[LinuxPPS] Which Gordeev patches to use?

[clemens at dwf.com]

> On Fri, Aug 06, 2010 at 01:34:16PM -0600, reg at dwf.com wrote:
> > There were several comments/corrections in the Re: lines for
> > the Gordeev patches.  Have these changes been incorporated
> > into the patch set anywhere, or are they generally, not needed?
> 
> There are some issues whose need fixes.
> 
Can I add one odd thing Im seeing to the list.

After running for a number of hours, I would expect ntpd to have settled down
to within 1us offset of the 'correct' time, and it does with your previous code.

However with the current patch set I see:

ntpdc> dmpeers
     remote           local      st poll reach  delay   offset    disp
===============================================
 LOCAL(0)        127.0.0.1        5   64   40 0.00000  0.000000 2.81778
.clemens-fw.dwf. 192.168.64.17    1   64   76 0.00027  0.000031 0.66222
.mythtv.dwf.com  192.168.64.17    1   64   77 0.00011  0.000097 0.43698
 orion.dwf.com   192.168.64.17   16   64    0 0.00000  0.000000 3.99217
*GPS_ONCORE(0)   127.0.0.1        0   16  377 0.00000  0.000100 0.01521
 mythtv-fe.dwf.c 192.168.64.17   16   64    0 0.00000  0.000000 3.99217

which shows ntpd off by 100us.  And it stays there.

Using my ppsapitest, which just prints out the ASSERT and CLEAR timestamps, I see

CLEAR	190.19
ASSERT    190.99
CLEAR      191.19
ASSERT    191.99
    ...

Now there is something wrong here.  Clear should follow Assert by about 0.2s
and its the other way around.  It would appear that Clear and Assert are being
mislabeled (exchanged).

Changing ASSERT -> CLEAR in the ntp config file, SEEMED to clear up the problem,
but watching it closer, the offset cycled between 0->100->0, and that I cant explain
at all without watching the individual timestamps go by in a bit more detail.

Something strange is wrong.

-- 
                                        Reg.Clemens
                                        reg at dwf.com