[LinuxPPS] LinuxPPS GPIO Generator

[Alexander Gordeev]

В Wed, 24 Aug 2011 13:43:22 +0200
Tom Kurmann <tkurmann at gmail.com> пишет:

> Hello,
> 
> For a project I need to to synchronize a handful of Cortex A8 based devices
> to a master clock so I can time stamp data which is coming from an ADC. The
> master clock should be synchronized to GPS and the slaves will be locally
> synchronized with PTP. The ADCs have a SYNC input which I would like to sync
> with a GPIO generating a PPS. Since I have been away from the Linux Kernel
> for a couple of years these questions may sound dumb, but a lot has changed
> since I last worked on the kernel...I need a synchronization of around 50 us
> in the network.
> 
> - Is it correct that the IRQ time stamping for say a GPS PPS happens before
> the ISR gets called? Since the interrupt latency on the cortex a8 is max.
> around 120 us and the jitter is quit high i would assume that the time
> stamping has to be done before the ISR gets called.

No, this is done in ISR.

> - I see that there are no drivers yet to generate a PPS on a GPIO, which
> would mean I could try to contribute something here :) Since hrtimers has
> been implemented I think it should be possible to generate an interrupt
> every full second for a gpio (the full second is to ensure that the clocks
> are synced), but again the problem with the interrupt latency comes up. If I
> have say an average of 100 us latency and a jitter of 50 us I would be at
> the limits of the specifications. Or is the hrtimer interrupt handled
> differently from external interrupts?

Please look at how pps-gen-parport and pps-parport work together. There
are some problems, but it works very well for me.

pps-gen-parport produces precise signal of 30 us duration where the
clear edge is the second boundary.

The key is that the interrupt in pps-parport is generated when the
signal goes down (assert edge) and then it polls for the signal to go
up again (clear edge). This way the accuracy of timestamping of the
clear edge is about 1-2us (time to read/write a port on x86) without
any interrupt-related latency.

-- 
  Alexander
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 490 bytes
Desc: not available
Url : http://ml.enneenne.com/pipermail/linuxpps/attachments/20110824/0350d4f6/attachment.pgp