
Quoted 5 lines
Then, if we are sampling this evsel shouldn't the backward setting match the leader? e.g.
if (attr->sample_freq) attr->write_backward = leader->core.attr.write_backward;Perhaps that should be a follow up change? This change is trying to make the behavior match the previous behavior.
Sure
Quoted 11 lines
Thanks,
IanQuoted 7 lines
Quoted 6 lines
tools/perf/util/record.c +2
-0+if ((term_types & (1 <<EVSEL__CONFIG_TERM_OVERWRITE )) ==0 ) + attr->write_backward =0 ;/* * We don't get a sample for slave events, we make them when delivering
Ian Rogers
reviewerRe: [PATCH v2 4/5] perf record: Don't clear event's period if set by a term
Quoted 76 lines · Adrian Hunter
Quoted 74 lines · Ian Rogers
Quoted 73 lines · Adrian Hunter
Quoted 5 lines · Ian Rogers
If events in a group explicitly set a frequency or period with leader sampling, don't disable the samples on those events.
Prior to 5.8:
perf record -e '{cycles/period=12345000/,instructions/period=6789000/}:S'Might be worth explaining this use-case some more. Perhaps add it to the leader sampling documentation for perf-list.
Quoted 49 lines
would clear the attributes then apply the config terms. In commit
5f34278867b7leader sampling configuration was moved to after applying the config terms, in the example, making the instructions' event have its period cleared.
This change makes it so that sampling is only disabled if configuration terms aren't present.Fixes 5f34278867b7 ("perf evlist: Move leader-sampling configuration")Signed-off-by Ian Rogers <[email protected]>--- tools/perf/util/record.c | 28 ++++++++++++++++++++-------- 1 file changed, 20 insertions(+), 8 deletions(-)tools/perf/util/record.c +9
-4@@ -2,6 +2,7 @@#include "debug.h" #include "evlist.h" #include "evsel.h" +#include "evsel_config.h" #include "parse-events.h" #include <errno.h> #include <limits.h> @@ -38,6 +39,9 @@ static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *evstruct perf_event_attr *attr = &evsel->core.attr;struct evsel *leader = evsel->leader;struct evsel *read_sampler; +struct evsel_config_term *term; +struct list_head *config_terms = &evsel->config_terms; +int term_types, freq_mask;if (!leader->sample_read)return ;@@ -47,16 +51,24 @@ static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *evif (evsel == read_sampler)return ; +/* Determine the evsel's config term types. */ + term_types =0 ; +list_for_each_entry (term, config_terms, list) { + term_types |=1 << term->type; + }/* -* Disable sampling for all group members other than the leader in -* case the leader 'leads' the sampling, except when the leader is an -* AUX area event, in which case the 2nd event in the group is the one -* that 'leads' the sampling. tools/perf/util/record.c static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *ev · line 65
Quoted 10 lines
tools/perf/util/record.c +5
-4+* Disable sampling for all group members except those with explicit +* config terms or the leader. In the case of an AUX area event, the 2nd +* event in the group is the one that 'leads' the sampling. */ - attr->freq =0 ; - attr->sample_freq =0 ; - attr->sample_period =0 ; - attr->write_backward =0 ; + freq_mask = (1 <<EVSEL__CONFIG_TERM_FREQ ) | (1 <<EVSEL__CONFIG_TERM_PERIOD ); +if ((term_types & freq_mask) ==0 ) {It would be nicer to have a helper e.g.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
if (!evsel__have_config_term(evsel, FREQ) && !evsel__have_config_term(evsel, PERIOD)) {Sure. The point of doing it this way was to avoid repeatedly iterating over the config term list.
But perhaps it is premature optimization
The alternative is more loc. I think we can bike shed on this but it's not really changing the substance of the change. I'm keen to try to be efficient where we can as we see issues at scale.
