All of lore.kernel.org
 help / color / mirror / Atom feed
From: John Stultz <john.stultz@linaro.org>
To: lkml <linux-kernel@vger.kernel.org>
Cc: John Stultz <john.stultz@linaro.org>, Ingo Molnar <mingo@elte.hu>,
	Thomas Gleixner <tglx@linutronix.de>,
	Eric Dumazet <eric.dumazet@gmail.com>,
	Richard Cochran <richardcochran@gmail.com>
Subject: [PATCH 1/9] time: Condense timekeeper.xtime into xtime_sec
Date: Thu,  1 Mar 2012 23:12:40 -0800	[thread overview]
Message-ID: <1330672368-32290-2-git-send-email-john.stultz@linaro.org> (raw)
In-Reply-To: <1330672368-32290-1-git-send-email-john.stultz@linaro.org>

The timekeeper struct has a xtime_nsec, which keeps the
sub-nanosecond remainder.  This ends up being somewhat
duplicative of the timekeeper.xtime.tv_nsec value, and we
have to do extra work to keep them apart, copying the full
nsec portion out and back in over and over.

This patch simplifies some of the logic by taking the timekeeper
xtime value and splitting it into timekeeper.xtime_sec and
reuses the timekeeper.xtime_nsec for the sub-second portion
(stored in higher res shifted nanoseconds).

This simplifies some of the accumulation logic. And will
allow for more accurate timekeeping once the vsyscall code
is updated to use the shifted nanosecond remainder.

CC: Ingo Molnar <mingo@elte.hu>
CC: Thomas Gleixner <tglx@linutronix.de>
CC: Eric Dumazet <eric.dumazet@gmail.com>
CC: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
---
 kernel/time/timekeeping.c |  170 ++++++++++++++++++++++++++++-----------------
 1 files changed, 107 insertions(+), 63 deletions(-)

diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
index 403c2a0..5434c6e 100644
--- a/kernel/time/timekeeping.c
+++ b/kernel/time/timekeeping.c
@@ -39,8 +39,11 @@ struct timekeeper {
 	/* Raw nano seconds accumulated per NTP interval. */
 	u32	raw_interval;
 
-	/* Clock shifted nano seconds remainder not stored in xtime.tv_nsec. */
+	/* Current CLOCK_REALTIME time in seconds */
+	u64	xtime_sec;
+	/* Clock shifted nano seconds */
 	u64	xtime_nsec;
+
 	/* Difference between accumulated time and NTP time in ntp
 	 * shifted nano seconds. */
 	s64	ntp_error;
@@ -48,8 +51,6 @@ struct timekeeper {
 	 * ntp shifted nano seconds. */
 	int	ntp_error_shift;
 
-	/* The current time */
-	struct timespec xtime;
 	/*
 	 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
 	 * for sub jiffie times) to get to monotonic time.  Monotonic is pegged
@@ -87,6 +88,36 @@ __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock);
 int __read_mostly timekeeping_suspended;
 
 
+static inline void tk_normalize_xtime(struct timekeeper *tk)
+{
+	while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) {
+		tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift;
+		tk->xtime_sec++;
+	}
+}
+
+static struct timespec tk_xtime(struct timekeeper *tk)
+{
+	struct timespec ts;
+
+	ts.tv_sec = tk->xtime_sec;
+	ts.tv_nsec = (long)(tk->xtime_nsec >> tk->shift);
+	return ts;
+}
+
+static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts)
+{
+	tk->xtime_sec = ts->tv_sec;
+	tk->xtime_nsec = ts->tv_nsec << tk->shift;
+}
+
+
+static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts)
+{
+	tk->xtime_sec += ts->tv_sec;
+	tk->xtime_nsec += ts->tv_nsec << tk->shift;
+}
+
 
 /**
  * timekeeper_setup_internals - Set up internals to use clocksource clock.
@@ -102,7 +133,9 @@ static void timekeeper_setup_internals(struct clocksource *clock)
 {
 	cycle_t interval;
 	u64 tmp, ntpinterval;
+	struct clocksource *old_clock;
 
+	old_clock = timekeeper.clock;
 	timekeeper.clock = clock;
 	clock->cycle_last = clock->read(clock);
 
@@ -124,7 +157,14 @@ static void timekeeper_setup_internals(struct clocksource *clock)
 	timekeeper.raw_interval =
 		((u64) interval * clock->mult) >> clock->shift;
 
