[ltt-dev] Timestamping for ARM

[ltt-dev] Timestamping for ARM

[Gaurav Singh]

Hi Matheiu/All,

I have managed to supply a timer to LTTng on my ST Nomadik (ARM based
chip) 8815 board. Luckily there are timers that I can use.
I am using another custom designed tool to find times for interrupt
handlers and bottom halves. After using the new timer for timestamping
I am getting similar results in LTT which probably proves the working
of the method.
However are there any other tests I can use to verify the working?
Also interrupt handler times are wrong using the LTTV version
(lttv-0.10.0-pre11-10032008). The patch supplied earlier is giving me
the correct times. Is the patch correct? What is the status of the
Interrupt handler time plugin in LTTV? I see that I have to load it
seperately - does that mean it is not fully supported?


Regards
Gaurav

On Mon, Sep 22, 2008 at 9:50 PM, Mathieu Desnoyers
<compudj at krystal.dyndns.org> wrote:
> Yeah, generic timestamping in LTTng is just a simple event counter and
> does not reflect the time elapsed between events.
>
> Sure, you can use a different time source if your specific ARM board
> supports it. It's just a matter of looking at what a patch like
> lttng-timestamp-powerpc.patch does for powerpc and do the same for ARM.
> You'll probably want to refer to your arch-specific documentation to
> find out if you have an high-precision fast time source available which
> is synchronised across CPUs.
>
> If you do it correctly (with the right ifdefs and HAVE_* dependency), I
> could pull this patch into LTTng so arm boards which have such time
> source could have such timing information.
>
> That said, it might be good to extend arch/arm/include/asm/timex.h to
> make get_cycles() support your timestamp on your specific board...
>
> Mathieu
>
> * Gaurav Singh (gausinghnsit at gmail.com) wrote:
>> Hi,
>> I am using an ARM based CPU. Using generic timestamping is giving
>> wrong time taken information as it shows most of the time taken in the
>> timer inteerupt handler. Is there any other way to present the correct
>> timestamping information. As I understand we can use another clock
>> source to provide timestamps.
>>
>> Regards
>> Gaurav Singh
>>
>> _______________________________________________
>> ltt-dev mailing list
>> ltt-dev at lists.casi.polymtl.ca
>> http://lists.casi.polymtl.ca/cgi-bin/mailman/listinfo/ltt-dev
>>
>
> --
> Mathieu Desnoyers
> OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68
>

[ltt-dev] Timestamping for ARM

[Mathieu Desnoyers]

Hi Gaurav,

I think comparing results you get from sampling to the results you get
from LTTV analysis is a good way to do it.

Since you seem to be interested in this view, you might want to have a
look at the average, std. dev. and other "more complex" data derived
from the min/max : you'll probably notice that there are still some big
rounding errors in there (1 / (unsigned long) is never a good idea). If
you have time to look into it, this would be helpful.

Thanks !

Mathieu

* Gaurav Singh (gausinghnsit at gmail.com) wrote:
> Hi Matheiu/All,
> 
> I have managed to supply a timer to LTTng on my ST Nomadik (ARM based
> chip) 8815 board. Luckily there are timers that I can use.
> I am using another custom designed tool to find times for interrupt
> handlers and bottom halves. After using the new timer for timestamping
> I am getting similar results in LTT which probably proves the working
> of the method.
> However are there any other tests I can use to verify the working?
> Also interrupt handler times are wrong using the LTTV version
> (lttv-0.10.0-pre11-10032008). The patch supplied earlier is giving me
> the correct times. Is the patch correct? What is the status of the
> Interrupt handler time plugin in LTTV? I see that I have to load it
> seperately - does that mean it is not fully supported?
> 
> 
> Regards
> Gaurav
> 
> On Mon, Sep 22, 2008 at 9:50 PM, Mathieu Desnoyers
> <compudj at krystal.dyndns.org> wrote:
> > Yeah, generic timestamping in LTTng is just a simple event counter and
> > does not reflect the time elapsed between events.
> >
> > Sure, you can use a different time source if your specific ARM board
> > supports it. It's just a matter of looking at what a patch like
> > lttng-timestamp-powerpc.patch does for powerpc and do the same for ARM.
> > You'll probably want to refer to your arch-specific documentation to
> > find out if you have an high-precision fast time source available which
> > is synchronised across CPUs.
> >
> > If you do it correctly (with the right ifdefs and HAVE_* dependency), I
> > could pull this patch into LTTng so arm boards which have such time
> > source could have such timing information.
> >
> > That said, it might be good to extend arch/arm/include/asm/timex.h to
> > make get_cycles() support your timestamp on your specific board...
> >
> > Mathieu
> >
> > * Gaurav Singh (gausinghnsit at gmail.com) wrote:
> >> Hi,
> >> I am using an ARM based CPU. Using generic timestamping is giving
> >> wrong time taken information as it shows most of the time taken in the
> >> timer inteerupt handler. Is there any other way to present the correct
> >> timestamping information. As I understand we can use another clock
> >> source to provide timestamps.
> >>
> >> Regards
> >> Gaurav Singh
> >>
> >> _______________________________________________
> >> ltt-dev mailing list
> >> ltt-dev at lists.casi.polymtl.ca
> >> http://lists.casi.polymtl.ca/cgi-bin/mailman/listinfo/ltt-dev
> >>
> >
> > --
> > Mathieu Desnoyers
> > OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68
> >
> 

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68