Thanks,
Ian
Quoted 34 lines
Quoted 11 lines
Quoted 11 lines
tools/perf/util/record.c static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *ev · line 68
Quoted 4 lines
tools/perf/util/record.c +3
-0+ attr->freq =0 ; + attr->sample_freq =0 ; + attr->sample_period =0 ;If we are not sampling, then maybe we should also put here:
attr->write_backward = 0;Quoted 1 line
tools/perf/util/record.c +1
-0+ }Quoted 8 lines
Quoted 5 lines
Then, if we are sampling this evsel shouldn't the backward setting match the leader? e.g.
if (attr->sample_freq) attr->write_backward = leader->core.attr.write_backward;Perhaps that should be a follow up change? This change is trying to make the behavior match the previous behavior.
Sure
Quoted 11 lines
Thanks,
IanQuoted 7 lines
Quoted 6 lines
tools/perf/util/record.c +2
-0+if ((term_types & (1 <<EVSEL__CONFIG_TERM_OVERWRITE )) ==0 ) + attr->write_backward =0 ;/* * We don't get a sample for slave events, we make them when delivering
Ian Rogers
reviewerRe: [PATCH v2 4/5] perf record: Don't clear event's period if set by a term
Quoted 84 lines · Ian Rogers
Quoted 82 lines · Adrian Hunter
Quoted 79 lines · Ian Rogers
Quoted 73 lines · Adrian Hunter
Quoted 5 lines · Ian Rogers
If events in a group explicitly set a frequency or period with leader sampling, don't disable the samples on those events.
Prior to 5.8:
perf record -e '{cycles/period=12345000/,instructions/period=6789000/}:S'Might be worth explaining this use-case some more. Perhaps add it to the leader sampling documentation for perf-list.
Quoted 49 lines
would clear the attributes then apply the config terms. In commit
5f34278867b7leader sampling configuration was moved to after applying the config terms, in the example, making the instructions' event have its period cleared.
This change makes it so that sampling is only disabled if configuration terms aren't present.Fixes 5f34278867b7 ("perf evlist: Move leader-sampling configuration")Signed-off-by Ian Rogers <[email protected]>--- tools/perf/util/record.c | 28 ++++++++++++++++++++-------- 1 file changed, 20 insertions(+), 8 deletions(-)tools/perf/util/record.c +9
-4@@ -2,6 +2,7 @@#include "debug.h" #include "evlist.h" #include "evsel.h" +#include "evsel_config.h" #include "parse-events.h" #include <errno.h> #include <limits.h> @@ -38,6 +39,9 @@ static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *evstruct perf_event_attr *attr = &evsel->core.attr;struct evsel *leader = evsel->leader;struct evsel *read_sampler; +struct evsel_config_term *term; +struct list_head *config_terms = &evsel->config_terms; +int term_types, freq_mask;if (!leader->sample_read)return ;@@ -47,16 +51,24 @@ static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *evif (evsel == read_sampler)return ; +/* Determine the evsel's config term types. */ + term_types =0 ; +list_for_each_entry (term, config_terms, list) { + term_types |=1 << term->type; + }/* -* Disable sampling for all group members other than the leader in -* case the leader 'leads' the sampling, except when the leader is an -* AUX area event, in which case the 2nd event in the group is the one -* that 'leads' the sampling. tools/perf/util/record.c static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *ev · line 65
Quoted 10 lines
tools/perf/util/record.c +5
-4+* Disable sampling for all group members except those with explicit +* config terms or the leader. In the case of an AUX area event, the 2nd +* event in the group is the one that 'leads' the sampling. */ - attr->freq =0 ; - attr->sample_freq =0 ; - attr->sample_period =0 ; - attr->write_backward =0 ; + freq_mask = (1 <<EVSEL__CONFIG_TERM_FREQ ) | (1 <<EVSEL__CONFIG_TERM_PERIOD ); +if ((term_types & freq_mask) ==0 ) {It would be nicer to have a helper e.g.
Quoted 2 lines
if (!evsel__have_config_term(evsel, FREQ) && !evsel__have_config_term(evsel, PERIOD)) {Sure. The point of doing it this way was to avoid repeatedly iterating over the config term list.
Quoted 8 lines
Quoted 1 line
But perhaps it is premature optimization
The alternative is more loc. I think we can bike shed on this but it's not really changing the substance of the change. I'm keen to try to be efficient where we can as we see issues at scale.
