Re: [PATCH v2 2/3] perf tools: Record sampling time for each entry

[Namhyung Kim <namhyung@kernel.org>]

Hi David,

On Tue, 03 Dec 2013 09:20:23 -0700, David Ahern wrote:
> On 12/3/13, 2:00 AM, Namhyung Kim wrote:
>> diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
>> index 20a7c653b74b..ac65fc67972c 100644
>> --- a/tools/perf/util/evsel.h
>> +++ b/tools/perf/util/evsel.h
>> @@ -69,6 +69,7 @@ struct perf_evsel {
>>   	struct hists		hists;
>>   	u64			first_timestamp;
>>   	u64			last_timestamp;
>> +	u64			*prev_timestamps;
>
>
> Why plural and why dynamically allocated? The allocation only does a
> single u64, not an array.

Nope, it'll be an array if the session was a system-wide one, so plural.

But, I think the current code won't work well if there're multiple
unrelated processes recorded - e.g. perf record -u `id -u` - since it'll
intermix all timestamps between the samples regardless of process.

Hmm.. I think there's not much thing we can do for this without help
from kernel side (PERF_SAMPLE_READ?).  So I'll just drop this unless I
can come up with a better idea.  But the patch 1/3 still makes some
sense and worth to merge by itself IMHO.

What do you think?

Thanks,
Namhyung