[LinuxPPS] http://lkml.org/lkml/2008/4/11/42

[Bernhard Schiffner]

Hi Rodolfo,

thanks for stepping up again. Seems the only opponent (of details) is Alan 
Cox now. Congrats so far.

But the details he is speaking of are IMHO crucial to linuxpps. That's why 
I'am writing this message.

If I understand Alan right, he wants to "to push up in the stack" the place 
where linuxpps gathers its raw data, here in case of DCD / serial.
This is in fundamental contradiction to the request of getting an as 
precise time information as possible.

James Boddington did a wounderful analysis about this theme and it fits 
best here:
http://ml.enneenne.com/pipermail/linuxpps/2008-April/001738.html
In short: every shift in stack adds delay and jitter.

I have no advice how to solve this problem in a general way.  May be there 
is no way at all.
But what about a ladder-type approach where to gather data? One of  these 
steps is good for a generalist like Alan, others are good for specialists 
like James. 

I have no clue on how to handle this mess. But I'd like you to ask others 
on this topic to get a well balanced opinion. I'am thinking of Thomas 
Gleixner, John Stultz, Roman Zippel , James Boddington to ask.

TIA!

Bernhard