ntp: Fix leap-second hrtimer livelock

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>
Signed-off-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

1 files changed, 7 insertions, 11 deletions

diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
index 5f45831..c444da0 100644
--- a/kernel/time/timekeeping.c
+++ b/kernel/time/timekeeping.c
@@ -169,15 +169,6 @@ static struct timespec raw_time;
 /* flag for if timekeeping is suspended */
 int __read_mostly timekeeping_suspended;
 
-/* must hold xtime_lock */
-void timekeeping_leap_insert(int leapsecond)
-{
-	xtime.tv_sec += leapsecond;
-	wall_to_monotonic.tv_sec -= leapsecond;
-	update_vsyscall(&xtime, &wall_to_monotonic, timekeeper.clock,
-			timekeeper.mult);
-}
-
 /**
  * timekeeping_forward_now - update clock to the current time
  *
@@ -828,9 +819,11 @@ static cycle_t logarithmic_accumulation(cycle_t offset, int shift)
 
 	timekeeper.xtime_nsec += timekeeper.xtime_interval << shift;
 	while (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;
 	}
 
 	/* Accumulate raw time */
@@ -936,9 +929,12 @@ static void update_wall_time(void)
 	 * xtime.tv_nsec isn't larger then NSEC_PER_SEC
 	 */
 	if (unlikely(xtime.tv_nsec >= NSEC_PER_SEC)) {
+		int leap;
 		xtime.tv_nsec -= NSEC_PER_SEC;
 		xtime.tv_sec++;
-		second_overflow();
+		leap = second_overflow(xtime.tv_sec);
+		xtime.tv_sec += leap;
+
 	}
 
 	/* check to see if there is a new clocksource to use */