ntp: Fix leap-second hrtimer livelock
RFC PATCH 1/1 ntp: Fix leap-second hrtimer livelock
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 1do_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.
--- include/linux/timex.h | 2 +- kernel/time/ntp.c | 128 ++++++++++++++------------------------------ kernel/time/timekeeping.c | 20 ++----- 3 files changed, 48 insertions(+), 102 deletions(-)
include/linux/timex.h +1 -1
@@ -252,7 +252,7 @@ extern void ntp_clear(void);/* Returns how long ticks are at present, in ns / 2^NTP_SCALE_SHIFT. */ extern u64 ntp_tick_length (void ); -extern void second_overflow (void ); +extern int second_overflow (unsigned long secs);extern int do_adjtimex (struct timex *);extern void hardpps (const struct timespec *,const struct timespec *);
kernel/time/ntp.c +41 -87
191 lines not on this page · raw message
kernel/time/timekeeping.c +6 -14
@@ -184,18 +184,6 @@ static void timekeeping_update(bool clearntp) } -void timekeeping_leap_insert (int leapsecond) -{ -unsigned long flags; - -write_seqlock_irqsave (&timekeeper.lock, flags); - timekeeper.xtime.tv_sec += leapsecond; - timekeeper.wall_to_monotonic.tv_sec -= leapsecond; -timekeeping_update (false); -write_sequnlock_irqrestore (&timekeeper.lock, flags); - -} -/** * timekeeping_forward_now - update clock to the current time *
@@ -962,9 +950,11 @@ 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; timekeeper.xtime.tv_sec++; -second_overflow (); + leap =second_overflow (timekeeper.xtime.tv_sec); + timekeeper.xtime.tv_sec += leap; }/* Accumulate raw time */
@@ -1075,9 +1065,11 @@ static void update_wall_time(void)* xtime.tv_nsec isn't larger then NSEC_PER_SEC */ if (unlikely (timekeeper.xtime.tv_nsec >=NSEC_PER_SEC )) { +int leap; timekeeper.xtime.tv_nsec -=NSEC_PER_SEC ; timekeeper.xtime.tv_sec++; -second_overflow (); + leap =second_overflow (timekeeper.xtime.tv_sec); + timekeeper.xtime.tv_sec += leap; }timekeeping_update (false);
Signature · 1 line
1.7.3.2.146.gca209
Quoted 3 lines · John Stultz
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.
Looks like the patch fixes the problem here, no other side effects observed.