[LinuxPPS] linuxpps-v5.3.2 is missing ioctl32 compatibility

Cirilo Bernardo cirilo.bernardo at gmail.com
Tue Oct 28 10:13:30 CET 2008


On Tue, Oct 28, 2008 at 11:02 AM, George Spelvin <linux at horizon.com> wrote:
> I'm staring at the code, trying to figure out why ntpd breaks linuxpps.
>
[snip]
> Anyway, it looks to me like it would be generally better design to
> associate the PPS params with the file descriptor rather than the
> device.  So you could have two users, one asking for CAPTUREASSERT
> and the other asking for CAPTURECLEAR.  And each user could have
> their own offsets and/or timestamp format.
>
> It would complicate the poll wakeup a bit, but sure give a cleaner
> programming model.
>


There has been some discussion before that a possible execution path
within ntpd may set the line discipline back to N_TTY (or perhaps the
request to change the line discipline is not handled correctly by the
serial driver?)

I don't see a point in being able to use both CAPTUREASSERT and
CAPTURECLEAR - usually only one of these provides a reliable timing
reference and the other will have a (usually) unspecified jitter.
Also, if the association is on a per-file descriptor basis, how do we
make the selection without trying to push more changes through NTPD?

- Cirilo



More information about the LinuxPPS mailing list