[PATCH 0/3] perf: PEBS/period freerunning fixes

[PATCH 0/3] perf: PEBS/period freerunning fixes

[Jiri Olsa]

hi,
Stephan reported that we don't support period for
enabling large PEBS data, which there's no reason
for. Sending fix for that plus related perf tool
period fixes Stephan reported.

thanks,
jirka


---
Jiri Olsa (3):
      perf tools: Fix period/freq terms setup
      perf record: Fix period option handling
      x86/events/intel/ds: Add PERF_SAMPLE_PERIOD into PEBS_FREERUNNING_FLAGS

 arch/x86/events/perf_event.h |  3 ++-
 tools/perf/builtin-record.c  |  3 ++-
 tools/perf/perf.h            |  1 +
 tools/perf/util/evsel.c      | 13 ++++++++++---
 4 files changed, 15 insertions(+), 5 deletions(-)

[PATCH 1/3] perf tools: Fix period/freq terms setup

[Jiri Olsa]

Stephane reported that we don't set properly PERIOD
sample type for events with period term defined.

Before:
  $ perf record -e cpu/cpu-cycles,period=1000/u ls
  $ perf evlist -v
  cpu/cpu-cycles,period=1000/u: ... sample_type: IP|TID|TIME|PERIOD, ...

After:
  $ perf record -e cpu/cpu-cycles,period=1000/u ls
  $ perf evlist -v
  cpu/cpu-cycles,period=1000/u: ... sample_type: IP|TID|TIME, ...

Setting PERIOD sample type based on period term setup.

Reported-by: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-anrtntkwfto5rqulegfwitn5@git.kernel.org
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
 tools/perf/util/evsel.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 66fa45198a11..f2f2eaafde6d 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -745,12 +745,14 @@ static void apply_config_terms(struct perf_evsel *evsel,
 			if (!(term->weak && opts->user_interval != ULLONG_MAX)) {
 				attr->sample_period = term->val.period;
 				attr->freq = 0;
+				perf_evsel__reset_sample_bit(evsel, PERIOD);
 			}
 			break;
 		case PERF_EVSEL__CONFIG_TERM_FREQ:
 			if (!(term->weak && opts->user_freq != UINT_MAX)) {
 				attr->sample_freq = term->val.freq;
 				attr->freq = 1;
+				perf_evsel__set_sample_bit(evsel, PERIOD);
 			}
 			break;
 		case PERF_EVSEL__CONFIG_TERM_TIME:
-- 
2.13.6

Re: [PATCH 1/3] perf tools: Fix period/freq terms setup

[Stephane Eranian]

Jiri,

On Thu, Feb 1, 2018 at 12:38 AM, Jiri Olsa <jolsa@kernel.org> wrote:
> Stephane reported that we don't set properly PERIOD
> sample type for events with period term defined.
>
> Before:
>   $ perf record -e cpu/cpu-cycles,period=1000/u ls
>   $ perf evlist -v
>   cpu/cpu-cycles,period=1000/u: ... sample_type: IP|TID|TIME|PERIOD, ...
>
> After:
>   $ perf record -e cpu/cpu-cycles,period=1000/u ls
>   $ perf evlist -v
>   cpu/cpu-cycles,period=1000/u: ... sample_type: IP|TID|TIME, ...
>
> Setting PERIOD sample type based on period term setup.
>
there is still one problem remaining here. It has to do with the handling
of cycles:pp or :p or :ppp. Suppose I want to set a period for it while I am
also sampling on other events: Something like:

$ perf record -e
cycles:pp,instructions,cpu/event=0xd0,umaks=0x81,period=100000/ .....

I want to set the period for cycles:pp, but not for instructions. I
cannot use -c because
it would also force a period on instructions. I could use the raw hw
raw event code for cycles:pp.
But that does not work because recent kernels prevent use of hw
filters on events programmed
for PEBS, e.g., cpu/event=0xc2,umask=0x1,cmask=16,inv/pp is rejected.
PEBS does not support filters.
It works in the case of cycles:pp simply by the nature on the
underlying event and the stalls.

To get precise cycles, the only event syntax you can use is cycles:pp,
but then you cannot specify
an event-specific period. This needs to be fixed as well.

I'd like to be able to say:

$ perf record -e
cycles:pp:period=10000001,cpu/event=0xd0,umaks=0x81,period=100000/

Or something equivalent.

Otherwise, I tested what you have written so far and it works.

Thanks.


> Reported-by: Stephane Eranian <eranian@google.com>
> Link: http://lkml.kernel.org/n/tip-anrtntkwfto5rqulegfwitn5@git.kernel.org
> Signed-off-by: Jiri Olsa <jolsa@kernel.org>
> ---
>  tools/perf/util/evsel.c | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
> index 66fa45198a11..f2f2eaafde6d 100644
> --- a/tools/perf/util/evsel.c
> +++ b/tools/perf/util/evsel.c
> @@ -745,12 +745,14 @@ static void apply_config_terms(struct perf_evsel *evsel,
>                         if (!(term->weak && opts->user_interval != ULLONG_MAX)) {
>                                 attr->sample_period = term->val.period;
>                                 attr->freq = 0;
> +                               perf_evsel__reset_sample_bit(evsel, PERIOD);
>                         }
>                         break;
>                 case PERF_EVSEL__CONFIG_TERM_FREQ:
>                         if (!(term->weak && opts->user_freq != UINT_MAX)) {
>                                 attr->sample_freq = term->val.freq;
>                                 attr->freq = 1;
> +                               perf_evsel__set_sample_bit(evsel, PERIOD);
>                         }
>                         break;
>                 case PERF_EVSEL__CONFIG_TERM_TIME:
> --
> 2.13.6
>

Re: [PATCH 1/3] perf tools: Fix period/freq terms setup

[Arnaldo Carvalho de Melo]

Em Fri, Feb 02, 2018 at 10:45:46AM -0800, Stephane Eranian escreveu:
> Jiri,
> 
> On Thu, Feb 1, 2018 at 12:38 AM, Jiri Olsa <jolsa@kernel.org> wrote:
> > Stephane reported that we don't set properly PERIOD
> > sample type for events with period term defined.
> >
> > Before:
> >   $ perf record -e cpu/cpu-cycles,period=1000/u ls
> >   $ perf evlist -v
> >   cpu/cpu-cycles,period=1000/u: ... sample_type: IP|TID|TIME|PERIOD, ...
> >
> > After:
> >   $ perf record -e cpu/cpu-cycles,period=1000/u ls
> >   $ perf evlist -v
> >   cpu/cpu-cycles,period=1000/u: ... sample_type: IP|TID|TIME, ...
> >
> > Setting PERIOD sample type based on period term setup.
> >
> there is still one problem remaining here. It has to do with the handling
> of cycles:pp or :p or :ppp. Suppose I want to set a period for it while I am
> also sampling on other events: Something like:
> 
> $ perf record -e
> cycles:pp,instructions,cpu/event=0xd0,umaks=0x81,period=100000/ .....
> 
> I want to set the period for cycles:pp, but not for instructions. I
> cannot use -c because
> it would also force a period on instructions. I could use the raw hw
> raw event code for cycles:pp.
> But that does not work because recent kernels prevent use of hw
> filters on events programmed
> for PEBS, e.g., cpu/event=0xc2,umask=0x1,cmask=16,inv/pp is rejected.
> PEBS does not support filters.
> It works in the case of cycles:pp simply by the nature on the
> underlying event and the stalls.
> 
> To get precise cycles, the only event syntax you can use is cycles:pp,
> but then you cannot specify
> an event-specific period. This needs to be fixed as well.
> 
> I'd like to be able to say:
> 
> $ perf record -e
> cycles:pp:period=10000001,cpu/event=0xd0,umaks=0x81,period=100000/
> 
> Or something equivalent.
> 
> Otherwise, I tested what you have written so far and it works.

So I take that as a Tested-by: Stephane and will apply the patches, Jiri
can continue working on these other aspects, right?

