diff options
author | Alex Frid <afrid@nvidia.com> | 2010-05-15 17:49:46 -0700 |
---|---|---|
committer | Gary King <gking@nvidia.com> | 2010-05-20 10:53:31 -0700 |
commit | b344cce5024b2742219db782aeeb38772a05325a (patch) | |
tree | 9285078fe841fe401cb0d8628126b4c01da41ff6 /kernel | |
parent | 4bc4768e06afa53392678197d59b6394cdab5e1c (diff) |
tegra timers: Added safety net for SMP tick handover.
Made sure that jiffies update duty is not stuck with off-line CPU: if this
situation is detected, re-assign the tick ownership respectively (should
fix bug 683277).
Replaced spin-locks in timer access code with local interrupt control:
local timers are "local", and do not need interprocessor protection.
Change-Id: Ib0d690d8558724f7da384aba8b5cdf3d5ea8fdbb
Reviewed-on: http://git-master/r/1415
Tested-by: Aleksandr Frid <afrid@nvidia.com>
Reviewed-by: Gary King <gking@nvidia.com>
Diffstat (limited to 'kernel')
-rw-r--r-- | kernel/time/tick-sched.c | 8 |
1 files changed, 7 insertions, 1 deletions
diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c index 4012da672821..6737d68fdb8d 100644 --- a/kernel/time/tick-sched.c +++ b/kernel/time/tick-sched.c @@ -639,9 +639,15 @@ static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer) * into a long sleep. If two cpus happen to assign themself to * this duty, then the jiffies update is still serialized by * xtime_lock. + * + * Also check if the do_timer duty is stuck with off-line cpu. + * Then, use the same remedy as for dropped duty: re-assign it + * to the cpu servicing this tick. */ - if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE)) + if ((unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE)) || + (!cpu_online(tick_do_timer_cpu))) { tick_do_timer_cpu = cpu; + } #endif /* Check, if the jiffies need an update */ |