Re: [PATCH v2 3/3] sched/preempt: fix cond_resched_lock() and cond_resched_softirq()

[Peter Zijlstra <peterz@infradead.org>]

On Wed, Jul 15, 2015 at 03:52:34PM +0300, Konstantin Khlebnikov wrote:
> On 15.07.2015 15:16, Eric Dumazet wrote:
> >On Wed, 2015-07-15 at 12:52 +0300, Konstantin Khlebnikov wrote:
> >>These functions check should_resched() before unlocking spinlock/bh-enable:
> >>preempt_count always non-zero => should_resched() always returns false.
> >>cond_resched_lock() worked iff spin_needbreak is set.
> >
> >Interesting, this definitely used to work (linux-3.11)
> >
> >Any idea of which commit broke things ?
> >
> 
> Searching... done
> 
> This one: bdb43806589096ac4272fe1307e789846ac08d7c in v3.13
> 
> before
> 
> -static inline int should_resched(void)
> -{
> -       return need_resched() && !(preempt_count() & PREEMPT_ACTIVE);
> -}
> 
> after
> 
> +static __always_inline bool should_resched(void)
> +{
> +       return unlikely(!*preempt_count_ptr());
> +}

Argh, indeed!

> 
> So,
> 
> Fixes: bdb438065890 ("sched: Extract the basic add/sub preempt_count
> modifiers")

Thanks!