- Arnaldo
 
> Thanks.
> 
> 
> > Reported-by: Stephane Eranian <eranian@google.com>
> > Link: http://lkml.kernel.org/n/tip-anrtntkwfto5rqulegfwitn5@git.kernel.org
> > Signed-off-by: Jiri Olsa <jolsa@kernel.org>
> > ---
> >  tools/perf/util/evsel.c | 2 ++
> >  1 file changed, 2 insertions(+)
> >
> > diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
> > index 66fa45198a11..f2f2eaafde6d 100644
> > --- a/tools/perf/util/evsel.c
> > +++ b/tools/perf/util/evsel.c
> > @@ -745,12 +745,14 @@ static void apply_config_terms(struct perf_evsel *evsel,
> >                         if (!(term->weak && opts->user_interval != ULLONG_MAX)) {
> >                                 attr->sample_period = term->val.period;
> >                                 attr->freq = 0;
> > +                               perf_evsel__reset_sample_bit(evsel, PERIOD);
> >                         }
> >                         break;
> >                 case PERF_EVSEL__CONFIG_TERM_FREQ:
> >                         if (!(term->weak && opts->user_freq != UINT_MAX)) {
> >                                 attr->sample_freq = term->val.freq;
> >                                 attr->freq = 1;
> > +                               perf_evsel__set_sample_bit(evsel, PERIOD);
> >                         }
> >                         break;
> >                 case PERF_EVSEL__CONFIG_TERM_TIME:
> > --
> > 2.13.6
> >

Re: [PATCH 1/3] perf tools: Fix period/freq terms setup

[Arnaldo Carvalho de Melo]

Em Fri, Feb 02, 2018 at 05:28:49PM -0300, Arnaldo Carvalho de Melo escreveu:
> Em Fri, Feb 02, 2018 at 10:45:46AM -0800, Stephane Eranian escreveu:
> > Otherwise, I tested what you have written so far and it works.
 
> So I take that as a Tested-by: Stephane and will apply the patches, Jiri
> can continue working on these other aspects, right?

I also added this for the casual reader to get up to speed more quickly,
please check that it makes sense.

    Committer note:
    
    When we use -c or a period=N term in the event definition, then we don't
    need to ask the kernel, via perf_event_attr.sample_type |=
    PERF_SAMPLE_PERIOD, to put the event period in each sample, as we know
    it already, it is in perf_event_attr.sample_period.

- Arnaldo

Re: [PATCH 1/3] perf tools: Fix period/freq terms setup

[Stephane Eranian]

On Fri, Feb 2, 2018 at 12:40 PM, Arnaldo Carvalho de Melo
<acme@kernel.org> wrote:
> Em Fri, Feb 02, 2018 at 05:28:49PM -0300, Arnaldo Carvalho de Melo escreveu:
>> Em Fri, Feb 02, 2018 at 10:45:46AM -0800, Stephane Eranian escreveu:
>> > Otherwise, I tested what you have written so far and it works.
>
>> So I take that as a Tested-by: Stephane and will apply the patches, Jiri
>> can continue working on these other aspects, right?
>
> I also added this for the casual reader to get up to speed more quickly,
> please check that it makes sense.
>
>     Committer note:
>
>     When we use -c or a period=N term in the event definition, then we don't
>     need to ask the kernel, via perf_event_attr.sample_type |=
>     PERF_SAMPLE_PERIOD, to put the event period in each sample, as we know
>     it already, it is in perf_event_attr.sample_period.
>
Not quite. It depends on how each event is setup. I can mix & match period
and frequency. The PERF_SAMPLE_PERIOD can be dropped only if all the
events use a fixed period either via period=N or -c.

I hope that perf report can deal with config mixing period and fixed
mode correctly.

Re: [PATCH 1/3] perf tools: Fix period/freq terms setup

[Jiri Olsa]

On Fri, Feb 02, 2018 at 01:04:34PM -0800, Stephane Eranian wrote:
> On Fri, Feb 2, 2018 at 12:40 PM, Arnaldo Carvalho de Melo
> <acme@kernel.org> wrote:
> > Em Fri, Feb 02, 2018 at 05:28:49PM -0300, Arnaldo Carvalho de Melo escreveu:
> >> Em Fri, Feb 02, 2018 at 10:45:46AM -0800, Stephane Eranian escreveu:
> >> > Otherwise, I tested what you have written so far and it works.
> >
> >> So I take that as a Tested-by: Stephane and will apply the patches, Jiri
> >> can continue working on these other aspects, right?
> >
> > I also added this for the casual reader to get up to speed more quickly,
> > please check that it makes sense.
> >
> >     Committer note:
> >
> >     When we use -c or a period=N term in the event definition, then we don't
> >     need to ask the kernel, via perf_event_attr.sample_type |=
> >     PERF_SAMPLE_PERIOD, to put the event period in each sample, as we know
> >     it already, it is in perf_event_attr.sample_period.
> >
> Not quite. It depends on how each event is setup. I can mix & match period
> and frequency. The PERF_SAMPLE_PERIOD can be dropped only if all the
> events use a fixed period either via period=N or -c.

I think you can have both period and freq based event in one session
if that's your concern..? what would be the problem?

jirka

> I hope that perf report can deal with config mixing period and fixed
> mode correctly.

Re: [PATCH 1/3] perf tools: Fix period/freq terms setup

[Stephane Eranian]

On Mon, Feb 5, 2018 at 7:17 AM, Jiri Olsa <jolsa@redhat.com> wrote:
> On Fri, Feb 02, 2018 at 01:04:34PM -0800, Stephane Eranian wrote:
>> On Fri, Feb 2, 2018 at 12:40 PM, Arnaldo Carvalho de Melo
>> <acme@kernel.org> wrote:
>> > Em Fri, Feb 02, 2018 at 05:28:49PM -0300, Arnaldo Carvalho de Melo escreveu:
>> >> Em Fri, Feb 02, 2018 at 10:45:46AM -0800, Stephane Eranian escreveu:
>> >> > Otherwise, I tested what you have written so far and it works.
>> >
>> >> So I take that as a Tested-by: Stephane and will apply the patches, Jiri
>> >> can continue working on these other aspects, right?
>> >
>> > I also added this for the casual reader to get up to speed more quickly,
>> > please check that it makes sense.
>> >
>> >     Committer note:
>> >
>> >     When we use -c or a period=N term in the event definition, then we don't
>> >     need to ask the kernel, via perf_event_attr.sample_type |=
>> >     PERF_SAMPLE_PERIOD, to put the event period in each sample, as we know
>> >     it already, it is in perf_event_attr.sample_period.
>> >
>> Not quite. It depends on how each event is setup. I can mix & match period
>> and frequency. The PERF_SAMPLE_PERIOD can be dropped only if all the
>> events use a fixed period either via period=N or -c.
>
> I think you can have both period and freq based event in one session
> if that's your concern..? what would be the problem?
>
My understanding was that perf only support configs where all events
have the same attr.sample_type. With frequency mode, you'd want the period
recorded in some cases.

> jirka
>
>> I hope that perf report can deal with config mixing period and fixed
>> mode correctly.

Re: [PATCH 1/3] perf tools: Fix period/freq terms setup

[Arnaldo Carvalho de Melo]

Em Mon, Feb 05, 2018 at 12:58:16PM -0800, Stephane Eranian escreveu:
> On Mon, Feb 5, 2018 at 7:17 AM, Jiri Olsa <jolsa@redhat.com> wrote:
> > On Fri, Feb 02, 2018 at 01:04:34PM -0800, Stephane Eranian wrote:
> >> On Fri, Feb 2, 2018 at 12:40 PM, Arnaldo Carvalho de Melo
> >> <acme@kernel.org> wrote:
> >> > Em Fri, Feb 02, 2018 at 05:28:49PM -0300, Arnaldo Carvalho de Melo escreveu:
> >> >> Em Fri, Feb 02, 2018 at 10:45:46AM -0800, Stephane Eranian escreveu:
> >> >> > Otherwise, I tested what you have written so far and it works.
> >> >
> >> >> So I take that as a Tested-by: Stephane and will apply the patches, Jiri
> >> >> can continue working on these other aspects, right?
> >> >
> >> > I also added this for the casual reader to get up to speed more quickly,
> >> > please check that it makes sense.
> >> >
> >> >     Committer note:
> >> >
> >> >     When we use -c or a period=N term in the event definition, then we don't
> >> >     need to ask the kernel, via perf_event_attr.sample_type |=
> >> >     PERF_SAMPLE_PERIOD, to put the event period in each sample, as we know
> >> >     it already, it is in perf_event_attr.sample_period.
> >> >
> >> Not quite. It depends on how each event is setup. I can mix & match period
> >> and frequency. The PERF_SAMPLE_PERIOD can be dropped only if all the
> >> events use a fixed period either via period=N or -c.

