GSoC Project | Basic Support for Trace Compass

Sebastian Huber sebastian.huber at embedded-brains.de
Fri Aug 2 10:56:13 UTC 2019


On 02/08/2019 10:45, Sebastian Huber wrote:
> On 02/08/2019 10:38, Sebastian Huber wrote:
>> On 31/07/2019 08:30, Ravindra Kumar Meena wrote:
>>>      > The bit field is documented in the manual. I would simplify 
>>> this to:
>>>      >
>>>      > cctx->switch_out_int[ item->cpu ].prev_state = ( ( ( item->data
>>>      >> 24 )
>>>      > & 0x7 ) == 1 ) ? TASK_IDLE : TASK_RUNNING;
>>>
>>>     Did you check how the Trace Compass behaviour changed with this?
>>>
>>> Yes. I imported the trace on Trace Compass. The CPU 5 Threads and CPU 
>>> 5 States lines are straight in the beginning because there is no 
>>> change in the cpu_id. At the end of the trace, there is a lot of 
>>> change in cpu_id due to which Trace Compass shows other CPU thread 
>>> and state lines.
>>>
>>> I have attached the screenshot of trace compass. Have a look.
>>
>> Yes, this is similar to what I see. I don't think this is what we 
>> want. The 09.. are all idle threads. What I want is something similar 
>> to the LTTNG output, see attached screen shot.
>>
>> It seems the prev_state is actually the state of the new thread. 
>> Doesn't really make sense to me. Could you please clarify this with 
>> the LTTNG mailing list?
> 
> It seems that there are some special cases for the swapper which has a 
> tid of 0. Could you please adjust the code so that idle threads get a 
> tid of 0.

Yes, this seems to achieve what we need. I updated the raw data so that 
more task switches are included.

With your current version the program doesn't terminate, I guess there 
is a bug in your new hash table code.

-- 
Sebastian Huber, embedded brains GmbH

Address : Dornierstr. 4, D-82178 Puchheim, Germany
Phone   : +49 89 189 47 41-16
Fax     : +49 89 189 47 41-09
E-Mail  : sebastian.huber at embedded-brains.de
PGP     : Public key available on request.

Diese Nachricht ist keine geschäftliche Mitteilung im Sinne des EHUG.



More information about the devel mailing list