[LinuxPPS] Experience with current linuxPPS kernel patch.

[Alexander Gordeev]

В Wed, 03 Mar 2010 20:03:55 -0700
clemens at dwf.com пишет:

> Alexander Gordeev has released a merged version of his and Rodolfo's
> patches as patch-2.6.33-pps536-ts12.tar.gz .

Yep, the link is:
http://lvk.cs.msu.su/~lasaine/timesync/patches/patch-2.6.33-pps536-ts12.tar.gz

> It is not obvious to me what version of Rodolfo's patch this
> incorporates, but I will assume it is one of the more recent.  (I
> tried diff'ing various pieces w/o success)

It was the most recent patchset that Rodolfo sent to LKML at the
moment. I think it is still the bleeding edge.

> In any case, this includes his kernel patch and his paralell port
> patch. I cant test the parallel port patch, but have been running the
> kernel patch and a serial PPS source.
> 
> And let me start by saying that this seems to work well with/without
> the kernel code tuned on, and on 32/64bit machines running Fedora11.
> I am not getting quite the accuracy I would expect (see below).
> 
> [[ I did have problems with previous versions, so if you are trying to
> run a previous version, you want this one. ]]
> 
> OK, (minor) patch errors:
> 
> Patching the file Documentation/ioctl/ioctl-number.txt fails
> its looking for an upper case A1-A4 and its a1-a4 in the file.
> 
> (thats it, nothing that will affect the compile).

This error appears because you apply the patch on top of the 2.6.32
kernel while it is for 2.6.33 :)

> OK, Compile errors:
> 
> In the file common.c (you dont have to know where it is) the include
> file pps_kernel.h is included TWICE, with lots of errors generated
> by gcc about 'error: redefinition of ...'
> 
> I didnt bother to find HOW it was being included twice, but just
> went to pps_kernel.h and put a
> 
>         #ifndef PPS_DEFINES
>         #define PPS_DEFINES
> 
>         ...
> 
>         #endif
> 
> arround the text in the file.

Yes, I know about this. This is a problem in the original PPS code. I
didn't add it to one of my commits to make the review process simpler.

> I still have the feeling that something is not quite right (but mabe
> elsewhere in
> the kernel code) since the accuracy is just about 1us.  That is, the
> 'fuzz' of the timestamps around the mean is just about 1us.  I would
> expect a factor of x10 better than that, and would not be surprised
> if there is not still some bad ns <-> us conversions in the kernel.

Still 1us precision is good enough for me. :)

-- 
  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/20100323/b19f2709/attachment.pgp