Interrupt latency in RTEMS (Zedboard)

Rehab Massoud rehab.massoud at gmail.com
Tue Mar 27 11:56:53 UTC 2018


Needless to say also you can't measure software execution time that
accurately with software. It's not only that it's intrusive but also would
never be cycle accurate. If you really need such cycle accuracy you might
think about hardware-based tracing techniques.
On Mar 27, 2018 13:52, "BRIARD Sebastien" <
sebastien.briard at thalesaleniaspace.com> wrote:

> Yeah, hardware limits to 1.5ns if we take in account the max frequency of
> the cortex. (That was the more or less 1ns =) )
>
>
>
> *De :* Rehab Massoud [mailto:rehab.massoud at gmail.com]
> *Envoyé :* mardi 27 mars 2018 13:49
> *À :* BRIARD Sebastien; rtems-users at rtems.org
> *Objet :* Re: Interrupt latency in RTEMS (Zedboard)
>
>
>
> Hi, the smaller theoretical delay accuracy you can measure (without adding
> to an ASIC cascaded FF) is one clock cycle. I think the maximum frequency
> that could be achieved on zedboard is not more than 800 MHz, and the
> maximum Zync's Cortex freq per Zedboard's datasheet is 667 MHz, which means
> you can't achieve 1 nanosecond accuracy even with hardware measurements,
> right?
>
> On Mar 27, 2018 13:30, "BRIARD Sebastien" <sebastien.briard@
> thalesaleniaspace.com> wrote:
>
>
>
>
>
> *De :* BRIARD Sebastien
> *Envoyé :* mardi 27 mars 2018 11:42
> *À :* 'users at rtems.org'
> *Objet :* Interrupt latency in RTEMS (Zedboard)
>
>
>
> Hi,
>
>
>
> I was trying to measure the interrupt latency in RTEMS with a Xilinx Zynq
> Zedboard (cortex A9).
> I modified the c file in classic_signal example to measure time in the
> main loop and right after entering the handler.
> That might seem artificial but well, it gives me a value with a relatively
> simple code.
>
> I used this code for 100, 10000, and one million ticks per second. I am
> trying to understand how the measure can quite equal with 100 ticks per
> second and a million.
> (I obtained values between 500ns and 1000ns).
> Is there another timer that is used for interrupt processing ?
>
> A subsequent question, is it possible to use a larger value than a tick
> per microsecond ?  Maybe I am confusing a little between frequency and tick
> clock in RTEMS but I would like to run test with more or less a tick per
> nanosecond.
>
> Thank you,
> Sébastien.
>
>
> _______________________________________________
> users mailing list
> users at rtems.org
> http://lists.rtems.org/mailman/listinfo/users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/users/attachments/20180327/5da4e957/attachment-0001.html>


More information about the users mailing list