[v3,0/3] perf: add new option '--workload-attr' to set workload sched_policy/priority/mask

  This adds a new option '--workload-attr' to set the sched_policy/priority/mask
of the workload to reduce system noise.

$ sudo perf stat --workload-attr fifo,40,0-3:7 -- ls

Please see patch 1/2 for more details.

v3:
  - replace taskset with --workload-attr option in documents and tests.
v2:
  - Use cpu list spec instead of cpu mask number.
  - Update documents.

Changbin Du (3):
  perf cpumap: Add __perf_cpu_map__new and perf_cpu_map__2_cpuset
  perf: add new option '--workload-attr' to set workload
    sched_policy/priority/mask
  perf: replace taskset with --workload-attr option

 tools/lib/perf/cpumap.c                       |  45 ++++++-
 tools/lib/perf/include/perf/cpumap.h          |   4 +
 tools/lib/perf/libperf.map                    |   2 +
 tools/perf/Documentation/intel-hybrid.txt     |   2 +-
 tools/perf/Documentation/perf-record.txt      |   7 ++
 tools/perf/Documentation/perf-stat.txt        |   8 +-
 tools/perf/builtin-record.c                   |  26 ++++
 tools/perf/builtin-stat.c                     |  18 +++
 tools/perf/tests/cpumap.c                     |  23 ++++
 .../tests/shell/stat_bpf_counters_cgrp.sh     |   2 +-
 tools/perf/tests/shell/test_arm_coresight.sh  |   2 +-
 tools/perf/tests/shell/test_data_symbol.sh    |   2 +-
 tools/perf/tests/shell/test_intel_pt.sh       |   2 +-
 tools/perf/util/evlist.c                      | 117 ++++++++++++++++++
 tools/perf/util/evlist.h                      |   3 +
 tools/perf/util/target.h                      |   9 ++
 16 files changed, 263 insertions(+), 9 deletions(-)
  

On 18/07/23 06:33, Changbin Du wrote:
> This adds a new option '--workload-attr' to set the sched_policy/priority/mask
> of the workload to reduce system noise.
> 
> $ sudo perf stat --workload-attr fifo,40,0-3:7 -- ls

Not really sold on the need for this, but maybe it could be
simpler.
What about just adding a hook for a command (e.g. script) to
run before exec'ing the workload e.g.

--configure-workload=blah.sh

results in perf doing system("blah.sh 12345") where 12345
is the workload PID.

Then maybe you could do:

--configure-workload="taskset -p 0x3"


> 
> Please see patch 1/2 for more details.
> 
> v3:
>   - replace taskset with --workload-attr option in documents and tests.
> v2:
>   - Use cpu list spec instead of cpu mask number.
>   - Update documents.
> 
> Changbin Du (3):
>   perf cpumap: Add __perf_cpu_map__new and perf_cpu_map__2_cpuset
>   perf: add new option '--workload-attr' to set workload
>     sched_policy/priority/mask
>   perf: replace taskset with --workload-attr option
> 
>  tools/lib/perf/cpumap.c                       |  45 ++++++-
>  tools/lib/perf/include/perf/cpumap.h          |   4 +
>  tools/lib/perf/libperf.map                    |   2 +
>  tools/perf/Documentation/intel-hybrid.txt     |   2 +-
>  tools/perf/Documentation/perf-record.txt      |   7 ++
>  tools/perf/Documentation/perf-stat.txt        |   8 +-
>  tools/perf/builtin-record.c                   |  26 ++++
>  tools/perf/builtin-stat.c                     |  18 +++
>  tools/perf/tests/cpumap.c                     |  23 ++++
>  .../tests/shell/stat_bpf_counters_cgrp.sh     |   2 +-
>  tools/perf/tests/shell/test_arm_coresight.sh  |   2 +-
>  tools/perf/tests/shell/test_data_symbol.sh    |   2 +-
>  tools/perf/tests/shell/test_intel_pt.sh       |   2 +-
>  tools/perf/util/evlist.c                      | 117 ++++++++++++++++++
>  tools/perf/util/evlist.h                      |   3 +
>  tools/perf/util/target.h                      |   9 ++
>  16 files changed, 263 insertions(+), 9 deletions(-)
>
  

On Thu, Jul 20, 2023 at 01:00:58PM +0300, Adrian Hunter wrote:
> On 18/07/23 06:33, Changbin Du wrote:
> > This adds a new option '--workload-attr' to set the sched_policy/priority/mask
> > of the workload to reduce system noise.
> > 
> > $ sudo perf stat --workload-attr fifo,40,0-3:7 -- ls
> 
> Not really sold on the need for this, but maybe it could be
> simpler.
> What about just adding a hook for a command (e.g. script) to
> run before exec'ing the workload e.g.
> 
> --configure-workload=blah.sh
> 
> results in perf doing system("blah.sh 12345") where 12345
> is the workload PID.
> 
> Then maybe you could do:
> 
> --configure-workload="taskset -p 0x3"
>
Acctually, we already have such option for perf-stat.