-	timekeeper.xtime_nsec = 0;
+	 /* if changing clocks, convert xtime_nsec shift units */
+	if (old_clock) {
+		int shift_change = clock->shift - old_clock->shift;
+		if (shift_change < 0)
+			timekeeper.xtime_nsec >>= -shift_change;
+		else
+			timekeeper.xtime_nsec <<= shift_change;
+	}
 	timekeeper.shift = clock->shift;
 
 	timekeeper.ntp_error = 0;
@@ -143,6 +183,7 @@ static inline s64 timekeeping_get_ns(void)
 {
 	cycle_t cycle_now, cycle_delta;
 	struct clocksource *clock;
+	s64 nsec;
 
 	/* read clocksource: */
 	clock = timekeeper.clock;
@@ -151,9 +192,8 @@ static inline s64 timekeeping_get_ns(void)
 	/* calculate the delta since the last update_wall_time: */
 	cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
 
-	/* return delta convert to nanoseconds using ntp adjusted mult. */
-	return clocksource_cyc2ns(cycle_delta, timekeeper.mult,
-				  timekeeper.shift);
+	nsec = cycle_delta * timekeeper.mult + timekeeper.xtime_nsec;
+	return nsec >> timekeeper.shift;
 }
 
 static inline s64 timekeeping_get_ns_raw(void)
@@ -175,11 +215,13 @@ static inline s64 timekeeping_get_ns_raw(void)
 /* must hold write on timekeeper.lock */
 static void timekeeping_update(bool clearntp)
 {
+	struct timespec xt;
 	if (clearntp) {
 		timekeeper.ntp_error = 0;
 		ntp_clear();
 	}
-	update_vsyscall(&timekeeper.xtime, &timekeeper.wall_to_monotonic,
+	xt = tk_xtime(&timekeeper);
+	update_vsyscall(&xt, &timekeeper.wall_to_monotonic,
 			 timekeeper.clock, timekeeper.mult);
 }
 
@@ -189,7 +231,7 @@ void timekeeping_leap_insert(int leapsecond)
 	unsigned long flags;
 
 	write_seqlock_irqsave(&timekeeper.lock, flags);
-	timekeeper.xtime.tv_sec += leapsecond;
+	timekeeper.xtime_sec += leapsecond;
 	timekeeper.wall_to_monotonic.tv_sec -= leapsecond;
 	timekeeping_update(false);
 	write_sequnlock_irqrestore(&timekeeper.lock, flags);
@@ -214,13 +256,12 @@ static void timekeeping_forward_now(void)
 	cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
 	clock->cycle_last = cycle_now;
 
-	nsec = clocksource_cyc2ns(cycle_delta, timekeeper.mult,
-				  timekeeper.shift);
+	timekeeper.xtime_nsec += cycle_delta * timekeeper.mult;
 
 	/* If arch requires, add in gettimeoffset() */
-	nsec += arch_gettimeoffset();
+	timekeeper.xtime_nsec += arch_gettimeoffset() << timekeeper.shift;
 
-	timespec_add_ns(&timekeeper.xtime, nsec);
+	tk_normalize_xtime(&timekeeper);
 
 	nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
 	timespec_add_ns(&timekeeper.raw_time, nsec);
@@ -235,15 +276,15 @@ static void timekeeping_forward_now(void)
 void getnstimeofday(struct timespec *ts)
 {
 	unsigned long seq;
-	s64 nsecs;
+	s64 nsecs = 0;
 
 	WARN_ON(timekeeping_suspended);
 
 	do {
 		seq = read_seqbegin(&timekeeper.lock);
 
-		*ts = timekeeper.xtime;
-		nsecs = timekeeping_get_ns();
+		ts->tv_sec = timekeeper.xtime_sec;
+		ts->tv_nsec = timekeeping_get_ns();
 
 		/* If arch requires, add in gettimeoffset() */
 		nsecs += arch_gettimeoffset();
@@ -264,11 +305,10 @@ ktime_t ktime_get(void)
 
 	do {
 		seq = read_seqbegin(&timekeeper.lock);
-		secs = timekeeper.xtime.tv_sec +
+		secs = timekeeper.xtime_sec +
 				timekeeper.wall_to_monotonic.tv_sec;
-		nsecs = timekeeper.xtime.tv_nsec +
+		nsecs = timekeeping_get_ns() +
 				timekeeper.wall_to_monotonic.tv_nsec;
-		nsecs += timekeeping_get_ns();
 		/* If arch requires, add in gettimeoffset() */
 		nsecs += arch_gettimeoffset();
 
