On Fri, Mar 22, 2024 at 12:24:13PM +0100, Sebastian Andrzej Siewior wrote:
> On 2024-03-19 13:44:34 [-0700], Yan Zhai wrote:
> > index 16f519914415..17d7ed5f3ae6 100644
> > --- a/include/linux/rcupdate.h
> > +++ b/include/linux/rcupdate.h
> > @@ -247,6 +247,37 @@ do { \
> > cond_resched(); \
> > } while (0)
> >
> > +/**
> > + * rcu_softirq_qs_periodic - Report RCU and RCU-Tasks quiescent states
> > + * @old_ts: jiffies at start of processing.
> > + *
> > + * This helper is for long-running softirq handlers, such as NAPI threads in
> > + * networking. The caller should initialize the variable passed in as @old_ts
> > + * at the beginning of the softirq handler. When invoked frequently, this macro
> > + * will invoke rcu_softirq_qs() every 100 milliseconds thereafter, which will
> > + * provide both RCU and RCU-Tasks quiescent states. Note that this macro
> > + * modifies its old_ts argument.
> > + *
> > + * Because regions of code that have disabled softirq act as RCU read-side
> > + * critical sections, this macro should be invoked with softirq (and
> > + * preemption) enabled.
> > + *
> > + * The macro is not needed when CONFIG_PREEMPT_RT is defined. RT kernels would
> > + * have more chance to invoke schedule() calls and provide necessary quiescent
> > + * states. As a contrast, calling cond_resched() only won't achieve the same
> > + * effect because cond_resched() does not provide RCU-Tasks quiescent states.
> > + */
>
> Paul, so CONFIG_PREEMPTION is affected but CONFIG_PREEMPT_RT is not.
> Why does RT have more scheduling points?
In RT, isn't BH-disabled code preemptible? But yes, this would not help
RCU Tasks.
> The RCU-Tasks thread is starving and yet there is no wake-up, correct?
> Shouldn't cond_resched() take care of RCU-Tasks's needs, too?
> This function is used by napi_threaded_poll() which is not invoked in
> softirq it is a simple thread which does disable BH but this is it. Any
> pending softirqs are served before the cond_resched().
>
> This napi_threaded_poll() case _basically_ a busy thread doing a lot of
> work and delaying RCU-Tasks as far as I understand. The same may happen
> to other busy-worker which have cond_resched() between works, such as
> the kworker. Therefore I would expect to have some kind of timeout at
> which point NEED_RESCHED is set so that cond_resched() can do its work.
A NEED_RESCHED with a cond_resched() would still be counted as a
preemption. If we were intending to keep cond_resched(), I would
be thinking in terms of changing that, but only for Tasks RCU.
Given no cond_resched(), I would be thinking in terms of removing
the check for CONFIG_PREEMPT_RT.
Thoughts?
Thanx, Paul
> > +#define rcu_softirq_qs_periodic(old_ts) \
> > +do { \
> > + if (!IS_ENABLED(CONFIG_PREEMPT_RT) && \
> > + time_after(jiffies, (old_ts) + HZ / 10)) { \
> > + preempt_disable(); \
> > + rcu_softirq_qs(); \
> > + preempt_enable(); \
> > + (old_ts) = jiffies; \
> > + } \
> > +} while (0)
> > +
> > /*
> > * Infrastructure to implement the synchronize_() primitives in
> > * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
>
> Sebastian