	--post <command>  command to run after to the measured command
        --pre <command>   command to run prior to the measured command

By involving a shell script we can do more complex setup. But sometimes I just
need to set sched attributes. For example, to investigate the impact of
various compiler optimizations. In this case, I don't want a script. This is the
original purpose I try to add this new option.

> 
> > 
> > Please see patch 1/2 for more details.
> > 
> > v3:
> >   - replace taskset with --workload-attr option in documents and tests.
> > v2:
> >   - Use cpu list spec instead of cpu mask number.
> >   - Update documents.
> > 
> > Changbin Du (3):
> >   perf cpumap: Add __perf_cpu_map__new and perf_cpu_map__2_cpuset
> >   perf: add new option '--workload-attr' to set workload
> >     sched_policy/priority/mask
> >   perf: replace taskset with --workload-attr option
> > 
> >  tools/lib/perf/cpumap.c                       |  45 ++++++-
> >  tools/lib/perf/include/perf/cpumap.h          |   4 +
> >  tools/lib/perf/libperf.map                    |   2 +
> >  tools/perf/Documentation/intel-hybrid.txt     |   2 +-
> >  tools/perf/Documentation/perf-record.txt      |   7 ++
> >  tools/perf/Documentation/perf-stat.txt        |   8 +-
> >  tools/perf/builtin-record.c                   |  26 ++++
> >  tools/perf/builtin-stat.c                     |  18 +++
> >  tools/perf/tests/cpumap.c                     |  23 ++++
> >  .../tests/shell/stat_bpf_counters_cgrp.sh     |   2 +-
> >  tools/perf/tests/shell/test_arm_coresight.sh  |   2 +-
> >  tools/perf/tests/shell/test_data_symbol.sh    |   2 +-
> >  tools/perf/tests/shell/test_intel_pt.sh       |   2 +-
> >  tools/perf/util/evlist.c                      | 117 ++++++++++++++++++
> >  tools/perf/util/evlist.h                      |   3 +
> >  tools/perf/util/target.h                      |   9 ++
> >  16 files changed, 263 insertions(+), 9 deletions(-)
> > 
>
  

On 24/07/23 07:02, Changbin Du wrote:
> On Thu, Jul 20, 2023 at 01:00:58PM +0300, Adrian Hunter wrote:
>> On 18/07/23 06:33, Changbin Du wrote:
>>> This adds a new option '--workload-attr' to set the sched_policy/priority/mask
>>> of the workload to reduce system noise.
>>>
>>> $ sudo perf stat --workload-attr fifo,40,0-3:7 -- ls
>>
>> Not really sold on the need for this, but maybe it could be
>> simpler.
>> What about just adding a hook for a command (e.g. script) to
>> run before exec'ing the workload e.g.
>>
>> --configure-workload=blah.sh
>>
>> results in perf doing system("blah.sh 12345") where 12345
>> is the workload PID.
>>
>> Then maybe you could do:
>>
>> --configure-workload="taskset -p 0x3"
>>
> Acctually, we already have such option for perf-stat.
> 
> 	--post <command>  command to run after to the measured command
>         --pre <command>   command to run prior to the measured command
> 
> By involving a shell script we can do more complex setup. But sometimes I just
> need to set sched attributes. For example, to investigate the impact of
> various compiler optimizations. In this case, I don't want a script. This is the
> original purpose I try to add this new option.

There is also command schedtool, but what exactly is the problem
with a script?

