Re: [PATCH v6 3/7] perf jevents: Support more event fields

From: Jing Zhang
Date: Tue Aug 15 2023 - 08:26:30 EST




在 2023/8/15 上午6:31, Ian Rogers 写道:
> On Mon, Aug 7, 2023 at 12:51 AM Jing Zhang <renyu.zj@xxxxxxxxxxxxxxxxx> wrote:
>>
>> The usual event descriptions are "event=xxx" or "config=xxx", while the
>> event descriptions of CMN are "type=xxx, eventid=xxx" or more complex.
>
> I found this difficult to read in relation to the code. Perhaps:
>
> The previous code assumes an event has either an "event=" or "config"
> field at the beginning. For CMN neither of these may be present, as an
> event is typically "type=xx,eventid=xxx".
>

Thank you for providing a more accurate and readable description.
I was indeed struggling with how to describe the problem more accurately before.

> I think the use of the name "type" here is unfortunate. It conflicts
> with the PMU's type as defined in perf_event_attr.
>

I agree, but it would require modifying the driver, which is not currently being
considered. In the meantime, I can describe the event_field as ('NodeType', 'type=').

> In general I think the jevents.py code needs improving, the
> event_fields dictionary is convoluted, we shouldn't be afraid to
> change the event json for example to get rid of things like ExtSel, we
> should really ensure the formats in the events are valid for the PMU
> they are for.
>
>> $cat /sys/bus/event_source/devices/arm_cmn_0/events/hnf_cache_fill
>> type=0x5,eventid=0x3
>>
>> When adding aliases for events described as "event=xxx" or "config=xxx",
>> EventCode or ConfigCode can be used in the JSON files to describe the
>> events. But "eventid=xxx, type=xxx" cannot be supported at present.
>>
>> If EventCode and ConfigCode is not added in the alias JSON file, the
>> event description will add "event=0" by default. So, even if the event
>> field is added to supplement "eventid=xxx" and "type=xxx", the final
>> parsing result will be "event=0, eventid=xxx, type=xxx".
>>
>> Therefore, when EventCode and ConfigCode are missing in JSON, "event=0" is
>> no longer added by default. EventIdCode and Type are added to the event
>> field, and ConfigCode is moved into the event_field array which can also
>> guarantee its original function.
>
> A useful test can be to build with JEVENTS_ARCH=all and confirm the
> before and after change generated pmu-events.c is the same.
>

Okay, I will test it right away.

>> Signed-off-by: Jing Zhang <renyu.zj@xxxxxxxxxxxxxxxxx>
>> ---
>> tools/perf/pmu-events/jevents.py | 21 +++++++++++++--------
>> 1 file changed, 13 insertions(+), 8 deletions(-)
>>
>> diff --git a/tools/perf/pmu-events/jevents.py b/tools/perf/pmu-events/jevents.py
>> index f57a8f2..9c0f63a 100755
>> --- a/tools/perf/pmu-events/jevents.py
>> +++ b/tools/perf/pmu-events/jevents.py
>> @@ -275,12 +275,6 @@ class JsonEvent:
>> }
>> return table[unit] if unit in table else f'uncore_{unit.lower()}'
>>
>> - eventcode = 0
>> - if 'EventCode' in jd:
>> - eventcode = int(jd['EventCode'].split(',', 1)[0], 0)
>> - if 'ExtSel' in jd:
>> - eventcode |= int(jd['ExtSel']) << 8
>> - configcode = int(jd['ConfigCode'], 0) if 'ConfigCode' in jd else None
>> self.name = jd['EventName'].lower() if 'EventName' in jd else None
>> self.topic = ''
>> self.compat = jd.get('Compat')
>> @@ -317,7 +311,15 @@ class JsonEvent:
>> if precise and self.desc and '(Precise Event)' not in self.desc:
>> extra_desc += ' (Must be precise)' if precise == '2' else (' (Precise '
>> 'event)')
>> - event = f'config={llx(configcode)}' if configcode is not None else f'event={llx(eventcode)}'
>> + eventcode = None
>> + if 'EventCode' in jd:
>> + eventcode = int(jd['EventCode'].split(',', 1)[0], 0)
>> + if 'ExtSel' in jd:
>> + if eventcode is None:
>> + eventcode = int(jd['ExtSel']) << 8
>> + else:
>> + eventcode |= int(jd['ExtSel']) << 8
>> + event = f'event={llx(eventcode)}' if eventcode is not None else None
>> event_fields = [
>> ('AnyThread', 'any='),
>> ('PortMask', 'ch_mask='),
>> @@ -327,10 +329,13 @@ class JsonEvent:
>> ('Invert', 'inv='),
>> ('SampleAfterValue', 'period='),
>> ('UMask', 'umask='),
>> + ('ConfigCode', 'config='),
>
> This loses the int and potential base conversion of ConfigCode.
> Clearly the code was taking care to maintain this behavior so I
> suspect this change has broken something. JEVENTS_ARCH=all should
> reveal the answer.
>

You are correct, I compared the generated pmu-events.c files before and after,
and they are indeed different, with before:config=0x5 vs after:config=0x05. I will keep the
original way of handling ConfigCode in the next version.

>> + ('Type', 'type='),
>> + ('EventIdCode', 'eventid='),
>> ]
>> for key, value in event_fields:
>> if key in jd and jd[key] != '0':
>> - event += ',' + value + jd[key]
>> + event = event + ',' + value + jd[key] if event is not None else value + jd[key]
>
> Perhaps initialize event above to the empty string then:
>
> if key in jd and jd[key] != '0':
> if event:
> event += ','
> event += value + jd[key]
>

If the event is None, the statement event += value + jd[key] would result in an error.
So, maybe I can use the following way:

if event:
event += ',' + value + jd[key]
else:
event = value + jd[key]


Thanks,
Jing