> > I think you can have both period and freq based event in one session
> > if that's your concern..? what would be the problem?

> My understanding was that perf only support configs where all events
> have the same attr.sample_type. With frequency mode, you'd want the period
> recorded in some cases.

[root@jouet ~]# perf record -e cycles/period=2/,instructions/freq=100/
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 1.895 MB perf.data (122 samples) ]

[root@jouet ~]# perf report
[root@jouet ~]# perf evlist -v
cycles/period=2/: size: 112, { sample_period, sample_freq }: 2, sample_type: IP|TID|TIME|CPU|IDENTIFIER, read_format: ID, disabled: 1, inherit: 1, mmap: 1, comm: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1
instructions/freq=100/: size: 112, config: 0x1, { sample_period, sample_freq }: 100, sample_type: IP|TID|TIME|CPU|PERIOD|IDENTIFIER, read_format: ID, disabled: 1, inherit: 1, freq: 1, sample_id_all: 1, exclude_guest: 1
[root@jouet ~]# 

commit ff3d527cebc1fa3707c617bfe9e74f53fcfb0955
Author: Adrian Hunter <adrian.hunter@intel.com>
Date:   Tue Aug 27 11:23:07 2013 +0300

    perf: make events stream always parsable
    
    The event stream is not always parsable because the format of a sample
    is dependent on the sample_type of the selected event.  When there is
    more than one selected event and the sample_types are not the same then
    parsing becomes problematic.  A sample can be matched to its selected
    event using the ID that is allocated when the event is opened.
    Unfortunately, to get the ID from the sample means first parsing it.
    
    This patch adds a new sample format bit PERF_SAMPLE_IDENTIFER that puts
    the ID at a fixed position so that the ID can be retrieved without
    parsing the sample.  For sample events, that is the first position
    immediately after the header.  For non-sample events, that is the last
    position.
    
    In this respect parsing samples requires that the sample_type and ID
    values are recorded.  For example, perf tools records struct
    perf_event_attr and the IDs within the perf.data file.  Those must be
    read first before it is possible to parse samples found later in the
    perf.data file.
    
    Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
    Tested-by: Stephane Eranian <eranian@google.com>
    Acked-by: Peter Zijlstra <peterz@infradead.org>

Re: [PATCH 1/3] perf tools: Fix period/freq terms setup

[Stephane Eranian]

On Mon, Feb 5, 2018 at 1:13 PM, Arnaldo Carvalho de Melo
<acme@kernel.org> wrote:
> Em Mon, Feb 05, 2018 at 12:58:16PM -0800, Stephane Eranian escreveu:
>> On Mon, Feb 5, 2018 at 7:17 AM, Jiri Olsa <jolsa@redhat.com> wrote:
>> > On Fri, Feb 02, 2018 at 01:04:34PM -0800, Stephane Eranian wrote:
>> >> On Fri, Feb 2, 2018 at 12:40 PM, Arnaldo Carvalho de Melo
>> >> <acme@kernel.org> wrote:
>> >> > Em Fri, Feb 02, 2018 at 05:28:49PM -0300, Arnaldo Carvalho de Melo escreveu:
>> >> >> Em Fri, Feb 02, 2018 at 10:45:46AM -0800, Stephane Eranian escreveu:
>> >> >> > Otherwise, I tested what you have written so far and it works.
>> >> >
>> >> >> So I take that as a Tested-by: Stephane and will apply the patches, Jiri
>> >> >> can continue working on these other aspects, right?
>> >> >
>> >> > I also added this for the casual reader to get up to speed more quickly,
>> >> > please check that it makes sense.
>> >> >
>> >> >     Committer note:
>> >> >
>> >> >     When we use -c or a period=N term in the event definition, then we don't
>> >> >     need to ask the kernel, via perf_event_attr.sample_type |=
>> >> >     PERF_SAMPLE_PERIOD, to put the event period in each sample, as we know
>> >> >     it already, it is in perf_event_attr.sample_period.
>> >> >
>> >> Not quite. It depends on how each event is setup. I can mix & match period
>> >> and frequency. The PERF_SAMPLE_PERIOD can be dropped only if all the
>> >> events use a fixed period either via period=N or -c.
>
>> > I think you can have both period and freq based event in one session
>> > if that's your concern..? what would be the problem?
>
>> My understanding was that perf only support configs where all events
>> have the same attr.sample_type. With frequency mode, you'd want the period
>> recorded in some cases.
>
> [root@jouet ~]# perf record -e cycles/period=2/,instructions/freq=100/
> ^C[ perf record: Woken up 1 times to write data ]
> [ perf record: Captured and wrote 1.895 MB perf.data (122 samples) ]
>
> [root@jouet ~]# perf report
> [root@jouet ~]# perf evlist -v
> cycles/period=2/: size: 112, { sample_period, sample_freq }: 2, sample_type: IP|TID|TIME|CPU|IDENTIFIER, read_format: ID, disabled: 1, inherit: 1, mmap: 1, comm: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1
> instructions/freq=100/: size: 112, config: 0x1, { sample_period, sample_freq }: 100, sample_type: IP|TID|TIME|CPU|PERIOD|IDENTIFIER, read_format: ID, disabled: 1, inherit: 1, freq: 1, sample_id_all: 1, exclude_guest: 1
> [root@jouet ~]#
>
Looks like this is working then, great!

Now, related to profiling and reporting. There is still an issue I
keep running into
with grouping. I want to sample on N events, where N > number of  hw counters.
Yet I want the same output as perf report --group, i.e., side-by-side
profiles as
opposed to showing me one event profile at a time (which is not very useful).

You should not require events to belong to the same group to support this. Many
other tools support such output (e.g., VTUNE, Gooda). It is still very
valuable even
though events may not have been measured at the same time.

Let me use a simple (and silly but portable) example.
Today if I do on  Intel x86:

 $ perf record -e branches,branches,branches,branches,branches my_test

And I do:

$ perf report --group
It will show me 5 distinct profiles.

I would like perf to show me a single profile where the 5 events are
side-by-side.

Similar to what I get if I do instead:
$ perf record -e '{branches,branches,branches,branches}' my_test
$ perf report --group

But here, I would have to ensure all events fits in a group to allow
the reporting
I want. So that would limit me to 4 events.

I think perf report --group should work regardless of how the events
were grouped.
Is there already a way to work around this?

Thanks.

> commit ff3d527cebc1fa3707c617bfe9e74f53fcfb0955
> Author: Adrian Hunter <adrian.hunter@intel.com>
> Date:   Tue Aug 27 11:23:07 2013 +0300
>
>     perf: make events stream always parsable
>
>     The event stream is not always parsable because the format of a sample
>     is dependent on the sample_type of the selected event.  When there is
>     more than one selected event and the sample_types are not the same then
>     parsing becomes problematic.  A sample can be matched to its selected
>     event using the ID that is allocated when the event is opened.
>     Unfortunately, to get the ID from the sample means first parsing it.
>
>     This patch adds a new sample format bit PERF_SAMPLE_IDENTIFER that puts
>     the ID at a fixed position so that the ID can be retrieved without
>     parsing the sample.  For sample events, that is the first position
>     immediately after the header.  For non-sample events, that is the last
>     position.
>
>     In this respect parsing samples requires that the sample_type and ID
>     values are recorded.  For example, perf tools records struct
>     perf_event_attr and the IDs within the perf.data file.  Those must be
>     read first before it is possible to parse samples found later in the
>     perf.data file.
>
>     Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
>     Tested-by: Stephane Eranian <eranian@google.com>
>     Acked-by: Peter Zijlstra <peterz@infradead.org>
>

Re: [PATCH 1/3] perf tools: Fix period/freq terms setup

[Jiri Olsa]

On Mon, Feb 05, 2018 at 06:51:05PM -0800, Stephane Eranian wrote:

