Re: [PATCH] mm/zsmalloc: add trace events for zs_compact

[Ganesh Mahendran <opensource.ganesh@gmail.com>]

2016-06-13 12:42 GMT+08:00 Minchan Kim <minchan@kernel.org>:
> On Wed, Jun 08, 2016 at 02:39:19PM +0800, Ganesh Mahendran wrote:
>
> <snip>
>
>> zsmalloc is not only used by zram, but also zswap. Maybe
>> others in the future.
>>
>> I tried to use function_graph. It seems there are too much log
>> printed:
>> ------
>> root@leo-test:/sys/kernel/debug/tracing# cat trace
>> # tracer: function_graph
>> #
>> # CPU  DURATION                  FUNCTION CALLS
>> # |     |   |                     |   |   |   |
>>  2)               |  zs_compact [zsmalloc]() {
>>  2)               |  /* zsmalloc_compact_start: pool zram0 */
>>  2)   0.889 us    |    _raw_spin_lock();
>>  2)   0.896 us    |    isolate_zspage [zsmalloc]();
>>  2)   0.938 us    |    _raw_spin_lock();
>>  2)   0.875 us    |    isolate_zspage [zsmalloc]();
>>  2)   0.942 us    |    _raw_spin_lock();
>>  2)   0.962 us    |    isolate_zspage [zsmalloc]();
>> ...
>>  2)   0.879 us    |      insert_zspage [zsmalloc]();
>>  2)   4.520 us    |    }
>>  2)   0.975 us    |    _raw_spin_lock();
>>  2)   0.890 us    |    isolate_zspage [zsmalloc]();
>>  2)   0.882 us    |    _raw_spin_lock();
>>  2)   0.894 us    |    isolate_zspage [zsmalloc]();
>>  2)               |  /* zsmalloc_compact_end: pool zram0: 0 pages
>> compacted(total 0) */
>>  2) # 1351.241 us |  }
>> ------
>> => 1351.241 us used
>>
>> And it seems the overhead of function_graph is bigger than trace event.
>>
>> bash-3682  [002] ....  1439.180646: zsmalloc_compact_start: pool zram0
>> bash-3682  [002] ....  1439.180659: zsmalloc_compact_end: pool zram0:
>> 0 pages compacted(total 0)
>> => 13 us > 1351.241 us
>
> You could use  to cut out.
>
> To introduce new event trace to get a elasped time, it's pointless,
> I think.

Agree.

>
> It should have more like pool name you mentioned.
> Like saying other thread, It would be better to show
> [pool name, compact size_class,
> the number of object moved, the number of freed page], IMO.

Thanks for you suggestion!
I would be useful to see compact details for each class.
I will send another patch to do this.

Thanks.

>
> Thanks.