Re: [PATCH 20/21] rcu/context_tracking: Merge dynticks counter and context tracking states

[nicolas saenz julienne <nsaenz@kernel.org>]

Hi Frederic,

On Wed, 2022-06-08 at 16:29 +0200, Frederic Weisbecker wrote:
> On Tue, May 31, 2022 at 06:15:36PM +0200, nicolas saenz julienne wrote:
> > On Tue, 2022-05-31 at 16:23 +0200, Frederic Weisbecker wrote:
> > > But idle at least is an exception and CONTEXT_IDLE will remain during the
> > > interrupt handling. It's not that trivial to handle the idle case because
> > > ct_irq_exit() needs to know that it is called between ct_idle_enter() and
> > > ct_idle_exit().
> >
> > Just for the record, this behaviour was already here regardless of this series,
> > so it's not something it needs to fix.
>
> Right.
>
> >
> > Something like this should work, right?
> >
> >      ct_idle_enter()
> >          //IRQ or NMI
> > 	 if (__ct_state() == CONTEXT_IDLE)
> > 	     ct_idle_exit()
>
> Right but that's one more costly operation (atomic_add_return())
>
> >          ct_irq_enter()
>
> Ideally this should increment by RCU_DYNTICKS_IDX - CONTEXT_IDLE
>
> > 	    ...
> >          ct_irq_exit()
>
> And this should increment by RCU_DYNTICKS_IDX + CONTEXT_IDLE
>
> I guess the CONTEXT_IDLE state should be remembered on some per cpu
> variable somewhere.
>
> BTW one interesting optimization to do when an idle interrupt leads to
> setting need_resched() would be to have:
>
>     idle_loop() {
>         while (!need_resched) {
>             rcu_idle_enter();
> 	    mwait();
> 	    //IRQ {
> 	        rcu_irq_enter();
> 		do_irq()... //set need_resched()
> 		rcu_irq_exit() // but no need to do the atomic_add_return() here
> 		               // since we want to keep RCU watching as we'll
>                                // escape from idle
>             }
> 	    rcu_idle_exit() // and no need to do the atomic_add_return() here either
>
>
> That's two expensive operations spared for a pretty common event.

>
>
>
> > 	 if (needs_update_state()) //using irqentry_state_t for ex.
> > 	     ct_idle_entry()
> >      ct_idle_exit()
> >
> > Note that it's not a big issue as we can work around this behaviour by checking
> > through dynticks whether a CPU is really idle.
> >
> > Do you think it's worth fixing nonetheless?
>
> Nothing urgent for sure.

Thanks for the feedback, I'll think about it.

BTW if you're patient I'll try to make a last test run on v4 of the series next
week.

Regards,
Nicolas