Re: [PATCH v1 13/20] perf jevents: Add cycles breakdown metric for Intel

[Liang, Kan]

On 2024-02-29 7:48 p.m., Ian Rogers wrote:
> On Thu, Feb 29, 2024 at 1:30 PM Liang, Kan <kan.liang@xxxxxxxxxxxxxxx> wrote:
>>
>>
>>
>> On 2024-02-28 7:17 p.m., Ian Rogers wrote:
>>> Breakdown cycles to user, kernel and guest.
>>>
>>> Signed-off-by: Ian Rogers <irogers@xxxxxxxxxx>
>>> ---
>>>  tools/perf/pmu-events/intel_metrics.py | 18 ++++++++++++++++++
>>>  1 file changed, 18 insertions(+)
>>>
>>> diff --git a/tools/perf/pmu-events/intel_metrics.py b/tools/perf/pmu-events/intel_metrics.py
>>> index dae44d296861..fef40969a4b8 100755
>>> --- a/tools/perf/pmu-events/intel_metrics.py
>>> +++ b/tools/perf/pmu-events/intel_metrics.py
>>> @@ -26,6 +26,23 @@ core_cycles = Event("CPU_CLK_UNHALTED.THREAD_P_ANY",
>>>  smt_cycles = Select(core_cycles / 2, Literal("#smt_on"), core_cycles)
>>>
>>>
>>> +def Cycles() -> MetricGroup:
>>> +  cyc_k = Event("cycles:kHh")
>>> +  cyc_g = Event("cycles:G")
>>> +  cyc_u = Event("cycles:uH")
>>> +  cyc = cyc_k + cyc_g + cyc_u
>>> +
>>> +  return MetricGroup("cycles", [
>>> +      Metric("cycles_total", "Total number of cycles", cyc, "cycles"),
>>> +      Metric("cycles_user", "User cycles as a percentage of all cycles",
>>> +             d_ratio(cyc_u, cyc), "100%"),
>>> +      Metric("cycles_kernel", "Kernel cycles as a percentage of all cycles",
>>> +             d_ratio(cyc_k, cyc), "100%"),
>>> +      Metric("cycles_guest", "Hypervisor guest cycles as a percentage of all cycles",
>>> +             d_ratio(cyc_g, cyc), "100%"),
>>> +  ], description = "cycles breakdown per privilege level (users, kernel, guest)")
>>> +
>>> +
>>>  def Idle() -> Metric:
>>>    cyc = Event("msr/mperf/")
>>>    tsc = Event("msr/tsc/")
>>> @@ -770,6 +787,7 @@ def IntelLdSt() -> Optional[MetricGroup]:
>>>
>>>
>>>  all_metrics = MetricGroup("", [
>>> +    Cycles(),
>>
>> The metric group seem exactly the same on AMD and ARM. Maybe we can have
>> tools/perf/pmu-events/common_metrics.py for all the common metrics.
> 
> Agreed. I think we can drop cycles in the three sets and then once
> then do the common_metrics.py as a follow up.
>

Sounds good to me.

Thanks,
Kan

> Thanks,
> Ian
> 
>> Thanks,
>> Kan
>>
>>>      Idle(),
>>>      Rapl(),
>>>      Smi(),
>