timekeeping: Rework frequency adjustments to work better w/ nohz

The existing timekeeping_adjust logic has always been complicated
to understand. Further, since it was developed prior to NOHZ becoming
common, its not surprising it performs poorly when NOHZ is enabled.

Since Miroslav pointed out the problematic nature of the existing code
in the NOHZ case, I've tried to refactor the code to perform better.

The problem with the previous approach was that it tried to adjust
for the total cumulative error using a scaled dampening factor. This
resulted in large errors to be corrected slowly, while small errors
were corrected quickly. With NOHZ the timekeeping code doesn't know
how far out the next tick will be, so this results in bad
over-correction to small errors, and insufficient correction to large
errors.

Inspired by Miroslav's patch, I've refactored the code to try to
address the correction in two steps.

1) Check the future freq error for the next tick, and if the frequency
error is large, try to make sure we correct it so it doesn't cause
much accumulated error.

2) Then make a small single unit adjustment to correct any cumulative
error that has collected over time.

This method performs fairly well in the simulator Miroslav created.

Major credit to Miroslav for pointing out the issue, providing the
original patch to resolve this, a simulator for testing, as well as
helping debug and resolve issues in my implementation so that it
performed closer to his original implementation.

Cc: Miroslav Lichvar <mlichvar@redhat.com>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Prarit Bhargava <prarit@redhat.com>
Reported-by: NMiroslav Lichvar <mlichvar@redhat.com>
Signed-off-by: NJohn Stultz <john.stultz@linaro.org>

include/linux/timekeeper_internal.h

@@ -92,6 +92,7 @@ struct timekeeper {
 	u32			raw_interval;
 	s64			ntp_error;
 	u32			ntp_error_shift;
+	u32			ntp_err_mult;
 };
 
 #ifdef CONFIG_GENERIC_TIME_VSYSCALL

kernel/time/timekeeping.c

@@ -178,6 +178,7 @@ static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
 	 * to counteract clock drifting.
 	 */
 	tk->tkr.mult = clock->mult;
+	tk->ntp_err_mult = 0;
 }
 
 /* Timekeeper helper functions. */
@@ -1257,125 +1258,34 @@ static int __init timekeeping_init_ops(void)
 	register_syscore_ops(&timekeeping_syscore_ops);
 	return 0;
 }
-
 device_initcall(timekeeping_init_ops);
 
 /*
- * If the error is already larger, we look ahead even further
- * to compensate for late or lost adjustments.
- */
-static __always_inline int timekeeping_bigadjust(struct timekeeper *tk,
-						 s64 error, s64 *interval,
-						 s64 *offset)
-{
-	s64 tick_error, i;
-	u32 look_ahead, adj;
-	s32 error2, mult;
-
-	/*
-	 * Use the current error value to determine how much to look ahead.
-	 * The larger the error the slower we adjust for it to avoid problems
-	 * with losing too many ticks, otherwise we would overadjust and
-	 * produce an even larger error.  The smaller the adjustment the
-	 * faster we try to adjust for it, as lost ticks can do less harm
-	 * here.  This is tuned so that an error of about 1 msec is adjusted
-	 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
-	 */
-	error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
-	error2 = abs(error2);
-	for (look_ahead = 0; error2 > 0; look_ahead++)
-		error2 >>= 2;
-
-	/*
-	 * Now calculate the error in (1 << look_ahead) ticks, but first
-	 * remove the single look ahead already included in the error.
-	 */
-	tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1);
-	tick_error -= tk->xtime_interval >> 1;
-	error = ((error - tick_error) >> look_ahead) + tick_error;
-
-	/* Finally calculate the adjustment shift value.  */
-	i = *interval;
-	mult = 1;
-	if (error < 0) {
-		error = -error;
-		*interval = -*interval;
-		*offset = -*offset;
-		mult = -1;
-	}
-	for (adj = 0; error > i; adj++)
-		error >>= 1;
-
-	*interval <<= adj;
-	*offset <<= adj;
-	return mult << adj;
-}
-
-/*
- * Adjust the multiplier to reduce the error value,
- * this is optimized for the most common adjustments of -1,0,1,
- * for other values we can do a bit more work.
+ * Apply a multiplier adjustment to the timekeeper
  */
-static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
+static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk,
+							 s64 offset,
+							 bool negative,
+							 int adj_scale)
 {
-	s64 error, interval = tk->cycle_interval;
-	int adj;
+	s64 interval = tk->cycle_interval;
+	s32 mult_adj = 1;
 
