[LinuxPPS] GPIO PPS Generator

[Igor Plyatov]

Hello Drasko!

> On Thu, May 23, 2013 at 12:03 PM, Drasko DRASKOVIC
> <drasko.draskovic at gmail.com> wrote:
>> Hello Igor,
>> think you very much for the code and the instructions. It has been
>> extremely helpful, and I integrated
>> your code in Ubuntu Precise 3.2 Linux kernel for my Pandaboard.
>>
>> Everything seems to work, and I can see PPS on the oscilloscope.
>>
>> My use case is not a standard PPS generator, though. I thought that I
>> can use this mechanism to observe clock synchronicity (offest +
>> jitter) between two Pandaboards.
>>
>> However, driver's hrtimer is started at the moment the driver is
>> inserted into the kernel, and keeps pulsing every second. While this
>> can be adequate for observing clock jitter between two boards, it is
>> impossible to have offset observation, as initial offset will be
>> unpredictable.
>>
>> What I am looking now is to have a pulse at exact sharp second in the
>> sense : 1:00, 2:00, etc..., so, zero nanoseconds after round sharp
>> second in local time.
>>
>>  From what I can see in the code, we are always setting hrtimer to
>> expire a little before and using getnstimeofday() to get our (local)
>> time and conclude second. That would mean (I think) that if we can
>> just start our hrtimer on exact round second, then after this it would
>> continue pulsing on each second exactly.
>>
>> Do you have some suggestions how this can be done ?
>>
>> Thanks and best regards,
>> Drasko
> Hmmm... On the more carefull inspection of the code, it seems that
> this driver does exectly what I want, i.e. that it "fires" on the
> exact absolute second.

Yes.

> Igor, can you please confirm me that GPIO is toogled every round
> second (i.e. 1:00, 2:00, i.e. when mS, uS and nS are zero), so that
> it's phase is zero, and not with offset from this zeo-phase (i.e.
> 1s:25ms:52us, 2s:25ms:52us, etc...).

Yes.

> Can you also please explain why this minimal hardcoded 10ms
> (ts.tv_nsec > 990 * NSEC_PER_MSEC) :
> static inline ktime_t next_intr_time(struct pps_generator_gpio_data *data)
> {
> 	struct timespec ts;
>
> 	getnstimeofday(&ts);
> 	return ktime_set(ts.tv_sec +
> 			((ts.tv_nsec > 990 * NSEC_PER_MSEC) ? 1 : 0),
> 			NSEC_PER_SEC - (send_delay +
> 			data->port_write_time + 3 * SAFETY_INTERVAL));
> }
>
> Thanks and best regards,
> Drasko

Don't know exactly why barrier is 990 ms, but this function makes set up 
of HR Timer to moment just before beginning of next second.
When HR Timer counts to this moment, then hrtimer_event() started and 
wait for exact beginning of second to assert GPIO.
...
     /* busy loop until the time is right for an assert edge */
     do {
         getnstimeofday(&ts2);
     } while (expire_time.tv_sec == ts2.tv_sec && ts2.tv_nsec < lim);
     gpio_set_value(data->pdata->gpio, 1); /* set the signal */
...

Best regards!

-- 
Igor Plyatov