Re: [PATCH v6 1/4] rcu: Make call_rcu() lazy to save power

From: Joel Fernandes
Date: Mon Sep 26 2022 - 21:16:30 EST


On Mon, Sep 26, 2022 at 04:57:55PM -0700, Paul E. McKenney wrote:
[..]
> > >>>>>> diff --git a/include/linux/rcupdate.h b/include/linux/rcupdate.h
> > >>>>>> index 08605ce7379d..40ae36904825 100644
> > >>>>>> --- a/include/linux/rcupdate.h
> > >>>>>> +++ b/include/linux/rcupdate.h
> > >>>>>> @@ -108,6 +108,13 @@ static inline int rcu_preempt_depth(void)
> > >>>>>>
> > >>>>>> #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
> > >>>>>>
> > >>>>>> +#ifdef CONFIG_RCU_LAZY
> > >>>>>> +void call_rcu_flush(struct rcu_head *head, rcu_callback_t func);
> > >>>>>> +#else
> > >>>>>> +static inline void call_rcu_flush(struct rcu_head *head,
> > >>>>>> + rcu_callback_t func) { call_rcu(head, func); }
> > >>>>>> +#endif
> > >>>>>> +
> > >>>>>> /* Internal to kernel */
> > >>>>>> void rcu_init(void);
> > >>>>>> extern int rcu_scheduler_active;
> > >>>>>> diff --git a/kernel/rcu/Kconfig b/kernel/rcu/Kconfig
> > >>>>>> index f53ad63b2bc6..edd632e68497 100644
> > >>>>>> --- a/kernel/rcu/Kconfig
> > >>>>>> +++ b/kernel/rcu/Kconfig
> > >>>>>> @@ -314,4 +314,12 @@ config TASKS_TRACE_RCU_READ_MB
> > >>>>>> Say N here if you hate read-side memory barriers.
> > >>>>>> Take the default if you are unsure.
> > >>>>>>
> > >>>>>> +config RCU_LAZY
> > >>>>>> + bool "RCU callback lazy invocation functionality"
> > >>>>>> + depends on RCU_NOCB_CPU
> > >>>>>> + default n
> > >>>>>> + help
> > >>>>>> + To save power, batch RCU callbacks and flush after delay, memory
> > >>>>>> + pressure or callback list growing too big.
> > >>>>>> +
> > >>>>>>
> > >>>>> Do you think you need this kernel option? Can we just consider and make
> > >>>>> it a run-time configurable? For example much more users will give it a try,
> > >>>>> so it will increase a coverage. By default it can be off.
> > >>>>>
> > >>>>> Also you do not need to do:
> > >>>>>
> > >>>>> #ifdef LAZY
> > >>>>
> > >>>> How does the "LAZY" macro end up being runtime-configurable? That's static /
> > >>>> compile time. Did I miss something?
> > >>>>
> > >>> I am talking about removing if:
> > >>>
> > >>> config RCU_LAZY
> > >>>
> > >>> we might run into issues related to run-time switching though.
> > >>
> > >> When we started off, Paul said he wanted it kernel CONFIGurable. I will defer
> > >> to Paul on a decision for that. I prefer kernel CONFIG so people don't forget
> > >> to pass a boot param.
> > >
> > > I am fine with a kernel boot parameter for this one. You guys were the
> > > ones preferring Kconfig options. ;-)
> >
> > Yes I still prefer that.. ;-)
> >
> > > But in that case, the CONFIG_RCU_NOCB_CPU would come into play to handle
> > > the case where there is no bypass.
> >
> > If you don’t mind, let’s do both like we did for NOCB_CPU_ALL. In which
> > case, Vlad since this was your suggestion, would you be so kind to send a
> > patch adding a boot parameter on top of the series? ;-). I’ll include it
> > in the next version. I’d suggest keep the boot param default off and add
> > a CONFIG option that forces the boot param to be turned on.
>
> NOCB_CPU_ALL? If you are thinking in terms of laziness/flushing being
> done on a per-CPU basis among the rcu_nocbs CPUs, that sounds like
> something for later.

Oh, no, I was just trying to bring that up as an example of making boot
parameters and CONFIG options for the same thing.

> Are you thinking in terms of Kconfig options that allow: (1) No laziness.
> (2) Laziness on all rcu_nocbs CPUs, but only if specified by a boot
> parameter. (3) Laziness on all rcu_nocbs CPUs regardless of boot
> parameter. I could get behind that.

Sure agreed, or we could just make it CONFIG_RCU_LAZY_DEFAULT=y and if boot
param is specified, override the CONFIG. That will be the simplest and least
confusing IMO.

thanks :)

- Joel




>