-	/*
-	 * The point of this is to check if the error is greater than half
-	 * an interval.
-	 *
-	 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
-	 *
-	 * Note we subtract one in the shift, so that error is really error*2.
-	 * This "saves" dividing(shifting) interval twice, but keeps the
-	 * (error > interval) comparison as still measuring if error is
-	 * larger than half an interval.
-	 *
-	 * Note: It does not "save" on aggravation when reading the code.
-	 */
-	error = tk->ntp_error >> (tk->ntp_error_shift - 1);
-	if (error > interval) {
-		/*
-		 * We now divide error by 4(via shift), which checks if
-		 * the error is greater than twice the interval.
-		 * If it is greater, we need a bigadjust, if its smaller,
-		 * we can adjust by 1.
-		 */
-		error >>= 2;
-		if (likely(error <= interval))
-			adj = 1;
-		else
-			adj = timekeeping_bigadjust(tk, error, &interval, &offset);
-	} else {
-		if (error < -interval) {
-			/* See comment above, this is just switched for the negative */
-			error >>= 2;
-			if (likely(error >= -interval)) {
-				adj = -1;
-				interval = -interval;
-				offset = -offset;
-			} else {
-				adj = timekeeping_bigadjust(tk, error, &interval, &offset);
-			}
-		} else {
-			goto out_adjust;
-		}
+	if (negative) {
+		mult_adj = -mult_adj;
+		interval = -interval;
+		offset  = -offset;
 	}
+	mult_adj <<= adj_scale;
+	interval <<= adj_scale;
+	offset <<= adj_scale;
 
-	if (unlikely(tk->tkr.clock->maxadj &&
-		(tk->tkr.mult + adj > tk->tkr.clock->mult + tk->tkr.clock->maxadj))) {
-		printk_deferred_once(KERN_WARNING
-			"Adjusting %s more than 11%% (%ld vs %ld)\n",
-			tk->tkr.clock->name, (long)tk->tkr.mult + adj,
-			(long)tk->tkr.clock->mult + tk->tkr.clock->maxadj);
-	}
 	/*
 	 * So the following can be confusing.
 	 *
-	 * To keep things simple, lets assume adj == 1 for now.
+	 * To keep things simple, lets assume mult_adj == 1 for now.
 	 *
-	 * When adj != 1, remember that the interval and offset values
+	 * When mult_adj != 1, remember that the interval and offset values
 	 * have been appropriately scaled so the math is the same.
 	 *
 	 * The basic idea here is that we're increasing the multiplier
@@ -1419,12 +1329,76 @@ static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
 	 *
 	 * XXX - TODO: Doc ntp_error calculation.
 	 */
-	tk->tkr.mult += adj;
+	tk->tkr.mult += mult_adj;
 	tk->xtime_interval += interval;
 	tk->tkr.xtime_nsec -= offset;
 	tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
+}
+
+/*
+ * Calculate the multiplier adjustment needed to match the frequency
+ * specified by NTP
+ */
+static __always_inline void timekeeping_freqadjust(struct timekeeper *tk,
+							s64 offset)
+{
+	s64 interval = tk->cycle_interval;
+	s64 xinterval = tk->xtime_interval;
+	s64 tick_error;
+	bool negative;
+	u32 adj;
+
+	/* Remove any current error adj from freq calculation */
+	if (tk->ntp_err_mult)
+		xinterval -= tk->cycle_interval;
+
+	/* Calculate current error per tick */
+	tick_error = ntp_tick_length() >> tk->ntp_error_shift;
+	tick_error -= (xinterval + tk->xtime_remainder);
+
+	/* Don't worry about correcting it if its small */
+	if (likely((tick_error >= 0) && (tick_error <= interval)))
+		return;
+
+	/* preserve the direction of correction */
+	negative = (tick_error < 0);
+
+	/* Sort out the magnitude of the correction */
+	tick_error = abs(tick_error);
+	for (adj = 0; tick_error > interval; adj++)
+		tick_error >>= 1;
+
+	/* scale the corrections */
+	timekeeping_apply_adjustment(tk, offset, negative, adj);
+}
+
+/*
+ * Adjust the timekeeper's multiplier to the correct frequency
+ * and also to reduce the accumulated error value.
+ */
+static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
+{
+	/* Correct for the current frequency error */
+	timekeeping_freqadjust(tk, offset);
+
+	/* Next make a small adjustment to fix any cumulative error */
+	if (!tk->ntp_err_mult && (tk->ntp_error > 0)) {
+		tk->ntp_err_mult = 1;
+		timekeeping_apply_adjustment(tk, offset, 0, 0);
+	} else if (tk->ntp_err_mult && (tk->ntp_error <= 0)) {
+		/* Undo any existing error adjustment */
+		timekeeping_apply_adjustment(tk, offset, 1, 0);
+		tk->ntp_err_mult = 0;
+	}
+
+	if (unlikely(tk->tkr.clock->maxadj &&
+		(tk->tkr.mult > tk->tkr.clock->mult + tk->tkr.clock->maxadj))) {
+		printk_once(KERN_WARNING
+			"Adjusting %s more than 11%% (%ld vs %ld)\n",
+			tk->tkr.clock->name, (long)tk->tkr.mult,
+			(long)tk->tkr.clock->mult + tk->tkr.clock->maxadj);
+	}
 
-out_adjust:
 	/*
 	 * It may be possible that when we entered this function, xtime_nsec
 	 * was very small.  Further, if we're slightly speeding the clocksource
@@ -1444,7 +1418,6 @@ static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
 		tk->tkr.xtime_nsec = 0;
 		tk->ntp_error += neg << tk->ntp_error_shift;
 	}
-
 }
 
 /**