diff mbox

[01/11] 2.6.32.x: ntp: Fix leap-second hrtimer livelock

Message ID 1342562735-55870-2-git-send-email-johnstul@us.ibm.com
State New
Headers show

Commit Message

john stultz July 17, 2012, 10:05 p.m. UTC
From: John Stultz <john.stultz@linaro.org>

This is a backport of 6b43ae8a619d17c4935c3320d2ef9e92bdeed05d

This should have been backported when it was commited, but I
mistook the problem as requiring the ntp_lock changes
that landed in 3.4 in order for it to occur.

Unfortunately the same issue can happen (with only one cpu)
as follows:
do_adjtimex()
 write_seqlock_irq(&xtime_lock);
  process_adjtimex_modes()
   process_adj_status()
    ntp_start_leap_timer()
     hrtimer_start()
      hrtimer_reprogram()
       tick_program_event()
        clockevents_program_event()
         ktime_get()
          seq = req_seqbegin(xtime_lock); [DEADLOCK]

This deadlock will no always occur, as it requires the
leap_timer to force a hrtimer_reprogram which only happens
if its set and there's no sooner timer to expire.

NOTE: This patch, being faithful to the original commit,
introduces a bug (we don't update wall_to_monotonic),
which will be resovled by backporting a following fix.

Original commit message below:

Since commit 7dffa3c673fbcf835cd7be80bb4aec8ad3f51168 the ntp
subsystem has used an hrtimer for triggering the leapsecond
adjustment. However, this can cause a potential livelock.

Thomas diagnosed this as the following pattern:
CPU 0                                                    CPU 1
do_adjtimex()
  spin_lock_irq(&ntp_lock);
    process_adjtimex_modes();				 timer_interrupt()
      process_adj_status();                                do_timer()
        ntp_start_leap_timer();                             write_lock(&xtime_lock);
          hrtimer_start();                                  update_wall_time();
             hrtimer_reprogram();                            ntp_tick_length()
               tick_program_event()                            spin_lock(&ntp_lock);
                 clockevents_program_event()
		   ktime_get()
                     seq = req_seqbegin(xtime_lock);

This patch tries to avoid the problem by reverting back to not using
an hrtimer to inject leapseconds, and instead we handle the leapsecond
processing in the second_overflow() function.

The downside to this change is that on systems that support highres
timers, the leap second processing will occur on a HZ tick boundary,
(ie: ~1-10ms, depending on HZ)  after the leap second instead of
possibly sooner (~34us in my tests w/ x86_64 lapic).

This patch applies on top of tip/timers/core.

CC: Sasha Levin <levinsasha928@gmail.com>
CC: Thomas Gleixner <tglx@linutronix.de>
Reported-by: Sasha Levin <levinsasha928@gmail.com>
Diagnoised-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Sasha Levin <levinsasha928@gmail.com>
Cc: Prarit Bhargava <prarit@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Linux Kernel <linux-kernel@vger.kernel.org>
Signed-off-by: John Stultz <john.stultz@linaro.org>
---
 include/linux/timex.h     |    2 +-
 kernel/time/ntp.c         |  122 +++++++++++++++------------------------------
 kernel/time/timekeeping.c |   12 ++---
 3 files changed, 44 insertions(+), 92 deletions(-)
diff mbox

Patch

diff --git a/include/linux/timex.h b/include/linux/timex.h
index e6967d1..3b587b4 100644
--- a/include/linux/timex.h
+++ b/include/linux/timex.h
@@ -271,7 +271,7 @@  static inline int ntp_synced(void)
 /* Returns how long ticks are at present, in ns / 2^NTP_SCALE_SHIFT. */
 extern u64 tick_length;
 
-extern void second_overflow(void);
+extern int second_overflow(unsigned long secs);
 extern void update_ntp_one_tick(void);
 extern int do_adjtimex(struct timex *);
 
diff --git a/kernel/time/ntp.c b/kernel/time/ntp.c
index 4800f93..dc76c9a 100644
--- a/kernel/time/ntp.c
+++ b/kernel/time/ntp.c
@@ -28,8 +28,6 @@  unsigned long			tick_nsec;
 u64				tick_length;
 static u64			tick_length_base;
 
-static struct hrtimer		leap_timer;
-
 #define MAX_TICKADJ		500LL		/* usecs */
 #define MAX_TICKADJ_SCALED \
 	(((MAX_TICKADJ * NSEC_PER_USEC) << NTP_SCALE_SHIFT) / NTP_INTERVAL_FREQ)
@@ -180,60 +178,60 @@  void ntp_clear(void)
 }
 
 /*
- * 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.
+ * 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 (Mills@UDEL.EDU) of NTP fame.
+ * They were originally developed for SUN and DEC kernels.
+ * All the kudos should go to Dave for this stuff.
+ *
+ * Also handles leap second processing, and returns leap offset
  */
-static enum hrtimer_restart ntp_leap_second(struct hrtimer *timer)
+int second_overflow(unsigned long secs)
 {
-	enum hrtimer_restart res = HRTIMER_NORESTART;
-
-	write_seqlock(&xtime_lock);
+	int leap = 0;
+	s64 delta;
 
