[RFC PATCH v2 0/2] perf tools: Don't set inherit bit for system wide evsel

[RFC PATCH v2 0/2] perf tools: Don't set inherit bit for system wide evsel

[Wang Nan]

Previous version can be found from [1].

To make bpf_perf_event_output() and bpf_perf_event_read() easier to be
used, these two patches set inherit to 0 for system wide evsels.
After applying them, users are possible to pass system wide events to
BPF programs in following command, and don't need to consider the
setting of inherit:

 # perf record -a -e evt=cycles -e ./test__map.c/maps.pmu_map.event=evt/ ...

Patch 1/2 set evsel->system_wide for system wide target. I have checked
in current evlist.c and evsel.c. The only behavior change I found is in
evlist.c:

		/*
		 * The system_wide flag causes a selected event to be opened
		 * always without a pid.  Consequently it will never get a
		 * POLLHUP, but it is used for tracking in combination with
		 * other events, so it should not need to be polled anyway.
		 * Therefore don't add it for polling.
		 */
		if (!evsel->system_wide &&
		    __perf_evlist__add_pollfd(evlist, fd, idx) < 0) {
			perf_evlist__mmap_put(evlist, idx);
			return -1;
		}

With patch 1/2, __perf_evlist__add_pollfd() won't be called. From the
comment I pasted this modification looks harmless.

[1] http://lkml.kernel.org/r/1445597029-133332-1-git-send-email-wangnan0@huawei.com

Wang Nan (2):
  perf tools: Set evsel->system_wide field for global system wide
    recording
  perf tools: Don't set inherit bit for system wide evsel

 tools/perf/util/evsel.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

-- 
1.8.3.4

[RFC PATCH v2 1/2] perf tools: Set evsel->system_wide field for global system wide recording

[Wang Nan]

evsel->system_wide is introduced by commit bf8e8f4b832972c76d64ab2e2837
(perf evlist: Add 'system_wide' option), which is used for mixing evsels
that aren't system-wide with ones that are [1]. However, for global
system wide recording (perf record -a ...), evsel->system_wide is set
to false, which is confusion.

This patch set evsel->system_wide to true if the target.system_wide is
set, which makes evsel->system_wide a reliable way to describe whether
itself is system_wide or not.

[1] http://lkml.kernel.org/r/562DF19B.2080608@intel.com

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Li Zefan <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/n/ebpf-qm3gtwidc1o5ktjd9tgjex25@git.kernel.org
---
 tools/perf/util/evsel.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 3ac4ee9c..36ecf0e 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -734,6 +734,7 @@ void perf_evsel__config(struct perf_evsel *evsel, struct record_opts *opts)
 	int track = evsel->tracking;
 	bool per_cpu = opts->target.default_per_cpu && !opts->target.per_thread;
 
+	evsel->system_wide = opts->target.system_wide;
 	attr->sample_id_all = perf_missing_features.sample_id_all ? 0 : 1;
 	attr->inherit	    = !opts->no_inherit;
 
-- 
1.8.3.4

[RFC PATCH v2 2/2] perf tools: Don't set inherit bit for system wide evsel

[Wang Nan]

Inherit bit is useless for a system wide evsel [1]. Further kernel
improvements are giving more constrain [2] on inherit events. This
patch set inherit bit to 0 for system wide evsels to avoid potential
constrains.

[1] http://lkml.kernel.org/r/20151022124142.GQ17308@twins.programming.kicks-ass.net
[2] http://lkml.kernel.org/r/1445559014-4667-1-git-send-email-ast@kernel.org

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Li Zefan <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/n/ebpf-0xsuv8wc1uwgj7qfylfolgqs@git.kernel.org
---
 tools/perf/util/evsel.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 36ecf0e..2a5f0f8 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -736,7 +736,7 @@ void perf_evsel__config(struct perf_evsel *evsel, struct record_opts *opts)
 
 	evsel->system_wide = opts->target.system_wide;
 	attr->sample_id_all = perf_missing_features.sample_id_all ? 0 : 1;
-	attr->inherit	    = !opts->no_inherit;
+	attr->inherit	    = !opts->no_inherit && !evsel->system_wide;
 
 	perf_evsel__set_sample_bit(evsel, IP);
 	perf_evsel__set_sample_bit(evsel, TID);
-- 
1.8.3.4

Re: [RFC PATCH v2 1/2] perf tools: Set evsel->system_wide field for global system wide recording

[Adrian Hunter]