SNIP

> >
> Looks like this is working then, great!
> 
> Now, related to profiling and reporting. There is still an issue I
> keep running into
> with grouping. I want to sample on N events, where N > number of  hw counters.
> Yet I want the same output as perf report --group, i.e., side-by-side
> profiles as
> opposed to showing me one event profile at a time (which is not very useful).
> 
> You should not require events to belong to the same group to support this. Many
> other tools support such output (e.g., VTUNE, Gooda). It is still very
> valuable even
> though events may not have been measured at the same time.
> 
> Let me use a simple (and silly but portable) example.
> Today if I do on  Intel x86:
> 
>  $ perf record -e branches,branches,branches,branches,branches my_test
> 
> And I do:
> 
> $ perf report --group
> It will show me 5 distinct profiles.
> 
> I would like perf to show me a single profile where the 5 events are
> side-by-side.
> 
> Similar to what I get if I do instead:
> $ perf record -e '{branches,branches,branches,branches}' my_test
> $ perf report --group
> 
> But here, I would have to ensure all events fits in a group to allow
> the reporting
> I want. So that would limit me to 4 events.
> 
> I think perf report --group should work regardless of how the events
> were grouped.
> Is there already a way to work around this?

no workaround.. please try attached patch, it seems
to work for what you described

thanks,
jirka


---
diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
index 4ad5dc649716..35a013992092 100644
--- a/tools/perf/builtin-report.c
+++ b/tools/perf/builtin-report.c
@@ -937,6 +937,7 @@ int cmd_report(int argc, const char **argv)
 		"perf report [<options>]",
 		NULL
 	};
+	bool group_set = false;
 	struct report report = {
 		.tool = {
 			.sample		 = process_sample_event,
@@ -1056,7 +1057,7 @@ int cmd_report(int argc, const char **argv)
 		   "Specify disassembler style (e.g. -M intel for intel syntax)"),
 	OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
 		    "Show a column with the sum of periods"),
-	OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
+	OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &group_set,
 		    "Show event group information together"),
 	OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
 		    "use branch records for per branch histogram filling",
@@ -1173,6 +1174,9 @@ int cmd_report(int argc, const char **argv)
 	has_br_stack = perf_header__has_feat(&session->header,
 					     HEADER_BRANCH_STACK);
 
+	if (group_set && !session->evlist->nr_groups)
+		perf_evlist__set_leader(session->evlist);
+
 	if (itrace_synth_opts.last_branch)
 		has_br_stack = true;
 

Re: [PATCH 1/3] perf tools: Fix period/freq terms setup

[Stephane Eranian]

On Tue, Feb 6, 2018 at 1:35 AM, Jiri Olsa <jolsa@redhat.com> wrote:
> On Mon, Feb 05, 2018 at 06:51:05PM -0800, Stephane Eranian wrote:
>
> SNIP
>
>> >
>> Looks like this is working then, great!
>>
>> Now, related to profiling and reporting. There is still an issue I
>> keep running into
>> with grouping. I want to sample on N events, where N > number of  hw counters.
>> Yet I want the same output as perf report --group, i.e., side-by-side
>> profiles as
>> opposed to showing me one event profile at a time (which is not very useful).
>>
>> You should not require events to belong to the same group to support this. Many
>> other tools support such output (e.g., VTUNE, Gooda). It is still very
>> valuable even
>> though events may not have been measured at the same time.
>>
>> Let me use a simple (and silly but portable) example.
>> Today if I do on  Intel x86:
>>
>>  $ perf record -e branches,branches,branches,branches,branches my_test
>>
>> And I do:
>>
>> $ perf report --group
>> It will show me 5 distinct profiles.
>>
>> I would like perf to show me a single profile where the 5 events are
>> side-by-side.
>>
>> Similar to what I get if I do instead:
>> $ perf record -e '{branches,branches,branches,branches}' my_test
>> $ perf report --group
>>
>> But here, I would have to ensure all events fits in a group to allow
>> the reporting
>> I want. So that would limit me to 4 events.
>>
>> I think perf report --group should work regardless of how the events
>> were grouped.
>> Is there already a way to work around this?
>
> no workaround.. please try attached patch, it seems
> to work for what you described
>
Works for me. That's great!
Thanks.

Tested-By: Stephane Eranian <eranian@google.com>

> thanks,
> jirka
>
>
> ---
> diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
> index 4ad5dc649716..35a013992092 100644
> --- a/tools/perf/builtin-report.c
> +++ b/tools/perf/builtin-report.c
> @@ -937,6 +937,7 @@ int cmd_report(int argc, const char **argv)
>                 "perf report [<options>]",
>                 NULL
>         };
> +       bool group_set = false;
>         struct report report = {
>                 .tool = {
>                         .sample          = process_sample_event,
> @@ -1056,7 +1057,7 @@ int cmd_report(int argc, const char **argv)
>                    "Specify disassembler style (e.g. -M intel for intel syntax)"),
>         OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
>                     "Show a column with the sum of periods"),
> -       OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
> +       OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &group_set,
>                     "Show event group information together"),
>         OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
>                     "use branch records for per branch histogram filling",
> @@ -1173,6 +1174,9 @@ int cmd_report(int argc, const char **argv)
>         has_br_stack = perf_header__has_feat(&session->header,
>                                              HEADER_BRANCH_STACK);
>
> +       if (group_set && !session->evlist->nr_groups)
> +               perf_evlist__set_leader(session->evlist);
> +
>         if (itrace_synth_opts.last_branch)
>                 has_br_stack = true;
>

[PATCH] perf report: Add support to display group output for non group events

[Jiri Olsa]

On Wed, Feb 07, 2018 at 10:52:35AM -0800, Stephane Eranian wrote:

SNIP

> >> Similar to what I get if I do instead:
> >> $ perf record -e '{branches,branches,branches,branches}' my_test
> >> $ perf report --group
> >>
> >> But here, I would have to ensure all events fits in a group to allow
> >> the reporting
> >> I want. So that would limit me to 4 events.
> >>
> >> I think perf report --group should work regardless of how the events
> >> were grouped.
> >> Is there already a way to work around this?
> >
> > no workaround.. please try attached patch, it seems
> > to work for what you described
> >
> Works for me. That's great!
> Thanks.
> 
> Tested-By: Stephane Eranian <eranian@google.com>

thanks, full patch attached

jirka


---
Add support to display group output for if non grouped events
are detected and user forces --group option. Now for non-group
events recorded like:

  $ perf record -e 'cycles,instructions' ls

you can still get group output by using --group option
in report:

  $ perf report --group --stdio
  ...
  #         Overhead  Command  Shared Object     Symbol
  # ................  .......  ................  ......................
  #
      17.67%   0.00%  ls       libc-2.25.so      [.] _IO_do_write@@GLIB
      15.59%  25.94%  ls       ls                [.] calculate_columns
      15.41%  31.35%  ls       libc-2.25.so      [.] __strcoll_l
  ...

Requested-and-Tested-by: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-m1ffikw8c3a55b3uaxrmk5w3@git.kernel.org
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
 tools/perf/Documentation/perf-report.txt | 3 ++-
 tools/perf/builtin-report.c              | 6 +++++-
 2 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/tools/perf/Documentation/perf-report.txt b/tools/perf/Documentation/perf-report.txt
index 907e505b6309..a76b871f78a6 100644
--- a/tools/perf/Documentation/perf-report.txt
+++ b/tools/perf/Documentation/perf-report.txt
@@ -354,7 +354,8 @@ OPTIONS
         Path to objdump binary.
 
 --group::
-	Show event group information together.
+	Show event group information together. It forces group output also
+	if there are no groups defined in data file.
 
