On Wed, 23 Apr 2014 09:14:33 +0200 Sebastian Andrzej Siewior <[email protected]> wrote:
> On 04/22/2014 08:16 PM, Steven Rostedt wrote: > > --- a/kernel/timer.c > > +++ b/kernel/timer.c > > @@ -1447,6 +1447,12 @@ static void run_timer_softirq(struct softirq_action > > *h) > > __run_timers(base); > > } > > > > +#ifdef CONFIG_SMP > > +#define timer_should_raise_softirq(lock) !spin_do_trylock(lock) > > +#else > > +#define timer_should_raise_softirq(lock) 1 > > +#endif > > + > > No. The lock may be taken but it also may be available no matter if UP > or not. With this patch applied the lockdep splat will go away but the > FULL_NO_HZ people will come back because the timer softirq is scheduled > even if no timer has expired. Although, as Peter said FULL_NO_HZ is pretty pointless on UP, but I've been thinking this from a non PREEMPT_RT viewpoint. In non PREEMPT_RT, spin_locks() are nops, where this is an issue, but in PREEMPT_RT, they are mutexes, BUT! The rt_mutex use wait_lock to protect its internal state, which is a raw spinlock, and in UP it's a nop. That means there's nothing preventing an interrupt here from corrupting the rtmutex's internal state. I see this is still an issue, and the warning is still valid. -- Steve -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
