Re: [PATCH v2 3/6] rcu: limit PREEMPT_RCU configurations

[Frederic Weisbecker <frederic@kernel.org>]

Le Mon, Nov 25, 2024 at 01:40:39PM -0800, Ankur Arora a écrit :
> 
> Frederic Weisbecker <frederic@kernel.org> writes:
> 
> > Le Wed, Nov 06, 2024 at 12:17:55PM -0800, Ankur Arora a écrit :
> >> PREEMPT_LAZY can be enabled stand-alone or alongside PREEMPT_DYNAMIC
> >> which allows for dynamic switching of preemption models.
> >>
> >> The choice of PREEMPT_RCU or not, however, is fixed at compile time.
> >>
> >> Given that PREEMPT_RCU makes some trade-offs to optimize for latency
> >> as opposed to throughput, configurations with limited preemption
> >> might prefer the stronger forward-progress guarantees of PREEMPT_RCU=n.
> >>
> >> Accordingly, explicitly limit PREEMPT_RCU=y to the latency oriented
> >> preemption models: PREEMPT, PREEMPT_RT, and the runtime configurable
> >> model PREEMPT_DYNAMIC.
> >>
> >> This means the throughput oriented models, PREEMPT_NONE,
> >> PREEMPT_VOLUNTARY and PREEMPT_LAZY will run with PREEMPT_RCU=n.
> >>
> >> Cc: Paul E. McKenney <paulmck@kernel.org>
> >> Cc: Peter Zijlstra <peterz@infradead.org>
> >> Signed-off-by: Ankur Arora <ankur.a.arora@oracle.com>
> >> ---
> >>  kernel/rcu/Kconfig | 2 +-
> >>  1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/kernel/rcu/Kconfig b/kernel/rcu/Kconfig
> >> index 5a7ff5e1cdcb..9d52f87fac27 100644
> >> --- a/kernel/rcu/Kconfig
> >> +++ b/kernel/rcu/Kconfig
> >> @@ -18,7 +18,7 @@ config TREE_RCU
> >>
> >>  config PREEMPT_RCU
> >>  	bool
> >> -	default y if PREEMPTION
> >> +	default y if (PREEMPT || PREEMPT_RT || PREEMPT_DYNAMIC)
> >>  	select TREE_RCU
> >>  	help
> >>  	  This option selects the RCU implementation that is
> >
> > Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
> >
> > But looking at !CONFIG_PREEMPT_RCU code on tree_plugin.h, I see
> > some issues now that the code can be preemptible. Well I think
> > it has always been preemptible but PREEMPTION && !PREEMPT_RCU
> > has seldom been exerciced (or was it even possible?).
> >
> > For example rcu_read_unlock_strict() can be called with preemption
> > enabled so we need the following otherwise the rdp is unstable, the
> > norm value becomes racy (though automagically fixed in rcu_report_qs_rdp())
> > and rcu_report_qs_rdp() might warn.
> >
> > diff --git a/include/linux/rcupdate.h b/include/linux/rcupdate.h
> > index 58d84c59f3dd..368f00267d4e 100644
> > --- a/include/linux/rcupdate.h
> > +++ b/include/linux/rcupdate.h
> > @@ -95,9 +95,9 @@ static inline void __rcu_read_lock(void)
> >
> >  static inline void __rcu_read_unlock(void)
> >  {
> > -	preempt_enable();
> >  	if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD))
> >  		rcu_read_unlock_strict();
> > +	preempt_enable();
> >  }
> >
> >  static inline int rcu_preempt_depth(void)
> 
> Based on the discussion on the thread, how about keeping this and
> changing the preempt_count check in rcu_read_unlock_strict() instead?
> 
> diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
> index 1c7cbd145d5e..8fc67639d3a7 100644
> --- a/kernel/rcu/tree_plugin.h
> +++ b/kernel/rcu/tree_plugin.h
> @@ -831,8 +831,15 @@ dump_blkd_tasks(struct rcu_node *rnp, int ncheck)
>  void rcu_read_unlock_strict(void)
>  {
>         struct rcu_data *rdp;
> +       int pc = ((preempt_count() & PREEMPT_MASK) >> PREEMPT_SHIFT);

This should be in_atomic_preempt_off(), otherwise softirqs and IRQs are
spuriously accounted as quiescent states.

Thanks.

> 
> -       if (irqs_disabled() || preempt_count() || !rcu_state.gp_kthread)
> +       /*
> +        * rcu_report_qs_rdp() can only be invoked with a stable rdp and
> +        * and from the local CPU.
> +        * With CONFIG_PREEMPTION=y, do this while holding the last
> +        * preempt_count which gets dropped after __rcu_read_unlock().
> +        */
> +       if (irqs_disabled() || pc > 1 || !rcu_state.gp_kthread)
>                 return;
>         rdp = this_cpu_ptr(&rcu_data);
>         rdp->cpu_no_qs.b.norm = false;
> 
> 
> Thanks
> 
> --
> ankur