Ulrich Windl writes on linux-kernel:
I just uploaded “PPSkit-0.6-i386-pre5.diff.gz” to the usual
This snapshot features clock_getres() similar to POSIX.4, and
the routine is used to set time_precision. Debugging messages in
the problem candidate, hardpps(), were improved, and a quick and
dirty fix seems to work well (ntptime and ntp4.0.91f: stability
0.082ppm, jitter <5?s, precision 7.5?s, estimated error
I also moved some time routines from i386/kernel/time.c to
kernel/time.c, because they do not make much sense in the
architecture dependent file.
Stickyness of TIME_ERROR has also been fixed. Setting the time
no longer sets TIME_ERROR, instead it is deduced from
I have completed most of the important parts, but I’ll have to
check some details of hardpps with Dave Mills. I’ll use this
weekend to update documentation and utilities, so that a real
release comes definitely closer.
I’ll describe the changes made to the time stuff, requesting
gurus for the other architectures to implement nanosecond
timeoffsets. I’ll try to merge most of the changes ready for use,
but I can’t test these architectures.
One current issue is: For compatibility includes in my snapshot
now. Therefore any change in the time stuff recompiles about
everything (e.g. drivers/scsi/sd.c). As many parts don’t really
need the , I’m considering to take it away from . Maybe I’ll have
to add it explicitly for some files. A compilation will show.
The time code changes are huge, but I think it was time for a
major cleanup. Probably more a candidate for 2.3 than for 2.2, but
I think we can extract some goodies for the stable kernel (if