Skip to contentSkip to conversation

timekeeping: Update multiplier when NTP frequency is set directly

PATCH Superseded Miroslav Lichvar 2 participants 3 messages May 29, 2018 – Jun 4, 2018 Timekeeping, clocksource core, NTP, alarmtimer

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Miroslav Lichvar

author reviewer
#0 lore raw

PATCH 1/1 timekeeping: Update multiplier when NTP frequency is set directly

When the NTP frequency is set directly from userspace using the ADJ_FREQUENCY or ADJ_TICK timex mode, immediately update the timekeeper's multiplier instead of waiting for the next tick.

This removes a hidden non-deterministic delay in setting of the frequency and allows an extremely tight control of the system clock with update rates close to or even exceeding the kernel HZ.

Cc Thomas Gleixner <[email protected]>
Cc John Stultz <[email protected]>
Cc Richard Cochran <[email protected]>
Cc Prarit Bhargava <[email protected]>

---

Notes:
RFC->v1:
- added a new parameter to force the update of the timekeeper to the current NTP tick length only from adjtimex()
- added timekeeping_advance() to keep the parameter local to timekeeping.c

 kernel/time/timekeeping.c | 23 ++++++++++++++++++-----
 1 file changed, 18 insertions(+), 5 deletions(-)
kernel/time/timekeeping.c +18 -5
index 49cbceef5deb..5524c07d43e3 100644
@@ -2021,11 +2021,11 @@ static u64 logarithmic_accumulation(struct timekeeper *tk, u64 offset,
 	return offset;
 }
 
-/**
- * update_wall_time - Uses the current clocksource to increment the wall time
- *
+/*
+ * timekeeping_advance - Updates the timekeeper to the current time and
+ * current NTP tick length
  */
-void update_wall_time(void)
+static void timekeeping_advance(bool force_update)
 {
 	struct timekeeper *real_tk = &tk_core.timekeeper;
 	struct timekeeper *tk = &shadow_timekeeper;
@@ -2048,7 +2048,7 @@ void update_wall_time(void)
 #endif
 
 	/* Check if there's really nothing to do */
-	if (offset < real_tk->cycle_interval)
+	if (offset < real_tk->cycle_interval && !force_update)
 		goto out;
 
 	/* Do some additional sanity checking */
@@ -2105,6 +2105,15 @@ void update_wall_time(void)
 		clock_was_set_delayed();
 }
 
+/**
+ * update_wall_time - Uses the current clocksource to increment the wall time
+ *
+ */
+void update_wall_time(void)
+{
+	timekeeping_advance(false);
+}
+
 /**
  * getboottime64 - Return the real time of system boot.
  * @ts:		pointer to the timespec64 to be set
@@ -2332,6 +2341,10 @@ int do_adjtimex(struct timex *txc)
 	write_seqcount_end(&tk_core.seq);
 	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
 
+	/* Update the multiplier immediately if frequency was set directly */
+	if (txc->modes & (ADJ_FREQUENCY | ADJ_TICK))
+		timekeeping_advance(true);
+
 	if (tai != orig_tai)
 		clock_was_set();
 
Signature · 1 line
2.14.3
#1 lore raw
Quoted 31 lines · Miroslav Lichvar

When the NTP frequency is set directly from userspace using the ADJ_FREQUENCY or ADJ_TICK timex mode, immediately update the timekeeper's multiplier instead of waiting for the next tick.

This removes a hidden non-deterministic delay in setting of the frequency and allows an extremely tight control of the system clock with update rates close to or even exceeding the kernel HZ.

Cc Thomas Gleixner <[email protected]>
Cc John Stultz <[email protected]>
Cc Richard Cochran <[email protected]>
Cc Prarit Bhargava <[email protected]>

---

Notes:
RFC->v1:
- added a new parameter to force the update of the timekeeper to the current NTP tick length only from adjtimex()
- added timekeeping_advance() to keep the parameter local to timekeeping.c

 kernel/time/timekeeping.c | 23 ++++++++++++++++++-----
 1 file changed, 18 insertions(+), 5 deletions(-)
kernel/time/timekeeping.c +0 -0
index 49cbceef5deb..5524c07d43e3 100644
@@ -2021,11 +2021,11 @@ static u64 logarithmic_accumulation(struct timekeeper *tk, u64 offset,
        return offset;
 }

kernel/time/timekeeping.c static u64 logarithmic_accumulation(struct timekeeper *tk, u64 offset, · line 2028

Quoted 10 lines
kernel/time/timekeeping.c +4 -4
 
-/**
- * update_wall_time - Uses the current clocksource to increment the wall time
- *
+/*
+ * timekeeping_advance - Updates the timekeeper to the current time and
+ * current NTP tick length
  */
-void update_wall_time(void)
+static void timekeeping_advance(bool force_update)

This is kind of a nit, but mind switching out bool for an enum? Using something like TK_ADV_NORMAL and TK_ADV_FORCE?

kernel/time/timekeeping.c void update_wall_time(void) · line 2115

Quoted 4 lines
kernel/time/timekeeping.c +4 -0
+void update_wall_time(void)
+{
+       timekeeping_advance(false);
+}

The enum makes usage like timekeeping_advance(false) a little less opaque to the reader ("Wait, don't advance? Let me go look at the function").

We got bitten by this earlier when we had the old "timekeeping_update(tk, true, false, true)" usage.

thanks
-john

author reviewer
#2 lore raw
Quoted 16 lines · John Stultz

kernel/time/timekeeping.c static u64 logarithmic_accumulation(struct timekeeper *tk, u64 offset, · line 2028

Quoted 2 lines · Miroslav Lichvar
kernel/time/timekeeping.c +1 -1
-void update_wall_time(void)
+static void timekeeping_advance(bool force_update)

This is kind of a nit, but mind switching out bool for an enum? Using something like TK_ADV_NORMAL and TK_ADV_FORCE?

kernel/time/timekeeping.c void update_wall_time(void) · line 2115

Quoted 4 lines
kernel/time/timekeeping.c +4 -0
+void update_wall_time(void)
+{
+       timekeeping_advance(false);
+}

The enum makes usage like timekeeping_advance(false) a little less opaque to the reader ("Wait, don't advance? Let me go look at the function").

Quoted 2 lines

We got bitten by this earlier when we had the old "timekeeping_update(tk, true, false, true)" usage.

Ok. That make sense. I'll send a v2.

Thanks,

Signature · 1 line
Miroslav Lichvar