On Fri, Sep 04, 2020 at 06:41:42AM -0700, Paul E. McKenney wrote:
> On Fri, Sep 04, 2020 at 12:05:34PM +0800, Boqun Feng wrote:
> > Hi Paul,
> >
> > On Mon, Aug 31, 2020 at 11:11:12AM -0700, paulmck@xxxxxxxxxx wrote:
> > > From: "Paul E. McKenney" <paulmck@xxxxxxxxxx>
> > >
> > > The ->rcu_read_unlock_special.b.need_qs field in the task_struct
> > > structure indicates that the RCU core needs a quiscent state from the
> > > corresponding task. The __rcu_read_unlock() function checks this (via
> > > an eventual call to rcu_preempt_deferred_qs_irqrestore()), and if set
> > > reports a quiscent state immediately upon exit from the outermost RCU
> > > read-side critical section.
> > >
> > > Currently, this flag is only set when the scheduling-clock interrupt
> > > decides that the current RCU grace period is too old, as in about
> > > one full second too old. But if the kernel has been built with
> > > CONFIG_RCU_STRICT_GRACE_PERIOD=y, we clearly do not want to wait that
> > > long. This commit therefore sets the .need_qs field immediately at the
> > > start of the RCU read-side critical section from within __rcu_read_lock()
> > > in order to unconditionally enlist help from __rcu_read_unlock().
> > >
> >
> > So why not make rcu_preempt_deferred_qs_irqrestore() always treat
> > need_qs is true if CONFIG_RCU_STRICT_GRACE_PERIOD = y? IOW:
> >
> > diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
> > index 982fc5be5269..2a9f31545453 100644
> > --- a/kernel/rcu/tree_plugin.h
> > +++ b/kernel/rcu/tree_plugin.h
> > @@ -449,6 +449,8 @@ rcu_preempt_deferred_qs_irqrestore(struct task_struct *t, unsigned long flags)
> > * t->rcu_read_unlock_special cannot change.
> > */
> > special = t->rcu_read_unlock_special;
> > + if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD) && rcu_state.gp_kthread)
> > + special.b.need_qs = true;
> > rdp = this_cpu_ptr(&rcu_data);
> > if (!special.s && !rdp->exp_deferred_qs) {
> > local_irq_restore(flags);
> >
> > , and in this way, you can save one store for each rcu_read_lock() ;-)
>
> Because unless I am missing something subtle, if the .need_qs
> flag is not set, execution is not guaranteed to reach
> rcu_preempt_deferred_qs_irqrestore().
>
Fair enough. Although I think we can also add IS_ENABLED(...) check to
make the outermost rcu_read_unlock() to call rcu_read_unlock_special()
unconditionally, but that's too much I think.
Regards,
Boqun
> Thanx, Paul
>
> > Regards,
> > Boqun
> >
> > > But note the additional check for rcu_state.gp_kthread, which prevents
> > > attempts to awaken RCU's grace-period kthread during early boot before
> > > there is a scheduler. Leaving off this check results in early boot hangs.
> > > So early that there is no console output. Thus, this additional check
> > > fails until such time as RCU's grace-period kthread has been created,
> > > avoiding these empty-console hangs.
> > >
> > > Reported-by Jann Horn <jannh@xxxxxxxxxx>
> > > Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
> > > ---
> > > kernel/rcu/tree_plugin.h | 2 ++
> > > 1 file changed, 2 insertions(+)
> > >
> > > diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
> > > index 44cf77d..668bbd2 100644
> > > --- a/kernel/rcu/tree_plugin.h
> > > +++ b/kernel/rcu/tree_plugin.h
> > > @@ -376,6 +376,8 @@ void __rcu_read_lock(void)
> > > rcu_preempt_read_enter();
> > > if (IS_ENABLED(CONFIG_PROVE_LOCKING))
> > > WARN_ON_ONCE(rcu_preempt_depth() > RCU_NEST_PMAX);
> > > + if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD) && rcu_state.gp_kthread)
> > > + WRITE_ONCE(current->rcu_read_unlock_special.b.need_qs, true);
> > > barrier(); /* critical section after entry code. */
> > > }
> > > EXPORT_SYMBOL_GPL(__rcu_read_lock);
> > > --
> > > 2.9.5
> > >