Thanks,
Ian
Ping. Do we want to turn this into multiple O(N) searches using a helper rather than 1 as coded here?
Thanks,
Ian
Quoted 34 lines
Quoted 34 lines
Quoted 11 lines
Quoted 11 lines
tools/perf/util/record.c static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *ev · line 68
Quoted 4 lines
tools/perf/util/record.c +3
-0+ attr->freq =0 ; + attr->sample_freq =0 ; + attr->sample_period =0 ;If we are not sampling, then maybe we should also put here:
attr->write_backward = 0;Quoted 1 line
tools/perf/util/record.c +1
-0+ }Quoted 8 lines
Quoted 5 lines
Then, if we are sampling this evsel shouldn't the backward setting match the leader? e.g.
if (attr->sample_freq) attr->write_backward = leader->core.attr.write_backward;Perhaps that should be a follow up change? This change is trying to make the behavior match the previous behavior.
Sure
Quoted 11 lines
Thanks,
IanQuoted 7 lines
Quoted 6 lines
tools/perf/util/record.c +2
-0+if ((term_types & (1 <<EVSEL__CONFIG_TERM_OVERWRITE )) ==0 ) + attr->write_backward =0 ;/* * We don't get a sample for slave events, we make them when delivering
Adrian Hunter
reviewerRe: [PATCH v2 4/5] perf record: Don't clear event's period if set by a term
Quoted 79 lines · Ian Rogers
Quoted 78 lines · Ian Rogers
Quoted 77 lines · Adrian Hunter
Quoted 71 lines · Ian Rogers
Quoted 70 lines · Adrian Hunter
Quoted 5 lines · Ian Rogers
If events in a group explicitly set a frequency or period with leader sampling, don't disable the samples on those events.
Prior to 5.8:
perf record -e '{cycles/period=12345000/,instructions/period=6789000/}:S'Might be worth explaining this use-case some more. Perhaps add it to the leader sampling documentation for perf-list.
Quoted 49 lines
would clear the attributes then apply the config terms. In commit
5f34278867b7leader sampling configuration was moved to after applying the config terms, in the example, making the instructions' event have its period cleared.
This change makes it so that sampling is only disabled if configuration terms aren't present.Fixes 5f34278867b7 ("perf evlist: Move leader-sampling configuration")Signed-off-by Ian Rogers <[email protected]>--- tools/perf/util/record.c | 28 ++++++++++++++++++++-------- 1 file changed, 20 insertions(+), 8 deletions(-)tools/perf/util/record.c +9
-4@@ -2,6 +2,7 @@#include "debug.h" #include "evlist.h" #include "evsel.h" +#include "evsel_config.h" #include "parse-events.h" #include <errno.h> #include <limits.h> @@ -38,6 +39,9 @@ static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *evstruct perf_event_attr *attr = &evsel->core.attr;struct evsel *leader = evsel->leader;struct evsel *read_sampler; +struct evsel_config_term *term; +struct list_head *config_terms = &evsel->config_terms; +int term_types, freq_mask;if (!leader->sample_read)return ;@@ -47,16 +51,24 @@ static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *evif (evsel == read_sampler)return ; +/* Determine the evsel's config term types. */ + term_types =0 ; +list_for_each_entry (term, config_terms, list) { + term_types |=1 << term->type; + }/* -* Disable sampling for all group members other than the leader in -* case the leader 'leads' the sampling, except when the leader is an -* AUX area event, in which case the 2nd event in the group is the one -* that 'leads' the sampling. tools/perf/util/record.c static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *ev · line 65
Quoted 10 lines
tools/perf/util/record.c +5
-4+* Disable sampling for all group members except those with explicit +* config terms or the leader. In the case of an AUX area event, the 2nd +* event in the group is the one that 'leads' the sampling. */ - attr->freq =0 ; - attr->sample_freq =0 ; - attr->sample_period =0 ; - attr->write_backward =0 ; + freq_mask = (1 <<EVSEL__CONFIG_TERM_FREQ ) | (1 <<EVSEL__CONFIG_TERM_PERIOD ); +if ((term_types & freq_mask) ==0 ) {It would be nicer to have a helper e.g.
