Remove the leap second handling from second_overflow(), which doesn't
has to check for it every second anymore. With CONFIG_NO_HZ this also
makes sure the leap second is handled close to the full second.
Additionally this makes it possible to abort a leap second properly
by resetting the STA_INS/STA_DEL status bits.
Signed-off-by: Roman Zippel <[email protected]>
---
include/linux/clocksource.h | 2
include/linux/timex.h | 1
kernel/time/ntp.c | 133 +++++++++++++++++++++++++++++---------------
kernel/time/timekeeping.c | 4 -
4 files changed, 95 insertions(+), 45 deletions(-)
Index: linux-2.6/include/linux/timex.h
===================================================================
--- linux-2.6.orig/include/linux/timex.h 2008-03-13 10:32:07.000000000 +0100
+++ linux-2.6/include/linux/timex.h 2008-03-13 10:33:24.000000000 +0100
@@ -210,6 +210,7 @@ extern long time_esterror; /* estimated
extern long time_adjust; /* The amount of adjtime left */
+extern void ntp_init(void);
extern void ntp_clear(void);
/**
Index: linux-2.6/kernel/time/ntp.c
===================================================================
--- linux-2.6.orig/kernel/time/ntp.c 2008-03-13 10:32:07.000000000 +0100
+++ linux-2.6/kernel/time/ntp.c 2008-03-13 10:33:24.000000000 +0100
@@ -16,6 +16,7 @@
#include <linux/hrtimer.h>
#include <linux/capability.h>
#include <linux/math64.h>
+#include <linux/clocksource.h>
#include <asm/timex.h>
/*
@@ -26,6 +27,8 @@ unsigned long tick_nsec; /* ACTHZ peri
u64 tick_length;
static u64 tick_length_base;
+static struct hrtimer leap_timer;
+
#define MAX_TICKADJ 500 /* microsecs */
#define MAX_TICKADJ_SCALED (((u64)(MAX_TICKADJ * NSEC_PER_USEC) << \
NTP_SCALE_SHIFT) / NTP_INTERVAL_FREQ)
@@ -120,64 +123,70 @@ void ntp_clear(void)
}
/*
- * this routine handles the overflow of the microsecond field
- *
- * The tricky bits of code to handle the accurate clock support
- * were provided by Dave Mills ([email protected]) of NTP fame.
- * They were originally developed for SUN and DEC kernels.
- * All the kudos should go to Dave for this stuff.
+ * Leap second processing. If in leap-insert state at the end of the
+ * day, the system clock is set back one second; if in leap-delete
+ * state, the system clock is set ahead one second.
*/
-void second_overflow(void)
+static enum hrtimer_restart ntp_leap_second(struct hrtimer *timer)
{
- s64 time_adj;
+ enum hrtimer_restart res = HRTIMER_NORESTART;
- /* Bump the maxerror field */
- time_maxerror += MAXFREQ / NSEC_PER_USEC;
- if (time_maxerror > NTP_PHASE_LIMIT) {
- time_maxerror = NTP_PHASE_LIMIT;
- time_status |= STA_UNSYNC;
- }
+ write_seqlock_irq(&xtime_lock);
- /*
- * Leap second processing. If in leap-insert state at the end of the
- * day, the system clock is set back one second; if in leap-delete
- * state, the system clock is set ahead one second. The microtime()
- * routine or external clock driver will insure that reported time is
- * always monotonic. The ugly divides should be replaced.
