tc_windup() calls ntp_update_second() up to 200 times:
/*
* A large step happens on boot. This constant detects such steps.
* It is relatively small so that ntp_update_second gets called enough
* in the typical 'missed a couple of seconds' case, but doesn't loop
* forever when the time step is large.
*/
#define LARGE_STEP 200
ntp_update_second() subtracts some amount of time from adjtimedelta and
sets th->th_adjustment accordingly. However, if ntp_update_second is
called more than once in a row, only the last value of th->th_adjustment
is actually applied. Shouldn't the adjustments from all ntp_update_second
calls be added instead? I don't know how often this is actually hit in
practice, the normal case is that ntp_update_second is called only once.
diff --git sys/kern/kern_tc.c sys/kern/kern_tc.c
index 59b31dc..60ea119 100644
--- sys/kern/kern_tc.c
+++ sys/kern/kern_tc.c
@@ -425,8 +426,12 @@ tc_windup(void)
i = bt.sec - tho->th_microtime.tv_sec;
if (i > LARGE_STEP)
i = 2;
- for (; i > 0; i--)
- ntp_update_second(&th->th_adjustment, &bt.sec);
+ th->th_adjustment = 0;
+ for (; i > 0; i--) {
+ uint64_t adj;
+ ntp_update_second(&adj, &bt.sec);
+ th->th_adjustment += adj;
+ }
/* Update the UTC timestamps used by the get*() functions. */
/* XXX shouldn't do this here. Should force non-`get' versions. */
