Re: [RFC PATCH RT 3/4] rcu: unlock special: Treat irq and preempt disabled the same

From: Paul E. McKenney
Date: Fri Jun 21 2019 - 20:26:20 EST


On Thu, Jun 20, 2019 at 06:08:19PM -0500, Scott Wood wrote:
> On Thu, 2019-06-20 at 15:25 -0700, Paul E. McKenney wrote:
> > On Thu, Jun 20, 2019 at 04:59:30PM -0500, Scott Wood wrote:
> > > On Thu, 2019-06-20 at 14:10 -0700, Paul E. McKenney wrote:
> > > > On Tue, Jun 18, 2019 at 08:19:07PM -0500, Scott Wood wrote:
> > > > > [Note: Just before posting this I noticed that the invoke_rcu_core
> > > > > stuff
> > > > > is part of the latest RCU pull request, and it has a patch that
> > > > > addresses this in a more complicated way that appears to deal with
> > > > > the
> > > > > bare irq-disabled sequence as well.
> > > >
> > > > Far easier to deal with it than to debug the lack of it. ;-)
> > > >
> > > > > Assuming we need/want to support such sequences, is the
> > > > > invoke_rcu_core() call actually going to result in scheduling any
> > > > > sooner? resched_curr() just does the same setting of need_resched
> > > > > when it's the same cpu.
> > > > > ]
> > > >
> > > > Yes, invoke_rcu_core() can in some cases invoke the scheduler sooner.
> > > > Setting the CPU-local bits might not have effect until the next
> > > > interrupt.
> > >
> > > Maybe I'm missing something, but I don't see how (in the non-use_softirq
> > > case). It just calls wake_up_process(), which in resched_curr() will
> > > set
> > > need_resched but not do an IPI-to-self.
> >
> > The common non-rt case will be use_softirq. Or are you referring
> > specifically to this block of code in current -rcu?
> >
> > } else if (exp && irqs_were_disabled && !use_softirq &&
> > !t->rcu_read_unlock_special.b.deferred_qs) {
> > // Safe to awaken and we get no help from enabling
> > // irqs, unlike bh/preempt.
> > invoke_rcu_core();
>
> Yes, that one. If that block is removed the else path should be sufficient,
> now that an IPI-to-self has been added.

I will give it a try and let you know what happens.

> Also, shouldn't the IPI-to-self be conditioned on irqs_were_disabled?
> Besides that being the problem the IPI was meant to address, if irqs are
> enabled the IPI is likely to happen before preempt is re-enabled and thus it
> won't accomplish anything.

Plus if preempt is disabled, the later preempt_enable() will check
(ditto for local_bh_enable()). Unless the preempt_enable() is instead
a preempt_enable_no_resched(), of course. :-/

Thanx, Paul