- */
switch (time_state) {
case TIME_OK:
- if (time_status & STA_INS)
- time_state = TIME_INS;
- else if (time_status & STA_DEL)
- time_state = TIME_DEL;
break;
case TIME_INS:
- if (xtime.tv_sec % 86400 == 0) {
- xtime.tv_sec--;
- wall_to_monotonic.tv_sec++;
- time_state = TIME_OOP;
- printk(KERN_NOTICE "Clock: inserting leap second "
- "23:59:60 UTC\n");
- }
+ xtime.tv_sec--;
+ wall_to_monotonic.tv_sec++;
+ time_state = TIME_OOP;
+ printk(KERN_NOTICE "Clock: "
+ "inserting leap second 23:59:60 UTC\n");
+ leap_timer.expires = ktime_add_ns(leap_timer.expires,
+ NSEC_PER_SEC);
+ res = HRTIMER_RESTART;
break;
case TIME_DEL:
- if ((xtime.tv_sec + 1) % 86400 == 0) {
- xtime.tv_sec++;
- time_tai--;
- wall_to_monotonic.tv_sec--;
- time_state = TIME_WAIT;
- printk(KERN_NOTICE "Clock: deleting leap second "
- "23:59:59 UTC\n");
- }
+ xtime.tv_sec++;
+ time_tai--;
+ wall_to_monotonic.tv_sec--;
+ time_state = TIME_WAIT;
+ printk(KERN_NOTICE "Clock: "
+ "deleting leap second 23:59:59 UTC\n");
break;
case TIME_OOP:
time_tai++;
time_state = TIME_WAIT;
- break;
+ /* fall through */
case TIME_WAIT:
if (!(time_status & (STA_INS | STA_DEL)))
time_state = TIME_OK;
+ break;
+ }
+ update_vsyscall(&xtime, clock);
+
+ write_sequnlock_irq(&xtime_lock);
+
+ return res;
+}
+
+/*
+ * this routine handles the overflow of the microsecond field
+ *
+ * The tricky bits of code to handle the accurate clock support
+ * were provided by Dave Mills ([email protected]) of NTP fame.
+ * They were originally developed for SUN and DEC kernels.
+ * All the kudos should go to Dave for this stuff.
+ */
+void second_overflow(void)
+{
+ s64 time_adj;
+
+ /* Bump the maxerror field */
+ time_maxerror += MAXFREQ / NSEC_PER_USEC;
+ if (time_maxerror > NTP_PHASE_LIMIT) {
+ time_maxerror = NTP_PHASE_LIMIT;
+ time_status |= STA_UNSYNC;
}
/*
@@ -268,7 +277,7 @@ static inline void notify_cmos_timer(voi
int do_adjtimex(struct timex *txc)
{
struct timespec ts;
- long save_adjust;
+ long save_adjust, sec;
int result;
/* In order to modify anything, you gotta be super-user! */
@@ -289,6 +298,10 @@ int do_adjtimex(struct timex *txc)
txc->tick > 1100000/USER_HZ)
return -EINVAL;
+ if (time_state != TIME_OK && txc->modes & ADJ_STATUS)
+ hrtimer_cancel(&leap_timer);
+ getnstimeofday(&ts);
+
write_seqlock_irq(&xtime_lock);
/* Save for later - semantics of adjtime is to return old value */
@@ -305,6 +318,34 @@ int do_adjtimex(struct timex *txc)
/* only set allowed bits */
time_status &= STA_RONLY;
time_status |= txc->status & ~STA_RONLY;
+
+ switch (time_state) {
+ case TIME_OK:
+ start_timer:
+ sec = ts.tv_sec;
+ if (time_status & STA_INS) {
+ time_state = TIME_INS;
+ sec += 86400 - sec % 86400;
+ hrtimer_start(&leap_timer, ktime_set(sec, 0), HRTIMER_MODE_ABS);
+ } else if (time_status & STA_DEL) {
+ time_state = TIME_DEL;
+ sec += 86400 - (sec + 1) % 86400;
+ hrtimer_start(&leap_timer, ktime_set(sec, 0), HRTIMER_MODE_ABS);
+ }
+ break;
+ case TIME_INS:
+ case TIME_DEL:
+ time_state = TIME_OK;
+ goto start_timer;
+ break;
+ case TIME_WAIT:
+ if (!(time_status & (STA_INS | STA_DEL)))