 --demangle::
 	Demangle symbol names to human readable form. It's enabled by default,
diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
index 8ef71669e7a0..1eedb1815c4c 100644
--- a/tools/perf/builtin-report.c
+++ b/tools/perf/builtin-report.c
@@ -938,6 +938,7 @@ int cmd_report(int argc, const char **argv)
 		"perf report [<options>]",
 		NULL
 	};
+	bool group_set = false;
 	struct report report = {
 		.tool = {
 			.sample		 = process_sample_event,
@@ -1057,7 +1058,7 @@ int cmd_report(int argc, const char **argv)
 		   "Specify disassembler style (e.g. -M intel for intel syntax)"),
 	OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
 		    "Show a column with the sum of periods"),
-	OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
+	OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &group_set,
 		    "Show event group information together"),
 	OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
 		    "use branch records for per branch histogram filling",
@@ -1174,6 +1175,9 @@ int cmd_report(int argc, const char **argv)
 	has_br_stack = perf_header__has_feat(&session->header,
 					     HEADER_BRANCH_STACK);
 
+	if (group_set && !session->evlist->nr_groups)
+		perf_evlist__set_leader(session->evlist);
+
 	if (itrace_synth_opts.last_branch)
 		has_br_stack = true;
 
-- 
2.13.6

Re: [PATCH] perf report: Add support to display group output for non group events

[Arnaldo Carvalho de Melo]

Em Fri, Feb 09, 2018 at 10:27:34AM +0100, Jiri Olsa escreveu:
> On Wed, Feb 07, 2018 at 10:52:35AM -0800, Stephane Eranian wrote:
> 
> SNIP
> 
> > >> Similar to what I get if I do instead:
> > >> $ perf record -e '{branches,branches,branches,branches}' my_test
> > >> $ perf report --group
> > >>
> > >> But here, I would have to ensure all events fits in a group to allow
> > >> the reporting
> > >> I want. So that would limit me to 4 events.
> > >>
> > >> I think perf report --group should work regardless of how the events
> > >> were grouped.
> > >> Is there already a way to work around this?
> > >
> > > no workaround.. please try attached patch, it seems
> > > to work for what you described
> > >
> > Works for me. That's great!
> > Thanks.
> > 
> > Tested-By: Stephane Eranian <eranian@google.com>
> 
> thanks, full patch attached
> 
> jirka

Humm, its a nice hack, but it would be even better if it didn't showed
it as if it was really a group:

  Samples: 20  of event 'anon group { cycles, instructions }', Event count (approx.): 4712980

It would be better to instead add another condition to the evlist that
would trigger the view with all the examples...

I'm applying it anyway, as it is useful, but would be nice to have the
same output except for that header, that should read instead:

  Samples: 20  of non grouped events: cycles, instructions, Event count (approx.): 4712980

- Arnaldo

 
> 
> ---
> Add support to display group output for if non grouped events
> are detected and user forces --group option. Now for non-group
> events recorded like:
> 
>   $ perf record -e 'cycles,instructions' ls
> 
> you can still get group output by using --group option
> in report:
> 
>   $ perf report --group --stdio
>   ...
>   #         Overhead  Command  Shared Object     Symbol
>   # ................  .......  ................  ......................
>   #
>       17.67%   0.00%  ls       libc-2.25.so      [.] _IO_do_write@@GLIB
>       15.59%  25.94%  ls       ls                [.] calculate_columns
>       15.41%  31.35%  ls       libc-2.25.so      [.] __strcoll_l
>   ...
> 
> Requested-and-Tested-by: Stephane Eranian <eranian@google.com>
> Link: http://lkml.kernel.org/n/tip-m1ffikw8c3a55b3uaxrmk5w3@git.kernel.org
> Signed-off-by: Jiri Olsa <jolsa@kernel.org>
> ---
>  tools/perf/Documentation/perf-report.txt | 3 ++-
>  tools/perf/builtin-report.c              | 6 +++++-
>  2 files changed, 7 insertions(+), 2 deletions(-)
> 
> diff --git a/tools/perf/Documentation/perf-report.txt b/tools/perf/Documentation/perf-report.txt
> index 907e505b6309..a76b871f78a6 100644
> --- a/tools/perf/Documentation/perf-report.txt
> +++ b/tools/perf/Documentation/perf-report.txt
> @@ -354,7 +354,8 @@ OPTIONS
>          Path to objdump binary.
>  
>  --group::
> -	Show event group information together.
> +	Show event group information together. It forces group output also
> +	if there are no groups defined in data file.
>  
>  --demangle::
>  	Demangle symbol names to human readable form. It's enabled by default,
> diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
> index 8ef71669e7a0..1eedb1815c4c 100644
> --- a/tools/perf/builtin-report.c
> +++ b/tools/perf/builtin-report.c
> @@ -938,6 +938,7 @@ int cmd_report(int argc, const char **argv)
>  		"perf report [<options>]",
>  		NULL
>  	};
> +	bool group_set = false;
>  	struct report report = {
>  		.tool = {
>  			.sample		 = process_sample_event,
> @@ -1057,7 +1058,7 @@ int cmd_report(int argc, const char **argv)
>  		   "Specify disassembler style (e.g. -M intel for intel syntax)"),
>  	OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
>  		    "Show a column with the sum of periods"),
> -	OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
> +	OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &group_set,
>  		    "Show event group information together"),
>  	OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
>  		    "use branch records for per branch histogram filling",
> @@ -1174,6 +1175,9 @@ int cmd_report(int argc, const char **argv)
>  	has_br_stack = perf_header__has_feat(&session->header,
>  					     HEADER_BRANCH_STACK);
>  
> +	if (group_set && !session->evlist->nr_groups)
> +		perf_evlist__set_leader(session->evlist);
> +
>  	if (itrace_synth_opts.last_branch)
>  		has_br_stack = true;
>  
> -- 
> 2.13.6

Re: [PATCH] perf report: Add support to display group output for non group events

[Arnaldo Carvalho de Melo]

Em Fri, Feb 09, 2018 at 03:37:11PM -0300, Arnaldo Carvalho de Melo escreveu:
> Em Fri, Feb 09, 2018 at 10:27:34AM +0100, Jiri Olsa escreveu:
> > On Wed, Feb 07, 2018 at 10:52:35AM -0800, Stephane Eranian wrote:
> > 
> > SNIP
> > 
> > > >> Similar to what I get if I do instead:
> > > >> $ perf record -e '{branches,branches,branches,branches}' my_test
> > > >> $ perf report --group
> > > >>
> > > >> But here, I would have to ensure all events fits in a group to allow
> > > >> the reporting
> > > >> I want. So that would limit me to 4 events.
> > > >>
> > > >> I think perf report --group should work regardless of how the events
> > > >> were grouped.
> > > >> Is there already a way to work around this?
> > > >
> > > > no workaround.. please try attached patch, it seems
> > > > to work for what you described
> > > >
> > > Works for me. That's great!
> > > Thanks.
> > > 
> > > Tested-By: Stephane Eranian <eranian@google.com>
> > 
> > thanks, full patch attached
> > 
> > jirka
> 
> Humm, its a nice hack, but it would be even better if it didn't showed
> it as if it was really a group:
> 
>   Samples: 20  of event 'anon group { cycles, instructions }', Event count (approx.): 4712980
> 
> It would be better to instead add another condition to the evlist that
> would trigger the view with all the examples...
> 
> I'm applying it anyway, as it is useful, but would be nice to have the
> same output except for that header, that should read instead:
> 
>   Samples: 20  of non grouped events: cycles, instructions, Event count (approx.): 4712980

Till then, that documentation hunk should help understanding this issue. :-)

- Arnaldo

Re: [PATCH] perf report: Add support to display group output for non group events

[Jiri Olsa]

On Fri, Feb 09, 2018 at 03:37:11PM -0300, Arnaldo Carvalho de Melo wrote:
> Em Fri, Feb 09, 2018 at 10:27:34AM +0100, Jiri Olsa escreveu:
> > On Wed, Feb 07, 2018 at 10:52:35AM -0800, Stephane Eranian wrote:
> > 
> > SNIP
> > 
> > > >> Similar to what I get if I do instead:
> > > >> $ perf record -e '{branches,branches,branches,branches}' my_test
> > > >> $ perf report --group
> > > >>
> > > >> But here, I would have to ensure all events fits in a group to allow
> > > >> the reporting
> > > >> I want. So that would limit me to 4 events.
> > > >>
> > > >> I think perf report --group should work regardless of how the events
> > > >> were grouped.
> > > >> Is there already a way to work around this?
> > > >
> > > > no workaround.. please try attached patch, it seems
> > > > to work for what you described
> > > >
> > > Works for me. That's great!
> > > Thanks.
> > > 
> > > Tested-By: Stephane Eranian <eranian@google.com>
> > 
> > thanks, full patch attached
> > 
> > jirka
> 
> Humm, its a nice hack, but it would be even better if it didn't showed
> it as if it was really a group:
> 
>   Samples: 20  of event 'anon group { cycles, instructions }', Event count (approx.): 4712980
> 
> It would be better to instead add another condition to the evlist that
> would trigger the view with all the examples...
> 
> I'm applying it anyway, as it is useful, but would be nice to have the
> same output except for that header, that should read instead:
> 
>   Samples: 20  of non grouped events: cycles, instructions, Event count (approx.): 4712980

right, I'll try to post something like that

thanks,
jirka

Re: [PATCH] perf report: Add support to display group output for non group events