On 26/10/15 13:41, Wang Nan wrote:
> evsel->system_wide is introduced by commit bf8e8f4b832972c76d64ab2e2837
> (perf evlist: Add 'system_wide' option), which is used for mixing evsels
> that aren't system-wide with ones that are [1]. However, for global
> system wide recording (perf record -a ...), evsel->system_wide is set
> to false, which is confusion.
> 
> This patch set evsel->system_wide to true if the target.system_wide is
> set, which makes evsel->system_wide a reliable way to describe whether
> itself is system_wide or not.
> 
> [1] http://lkml.kernel.org/r/562DF19B.2080608@intel.com
> 
> Signed-off-by: Wang Nan <wangnan0@huawei.com>
> Cc: Adrian Hunter <adrian.hunter@intel.com>
> Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
> Cc: Alexei Starovoitov <ast@plumgrid.com>
> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
> Cc: Li Zefan <lizefan@huawei.com>
> Cc: pi3orama@163.com
> Link: http://lkml.kernel.org/n/ebpf-qm3gtwidc1o5ktjd9tgjex25@git.kernel.org
> ---
>  tools/perf/util/evsel.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
> index 3ac4ee9c..36ecf0e 100644
> --- a/tools/perf/util/evsel.c
> +++ b/tools/perf/util/evsel.c
> @@ -734,6 +734,7 @@ void perf_evsel__config(struct perf_evsel *evsel, struct record_opts *opts)
>  	int track = evsel->tracking;
>  	bool per_cpu = opts->target.default_per_cpu && !opts->target.per_thread;
>  
> +	evsel->system_wide = opts->target.system_wide;

Well that breaks the way evsel->system_wide is used i.e. it is a parameter
to the evsel and here you just overwrote it.

>  	attr->sample_id_all = perf_missing_features.sample_id_all ? 0 : 1;
>  	attr->inherit	    = !opts->no_inherit;
>  
> 

Re: [RFC PATCH v2 1/2] perf tools: Set evsel->system_wide field for global system wide recording

[Wangnan (F)]

On 2015/10/26 20:24, Adrian Hunter wrote:
> On 26/10/15 13:41, Wang Nan wrote:
>> evsel->system_wide is introduced by commit bf8e8f4b832972c76d64ab2e2837
>> (perf evlist: Add 'system_wide' option), which is used for mixing evsels
>> that aren't system-wide with ones that are [1]. However, for global
>> system wide recording (perf record -a ...), evsel->system_wide is set
>> to false, which is confusion.
>>
>> This patch set evsel->system_wide to true if the target.system_wide is
>> set, which makes evsel->system_wide a reliable way to describe whether
>> itself is system_wide or not.
>>
>> [1] http://lkml.kernel.org/r/562DF19B.2080608@intel.com
>>
>> Signed-off-by: Wang Nan <wangnan0@huawei.com>
>> Cc: Adrian Hunter <adrian.hunter@intel.com>
>> Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
>> Cc: Alexei Starovoitov <ast@plumgrid.com>
>> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
>> Cc: Li Zefan <lizefan@huawei.com>
>> Cc: pi3orama@163.com
>> Link: http://lkml.kernel.org/n/ebpf-qm3gtwidc1o5ktjd9tgjex25@git.kernel.org
>> ---
>>   tools/perf/util/evsel.c | 1 +
>>   1 file changed, 1 insertion(+)
>>
>> diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
>> index 3ac4ee9c..36ecf0e 100644
>> --- a/tools/perf/util/evsel.c
>> +++ b/tools/perf/util/evsel.c
>> @@ -734,6 +734,7 @@ void perf_evsel__config(struct perf_evsel *evsel, struct record_opts *opts)
>>   	int track = evsel->tracking;
>>   	bool per_cpu = opts->target.default_per_cpu && !opts->target.per_thread;
>>   
>> +	evsel->system_wide = opts->target.system_wide;
> Well that breaks the way evsel->system_wide is used i.e. it is a parameter
> to the evsel and here you just overwrote it.

Currently the only user of evsel->system_wide is IPT:

auxtrace_record__options -> intel_pt_recording_options

and it only set it to true.

So I think changing to this should make it safe:

evsel->system_wide = (evsel->system_wide || opt->target.system_wide);

Thought?

If we want to add further config terms we can put it to 
apply_config_terms(),
where we can implement something like:

  # perf record -e cycles/system-wide/ -e instruction/no-system-wide/ ...

But currently I don't have such requirement.

Thank you.

Re: [RFC PATCH v2 1/2] perf tools: Set evsel->system_wide field for global system wide recording

[Adrian Hunter]

