Messages in this thread | | | Date | Tue, 7 Apr 2015 23:38:48 +0200 (CEST) | From | Thomas Gleixner <> | Subject | Re: [PATCH v2 1/2] rtmutex Real-Time Linux: Fixing kernel BUG at kernel/locking/rtmutex.c:997! |
| |
On Tue, 7 Apr 2015, Jason Low wrote: > On Tue, 2015-04-07 at 21:17 +0200, Thomas Gleixner wrote: > > On Tue, 7 Apr 2015, Jason Low wrote: > > > The lock shouldn't be used in get_next_timer_interrupt() either right? > > > > > > unsigned long get_next_timer_interrupt(unsigned long now) > > > { > > > ... > > > > > > #ifdef CONFIG_PREEMPT_RT_FULL > > > /* > > > * On PREEMPT_RT we cannot sleep here. If the trylock does not > > > * succeed then we return the worst-case 'expires in 1 tick' > > > * value. We use the rt functions here directly to avoid a > > > * migrate_disable() call. > > > */ > > > if (!spin_do_trylock(&base->lock)) > > > return now + 1; > > > #else > > > > And how do you protect the walk of the timer wheel against a > > concurrent insertion/removal? > > So I just wanted to mention that the issue also applies to > get_next_timer_interrupt(), in addition to run_local_timers(), but if we > really want to remove the lock there, can we always return "now + 1" for > PREEMPT_RT_FULL?
There is a simpler solution for this: Disable NOHZ for RT. Because that's what 'always return "now + 1"' does at runtime just with more overhead.
We need a smarter solution for RT to avoid the lock completely.
Thanks,
tglx
| |