+ time_state = TIME_OK;
+ break;
+ case TIME_OOP:
+ hrtimer_restart(&leap_timer);
+ break;
+ }
}
if (txc->modes & ADJ_NANO)
@@ -384,7 +425,6 @@ int do_adjtimex(struct timex *txc)
txc->stbcnt = 0;
write_sequnlock_irq(&xtime_lock);
- getnstimeofday(&ts);
txc->time.tv_sec = ts.tv_sec;
txc->time.tv_usec = ts.tv_nsec;
if (!(time_status & STA_NANO))
@@ -402,3 +442,10 @@ static int __init ntp_tick_adj_setup(cha
}
__setup("ntp_tick_adj=", ntp_tick_adj_setup);
+
+void __init ntp_init(void)
+{
+ ntp_clear();
+ hrtimer_init(&leap_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
+ leap_timer.function = ntp_leap_second;
+}
Index: linux-2.6/kernel/time/timekeeping.c
===================================================================
--- linux-2.6.orig/kernel/time/timekeeping.c 2008-03-13 10:32:07.000000000 +0100
+++ linux-2.6/kernel/time/timekeeping.c 2008-03-13 10:33:24.000000000 +0100
@@ -53,7 +53,7 @@ void update_xtime_cache(u64 nsec)
timespec_add_ns(&xtime_cache, nsec);
}
-static struct clocksource *clock; /* pointer to current clocksource */
+struct clocksource *clock;
#ifdef CONFIG_GENERIC_TIME
@@ -241,7 +241,7 @@ void __init timekeeping_init(void)
write_seqlock_irqsave(&xtime_lock, flags);
- ntp_clear();
+ ntp_init();
clock = clocksource_get_next();
clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH);
Index: linux-2.6/include/linux/clocksource.h
===================================================================
--- linux-2.6.orig/include/linux/clocksource.h 2008-03-10 19:50:33.000000000 +0100
+++ linux-2.6/include/linux/clocksource.h 2008-03-13 10:33:24.000000000 +0100
@@ -93,6 +93,8 @@ struct clocksource {
#endif
};
+extern struct clocksource *clock; /* current clocksource */
+
/*
* Clock source flags bits::
*/
--
On Fri, 2008-03-14 at 19:40 +0100, [email protected] wrote:
> plain text document attachment (leap_sec)
> Remove the leap second handling from second_overflow(), which doesn't
> has to check for it every second anymore. With CONFIG_NO_HZ this also
> makes sure the leap second is handled close to the full second.
> Additionally this makes it possible to abort a leap second properly
> by resetting the STA_INS/STA_DEL status bits.
>
> Signed-off-by: Roman Zippel <[email protected]>
Looks interesting. Small comment below.
> -void second_overflow(void)
> +static enum hrtimer_restart ntp_leap_second(struct hrtimer *timer)
> {
> - s64 time_adj;
> + enum hrtimer_restart res = HRTIMER_NORESTART;
>
> - /* Bump the maxerror field */
> - time_maxerror += MAXFREQ / NSEC_PER_USEC;
> - if (time_maxerror > NTP_PHASE_LIMIT) {
> - time_maxerror = NTP_PHASE_LIMIT;
> - time_status |= STA_UNSYNC;
> - }
> + write_seqlock_irq(&xtime_lock);
>
> - /*
> - * Leap second processing. If in leap-insert state at the end of the
> - * day, the system clock is set back one second; if in leap-delete
> - * state, the system clock is set ahead one second. The microtime()
> - * routine or external clock driver will insure that reported time is
> - * always monotonic. The ugly divides should be replaced.
