timekeeping: Use cached ntp_tick_length when accumulating error

[deliverable/linux.git] / include / linux / timekeeper_internal.h

diff --git a/include/linux/timekeeper_internal.h b/include/linux/timekeeper_internal.h
index f7ac48d2edf50e3886b37cd22d3ba64fe3b919f5..e9660e52dc090610cd43a368d03e1cc442114dfe 100644 (file)
--- a/include/linux/timekeeper_internal.h
+++ b/include/linux/timekeeper_internal.h
@@ -90,6 +90,15 @@ struct timekeeper {
        u64                     xtime_interval;
        s64                     xtime_remainder;
        u32                     raw_interval;
+       /* The ntp_tick_length() value currently being used.
+        * This cached copy ensures we consistently apply the tick
+        * length for an entire tick, as ntp_tick_length may change
+        * mid-tick, and we don't want to apply that new value to
+        * the tick in progress.
+        */
+       u64                     ntp_tick;
+       /* Difference between accumulated time and NTP time in ntp
+        * shifted nano seconds. */
        s64                     ntp_error;
        u32                     ntp_error_shift;
        u32                     ntp_err_mult;