@@ -293,22 +333,21 @@ void ktime_get_ts(struct timespec *ts)
 {
 	struct timespec tomono;
 	unsigned int seq;
-	s64 nsecs;
 
 	WARN_ON(timekeeping_suspended);
 
 	do {
 		seq = read_seqbegin(&timekeeper.lock);
-		*ts = timekeeper.xtime;
+		ts->tv_sec = timekeeper.xtime_sec;
+		ts->tv_nsec = timekeeping_get_ns();
 		tomono = timekeeper.wall_to_monotonic;
-		nsecs = timekeeping_get_ns();
 		/* If arch requires, add in gettimeoffset() */
-		nsecs += arch_gettimeoffset();
+		ts->tv_nsec += arch_gettimeoffset();
 
 	} while (read_seqretry(&timekeeper.lock, seq));
 
 	set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
-				ts->tv_nsec + tomono.tv_nsec + nsecs);
+				ts->tv_nsec + tomono.tv_nsec);
 }
 EXPORT_SYMBOL_GPL(ktime_get_ts);
 
@@ -336,7 +375,8 @@ void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
 		seq = read_seqbegin(&timekeeper.lock);
 
 		*ts_raw = timekeeper.raw_time;
-		*ts_real = timekeeper.xtime;
+		ts_real->tv_sec = timekeeper.xtime_sec;
+		ts_real->tv_nsec = 0;
 
 		nsecs_raw = timekeeping_get_ns_raw();
 		nsecs_real = timekeeping_get_ns();
@@ -379,7 +419,7 @@ EXPORT_SYMBOL(do_gettimeofday);
  */
 int do_settimeofday(const struct timespec *tv)
 {
-	struct timespec ts_delta;
+	struct timespec ts_delta, xt;
 	unsigned long flags;
 
 	if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
@@ -389,12 +429,15 @@ int do_settimeofday(const struct timespec *tv)
 
 	timekeeping_forward_now();
 
-	ts_delta.tv_sec = tv->tv_sec - timekeeper.xtime.tv_sec;
-	ts_delta.tv_nsec = tv->tv_nsec - timekeeper.xtime.tv_nsec;
+	xt = tk_xtime(&timekeeper);
+	ts_delta.tv_sec = tv->tv_sec - xt.tv_sec;
+	ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec;
+
 	timekeeper.wall_to_monotonic =
 			timespec_sub(timekeeper.wall_to_monotonic, ts_delta);
 
-	timekeeper.xtime = *tv;
+	tk_set_xtime(&timekeeper, tv);
+
 	timekeeping_update(true);
 
 	write_sequnlock_irqrestore(&timekeeper.lock, flags);
@@ -425,7 +468,8 @@ int timekeeping_inject_offset(struct timespec *ts)
 
 	timekeeping_forward_now();
 
-	timekeeper.xtime = timespec_add(timekeeper.xtime, *ts);
+
+	tk_xtime_add(&timekeeper, ts);
 	timekeeper.wall_to_monotonic =
 				timespec_sub(timekeeper.wall_to_monotonic, *ts);
 
@@ -601,14 +645,12 @@ void __init timekeeping_init(void)
 		clock->enable(clock);
 	timekeeper_setup_internals(clock);
 
-	timekeeper.xtime.tv_sec = now.tv_sec;
-	timekeeper.xtime.tv_nsec = now.tv_nsec;
+	tk_set_xtime(&timekeeper, &now);
 	timekeeper.raw_time.tv_sec = 0;
 	timekeeper.raw_time.tv_nsec = 0;
-	if (boot.tv_sec == 0 && boot.tv_nsec == 0) {
-		boot.tv_sec = timekeeper.xtime.tv_sec;
-		boot.tv_nsec = timekeeper.xtime.tv_nsec;
-	}
+	if (boot.tv_sec == 0 && boot.tv_nsec == 0)
+		boot = tk_xtime(&timekeeper);
+
 	set_normalized_timespec(&timekeeper.wall_to_monotonic,
 				-boot.tv_sec, -boot.tv_nsec);
 	timekeeper.total_sleep_time.tv_sec = 0;
@@ -634,7 +676,7 @@ static void __timekeeping_inject_sleeptime(struct timespec *delta)
 		return;
 	}
 
