To reduce RCU noise for nohz_full configurations, osnoise depends
on cond_resched() providing quiescent states for PREEMPT_RCU=n
configurations. And, for PREEMPT_RCU=y configurations does this
by directly calling rcu_momentary_eqs().

With PREEMPT_LAZY=y, however, we can have configurations with
(PREEMPTION=y, PREEMPT_RCU=n), which means neither of the above
can help.

Handle that by fallback to the explicit quiescent states via
rcu_momentary_eqs().

Cc: Paul E. McKenney <paulmck@kernel.org>
Cc: Daniel Bristot de Oliveira <bristot@kernel.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Suggested-by: Paul E. McKenney <paulmck@kernel.org>
Acked-by: Daniel Bristot de Oliveira <bristot@kernel.org>
Signed-off-by: Ankur Arora <ankur.a.arora@oracle.com>
---
 kernel/trace/trace_osnoise.c | 22 ++++++++++++----------
 1 file changed, 12 insertions(+), 10 deletions(-)

diff --git a/kernel/trace/trace_osnoise.c b/kernel/trace/trace_osnoise.c
index 1439064f65d6..4f4ebf3f15f0 100644
--- a/kernel/trace/trace_osnoise.c
+++ b/kernel/trace/trace_osnoise.c
@@ -1538,18 +1538,20 @@ static int run_osnoise(void)
 		/*
 		 * In some cases, notably when running on a nohz_full CPU with
 		 * a stopped tick PREEMPT_RCU has no way to account for QSs.
-		 * This will eventually cause unwarranted noise as PREEMPT_RCU
-		 * will force preemption as the means of ending the current
-		 * grace period. We avoid this problem by calling
-		 * rcu_momentary_eqs(), which performs a zero duration
-		 * EQS allowing PREEMPT_RCU to end the current grace period.
-		 * This call shouldn't be wrapped inside an RCU critical
-		 * section.
+		 * This will eventually cause unwarranted noise as RCU forces
+		 * preemption as the means of ending the current grace period.
+		 * We avoid this by calling rcu_momentary_eqs(), which performs
+		 * a zero duration EQS allowing RCU to end the current grace
+		 * period. This call shouldn't be wrapped inside an RCU
+		 * critical section.
 		 *
-		 * Note that in non PREEMPT_RCU kernels QSs are handled through
-		 * cond_resched()
+		 * For non-PREEMPT_RCU kernels with cond_resched() (non-
+		 * PREEMPT_LAZY configurations), QSs are handled through
+		 * cond_resched(). For PREEMPT_LAZY kernels, we fallback to
+		 * the zero duration QS via rcu_momentary_eqs().
 		 */
-		if (IS_ENABLED(CONFIG_PREEMPT_RCU)) {
+		if (IS_ENABLED(CONFIG_PREEMPT_RCU) ||
+		    (!IS_ENABLED(CONFIG_PREEMPT_RCU) && IS_ENABLED(CONFIG_PREEMPTION))) {
 			if (!disable_irq)
 				local_irq_disable();
 
-- 
2.43.5

On 2024-10-09 09:54:10 [-0700], Ankur Arora wrote:
> To reduce RCU noise for nohz_full configurations, osnoise depends
> on cond_resched() providing quiescent states for PREEMPT_RCU=n
> configurations. And, for PREEMPT_RCU=y configurations does this
> by directly calling rcu_momentary_eqs().
> 
> With PREEMPT_LAZY=y, however, we can have configurations with
> (PREEMPTION=y, PREEMPT_RCU=n), which means neither of the above
> can help.

PREEMPTION=y, PREEMPT_RCU=n should not be possible.

> Handle that by fallback to the explicit quiescent states via
> rcu_momentary_eqs().

Sebastian

Sebastian Andrzej Siewior <bigeasy@linutronix.de> writes:

> On 2024-10-09 09:54:10 [-0700], Ankur Arora wrote:
>> To reduce RCU noise for nohz_full configurations, osnoise depends
>> on cond_resched() providing quiescent states for PREEMPT_RCU=n
>> configurations. And, for PREEMPT_RCU=y configurations does this
>> by directly calling rcu_momentary_eqs().
>>
>> With PREEMPT_LAZY=y, however, we can have configurations with
>> (PREEMPTION=y, PREEMPT_RCU=n), which means neither of the above
>> can help.
>
> PREEMPTION=y, PREEMPT_RCU=n should not be possible.

That's a statement. Is there an argument here?

--
ankur

On 2024-10-10 10:50:29 [-0700], Ankur Arora wrote:
> > PREEMPTION=y, PREEMPT_RCU=n should not be possible.
> 
> That's a statement. Is there an argument here?

For my taste you should describe in your cover letter the actual problem
and what you intend to do about it. Then you should a series addressing
this issue which would probably qualify for all patches in your series
but 7/7 (the PPC bits for lazy preempt). 7/7 should have been the part
where you make possible to make PREEMPT_RCU selectable.

Based on my understanding so far you (or Paul) want to make PREEMPT_RCU
selectable if PREEMPT_LAZY is enabled _or_ if DYNAMIC_PREEMPT is enabled
with the NONE or VOLUNTARY model.

This series as-is made no sense to me until Peter made the snippet where
you could indeed make PREEMPT_RCU selectable.

Sebastian

Sebastian Andrzej Siewior <bigeasy@linutronix.de> writes:

> On 2024-10-10 10:50:29 [-0700], Ankur Arora wrote:
>> > PREEMPTION=y, PREEMPT_RCU=n should not be possible.
>>
>> That's a statement. Is there an argument here?
>
> For my taste you should describe in your cover letter the actual problem
> and what you intend to do about it.

Yes, that's a good point. It has been discussed a few times before but
I clearly didn't make a case for it here.

Thanks
Ankur

On Thu, Oct 10, 2024 at 08:53:38AM +0200, Sebastian Andrzej Siewior wrote:
> On 2024-10-09 09:54:10 [-0700], Ankur Arora wrote:
> > To reduce RCU noise for nohz_full configurations, osnoise depends
> > on cond_resched() providing quiescent states for PREEMPT_RCU=n
> > configurations. And, for PREEMPT_RCU=y configurations does this
> > by directly calling rcu_momentary_eqs().
> > 
> > With PREEMPT_LAZY=y, however, we can have configurations with
> > (PREEMPTION=y, PREEMPT_RCU=n), which means neither of the above
> > can help.
> 
> PREEMPTION=y, PREEMPT_RCU=n should not be possible.

PREEMPT_LAZY=y, PREEMPT_RCU=n really does need to be possible in order to
avoid OOM on systems not abundantly endowed with memory.  This gets the
improvement in tail latencies from PREEMPT_LAZY but the OOM resistance
of PREEMPT_RCU=n.  Such systems really do not want to be preempting
RCU readers.

							Thanx, Paul

> > Handle that by fallback to the explicit quiescent states via
> > rcu_momentary_eqs().
> 
> Sebastian