> does anyone know of a successful deployment of LinuxPPS with newer
> kernel (4.4), ntp (4.2.8p?) and PPS over serial interface on x86-based
> PCs? Or of any known issues that linuxpps has with such a configuration?
I am running ntp 4.2.8p9 successfully on a Raspberry Pi with kernel 4.4.39
and kernel-bound PPS, albeit via a GPIO input rather than a serial
interface.
I think your issue may be that you are building NTP with a local copy of
linuxpps rather than the version now integrated into newer kernels. When
I execute ./configure prior to compiling ntp, I get:
configure:28418: checking for sys/timepps.h
configure:28418: gcc -c -g -O2 conftest.c >&5
configure:28418: $? = 0
configure:28418: result: yes
:
configure:29839: checking timepps.h usability
configure:29839: gcc -c -g -O2 conftest.c >&5
conftest.c:233:21: fatal error: timepps.h: No such file or directory
#include <timepps.h>
^
compilation terminated.
and in the generated config.h file I get:
/* Define to 1 if you have the <sys/timepps.h> header file. */
#define HAVE_SYS_TIMEPPS_H 1
:
/* Define to 1 if you have the <timepps.h> header file. */
/* #undef HAVE_TIMEPPS_H */
These 2 defines are used in file ntpd/ppsapi_timepps.h to select which
timepps.h
to include in the compile. <timepps.h> is used in preference to
<sys/timepps.h>
if the former exists. The <sys/timepps.h> include file (located in my
system at
/usr/include/sys/timepps.h) contains 2 versions of time_pps_kcbind():
ifdef PPS_KC_BIND
static __inline int time_pps_kcbind(pps_handle_t handle,
const int kernel_consumer,
const int edge, const int
tsformat)
{
struct pps_bind_args __bind_args;
__bind_args.tsformat = tsformat;
__bind_args.edge = edge;
__bind_args.consumer = kernel_consumer;
return ioctl(handle, PPS_KC_BIND, &__bind_args);
}
#else /* !PPS_KC_BIND */
static __inline int time_pps_kcbind(pps_handle_t handle,
const int kernel_consumer,
const int edge, const int
tsformat)
{
/* LinuxPPS doesn't implement kernel consumer feature */
errno = EOPNOTSUPP;
return -1;
}
#endif /* PPS_KC_BIND */
In my system, PPS_KC_BIND is defined in a file /usr/include/linux/pps.h and
so
the first version of the function, which does not return EOPNOTSUPP, is
used.
I would suggest you check that HAVE_TIMEPPS_H is not being defined in your
build (i.e. the 'checking timepps.h usability' test fails) and, if it is,
find your local copy of timepps.h and delete it.
There is a discussion of this part of the ntp config process on this mailing
list here:
http://lists.ntp.org/pipermail/questions/2014-December/039424.html
Good luck,
Richard
_______________________________________________
questions mailing list
questions@lists.ntp.org
http://lists.ntp.org/listinfo/questions