-	timekeeper.xtime = timespec_add(timekeeper.xtime, *delta);
+	tk_xtime_add(&timekeeper, delta);
 	timekeeper.wall_to_monotonic =
 			timespec_sub(timekeeper.wall_to_monotonic, *delta);
 	timekeeper.total_sleep_time = timespec_add(
@@ -731,7 +773,7 @@ static int timekeeping_suspend(void)
 	 * try to compensate so the difference in system time
 	 * and persistent_clock time stays close to constant.
 	 */
-	delta = timespec_sub(timekeeper.xtime, timekeeping_suspend_time);
+	delta = timespec_sub(tk_xtime(&timekeeper), timekeeping_suspend_time);
 	delta_delta = timespec_sub(delta, old_delta);
 	if (abs(delta_delta.tv_sec)  >= 2) {
 		/*
@@ -963,7 +1005,7 @@ static cycle_t logarithmic_accumulation(cycle_t offset, int shift)
 	timekeeper.xtime_nsec += timekeeper.xtime_interval << shift;
 	while (timekeeper.xtime_nsec >= nsecps) {
 		timekeeper.xtime_nsec -= nsecps;
-		timekeeper.xtime.tv_sec++;
+		timekeeper.xtime_sec++;
 		second_overflow();
 	}
 
@@ -997,6 +1039,7 @@ static void update_wall_time(void)
 	cycle_t offset;
 	int shift = 0, maxshift;
 	unsigned long flags;
+	s64 remainder;
 
 	write_seqlock_irqsave(&timekeeper.lock, flags);
 
@@ -1011,8 +1054,6 @@ static void update_wall_time(void)
 #else
 	offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
 #endif
-	timekeeper.xtime_nsec = (s64)timekeeper.xtime.tv_nsec <<
-						timekeeper.shift;
 
 	/*
 	 * With NO_HZ we may have to accumulate many cycle_intervals
@@ -1058,25 +1099,29 @@ static void update_wall_time(void)
 		timekeeper.ntp_error += neg << timekeeper.ntp_error_shift;
 	}
 
-
 	/*
-	 * Store full nanoseconds into xtime after rounding it up and
-	 * add the remainder to the error difference.
-	 */
-	timekeeper.xtime.tv_nsec = ((s64)timekeeper.xtime_nsec >>
-						timekeeper.shift) + 1;
-	timekeeper.xtime_nsec -= (s64)timekeeper.xtime.tv_nsec <<
-						timekeeper.shift;
-	timekeeper.ntp_error +=	timekeeper.xtime_nsec <<
-				timekeeper.ntp_error_shift;
+	* Store only full nanoseconds into xtime_nsec after rounding
+	* it up and add the remainder to the error difference.
+	* XXX - This is necessary to avoid small 1ns inconsistnecies caused
+	* by truncating the remainder in vsyscalls. However, it causes
+	* additional work to be done in timekeeping_adjust(). Once
+	* the vsyscall implementations are converted to use xtime_nsec
+	* (shifted nanoseconds), this can be killed.
+	*/
+	remainder = timekeeper.xtime_nsec & ((1<<timekeeper.shift)-1);
+	timekeeper.xtime_nsec -= remainder;
+	timekeeper.xtime_nsec += 1<<timekeeper.shift;
+	timekeeper.ntp_error += remainder <<
+					timekeeper.ntp_error_shift;
 
 	/*
 	 * Finally, make sure that after the rounding
 	 * xtime.tv_nsec isn't larger then NSEC_PER_SEC
 	 */
-	if (unlikely(timekeeper.xtime.tv_nsec >= NSEC_PER_SEC)) {
-		timekeeper.xtime.tv_nsec -= NSEC_PER_SEC;
-		timekeeper.xtime.tv_sec++;
+	if (unlikely(timekeeper.xtime_nsec >=
+			((u64)NSEC_PER_SEC << timekeeper.shift))) {
+		timekeeper.xtime_nsec -= (u64)NSEC_PER_SEC << timekeeper.shift;
+		timekeeper.xtime_sec++;
 		second_overflow();
 	}
 
@@ -1125,21 +1170,20 @@ void get_monotonic_boottime(struct timespec *ts)
 {
 	struct timespec tomono, sleep;
 	unsigned int seq;
-	s64 nsecs;
 