> - */
> switch (time_state) {
> case TIME_OK:
> - if (time_status & STA_INS)
> - time_state = TIME_INS;
> - else if (time_status & STA_DEL)
> - time_state = TIME_DEL;
> break;
> case TIME_INS:
> - if (xtime.tv_sec % 86400 == 0) {
> - xtime.tv_sec--;
> - wall_to_monotonic.tv_sec++;
> - time_state = TIME_OOP;
> - printk(KERN_NOTICE "Clock: inserting leap second "
> - "23:59:60 UTC\n");
> - }
> + xtime.tv_sec--;
> + wall_to_monotonic.tv_sec++;
> + time_state = TIME_OOP;
> + printk(KERN_NOTICE "Clock: "
> + "inserting leap second 23:59:60 UTC\n");
> + leap_timer.expires = ktime_add_ns(leap_timer.expires,
> + NSEC_PER_SEC);
> + res = HRTIMER_RESTART;
> break;
> case TIME_DEL:
> - if ((xtime.tv_sec + 1) % 86400 == 0) {
> - xtime.tv_sec++;
> - time_tai--;
> - wall_to_monotonic.tv_sec--;
> - time_state = TIME_WAIT;
> - printk(KERN_NOTICE "Clock: deleting leap second "
> - "23:59:59 UTC\n");
> - }
> + xtime.tv_sec++;
> + time_tai--;
> + wall_to_monotonic.tv_sec--;
> + time_state = TIME_WAIT;
> + printk(KERN_NOTICE "Clock: "
> + "deleting leap second 23:59:59 UTC\n");
> break;
> case TIME_OOP:
> time_tai++;
> time_state = TIME_WAIT;
> - break;
> + /* fall through */
> case TIME_WAIT:
> if (!(time_status & (STA_INS | STA_DEL)))
> time_state = TIME_OK;
> + break;
> + }
> + update_vsyscall(&xtime, clock);
> +
<snip>
> Index: linux-2.6/include/linux/clocksource.h
> ===================================================================
> --- linux-2.6.orig/include/linux/clocksource.h 2008-03-10 19:50:33.000000000 +0100
> +++ linux-2.6/include/linux/clocksource.h 2008-03-13 10:33:24.000000000 +0100
> @@ -93,6 +93,8 @@ struct clocksource {
> #endif
> };
>
> +extern struct clocksource *clock; /* current clocksource */
> +
> /*
> * Clock source flags bits::
> */
Instead of exporting the clocksource making it global, could you use a
timekeeping_insert/remove_second() style interface?
That would help remove the xtime/wall_to_monotonic references in ntp.c
as well, cleaning up things nicely.
thanks
-john
Hi,
On Fri, 14 Mar 2008, john stultz wrote:
> Instead of exporting the clocksource making it global, could you use a
> timekeeping_insert/remove_second() style interface?
Sounds good.
bye, Roman
On Sat, 15 Mar 2008 04:18:39 +0100 (CET)
Roman Zippel <[email protected]> wrote:
> Hi,
>
> On Fri, 14 Mar 2008, john stultz wrote:
>
> > Instead of exporting the clocksource making it global, could you use a
> > timekeeping_insert/remove_second() style interface?
>
> Sounds good.
>
It sounds like a few updates are in the pipeline, but I merged this series
as-is into -mm.
I'll normally push ntp changes through Thomas's git-hrt tree, however this
patch series has dependencies upon at least
introduce-explicit-signed-unsigned-64bit-divide.patch
convert-a-few-do_div-user.patch
rename-div64_64-to-div64_u64.patch
rename-div64_64-to-div64_u64-mm.patch
remove-div_long_long_rem.patch
so they can't go into git-hrt immediately.
The idealised algorithm is
- 2.6.26 opens
- Thomas merges git-hrt
- I merge the above patches
- I then send these ntp patches to Thomas
- He merges them into git-hrt
- He does another git-hrt -> Linus merge
And that's all OK, but various lags might cause us to miss the merge
window, as I merge the -mm stuff last, and some git-tree maintainers are
dawdlers.
So if Thomas wants to ack these I can merge them directly late in the
2.6.26 merge window.
On Mon, 17 Mar 2008, Andrew Morton wrote:
> It sounds like a few updates are in the pipeline, but I merged this series
> as-is into -mm.
>
> I'll normally push ntp changes through Thomas's git-hrt tree, however this
> patch series has dependencies upon at least
>
> introduce-explicit-signed-unsigned-64bit-divide.patch
> convert-a-few-do_div-user.patch
> rename-div64_64-to-div64_u64.patch
> rename-div64_64-to-div64_u64-mm.patch
> remove-div_long_long_rem.patch
>
> so they can't go into git-hrt immediately.
>
> The idealised algorithm is
>
> - 2.6.26 opens
> - Thomas merges git-hrt
> - I merge the above patches
> - I then send these ntp patches to Thomas
> - He merges them into git-hrt
> - He does another git-hrt -> Linus merge
>
> And that's all OK, but various lags might cause us to miss the merge
> window, as I merge the -mm stuff last, and some git-tree maintainers are
> dawdlers.
:)
> So if Thomas wants to ack these I can merge them directly late in the
> 2.6.26 merge window.
Fine for me either way.
Thanks,
tglx