On Wed, 8 Aug 2018 07:42:00 -0700
"Paul E. McKenney" <[email protected]> wrote:
> > There's also a local_inc() if you are using per cpu pointers, that is
> > suppose to guarantee atomicity for single cpu operations. That's what
> > the ftrace ring buffer uses.
>
> Good point, that becomes atomic_long_inc() or equivalent on most
> architectures, but an incl instruction (not locked) on x86. So updating
> my earlier still-untested thought:
>
> int __srcu_read_lock_nmi(struct srcu_struct *sp) /* UNTESTED. */
> {
> int idx;
>
> idx = READ_ONCE(sp->srcu_idx) & 0x1;
> local_inc(&sp->sda->srcu_lock_count[idx]);
> smp_mb__after_atomic(); /* B */ /* Avoid leaking critical section. */
> return idx;
> }
>
> void __srcu_read_unlock_nmi(struct srcu_struct *sp, int idx)
> {
> smp_mb__before_atomic(); /* C */ /* Avoid leaking critical section. */
> local_inc(&sp->sda->srcu_unlock_count[idx]);
> }
>
> Would that work, or is there a better way to handle this?
This would work much better than using the atomic ops, and I think it
would be doable.
I may need to revert the srcu for trace_*_rcidle() for now, as I want
most of the other changes in this merge window, and it's getting too
late to do it with these updates.
-- Steve