 	WARN_ON(timekeeping_suspended);
 
 	do {
 		seq = read_seqbegin(&timekeeper.lock);
-		*ts = timekeeper.xtime;
+		ts->tv_sec = timekeeper.xtime_sec;
+		ts->tv_nsec = timekeeping_get_ns();
 		tomono = timekeeper.wall_to_monotonic;
 		sleep = timekeeper.total_sleep_time;
-		nsecs = timekeeping_get_ns();
 
 	} while (read_seqretry(&timekeeper.lock, seq));
 
 	set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec + sleep.tv_sec,
-			ts->tv_nsec + tomono.tv_nsec + sleep.tv_nsec + nsecs);
+			ts->tv_nsec + tomono.tv_nsec + sleep.tv_nsec);
 }
 EXPORT_SYMBOL_GPL(get_monotonic_boottime);
 
@@ -1172,13 +1216,13 @@ EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
 
 unsigned long get_seconds(void)
 {
-	return timekeeper.xtime.tv_sec;
+	return timekeeper.xtime_sec;
 }
 EXPORT_SYMBOL(get_seconds);
 
 struct timespec __current_kernel_time(void)
 {
-	return timekeeper.xtime;
+	return tk_xtime(&timekeeper);
 }
 
 struct timespec current_kernel_time(void)
@@ -1189,7 +1233,7 @@ struct timespec current_kernel_time(void)
 	do {
 		seq = read_seqbegin(&timekeeper.lock);
 
-		now = timekeeper.xtime;
+		now = tk_xtime(&timekeeper);
 	} while (read_seqretry(&timekeeper.lock, seq));
 
 	return now;
@@ -1204,7 +1248,7 @@ struct timespec get_monotonic_coarse(void)
 	do {
 		seq = read_seqbegin(&timekeeper.lock);
 
-		now = timekeeper.xtime;
+		now = tk_xtime(&timekeeper);
 		mono = timekeeper.wall_to_monotonic;
 	} while (read_seqretry(&timekeeper.lock, seq));
 
@@ -1239,7 +1283,7 @@ void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim,
 
 	do {
 		seq = read_seqbegin(&timekeeper.lock);
-		*xtim = timekeeper.xtime;
+		*xtim = tk_xtime(&timekeeper);
 		*wtom = timekeeper.wall_to_monotonic;
 		*sleep = timekeeper.total_sleep_time;
 	} while (read_seqretry(&timekeeper.lock, seq));
-- 
1.7.3.2.146.gca209


  reply	other threads:[~2012-03-02  7:13 UTC|newest]

Thread overview: 15+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2012-03-02  7:12 [PATCH 0/9] Reduce timekeeping lock hold time v2 John Stultz
2012-03-02  7:12 ` John Stultz [this message]
2012-03-02 21:13   ` [PATCH 1/9] time: Condense timekeeper.xtime into xtime_sec Christoph Lameter
2012-03-03  5:54     ` Richard Cochran
2012-03-02  7:12 ` [PATCH 2/9] time: Rework timekeeping functions to take timekeeper ptr as argument John Stultz
2012-03-02  7:12 ` [PATCH 3/9] time: Split timekeeper lock into separate reader/writer locks John Stultz
2012-03-02  7:12 ` [PATCH 4/9] time: Update timekeeper structure using a local shadow John Stultz
2012-03-02  7:38   ` Ingo Molnar
2012-03-02  7:53     ` John Stultz
2012-03-02  8:07       ` Ingo Molnar
2012-03-02  7:12 ` [PATCH 5/9] time: Shadow cycle_last in timekeeper structure John Stultz
2012-03-02  7:12 ` [PATCH 6/9] time: Reduce timekeeper read lock hold time John Stultz
2012-03-02  7:12 ` [PATCH 7/9] time: Convert the timekeeper's wlock to a raw_spin_lock John Stultz
2012-03-02  7:12 ` [PATCH 8/9] time: Whitespace cleanups per Ingo's requests John Stultz
2012-03-02  7:12 ` [PATCH 9/9] time: Explicitly use u32 instead of int for shift values John Stultz

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1330672368-32290-2-git-send-email-john.stultz@linaro.org \
    --to=john.stultz@linaro.org \
    --cc=eric.dumazet@gmail.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mingo@elte.hu \
    --cc=richardcochran@gmail.com \
    --cc=tglx@linutronix.de \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.