Re: [PATCH v8 15/52] perf jevents: Add RAPL event metric for AMD zen models

[Sandipan Das <sandipan.das@amd.com>]

On 11/28/2025 2:50 PM, Ian Rogers wrote:
> On Tue, Nov 25, 2025 at 9:05 PM Sandipan Das <sandipan.das@amd.com> wrote:
>>
>> On 11/13/2025 8:50 AM, Ian Rogers wrote:
>>> Add power per second metrics based on RAPL.
>>>
>>> Signed-off-by: Ian Rogers <irogers@google.com>
>>> ---
>>>  tools/perf/pmu-events/amd_metrics.py | 31 +++++++++++++++++++++++++---
>>>  1 file changed, 28 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/tools/perf/pmu-events/amd_metrics.py b/tools/perf/pmu-events/amd_metrics.py
>>> index bc91d9c120fa..b6cdeb4f09fe 100755
>>> --- a/tools/perf/pmu-events/amd_metrics.py
>>> +++ b/tools/perf/pmu-events/amd_metrics.py
>>> @@ -1,13 +1,36 @@
>>>  #!/usr/bin/env python3
>>>  # SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
>>>  import argparse
>>> +import math
>>>  import os
>>> -from metric import (JsonEncodeMetric, JsonEncodeMetricGroupDescriptions, LoadEvents,
>>> -                    MetricGroup)
>>> +from metric import (d_ratio, has_event, Event, JsonEncodeMetric, JsonEncodeMetricGroupDescriptions,
>>> +                    LoadEvents, Metric, MetricGroup, Select)
>>>
>>>  # Global command line arguments.
>>>  _args = None
>>>
>>> +interval_sec = Event("duration_time")
>>> +
>>> +
>>> +def Rapl() -> MetricGroup:
>>> +    """Processor socket power consumption estimate.
>>> +
>>> +    Use events from the running average power limit (RAPL) driver.
>>> +    """
>>> +    # Watts = joules/second
>>> +    # Currently only energy-pkg is supported by AMD:
>>> +    # https://lore.kernel.org/lkml/20220105185659.643355-1-eranian@google.com/
>>> +    pkg = Event("power/energy\\-pkg/")
>>> +    cond_pkg = Select(pkg, has_event(pkg), math.nan)
>>> +    scale = 2.3283064365386962890625e-10
>>
>> It is unlikely that the scale factor will change, but would it still be safer to read
>> it from /sys/bus/event_source/devices/power/events/energy-pkg.scale?
> 
> Thanks Sandipan, I agree with the feedback but this isn't something
> the metrics currently support. I'll keep it in mind.
> 

Thanks Ian. Parsing the scale factor and unit will be a nice addition.

> Wrt, the other feedback, I'm wondering if we can get this series
> landed and drop for now the patches that you have commented on? I'll
> move them to a follow up series. That'd make the AMD patches here
> something like:
> 
> Keep: [PATCH v8 15/52] perf jevents: Add RAPL event metric for AMD zen models
> Keep: [PATCH v8 16/52] perf jevents: Add idle metric for AMD zen models
> Keep: [PATCH v8 17/52] perf jevents: Add upc metric for uops per cycle for AMD
> Keep: [PATCH v8 18/52] perf jevents: Add br metric group for branch
> statistics on AMD
> Drop: [PATCH v8 19/52] perf jevents: Add software prefetch (swpf)
> metric group for AMD
> Drop: [PATCH v8 20/52] perf jevents: Add hardware prefetch (hwpf)
> metric group for AMD
> Keep: [PATCH v8 21/52] perf jevents: Add itlb metric group for AMD
> Keep: [PATCH v8 22/52] perf jevents: Add dtlb metric group for AMD
> Drop (or perhaps keep and followup with miss latency): [PATCH v8
> 23/52] perf jevents: Add uncore l3 metric group for AMD
> Keep: [PATCH v8 24/52] perf jevents: Add load store breakdown metrics
> ldst for AMD
> Drop: [PATCH v8 25/52] perf jevents: Add ILP metrics for AMD
> Keep: [PATCH v8 26/52] perf jevents: Add context switch metrics for AMD
> Keep: [PATCH v8 27/52] perf jevents: Add uop cache hit/miss rates for AMD
> 
> Is it okay to use your reviewed-by tag on the kept patches? If I keep
> patch 23 for uncore l3, with a follow up on miss latency, then are you
> okay with that?
> 

I am fine with patch 23 currently not having a miss latency metric.

For all the patches tagged "Keep" except 27/52, I don't see any obvious
issues but to be sure that the interpretations are correct, I reached
out to our internal hardware and performance teams for clarity on the
behaviour of some of the events. I would like to wait until I have the
answers.

> 
> 
>>> +    metrics = [
>>> +        Metric("lpm_cpu_power_pkg", "",
>>> +               d_ratio(cond_pkg * scale, interval_sec), "Watts"),
>>> +    ]
>>> +
>>> +    return MetricGroup("lpm_cpu_power", metrics,
>>> +                       description="Processor socket power consumption estimates")
>>> +
>>>
>>>  def main() -> None:
>>>      global _args
>>> @@ -33,7 +56,9 @@ def main() -> None:
>>>      directory = f"{_args.events_path}/x86/{_args.model}/"
>>>      LoadEvents(directory)
>>>
>>> -    all_metrics = MetricGroup("", [])
>>> +    all_metrics = MetricGroup("", [
>>> +        Rapl(),
>>> +    ])
>>>
>>>      if _args.metricgroups:
>>>          print(JsonEncodeMetricGroupDescriptions(all_metrics))
>>