On Sat, 4 Dec 2004, YONETANI Tomokazu wrote:
>> I wonder, if you revert kern_clock.c:1.19 (i.e. set boottime =
>> basetime) do you still see your problem from April 2003? I can't
>> seem to reproduce it at all.
>
> I've confirmed that it doesn't. At lease uptime, ntpdate and ps are working
> correctly. I'm not 100% sure that this change affects other parts of the
> system that depend on this twisted boottime though.
I doubt they do -- I'm sure the 2x uptime is unintentional. Here
is a quick fix, could someone please commit something like this?
Index: kern_clock.c
===================================================================
RCS file: /u01/cvs-repositories/dcvs/src/sys/kern/kern_clock.c,v
retrieving revision 1.27
diff -u -r1.27 kern_clock.c
--- kern_clock.c 20 Nov 2004 20:25:09 -0000 1.27
+++ kern_clock.c 4 Dec 2004 15:48:59 -0000
@@ -227,7 +227,7 @@
basetime.tv_nsec += 1000000000;
--basetime.tv_sec;
}
- boottime.tv_sec = basetime.tv_sec - mycpu->gd_time_seconds;
+ boottime = basetime;
timedelta = 0;
crit_exit();
}
As far as I can tell, both boottime/basetime serve the same purpose
and one of the should be removed because as it stands now, boottime
is only adjusted by set_timeofday(). Finer clock adjustments ala
adjtime() do not currently adjust boottime (which they should)
....that is probably a patch for another day.
To see first hand your real uptime sliping away from you, here's a
short test program. Getting 0 and 0 for both differences is what
you should be getting (well, 99.99% of the time :-)
su-2.05b# cc -o x x.c
su-2.05b# ./x
Real time diff: 0
boottime diff: -262
su-2.05b# cat x.c
#include <sys/time.h>
#include <sys/sysctl.h>
#include <stdio.h>
int main() {
struct timeval tv1, tv2, b1, b2;
int l = sizeof(tv1);
sysctlbyname("kern.boottime", &b1, &l, NULL, 0);
gettimeofday(&tv1, NULL);
settimeofday(&tv1, NULL); // <--- This slides boottime back
gettimeofday(&tv2, NULL);
sysctlbyname("kern.boottime", &b2, &l, NULL, 0);
printf("Real time diff: %ld\n", tv2.tv_sec - tv1.tv_sec);
printf("boottime diff: %ld\n", b2.tv_sec - b1.tv_sec);
return 0;
}
-Paul.