Re: [PATCH 14/42] perf record: Add --index option for building index table

From: Adrian Hunter
Date: Mon Feb 02 2015 - 07:09:25 EST


On 02/02/15 12:05, Jiri Olsa wrote:
> On Mon, Feb 02, 2015 at 11:52:26AM +0200, Adrian Hunter wrote:
>> On 02/02/15 11:15, Jiri Olsa wrote:
>>> On Mon, Feb 02, 2015 at 10:34:50AM +0200, Adrian Hunter wrote:
>>>
>>> SNIP
>>>
>>>>> but how about bump up the header version for this feature? ;-)
>>>>>
>>>>> currently it's:
>>>>>
>>>>> struct perf_file_header {
>>>>> u64 magic;
>>>>> u64 size;
>>>>> u64 attr_size;
>>>>> struct perf_file_section attrs;
>>>>> struct perf_file_section data;
>>>>> /* event_types is ignored */
>>>>> struct perf_file_section event_types;
>>>>> DECLARE_BITMAP(adds_features, HEADER_FEAT_BITS);
>>>>> };
>>>>>
>>>>>
>>>>> - we already store attrs as a FEATURE so we could omit that
>>>>> - your patch stores only synthesized data into 'data' section (-1 idx)
>>>>> this could be stored into separate file and get merged with the rest
>>>>> - new header version would have 'features' section, so the features
>>>>> position wouldnt depend on the 'data' end as of now and we could
>>>>> easily store after all data is merged:
>>>>>
>>>>> struct perf_file_header {
>>>>> u64 magic;
>>>>> u64 size;
>>>>> u64 attr_size;
>>>>> struct perf_file_section features;
>>>>> DECLARE_BITMAP(adds_features, HEADER_FEAT_BITS);
>>>>> };
>>>>>
>>>>>
>>>>> thoughts?
>>>>
>>>> How come the features are being written before the sample data anyway?
>>>> I would have expected:
>>>> - write the data (update the index in memory)
>>>> - write the features (including index)
>>>>
>>>
>>> I think the problem is that the only way how to get features offset
>>> right now is via perf_file_header::data.offset + perf_file_headerdata.size,
>>> and we still use this section to carry 'sythesized' data, so it needs
>>> to have correct size.
>>
>> Why not make it the same as all the other data. i.e. find the start and size
>> via the index? And then just lump all the data together?
>
> thats what I suggested

No, I meant really lump it all together. i.e. perf_file_header.data.size =
total data size

>
>>
>>> I guess we could workaround that by storing the 'perf_file_header::data'
>>> as the last data section. That would require to treat it the same way as
>>> all other data sections, but we could keep current header layout.
>>
>> Would it need to be last? Logically it should precede the data that depends
>> on it.
>
> i suggested this as a workaround for having features at the end of the file
> while keeping the current perf data header

Which wouldn't be necessary if you lump it all together?

--
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/