On 26/10/15 14:39, Wangnan (F) wrote:
> 
> 
> On 2015/10/26 20:24, Adrian Hunter wrote:
>> On 26/10/15 13:41, Wang Nan wrote:
>>> evsel->system_wide is introduced by commit bf8e8f4b832972c76d64ab2e2837
>>> (perf evlist: Add 'system_wide' option), which is used for mixing evsels
>>> that aren't system-wide with ones that are [1]. However, for global
>>> system wide recording (perf record -a ...), evsel->system_wide is set
>>> to false, which is confusion.
>>>
>>> This patch set evsel->system_wide to true if the target.system_wide is
>>> set, which makes evsel->system_wide a reliable way to describe whether
>>> itself is system_wide or not.
>>>
>>> [1] http://lkml.kernel.org/r/562DF19B.2080608@intel.com
>>>
>>> Signed-off-by: Wang Nan <wangnan0@huawei.com>
>>> Cc: Adrian Hunter <adrian.hunter@intel.com>
>>> Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
>>> Cc: Alexei Starovoitov <ast@plumgrid.com>
>>> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
>>> Cc: Li Zefan <lizefan@huawei.com>
>>> Cc: pi3orama@163.com
>>> Link: http://lkml.kernel.org/n/ebpf-qm3gtwidc1o5ktjd9tgjex25@git.kernel.org
>>> ---
>>>   tools/perf/util/evsel.c | 1 +
>>>   1 file changed, 1 insertion(+)
>>>
>>> diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
>>> index 3ac4ee9c..36ecf0e 100644
>>> --- a/tools/perf/util/evsel.c
>>> +++ b/tools/perf/util/evsel.c
>>> @@ -734,6 +734,7 @@ void perf_evsel__config(struct perf_evsel *evsel,
>>> struct record_opts *opts)
>>>       int track = evsel->tracking;
>>>       bool per_cpu = opts->target.default_per_cpu &&
>>> !opts->target.per_thread;
>>>   +    evsel->system_wide = opts->target.system_wide;
>> Well that breaks the way evsel->system_wide is used i.e. it is a parameter
>> to the evsel and here you just overwrote it.
> 
> Currently the only user of evsel->system_wide is IPT:
> 
> auxtrace_record__options -> intel_pt_recording_options
> 
> and it only set it to true.
> 
> So I think changing to this should make it safe:
> 
> evsel->system_wide = (evsel->system_wide || opt->target.system_wide);
> 
> Thought?

That would work although it looks like a kludge.

Have you looked at properly validating opts->no_inherit ?

Also, aren't you neglecting target->cpu_list ?

> 
> If we want to add further config terms we can put it to apply_config_terms(),
> where we can implement something like:
> 
>  # perf record -e cycles/system-wide/ -e instruction/no-system-wide/ ...
> 
> But currently I don't have such requirement.
> 
> Thank you.
> 
> 

Re: [RFC PATCH v2 1/2] perf tools: Set evsel->system_wide field for global system wide recording

[Wangnan (F)]

On 2015/10/26 21:02, Adrian Hunter wrote:
> On 26/10/15 14:39, Wangnan (F) wrote:
>>
>> On 2015/10/26 20:24, Adrian Hunter wrote:
>>> On 26/10/15 13:41, Wang Nan wrote:
>>>> evsel->system_wide is introduced by commit bf8e8f4b832972c76d64ab2e2837
>>>> (perf evlist: Add 'system_wide' option), which is used for mixing evsels
>>>> that aren't system-wide with ones that are [1]. However, for global
>>>> system wide recording (perf record -a ...), evsel->system_wide is set
>>>> to false, which is confusion.
>>>>
>>>> This patch set evsel->system_wide to true if the target.system_wide is
>>>> set, which makes evsel->system_wide a reliable way to describe whether
>>>> itself is system_wide or not.
>>>>
>>>> [1] http://lkml.kernel.org/r/562DF19B.2080608@intel.com
>>>>
>>>> Signed-off-by: Wang Nan <wangnan0@huawei.com>
>>>> Cc: Adrian Hunter <adrian.hunter@intel.com>
>>>> Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
>>>> Cc: Alexei Starovoitov <ast@plumgrid.com>
>>>> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
>>>> Cc: Li Zefan <lizefan@huawei.com>
>>>> Cc: pi3orama@163.com
>>>> Link: http://lkml.kernel.org/n/ebpf-qm3gtwidc1o5ktjd9tgjex25@git.kernel.org
>>>> ---
>>>>    tools/perf/util/evsel.c | 1 +
>>>>    1 file changed, 1 insertion(+)
>>>>
>>>> diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
>>>> index 3ac4ee9c..36ecf0e 100644
>>>> --- a/tools/perf/util/evsel.c
>>>> +++ b/tools/perf/util/evsel.c
>>>> @@ -734,6 +734,7 @@ void perf_evsel__config(struct perf_evsel *evsel,
>>>> struct record_opts *opts)
>>>>        int track = evsel->tracking;
>>>>        bool per_cpu = opts->target.default_per_cpu &&
>>>> !opts->target.per_thread;
>>>>    +    evsel->system_wide = opts->target.system_wide;
>>> Well that breaks the way evsel->system_wide is used i.e. it is a parameter
>>> to the evsel and here you just overwrote it.
>> Currently the only user of evsel->system_wide is IPT:
>>
>> auxtrace_record__options -> intel_pt_recording_options
>>
>> and it only set it to true.
>>
>> So I think changing to this should make it safe:
>>
>> evsel->system_wide = (evsel->system_wide || opt->target.system_wide);
>>
>> Thought?
> That would work although it looks like a kludge.
>
> Have you looked at properly validating opts->no_inherit ?
>
> Also, aren't you neglecting target->cpu_list ?

I need this because Alexei's bpf_perf_event_output() helper needs perf side
support, and it only accept no-inherit perf event. I have already made a
patch[1] to support setting inherit bit per-event, so even without this and
next patch I can continue my work on bpf output. I'd like to rethink this
patch after I finish perf support for bpf output.

Thank you.

[1] 
http://lkml.kernel.org/g/1445847884-143981-1-git-send-email-wangnan0@huawei.com