time: Move xtime_nsec adjustment underflow handling timekeeping_adjust
authorJohn Stultz <john.stultz@linaro.org>
Fri, 13 Jul 2012 05:21:56 +0000 (01:21 -0400)
committerThomas Gleixner <tglx@linutronix.de>
Sun, 15 Jul 2012 08:39:07 +0000 (10:39 +0200)
When we make adjustments speeding up the clock, its possible
for xtime_nsec to underflow. We already handle this properly,
but we do so from update_wall_time() instead of the more logical
timekeeping_adjust(), where the possible underflow actually
occurs.

Thus, move the correction logic to the timekeeping_adjust, which
is the function that causes the issue. Making update_wall_time()
more readable.

Signed-off-by: John Stultz <johnstul@us.ibm.com>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Prarit Bhargava <prarit@redhat.com>
Link: http://lkml.kernel.org/r/1342156917-25092-8-git-send-email-john.stultz@linaro.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
kernel/time/timekeeping.c

index e43289df28c2789a019d81b096e2549dbf89dd05..aeeaab8cba6e8f39d56b73383fb0286ab803f2cd 100644 (file)
@@ -980,6 +980,27 @@ static void timekeeping_adjust(s64 offset)
        timekeeper.xtime_nsec -= offset;
        timekeeper.ntp_error -= (interval - offset) <<
                                timekeeper.ntp_error_shift;
+
+       /*
+        * It may be possible that when we entered this function, xtime_nsec
+        * was very small.  Further, if we're slightly speeding the clocksource
+        * in the code above, its possible the required corrective factor to
+        * xtime_nsec could cause it to underflow.
+        *
+        * Now, since we already accumulated the second, cannot simply roll
+        * the accumulated second back, since the NTP subsystem has been
+        * notified via second_overflow. So instead we push xtime_nsec forward
+        * by the amount we underflowed, and add that amount into the error.
+        *
+        * We'll correct this error next time through this function, when
+        * xtime_nsec is not as small.
+        */
+       if (unlikely((s64)timekeeper.xtime_nsec < 0)) {
+               s64 neg = -(s64)timekeeper.xtime_nsec;
+               timekeeper.xtime_nsec = 0;
+               timekeeper.ntp_error += neg << timekeeper.ntp_error_shift;
+       }
+
 }
 
 
@@ -1105,27 +1126,6 @@ static void update_wall_time(void)
        /* correct the clock when NTP error is too big */
        timekeeping_adjust(offset);
 
-       /*
-        * Since in the loop above, we accumulate any amount of time
-        * in xtime_nsec over a second into xtime.tv_sec, its possible for
-        * xtime_nsec to be fairly small after the loop. Further, if we're
-        * slightly speeding the clocksource up in timekeeping_adjust(),
-        * its possible the required corrective factor to xtime_nsec could
-        * cause it to underflow.
-        *
-        * Now, we cannot simply roll the accumulated second back, since
-        * the NTP subsystem has been notified via second_overflow. So
-        * instead we push xtime_nsec forward by the amount we underflowed,
-        * and add that amount into the error.
-        *
-        * We'll correct this error next time through this function, when
-        * xtime_nsec is not as small.
-        */
-       if (unlikely((s64)timekeeper.xtime_nsec < 0)) {
-               s64 neg = -(s64)timekeeper.xtime_nsec;
-               timekeeper.xtime_nsec = 0;
-               timekeeper.ntp_error += neg << timekeeper.ntp_error_shift;
-       }
 
        /*
        * Store only full nanoseconds into xtime_nsec after rounding