Quoted 4 lines
Quoted 2 lines
if (!evsel__have_config_term(evsel, FREQ) && !evsel__have_config_term(evsel, PERIOD)) {Sure. The point of doing it this way was to avoid repeatedly iterating over the config term list.
But perhaps it is premature optimization
Quoted 8 lines
Quoted 6 lines
The alternative is more loc. I think we can bike shed on this but it's not really changing the substance of the change. I'm keen to try to be efficient where we can as we see issues at scale.
Thanks,
IanPing. Do we want to turn this into multiple O(N) searches using a helper rather than 1 as coded here?
Actually max. 30 iterations vs max. 15 iterations for the 15 current possible config terms.
At least please don't open code the config term implementation details.
For example, make evsel__have_config_term() accept multiple terms or introduce
u64 term_mask = evsel__config_term_mask(evsel);
has_config_term(term_mask, FREQ) etc
or whatever.
Quoted 30 lines
Thanks,
IanQuoted 26 lines
Quoted 26 lines
Quoted 7 lines
Quoted 7 lines
tools/perf/util/record.c static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *ev · line 68
Quoted 3 lines
tools/perf/util/record.c +3
-0+ attr->freq =0 ; + attr->sample_freq =0 ; + attr->sample_period =0 ;If we are not sampling, then maybe we should also put here:
attr->write_backward = 0;Quoted 8 lines
Quoted 6 lines
tools/perf/util/record.c static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *ev · line 69
Quoted 1 line
tools/perf/util/record.c +1
-0+ }Then, if we are sampling this evsel shouldn't the backward setting match the leader? e.g.
if (attr->sample_freq) attr->write_backward = leader->core.attr.write_backward;Perhaps that should be a follow up change? This change is trying to make the behavior match the previous behavior.
Sure
Quoted 9 lines
Thanks,
IanQuoted 6 lines
Quoted 6 lines
tools/perf/util/record.c +2
-0+if ((term_types & (1 <<EVSEL__CONFIG_TERM_OVERWRITE )) ==0 ) + attr->write_backward =0 ;/* * We don't get a sample for slave events, we make them when delivering
Ian Rogers
reviewerPATCH v2 5/5 perf test: Leader sampling shouldn't clear sample period
In reply to Ian RogersAdd test that a sibling with leader sampling doesn't have its period cleared.
--- tools/perf/tests/attr/README | 1 + tools/perf/tests/attr/test-record-group2 | 29 ++++++++++++++++++++++++ 2 files changed, 30 insertions(+) create mode 100644 tools/perf/tests/attr/test-record-group2
tools/perf/tests/attr/README +1 -0
@@ -49,6 +49,7 @@ Following tests are defined (with perf commands): perf record --call-graph fp kill (test-record-graph-fp) perf record --group -e cycles,instructions kill (test-record-group) perf record -e '{cycles,instructions}' kill (test-record-group1) + perf record -e '{cycles/period=1/,instructions/period=2/}:S' kill (test-record-group2) perf record -D kill (test-record-no-delay) perf record -i kill (test-record-no-inherit) perf record -n kill (test-record-no-samples)
tools/perf/tests/attr/test-record-group2 new file +29 -0
@@ -0,0 +1,29 @@ +[config] +command = record +args = --no-bpf-event -e '{cycles/period=1234000/,instructions/period=6789000/}:S' kill >/dev/null 2>&1 +ret = 1 + +[event-1:base-record] +fd=1 +group_fd=-1 +config=0|1 +sample_period=1234000 +sample_type=87 +read_format=12 +inherit=0 +freq=0 + +[event-2:base-record] +fd=2 +group_fd=1 +config=0|1 +sample_period=6789000 +sample_type=87 +read_format=12 +disabled=0 +inherit=0 +mmap=0 +comm=0 +freq=0 +enable_on_exec=0 +task=0
Signature · 1 line
2.28.0.163.g6104cc2f0b6-goog