+	/*
+	 * 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.
+	 */
 	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:
-		timekeeping_leap_insert(-1);
-		time_state = TIME_OOP;
-		printk(KERN_NOTICE
-			"Clock: inserting leap second 23:59:60 UTC\n");
-		hrtimer_add_expires_ns(&leap_timer, NSEC_PER_SEC);
-		res = HRTIMER_RESTART;
+		if (secs % 86400 == 0) {
+			leap = -1;
+			time_state = TIME_OOP;
+			printk(KERN_NOTICE
+				"Clock: inserting leap second 23:59:60 UTC\n");
+		}
 		break;
 	case TIME_DEL:
-		timekeeping_leap_insert(1);
-		time_tai--;
-		time_state = TIME_WAIT;
-		printk(KERN_NOTICE
-			"Clock: deleting leap second 23:59:59 UTC\n");
+		if ((secs + 1) % 86400 == 0) {
+			leap = 1;
+			time_tai--;
+			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;
-		/* fall through */
+		break;
+
 	case TIME_WAIT:
 		if (!(time_status & (STA_INS | STA_DEL)))
 			time_state = TIME_OK;
 		break;
 	}
 
-	write_sequnlock(&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 (Mills@UDEL.EDU) 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 delta;
 
 	/* Bump the maxerror field */
 	time_maxerror += MAXFREQ / NSEC_PER_USEC;
@@ -253,23 +251,25 @@  void second_overflow(void)
 	tick_length	+= delta;
 
 	if (!time_adjust)
-		return;
+		goto out;
 
 	if (time_adjust > MAX_TICKADJ) {
 		time_adjust -= MAX_TICKADJ;
 		tick_length += MAX_TICKADJ_SCALED;
-		return;
+		goto out;
 	}
 
 	if (time_adjust < -MAX_TICKADJ) {
 		time_adjust += MAX_TICKADJ;
 		tick_length -= MAX_TICKADJ_SCALED;
-		return;
+		goto out;
 	}
 
 	tick_length += (s64)(time_adjust * NSEC_PER_USEC / NTP_INTERVAL_FREQ)
 							 << NTP_SCALE_SHIFT;
 	time_adjust = 0;
+out:
+	return leap;
 }
 
 #ifdef CONFIG_GENERIC_CMOS_UPDATE
@@ -331,27 +331,6 @@  static void notify_cmos_timer(void)
 static inline void notify_cmos_timer(void) { }
 #endif
 
-/*
- * Start the leap seconds timer:
- */
-static inline void ntp_start_leap_timer(struct timespec *ts)
-{
-	long now = ts->tv_sec;
-
-	if (time_status & STA_INS) {
-		time_state = TIME_INS;
-		now += 86400 - now % 86400;
-		hrtimer_start(&leap_timer, ktime_set(now, 0), HRTIMER_MODE_ABS);
-
-		return;
-	}
-
-	if (time_status & STA_DEL) {
-		time_state = TIME_DEL;
-		now += 86400 - (now + 1) % 86400;
-		hrtimer_start(&leap_timer, ktime_set(now, 0), HRTIMER_MODE_ABS);
-	}
-}
 
 /*
  * Propagate a new txc->status value into the NTP state:
@@ -374,22 +353,6 @@  static inline void process_adj_status(struct timex *txc, struct timespec *ts)
 	time_status &= STA_RONLY;
 	time_status |= txc->status & ~STA_RONLY;
 
-	switch (time_state) {
-	case TIME_OK:
-		ntp_start_leap_timer(ts);
-		break;
-	case TIME_INS:
-	case TIME_DEL:
-		time_state = TIME_OK;
-		ntp_start_leap_timer(ts);
-	case TIME_WAIT:
-		if (!(time_status & (STA_INS | STA_DEL)))
-			time_state = TIME_OK;
-		break;
-	case TIME_OOP:
-		hrtimer_restart(&leap_timer);
-		break;
-	}
 }
 /*
  * Called with the xtime lock held, so we can access and modify
@@ -469,9 +432,6 @@  int do_adjtimex(struct timex *txc)
 		    (txc->tick <  900000/USER_HZ ||
 		     txc->tick > 1100000/USER_HZ))
 			return -EINVAL;
-
-		if (txc->modes & ADJ_STATUS && time_state != TIME_OK)
-			hrtimer_cancel(&leap_timer);
 	}
 
 	getnstimeofday(&ts);
@@ -549,6 +509,4 @@  __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;
 }
diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
index 4a71cff..00e2fae 100644
--- a/kernel/time/timekeeping.c
+++ b/kernel/time/timekeeping.c
@@ -183,14 +183,6 @@  void update_xtime_cache(u64 nsec)
 	ACCESS_ONCE(xtime_cache) = ts;
 }
 
-/* must hold xtime_lock */
-void timekeeping_leap_insert(int leapsecond)
-{
-	xtime.tv_sec += leapsecond;
-	wall_to_monotonic.tv_sec -= leapsecond;
-	update_vsyscall(&xtime, timekeeper.clock, timekeeper.mult);
-}
-
 #ifdef CONFIG_GENERIC_TIME
 
 /**
@@ -783,9 +775,11 @@  void update_wall_time(void)
 
 		timekeeper.xtime_nsec += timekeeper.xtime_interval;
 		if (timekeeper.xtime_nsec >= nsecps) {
+			int leap;
 			timekeeper.xtime_nsec -= nsecps;
 			xtime.tv_sec++;
-			second_overflow();
+			leap = second_overflow(xtime.tv_sec);
+			xtime.tv_sec += leap;
 		}
 
 		raw_time.tv_nsec += timekeeper.raw_interval;