[Arnaldo Carvalho de Melo]

Em Fri, Feb 09, 2018 at 08:10:57PM +0100, Jiri Olsa escreveu:
> On Fri, Feb 09, 2018 at 03:37:11PM -0300, Arnaldo Carvalho de Melo wrote:
> > Em Fri, Feb 09, 2018 at 10:27:34AM +0100, Jiri Olsa escreveu:

> > Humm, its a nice hack, but it would be even better if it didn't showed
> > it as if it was really a group:

> >   Samples: 20  of event 'anon group { cycles, instructions }', Event count (approx.): 4712980

> > It would be better to instead add another condition to the evlist that
> > would trigger the view with all the examples...

> > I'm applying it anyway, as it is useful, but would be nice to have the
> > same output except for that header, that should read instead:

> >   Samples: 20  of non grouped events: cycles, instructions, Event count (approx.): 4712980

> right, I'll try to post something like that

I've pushed out with this patch, so please go on from there.

Thanks!

- Arnaldo

[tip:perf/core] perf report: Add support to display group output for non group events

[tip-bot for Jiri Olsa]

Commit-ID:  ad52b8cb4886f572b147b02f4c59a648bbf05f9c
Gitweb:     https://git.kernel.org/tip/ad52b8cb4886f572b147b02f4c59a648bbf05f9c
Author:     Jiri Olsa <jolsa@redhat.com>
AuthorDate: Fri, 9 Feb 2018 10:27:34 +0100
Committer:  Arnaldo Carvalho de Melo <acme@redhat.com>
CommitDate: Fri, 16 Feb 2018 10:09:24 -0300

perf report: Add support to display group output for non group events

Add support to display group output for if non grouped events are
detected and user forces --group option. Now for non-group events
recorded like:

  $ perf record -e 'cycles,instructions' ls

you can still get group output by using --group option
in report:

  $ perf report --group --stdio
  ...
  #         Overhead  Command  Shared Object     Symbol
  # ................  .......  ................  ......................
  #
      17.67%   0.00%  ls       libc-2.25.so      [.] _IO_do_write@@GLIB
      15.59%  25.94%  ls       ls                [.] calculate_columns
      15.41%  31.35%  ls       libc-2.25.so      [.] __strcoll_l
  ...

Committer note:

We should improve on this by making sure that the first line states that
this is not a group, but since the user doesn't have to force group view
when really using grouped events (e.g. '{cycles,instructions}'), the
user better know what is being done...

Requested-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Tested-by: Stephane Eranian <eranian@google.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180209092734.GB20449@krava
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/Documentation/perf-report.txt | 3 ++-
 tools/perf/builtin-report.c              | 6 +++++-
 2 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/tools/perf/Documentation/perf-report.txt b/tools/perf/Documentation/perf-report.txt
index 907e505..a76b871 100644
--- a/tools/perf/Documentation/perf-report.txt
+++ b/tools/perf/Documentation/perf-report.txt
@@ -354,7 +354,8 @@ OPTIONS
         Path to objdump binary.
 
 --group::
-	Show event group information together.
+	Show event group information together. It forces group output also
+	if there are no groups defined in data file.
 
 --demangle::
 	Demangle symbol names to human readable form. It's enabled by default,
diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
index 8ef7166..1eedb18 100644
--- a/tools/perf/builtin-report.c
+++ b/tools/perf/builtin-report.c
@@ -938,6 +938,7 @@ int cmd_report(int argc, const char **argv)
 		"perf report [<options>]",
 		NULL
 	};
+	bool group_set = false;
 	struct report report = {
 		.tool = {
 			.sample		 = process_sample_event,
@@ -1057,7 +1058,7 @@ int cmd_report(int argc, const char **argv)
 		   "Specify disassembler style (e.g. -M intel for intel syntax)"),
 	OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
 		    "Show a column with the sum of periods"),
-	OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
+	OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &group_set,
 		    "Show event group information together"),
 	OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
 		    "use branch records for per branch histogram filling",
@@ -1174,6 +1175,9 @@ repeat:
 	has_br_stack = perf_header__has_feat(&session->header,
 					     HEADER_BRANCH_STACK);
 
+	if (group_set && !session->evlist->nr_groups)
+		perf_evlist__set_leader(session->evlist);
+
 	if (itrace_synth_opts.last_branch)
 		has_br_stack = true;
 

Re: [PATCH 1/3] perf tools: Fix period/freq terms setup

[Jiri Olsa]

On Fri, Feb 02, 2018 at 10:45:46AM -0800, Stephane Eranian wrote:
> Jiri,
> 
> On Thu, Feb 1, 2018 at 12:38 AM, Jiri Olsa <jolsa@kernel.org> wrote:
> > Stephane reported that we don't set properly PERIOD
> > sample type for events with period term defined.
> >
> > Before:
> >   $ perf record -e cpu/cpu-cycles,period=1000/u ls
> >   $ perf evlist -v
> >   cpu/cpu-cycles,period=1000/u: ... sample_type: IP|TID|TIME|PERIOD, ...
> >
> > After:
> >   $ perf record -e cpu/cpu-cycles,period=1000/u ls
> >   $ perf evlist -v
> >   cpu/cpu-cycles,period=1000/u: ... sample_type: IP|TID|TIME, ...
> >
> > Setting PERIOD sample type based on period term setup.
> >
> there is still one problem remaining here. It has to do with the handling
> of cycles:pp or :p or :ppp. Suppose I want to set a period for it while I am
> also sampling on other events: Something like:
> 
> $ perf record -e
> cycles:pp,instructions,cpu/event=0xd0,umaks=0x81,period=100000/ .....
> 
> I want to set the period for cycles:pp, but not for instructions. I
> cannot use -c because
> it would also force a period on instructions. I could use the raw hw
> raw event code for cycles:pp.
> But that does not work because recent kernels prevent use of hw
> filters on events programmed
> for PEBS, e.g., cpu/event=0xc2,umask=0x1,cmask=16,inv/pp is rejected.
> PEBS does not support filters.
> It works in the case of cycles:pp simply by the nature on the
> underlying event and the stalls.
> 
> To get precise cycles, the only event syntax you can use is cycles:pp,
> but then you cannot specify
> an event-specific period. This needs to be fixed as well.

you can use p modifier behind like: cpu/.../pp

> 
> I'd like to be able to say:
> 
> $ perf record -e
> cycles:pp:period=10000001,cpu/event=0xd0,umaks=0x81,period=100000/
> 
> Or something equivalent.

and you can specify terms for hw events like 'cycles'

[jolsa@krava perf]$ sudo ./perf record -e 'cycles/period=10000001/pp,cpu/event=0xd0,umask=0x81,period=100000/'
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.579 MB perf.data (722 samples) ]

[jolsa@krava perf]$ ./perf evlist -v
cycles/period=10000001/pp: size: 112, { sample_period, sample_freq }: 10000001, sample_type: IP|TID|TIME|ID|CPU, read_format: ID, disabled: 1, inherit: 1, mmap: 1, comm: 1, task: 1, precise_ip: 2, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1
cpu/event=0xd0,umask=0x81,period=100000/: type: 4, size: 112, config: 0x81d0, { sample_period, sample_freq }: 100000, sample_type: IP|TID|TIME|ID|CPU, read_format: ID, disabled: 1, inherit: 1, sample_id_all: 1, exclude_guest: 1

jirka

Re: [PATCH 1/3] perf tools: Fix period/freq terms setup

[Stephane Eranian]

