[ltt-dev] trace-clock-userspace.patch for ARM/PowerPC ?

[Mathieu Desnoyers]

* Abbas Raza (abbas_raza at mentor.com) wrote:
> On 07/26/2011 11:02 PM, Mathieu Desnoyers wrote:
>> * Abbas Raza (abbas_raza at mentor.com) wrote:
>>> On 07/26/2011 10:17 PM, Mathieu Desnoyers wrote:
>>>> * Abbas Raza (abbas_raza at mentor.com) wrote:
>>>> [...]
>>>>> Hi Mathieu,
>>>>>
>>>>> Some confusion here :(
>>>>>
>>>>> In case of LTTng 0.x, when we take kernel traces then timestamps for
>>>>> these traces is based on TSC/TB . So after that we view these traces by
>>>>> lttv and it shows timestamps with all events which occurred.
>>>>> I print the timestamp in kernel ltt function 'ltt_trace_alloc()'  and it
>>>>> comes out to be 22944596907 (by calling trace_clock_read64() ). While
>>>>> lttv show Birth sec of first event to 367 second. How this conversion is
>>>>> made from TSC to the Birth sec/nsec value displayed in lttv? just want
>>>>> to confirm that whether timestamps shown in lttv (Birth sec/nsec) are
>>>>> based on TSC values... if yes then how TSC values are converted to Birth
>>>>> sec/nsec and i no then from where lttv gets timestamps to be displayed?
>>>>> Or in simple words, do kernel traces which we get contain timestamps
>>>>> based simply on TSC values or it is something else?
>>>> Yes, we write, in the lttng 0.x headers, the following information:
>>>>
>>>> struct ltt_subbuffer_header {
>>>>    [...]
>>>>           uint64_t start_freq;            /*
>>>>                                            * Frequency at trace start,
>>>>                                            * used all along the trace.
>>>>                                            */
>>>>           uint32_t freq_scale;            /* Frequency scaling (divisor) */
>>>> [...]
>>>> }
>>>>
>>>> so by using the start frequency and the scale, we can convert from TSC
>>>> values to nanoseconds. Please note that LTTng 0.x requires the clock
>>>> source to appear as if it has a constant rate.
>>>>
>>>> These values map to trace_clock_frequency() and trace_clock_freq_scale()
>>>> trace clock functions.
>>>>
>>>> Best regards,
>>>>
>>>> Mathieu
>>>>
>>>>> Thanks a lot again for help :)
>>>>>
>>>>>
>>>>> Abbas Raza
>>>>>
>>> Just to confirm that TSC's just have raw values of clock cycles and lttv
>>> converts these values into sec/nsec by reading ltt_subbuffer_header ?
>> Yes, that's it.
>>
>> Thanks,
>>
>> Mathieu
>>
>>>
>>> Thanks!
>>>
>>> Abbas Raza
>>>
>
> Hello Mathieu,
>
> Just wanted to ask that trace clock userspace patch for x86 adds  
> functionality to register posix clocks like
>
> register_posix_clock(CLOCK_TRACE, &clock_trace);
> register_posix_clock(CLOCK_TRACE_FREQ, &clock_trace_freq);
>
> What we understood is that these clocks are registered to provide system  
> call functionality for CLOCK_TRACE and CLOCK_TRACE_FREQ clocks just like  
> other posix clocks. Right?
>
> And same we will have to do for powerpc (even if we have got vdso  
> functionality for these clocks)?

Yes, exactly. The kernel always needs to provide a system call fallback
for vDSO, just in case the vDSO cannot be used.

Thanks

Mathieu

>
> Thanks !
>
> --
> Abbas Raza
>

-- 
Mathieu Desnoyers
Operating System Efficiency R&D Consultant
EfficiOS Inc.
http://www.efficios.com