Re: [PATCH V4 20/21] perf: make events stream always parsable

From: Adrian Hunter
Date: Thu Jul 11 2013 - 09:20:28 EST


On 05/07/13 16:24, Namhyung Kim wrote:
> Hi Adrian,
>
> On Thu, Jul 4, 2013 at 10:20 PM, Adrian Hunter <adrian.hunter@xxxxxxxxx> wrote:
>> The event stream is not always parsable because the format of a sample
>> is dependent on the sample_type of the selected event. When there
>> is more than one selected event and the sample_types are not the
>> same then parsing becomes problematic. A sample can be matched to its
>> selected event using the ID that is allocated when the event is opened.
>> Unfortunately, to get the ID from the sample means first parsing it.
>>
>> This patch adds a new sample format bit PERF_SAMPLE_IDENTIFER that puts
>> the ID at a fixed position so that the ID can be retrieved without
>> parsing the sample. For sample events, that is the first position
>> immediately after the header. For non-sample events, that is the last
>> position.
>
> Why do we have to keep another ID again?

It is a different place for the ID not another ID.

> If all we need is the sample_type, why not just saving it directly?

We really want to associate the event with the attribute so the ID is
needed, not just the sample type.

>
> And for the implementation, I guess it'll break old perf tools
> by parsing invalid position of data. IOW if it's recorded on a newer
> kernel/tool and then reported on an older perf tool (maybe on a
> different machine). In this case the old tool cannot recognize the
> PERF_SAMPLE_IDENTIFIER and try to parse it as a different
> type of data, right?

I have changed the code so that PERF_SAMPLE_IDENTIFIER is only used for
non-matching sample types. Since in those cases the data was anyway not
parsable, the scenario you outline cannot happen.

>
> So I suggest put the data at the end of sample data like other new
> sample types added. Older tools would simply ignore that part, and
> newer tools can access it directly by checking event.header.size
> and/or evsel->sample_size.

The ID will be on the end of non-sample (e.g. mmap, comm, task) events, so
if tools are parsing those events backwards (as perf tools does) then they
will anyway fail if they are not expecting PERF_SAMPLE_IDENTIFIER. However
that scenario only happens when the data is anyway not parsable as I
mentioned above.

>
> Thanks,
> Namhyung
>
>>
>> Signed-off-by: Adrian Hunter <adrian.hunter@xxxxxxxxx>
>> ---
>> include/uapi/linux/perf_event.h | 3 ++-
>> kernel/events/core.c | 11 ++++++++++-
>> 2 files changed, 12 insertions(+), 2 deletions(-)
>>
>> diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
>> index 0b1df41..6bb217e 100644
>> --- a/include/uapi/linux/perf_event.h
>> +++ b/include/uapi/linux/perf_event.h
>> @@ -134,8 +134,9 @@ enum perf_event_sample_format {
>> PERF_SAMPLE_STACK_USER = 1U << 13,
>> PERF_SAMPLE_WEIGHT = 1U << 14,
>> PERF_SAMPLE_DATA_SRC = 1U << 15,
>> + PERF_SAMPLE_IDENTIFIER = 1U << 16,
>>
>> - PERF_SAMPLE_MAX = 1U << 16, /* non-ABI */
>> + PERF_SAMPLE_MAX = 1U << 17, /* non-ABI */
>> };
>>
>> /*
>> diff --git a/kernel/events/core.c b/kernel/events/core.c
>> index 1db3af9..ca532f2 100644
>> --- a/kernel/events/core.c
>> +++ b/kernel/events/core.c
>> @@ -1203,6 +1203,9 @@ static void perf_event__id_header_size(struct perf_event *event)
>> if (sample_type & PERF_SAMPLE_TIME)
>> size += sizeof(data->time);
>>
>> + if (sample_type & PERF_SAMPLE_IDENTIFIER)
>> + size += sizeof(data->id);
>> +
>> if (sample_type & PERF_SAMPLE_ID)
>> size += sizeof(data->id);
>>
>> @@ -4229,7 +4232,7 @@ static void __perf_event_header__init_id(struct perf_event_header *header,
>> if (sample_type & PERF_SAMPLE_TIME)
>> data->time = perf_clock();
>>
>> - if (sample_type & PERF_SAMPLE_ID)
>> + if (sample_type & (PERF_SAMPLE_ID | PERF_SAMPLE_IDENTIFIER))
>> data->id = primary_event_id(event);
>>
>> if (sample_type & PERF_SAMPLE_STREAM_ID)
>> @@ -4268,6 +4271,9 @@ static void __perf_event__output_id_sample(struct perf_output_handle *handle,
>>
>> if (sample_type & PERF_SAMPLE_CPU)
>> perf_output_put(handle, data->cpu_entry);
>> +
>> + if (sample_type & PERF_SAMPLE_IDENTIFIER)
>> + perf_output_put(handle, data->id);
>> }
>>
>> void perf_event__output_id_sample(struct perf_event *event,
>> @@ -4380,6 +4386,9 @@ void perf_output_sample(struct perf_output_handle *handle,
>>
>> perf_output_put(handle, *header);
>>
>> + if (sample_type & PERF_SAMPLE_IDENTIFIER)
>> + perf_output_put(handle, data->id);
>> +
>> if (sample_type & PERF_SAMPLE_IP)
>> perf_output_put(handle, data->ip);
>>
>> --
>> 1.7.11.7
>>
>
>

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/