On Sat, Feb 3, 2018 at 7:30 AM, Jiri Olsa <jolsa@redhat.com> wrote:
> On Fri, Feb 02, 2018 at 10:45:46AM -0800, Stephane Eranian wrote:
>> Jiri,
>>
>> On Thu, Feb 1, 2018 at 12:38 AM, Jiri Olsa <jolsa@kernel.org> wrote:
>> > Stephane reported that we don't set properly PERIOD
>> > sample type for events with period term defined.
>> >
>> > Before:
>> >   $ perf record -e cpu/cpu-cycles,period=1000/u ls
>> >   $ perf evlist -v
>> >   cpu/cpu-cycles,period=1000/u: ... sample_type: IP|TID|TIME|PERIOD, ...
>> >
>> > After:
>> >   $ perf record -e cpu/cpu-cycles,period=1000/u ls
>> >   $ perf evlist -v
>> >   cpu/cpu-cycles,period=1000/u: ... sample_type: IP|TID|TIME, ...
>> >
>> > Setting PERIOD sample type based on period term setup.
>> >
>> there is still one problem remaining here. It has to do with the handling
>> of cycles:pp or :p or :ppp. Suppose I want to set a period for it while I am
>> also sampling on other events: Something like:
>>
>> $ perf record -e
>> cycles:pp,instructions,cpu/event=0xd0,umaks=0x81,period=100000/ .....
>>
>> I want to set the period for cycles:pp, but not for instructions. I
>> cannot use -c because
>> it would also force a period on instructions. I could use the raw hw
>> raw event code for cycles:pp.
>> But that does not work because recent kernels prevent use of hw
>> filters on events programmed
>> for PEBS, e.g., cpu/event=0xc2,umask=0x1,cmask=16,inv/pp is rejected.
>> PEBS does not support filters.
>> It works in the case of cycles:pp simply by the nature on the
>> underlying event and the stalls.
>>
>> To get precise cycles, the only event syntax you can use is cycles:pp,
>> but then you cannot specify
>> an event-specific period. This needs to be fixed as well.
>
> you can use p modifier behind like: cpu/.../pp
>
>>
>> I'd like to be able to say:
>>
>> $ perf record -e
>> cycles:pp:period=10000001,cpu/event=0xd0,umaks=0x81,period=100000/
>>
>> Or something equivalent.
>
> and you can specify terms for hw events like 'cycles'
>
> [jolsa@krava perf]$ sudo ./perf record -e 'cycles/period=10000001/pp,cpu/event=0xd0,umask=0x81,period=100000/'
> ^C[ perf record: Woken up 1 times to write data ]
> [ perf record: Captured and wrote 0.579 MB perf.data (722 samples) ]
>
Ok, I did not know about this syntax. It looks bizarre because you are
using an event name as a PMU instance.
Works for me.
Thanks.

> [jolsa@krava perf]$ ./perf evlist -v
> cycles/period=10000001/pp: size: 112, { sample_period, sample_freq }: 10000001, sample_type: IP|TID|TIME|ID|CPU, read_format: ID, disabled: 1, inherit: 1, mmap: 1, comm: 1, task: 1, precise_ip: 2, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1
> cpu/event=0xd0,umask=0x81,period=100000/: type: 4, size: 112, config: 0x81d0, { sample_period, sample_freq }: 100000, sample_type: IP|TID|TIME|ID|CPU, read_format: ID, disabled: 1, inherit: 1, sample_id_all: 1, exclude_guest: 1
>
> jirka

[tip:perf/urgent] perf evsel: Fix period/freq terms setup

[tip-bot for Jiri Olsa]

Commit-ID:  49c0ae80eb32426fa133246200628e529067c595
Gitweb:     https://git.kernel.org/tip/49c0ae80eb32426fa133246200628e529067c595
Author:     Jiri Olsa <jolsa@kernel.org>
AuthorDate: Thu, 1 Feb 2018 09:38:10 +0100
Committer:  Arnaldo Carvalho de Melo <acme@redhat.com>
CommitDate: Mon, 5 Feb 2018 12:11:58 -0300

perf evsel: Fix period/freq terms setup

Stephane reported that we don't set properly PERIOD sample type for
events with period term defined.

Before:
  $ perf record -e cpu/cpu-cycles,period=1000/u ls
  $ perf evlist -v
  cpu/cpu-cycles,period=1000/u: ... sample_type: IP|TID|TIME|PERIOD, ...

After:
  $ perf record -e cpu/cpu-cycles,period=1000/u ls
  $ perf evlist -v
  cpu/cpu-cycles,period=1000/u: ... sample_type: IP|TID|TIME, ...

Setting PERIOD sample type based on period term setup.

Committer note:

When we use -c or a period=N term in the event definition, then we don't
need to ask the kernel, for this event, via perf_event_attr.sample_type
|= PERF_SAMPLE_PERIOD, to put the event period in each sample for this
event, as we know it already, it is in perf_event_attr.sample_period.

Reported-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Stephane Eranian <eranian@google.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180201083812.11359-2-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/util/evsel.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 66fa451..f2f2eaa 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -745,12 +745,14 @@ static void apply_config_terms(struct perf_evsel *evsel,
 			if (!(term->weak && opts->user_interval != ULLONG_MAX)) {
 				attr->sample_period = term->val.period;
 				attr->freq = 0;
+				perf_evsel__reset_sample_bit(evsel, PERIOD);
 			}
 			break;
 		case PERF_EVSEL__CONFIG_TERM_FREQ:
 			if (!(term->weak && opts->user_freq != UINT_MAX)) {
 				attr->sample_freq = term->val.freq;
 				attr->freq = 1;
+				perf_evsel__set_sample_bit(evsel, PERIOD);
 			}
 			break;
 		case PERF_EVSEL__CONFIG_TERM_TIME:

[PATCH 2/3] perf record: Fix period option handling

[Jiri Olsa]

Stephan reported we don't unset PERIOD sample type
when --no-period is specified. Adding the unset
check and reset PERIOD if --no-period is specified.

Reported-by: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-b89ggszz9x26hm7ztor29b0s@git.kernel.org
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
 tools/perf/builtin-record.c |  3 ++-
 tools/perf/perf.h           |  1 +
 tools/perf/util/evsel.c     | 11 ++++++++---
 3 files changed, 11 insertions(+), 4 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index f251e824edac..907267206973 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -1566,7 +1566,8 @@ static struct option __record_options[] = {
 	OPT_BOOLEAN_SET('T', "timestamp", &record.opts.sample_time,
 			&record.opts.sample_time_set,
 			"Record the sample timestamps"),
-	OPT_BOOLEAN('P', "period", &record.opts.period, "Record the sample period"),
+	OPT_BOOLEAN_SET('P', "period", &record.opts.period, &record.opts.period_set,
+			"Record the sample period"),
 	OPT_BOOLEAN('n', "no-samples", &record.opts.no_samples,
 		    "don't sample"),
 	OPT_BOOLEAN_SET('N', "no-buildid-cache", &record.no_buildid_cache,
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index 2357f4ccc9c7..cfe46236a5e5 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -50,6 +50,7 @@ struct record_opts {
 	bool	     sample_time_set;
 	bool	     sample_cpu;
 	bool	     period;
+	bool	     period_set;
 	bool	     running_time;
 	bool	     full_auxtrace;
 	bool	     auxtrace_snapshot_mode;
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index f2f2eaafde6d..ff359c9ece2e 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -971,9 +971,6 @@ void perf_evsel__config(struct perf_evsel *evsel, struct record_opts *opts,
 	if (target__has_cpu(&opts->target) || opts->sample_cpu)
 		perf_evsel__set_sample_bit(evsel, CPU);
 
-	if (opts->period)
-		perf_evsel__set_sample_bit(evsel, PERIOD);
-
 	/*
 	 * When the user explicitly disabled time don't force it here.
 	 */
@@ -1075,6 +1072,14 @@ void perf_evsel__config(struct perf_evsel *evsel, struct record_opts *opts,
 	apply_config_terms(evsel, opts, track);
 
 	evsel->ignore_missing_thread = opts->ignore_missing_thread;
+
+	/* The --period option takes the precedence. */
+	if (opts->period_set) {
+		if (opts->period)
+			perf_evsel__set_sample_bit(evsel, PERIOD);
+		else
+			perf_evsel__reset_sample_bit(evsel, PERIOD);
+	}
 }
 
 static int perf_evsel__alloc_fd(struct perf_evsel *evsel, int ncpus, int nthreads)
-- 
2.13.6

[tip:perf/urgent] perf record: Fix period option handling

[tip-bot for Jiri Olsa]

Commit-ID:  f290aa1ffa45ed7e37599840878b4dae68269ee1
Gitweb:     https://git.kernel.org/tip/f290aa1ffa45ed7e37599840878b4dae68269ee1
Author:     Jiri Olsa <jolsa@kernel.org>
AuthorDate: Thu, 1 Feb 2018 09:38:11 +0100
Committer:  Arnaldo Carvalho de Melo <acme@redhat.com>
CommitDate: Mon, 5 Feb 2018 12:18:28 -0300

perf record: Fix period option handling

Stephan reported we don't unset PERIOD sample type when --no-period is
specified. Adding the unset check and reset PERIOD if --no-period is
specified.

