Re: [RFC PATCH 59/86] treewide: rcu: remove cond_resched()

From: Paul E. McKenney
Date: Mon Nov 20 2023 - 20:01:07 EST


On Tue, Nov 07, 2023 at 03:07:55PM -0800, Ankur Arora wrote:
> All the cond_resched() calls in the RCU interfaces here are to
> drive preemption once it has reported a potentially quiescent
> state, or to exit the grace period. With PREEMPTION=y that should
> happen implicitly.
>
> So we can remove these.
>
> [1] https://lore.kernel.org/lkml/20231107215742.363031-1-ankur.a.arora@xxxxxxxxxx/
>
> Cc: "Paul E. McKenney" <paulmck@xxxxxxxxxx>
> Cc: Frederic Weisbecker <frederic@xxxxxxxxxx>
> Cc: Ingo Molnar <mingo@xxxxxxxxxx>
> Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> Cc: Juri Lelli <juri.lelli@xxxxxxxxxx>
> Cc: Vincent Guittot <vincent.guittot@xxxxxxxxxx>
> Signed-off-by: Ankur Arora <ankur.a.arora@xxxxxxxxxx>
> ---
> include/linux/rcupdate.h | 6 ++----
> include/linux/sched.h | 7 ++++++-
> kernel/hung_task.c | 6 +++---
> kernel/rcu/tasks.h | 5 +----
> 4 files changed, 12 insertions(+), 12 deletions(-)
>
> diff --git a/include/linux/rcupdate.h b/include/linux/rcupdate.h
> index 7246ee602b0b..58f8c7faaa52 100644
> --- a/include/linux/rcupdate.h
> +++ b/include/linux/rcupdate.h
> @@ -238,14 +238,12 @@ static inline bool rcu_trace_implies_rcu_gp(void) { return true; }
> /**
> * cond_resched_tasks_rcu_qs - Report potential quiescent states to RCU
> *
> - * This macro resembles cond_resched(), except that it is defined to
> - * report potential quiescent states to RCU-tasks even if the cond_resched()
> - * machinery were to be shut off, as some advocate for PREEMPTION kernels.
> + * This macro resembles cond_resched(), in that it reports potential
> + * quiescent states to RCU-tasks.
> */
> #define cond_resched_tasks_rcu_qs() \
> do { \
> rcu_tasks_qs(current, false); \
> - cond_resched(); \

I am a bit nervous about dropping the cond_resched() in a few cases,
for example, the call from rcu_tasks_trace_pregp_step() only momentarily
enables interrupts. This should be OK given a scheduling-clock interrupt,
except that nohz_full CPUs don't necessarily have these. At least not
unless RCU happens to be in a grace period at the time.

> } while (0)
>
> /*
> diff --git a/include/linux/sched.h b/include/linux/sched.h
> index 199f8f7211f2..bae6eed534dd 100644
> --- a/include/linux/sched.h
> +++ b/include/linux/sched.h
> @@ -2145,7 +2145,12 @@ static inline void cond_resched_rcu(void)
> {
> #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) || !defined(CONFIG_PREEMPT_RCU)
> rcu_read_unlock();
> - cond_resched();
> +
> + /*
> + * Might reschedule here as we exit the RCU read-side
> + * critical section.
> + */
> +
> rcu_read_lock();

And here I am wondering if some of my nervousness about increased
grace-period latency due to removing cond_resched() might be addressed
by making preempt_enable() take over the help-RCU functionality currently
being provided by cond_resched()...

> #endif
> }
> diff --git a/kernel/hung_task.c b/kernel/hung_task.c
> index 9a24574988d2..4bdfad08a2e8 100644
> --- a/kernel/hung_task.c
> +++ b/kernel/hung_task.c
> @@ -153,8 +153,8 @@ static void check_hung_task(struct task_struct *t, unsigned long timeout)
> * To avoid extending the RCU grace period for an unbounded amount of time,
> * periodically exit the critical section and enter a new one.
> *
> - * For preemptible RCU it is sufficient to call rcu_read_unlock in order
> - * to exit the grace period. For classic RCU, a reschedule is required.
> + * Under a preemptive kernel, or with preemptible RCU, it is sufficient to
> + * call rcu_read_unlock in order to exit the grace period.
> */
> static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
> {
> @@ -163,7 +163,7 @@ static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
> get_task_struct(g);
> get_task_struct(t);
> rcu_read_unlock();
> - cond_resched();
> +
> rcu_read_lock();
> can_cont = pid_alive(g) && pid_alive(t);
> put_task_struct(t);
> diff --git a/kernel/rcu/tasks.h b/kernel/rcu/tasks.h
> index 8d65f7d576a3..fa1d9aa31b36 100644
> --- a/kernel/rcu/tasks.h
> +++ b/kernel/rcu/tasks.h
> @@ -541,7 +541,6 @@ static void rcu_tasks_invoke_cbs(struct rcu_tasks *rtp, struct rcu_tasks_percpu
> local_bh_disable();
> rhp->func(rhp);
> local_bh_enable();
> - cond_resched();

...and by local_bh_enable().

Thanx, Paul

> }
> raw_spin_lock_irqsave_rcu_node(rtpcp, flags);
> rcu_segcblist_add_len(&rtpcp->cblist, -len);
> @@ -974,10 +973,8 @@ static void check_all_holdout_tasks(struct list_head *hop,
> {
> struct task_struct *t, *t1;
>
> - list_for_each_entry_safe(t, t1, hop, rcu_tasks_holdout_list) {
> + list_for_each_entry_safe(t, t1, hop, rcu_tasks_holdout_list)
> check_holdout_task(t, needreport, firstreport);
> - cond_resched();
> - }
> }
>
> /* Finish off the Tasks-RCU grace period. */
> --
> 2.31.1
>