Re: [PATCH] rcu: Fix race in set and clear TICK_DEP_BIT_RCU_EXP bitmask

[Frederic Weisbecker]

On Sun, Jan 01, 2023 at 09:41:57AM +0000, Zhang, Qiang1 wrote:
> > > >On Sat, Dec 31, 2022 at 07:25:08PM +0100, Frederic Weisbecker wrote:
> 
> Yes,    and I have a question,  we forcing the tick dependency because the expedited grace period
> is not end for the first time, this means that it is not to set the tick dependency every time.
> if we set the tick dependency in rcu_exp_handler(), does this mean that every time the expedited
> grace period starts the tick dependency will be set unconditionally ?
> 
> Thoughts ?

Only if rcu_exp_handler() fails to report a quiescent state. Then it means we
must poll on the CPU looking for a future one.

In fact the tick dependency should be set when rdp->cpu_no_qs.b.exp is set to
true and cleared when rdp->cpu_no_qs.b.exp is set to false.

Thanks.

> 
> Thanks
> Zqiang
> 
> >
> >							Thanx, Paul