Committer notes:

Check the sample_type, it shouldn't have PERF_SAMPLE_PERIOD there when
--no-period is used.

Before:

  # perf record --no-period sleep 1
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.018 MB perf.data (7 samples) ]
  # perf evlist -v
  cycles:ppp: size: 112, { sample_period, sample_freq }: 4000, sample_type: IP|TID|TIME|PERIOD, disabled: 1, inherit: 1, mmap: 1, comm: 1, freq: 1, enable_on_exec: 1, task: 1, precise_ip: 3, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1
  #

After:

[root@jouet ~]# perf record --no-period sleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.019 MB perf.data (17 samples) ]
[root@jouet ~]# perf evlist -v
cycles:ppp: size: 112, { sample_period, sample_freq }: 4000, sample_type: IP|TID|TIME, disabled: 1, inherit: 1, mmap: 1, comm: 1, freq: 1, enable_on_exec: 1, task: 1, precise_ip: 3, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1
[root@jouet ~]#

Reported-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Tested-by: Stephane Eranian <eranian@google.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180201083812.11359-3-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/builtin-record.c |  3 ++-
 tools/perf/perf.h           |  1 +
 tools/perf/util/evsel.c     | 11 ++++++++---
 3 files changed, 11 insertions(+), 4 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 65681a1..bf4ca74 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -1566,7 +1566,8 @@ static struct option __record_options[] = {
 	OPT_BOOLEAN_SET('T', "timestamp", &record.opts.sample_time,
 			&record.opts.sample_time_set,
 			"Record the sample timestamps"),
-	OPT_BOOLEAN('P', "period", &record.opts.period, "Record the sample period"),
+	OPT_BOOLEAN_SET('P', "period", &record.opts.period, &record.opts.period_set,
+			"Record the sample period"),
 	OPT_BOOLEAN('n', "no-samples", &record.opts.no_samples,
 		    "don't sample"),
 	OPT_BOOLEAN_SET('N', "no-buildid-cache", &record.no_buildid_cache,
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index 2357f4cc..cfe4623 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -50,6 +50,7 @@ struct record_opts {
 	bool	     sample_time_set;
 	bool	     sample_cpu;
 	bool	     period;
+	bool	     period_set;
 	bool	     running_time;
 	bool	     full_auxtrace;
 	bool	     auxtrace_snapshot_mode;
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index f2f2eaa..ff359c9 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -971,9 +971,6 @@ void perf_evsel__config(struct perf_evsel *evsel, struct record_opts *opts,
 	if (target__has_cpu(&opts->target) || opts->sample_cpu)
 		perf_evsel__set_sample_bit(evsel, CPU);
 
-	if (opts->period)
-		perf_evsel__set_sample_bit(evsel, PERIOD);
-
 	/*
 	 * When the user explicitly disabled time don't force it here.
 	 */
@@ -1075,6 +1072,14 @@ void perf_evsel__config(struct perf_evsel *evsel, struct record_opts *opts,
 	apply_config_terms(evsel, opts, track);
 
 	evsel->ignore_missing_thread = opts->ignore_missing_thread;
+
+	/* The --period option takes the precedence. */
+	if (opts->period_set) {
+		if (opts->period)
+			perf_evsel__set_sample_bit(evsel, PERIOD);
+		else
+			perf_evsel__reset_sample_bit(evsel, PERIOD);
+	}
 }
 
 static int perf_evsel__alloc_fd(struct perf_evsel *evsel, int ncpus, int nthreads)

[PATCH 3/3] x86/events/intel/ds: Add PERF_SAMPLE_PERIOD into PEBS_FREERUNNING_FLAGS

[Jiri Olsa]

Stephan reported that we don't support period for
enabling large PEBS data, which there's no reason
for. Adding PERF_SAMPLE_PERIOD into freerunning
flags.

Reported-by: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-hx5yqua1c9u7iewm3h8frq5h@git.kernel.org
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
 arch/x86/events/perf_event.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/x86/events/perf_event.h b/arch/x86/events/perf_event.h
index 8e4ea143ed96..78f91ec1056e 100644
--- a/arch/x86/events/perf_event.h
+++ b/arch/x86/events/perf_event.h
@@ -93,7 +93,8 @@ struct amd_nb {
 	PERF_SAMPLE_ID | PERF_SAMPLE_CPU | PERF_SAMPLE_STREAM_ID | \
 	PERF_SAMPLE_DATA_SRC | PERF_SAMPLE_IDENTIFIER | \
 	PERF_SAMPLE_TRANSACTION | PERF_SAMPLE_PHYS_ADDR | \
-	PERF_SAMPLE_REGS_INTR | PERF_SAMPLE_REGS_USER)
+	PERF_SAMPLE_REGS_INTR | PERF_SAMPLE_REGS_USER | \
+	PERF_SAMPLE_PERIOD)
 
 #define PEBS_REGS \
 	(PERF_REG_X86_AX | \
-- 
2.13.6

[tip:perf/urgent] x86/events/intel/ds: Add PERF_SAMPLE_PERIOD into PEBS_FREERUNNING_FLAGS

[tip-bot for Jiri Olsa]

Commit-ID:  11974914e8e6d44387b4b23fa6dbd40c94baeb8d
Gitweb:     https://git.kernel.org/tip/11974914e8e6d44387b4b23fa6dbd40c94baeb8d
Author:     Jiri Olsa <jolsa@kernel.org>
AuthorDate: Thu, 1 Feb 2018 09:38:12 +0100
Committer:  Arnaldo Carvalho de Melo <acme@redhat.com>
CommitDate: Mon, 5 Feb 2018 13:48:44 -0300

x86/events/intel/ds: Add PERF_SAMPLE_PERIOD into PEBS_FREERUNNING_FLAGS

Stephane reported that we don't support period for enabling large PEBS
data, which there's no reason for. Adding PERF_SAMPLE_PERIOD into
freerunning flags.

Tested it with:

  # perf record -e cycles:P -c 100 --no-timestamp -C 0 --period

Reported-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Kan Liang <kan.liang@intel.com>
Tested-by: Stephane Eranian <eranian@google.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20180201083812.11359-4-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 arch/x86/events/perf_event.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/x86/events/perf_event.h b/arch/x86/events/perf_event.h
index 8e4ea143..78f91ec 100644
--- a/arch/x86/events/perf_event.h
+++ b/arch/x86/events/perf_event.h
@@ -93,7 +93,8 @@ struct amd_nb {
 	PERF_SAMPLE_ID | PERF_SAMPLE_CPU | PERF_SAMPLE_STREAM_ID | \
 	PERF_SAMPLE_DATA_SRC | PERF_SAMPLE_IDENTIFIER | \
 	PERF_SAMPLE_TRANSACTION | PERF_SAMPLE_PHYS_ADDR | \
-	PERF_SAMPLE_REGS_INTR | PERF_SAMPLE_REGS_USER)
+	PERF_SAMPLE_REGS_INTR | PERF_SAMPLE_REGS_USER | \
+	PERF_SAMPLE_PERIOD)
 
 #define PEBS_REGS \
 	(PERF_REG_X86_AX | \