> 
>>
>>>
>>> Please see patch 1/2 for more details.
>>>
>>> v3:
>>>   - replace taskset with --workload-attr option in documents and tests.
>>> v2:
>>>   - Use cpu list spec instead of cpu mask number.
>>>   - Update documents.
>>>
>>> Changbin Du (3):
>>>   perf cpumap: Add __perf_cpu_map__new and perf_cpu_map__2_cpuset
>>>   perf: add new option '--workload-attr' to set workload
>>>     sched_policy/priority/mask
>>>   perf: replace taskset with --workload-attr option
>>>
>>>  tools/lib/perf/cpumap.c                       |  45 ++++++-
>>>  tools/lib/perf/include/perf/cpumap.h          |   4 +
>>>  tools/lib/perf/libperf.map                    |   2 +
>>>  tools/perf/Documentation/intel-hybrid.txt     |   2 +-
>>>  tools/perf/Documentation/perf-record.txt      |   7 ++
>>>  tools/perf/Documentation/perf-stat.txt        |   8 +-
>>>  tools/perf/builtin-record.c                   |  26 ++++
>>>  tools/perf/builtin-stat.c                     |  18 +++
>>>  tools/perf/tests/cpumap.c                     |  23 ++++
>>>  .../tests/shell/stat_bpf_counters_cgrp.sh     |   2 +-
>>>  tools/perf/tests/shell/test_arm_coresight.sh  |   2 +-
>>>  tools/perf/tests/shell/test_data_symbol.sh    |   2 +-
>>>  tools/perf/tests/shell/test_intel_pt.sh       |   2 +-
>>>  tools/perf/util/evlist.c                      | 117 ++++++++++++++++++
>>>  tools/perf/util/evlist.h                      |   3 +
>>>  tools/perf/util/target.h                      |   9 ++
>>>  16 files changed, 263 insertions(+), 9 deletions(-)
>>>
>>
>
  

On Mon, Jul 24, 2023 at 08:44:12AM +0300, Adrian Hunter wrote:
> On 24/07/23 07:02, Changbin Du wrote:
> > On Thu, Jul 20, 2023 at 01:00:58PM +0300, Adrian Hunter wrote:
> >> On 18/07/23 06:33, Changbin Du wrote:
> >>> This adds a new option '--workload-attr' to set the sched_policy/priority/mask
> >>> of the workload to reduce system noise.
> >>>
> >>> $ sudo perf stat --workload-attr fifo,40,0-3:7 -- ls
> >>
> >> Not really sold on the need for this, but maybe it could be
> >> simpler.
> >> What about just adding a hook for a command (e.g. script) to
> >> run before exec'ing the workload e.g.
> >>
> >> --configure-workload=blah.sh
> >>
> >> results in perf doing system("blah.sh 12345") where 12345
> >> is the workload PID.
> >>
> >> Then maybe you could do:
> >>
> >> --configure-workload="taskset -p 0x3"
> >>
> > Acctually, we already have such option for perf-stat.
> > 
> > 	--post <command>  command to run after to the measured command
> >         --pre <command>   command to run prior to the measured command
> > 
> > By involving a shell script we can do more complex setup. But sometimes I just
> > need to set sched attributes. For example, to investigate the impact of
> > various compiler optimizations. In this case, I don't want a script. This is the
> > original purpose I try to add this new option.
> 
> There is also command schedtool, but what exactly is the problem
> with a script?
> 
> > 
There is no problem with a script, just a shortcut for convenience. When I want
to share perf investigation with somebody, I just need to share a single commandline.
Anyway nothing bad, right? :)
  

On 24/07/23 12:34, Changbin Du wrote:
> On Mon, Jul 24, 2023 at 08:44:12AM +0300, Adrian Hunter wrote:
>> On 24/07/23 07:02, Changbin Du wrote:
>>> On Thu, Jul 20, 2023 at 01:00:58PM +0300, Adrian Hunter wrote:
>>>> On 18/07/23 06:33, Changbin Du wrote:
>>>>> This adds a new option '--workload-attr' to set the sched_policy/priority/mask
>>>>> of the workload to reduce system noise.
>>>>>
>>>>> $ sudo perf stat --workload-attr fifo,40,0-3:7 -- ls
>>>>
>>>> Not really sold on the need for this, but maybe it could be
>>>> simpler.
>>>> What about just adding a hook for a command (e.g. script) to
>>>> run before exec'ing the workload e.g.
>>>>
>>>> --configure-workload=blah.sh
>>>>
>>>> results in perf doing system("blah.sh 12345") where 12345
>>>> is the workload PID.
>>>>
>>>> Then maybe you could do:
>>>>
>>>> --configure-workload="taskset -p 0x3"
>>>>
>>> Acctually, we already have such option for perf-stat.
>>>
>>> 	--post <command>  command to run after to the measured command
>>>         --pre <command>   command to run prior to the measured command
>>>
>>> By involving a shell script we can do more complex setup. But sometimes I just
>>> need to set sched attributes. For example, to investigate the impact of
>>> various compiler optimizations. In this case, I don't want a script. This is the
>>> original purpose I try to add this new option.
>>
>> There is also command schedtool, but what exactly is the problem
>> with a script?
>>
>>>
> There is no problem with a script, just a shortcut for convenience. When I want
> to share perf investigation with somebody, I just need to share a single commandline.
> Anyway nothing bad, right? :)

Also depends on what you call a single commandline, since you could
always create the script and run perf in one long line :-)

I would still go for the simpler option: easier to maintain and
potentially more flexible.