Re: [tip:sched/core] sched/core: Avoid _cond_resched() for PREEMPT=y

Re: [tip:sched/core] sched/core: Avoid _cond_resched() for PREEMPT=y

[Mikulas Patocka]

I've found a document that says that cond_resched() is needed on
preemptible kernels to mark RCU quiescent states:

https://lwn.net/Articles/603252/

Paul, is cond_resched() still needed on current RCU implementation? Or not?

Mikulas



On Thu, 22 Sep 2016, tip-bot for Peter Zijlstra wrote:

> Commit-ID:  35a773a07926a22bf19d77ee00024522279c4e68
> Gitweb:     http://git.kernel.org/tip/35a773a07926a22bf19d77ee00024522279c4e68
> Author:     Peter Zijlstra <peterz@infradead.org>
> AuthorDate: Mon, 19 Sep 2016 12:57:53 +0200
> Committer:  Ingo Molnar <mingo@kernel.org>
> CommitDate: Thu, 22 Sep 2016 14:53:46 +0200
> 
> sched/core: Avoid _cond_resched() for PREEMPT=y
> 
> On fully preemptible kernels _cond_resched() is pointless, so avoid
> emitting any code for it.
> 
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> Cc: Linus Torvalds <torvalds@linux-foundation.org>
> Cc: Mikulas Patocka <mpatocka@redhat.com>
> Cc: Oleg Nesterov <oleg@redhat.com>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Cc: linux-kernel@vger.kernel.org
> Signed-off-by: Ingo Molnar <mingo@kernel.org>
> ---
>  include/linux/sched.h | 4 ++++
>  kernel/sched/core.c   | 2 ++
>  2 files changed, 6 insertions(+)
> 
> diff --git a/include/linux/sched.h b/include/linux/sched.h
> index f00ee8e..b99fcd1 100644
> --- a/include/linux/sched.h
> +++ b/include/linux/sched.h
> @@ -3209,7 +3209,11 @@ static inline int signal_pending_state(long state, struct task_struct *p)
>   * cond_resched_lock() will drop the spinlock before scheduling,
>   * cond_resched_softirq() will enable bhs before scheduling.
>   */
> +#ifndef CONFIG_PREEMPT
>  extern int _cond_resched(void);
> +#else
> +static inline int _cond_resched(void) { return 0; }
> +#endif
>  
>  #define cond_resched() ({			\
>  	___might_sleep(__FILE__, __LINE__, 0);	\
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index b2ec53c..d7babcc 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -4883,6 +4883,7 @@ SYSCALL_DEFINE0(sched_yield)
>  	return 0;
>  }
>  
> +#ifndef CONFIG_PREEMPT
>  int __sched _cond_resched(void)
>  {
>  	if (should_resched(0)) {
> @@ -4892,6 +4893,7 @@ int __sched _cond_resched(void)
>  	return 0;
>  }
>  EXPORT_SYMBOL(_cond_resched);
> +#endif
>  
>  /*
>   * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
> 

Re: [tip:sched/core] sched/core: Avoid _cond_resched() for PREEMPT=y

[Paul E. McKenney]

On Thu, Sep 22, 2016 at 02:25:01PM -0400, Mikulas Patocka wrote:
> I've found a document that says that cond_resched() is needed on
> preemptible kernels to mark RCU quiescent states:
> 
> https://lwn.net/Articles/603252/
> 
> Paul, is cond_resched() still needed on current RCU implementation? Or not?

It can be if you loop in the kernel for a very long time.

But cond_resched_rcu_qs() is instead needed in cases where there is only
one runnable non-idle non-nohz task on a given CPU.

							Thanx, Paul

> Mikulas
> 
> 
> 
> On Thu, 22 Sep 2016, tip-bot for Peter Zijlstra wrote:
> 
> > Commit-ID:  35a773a07926a22bf19d77ee00024522279c4e68
> > Gitweb:     http://git.kernel.org/tip/35a773a07926a22bf19d77ee00024522279c4e68
> > Author:     Peter Zijlstra <peterz@infradead.org>
> > AuthorDate: Mon, 19 Sep 2016 12:57:53 +0200
> > Committer:  Ingo Molnar <mingo@kernel.org>
> > CommitDate: Thu, 22 Sep 2016 14:53:46 +0200
> > 
> > sched/core: Avoid _cond_resched() for PREEMPT=y
> > 
> > On fully preemptible kernels _cond_resched() is pointless, so avoid
> > emitting any code for it.
> > 
> > Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> > Cc: Linus Torvalds <torvalds@linux-foundation.org>
> > Cc: Mikulas Patocka <mpatocka@redhat.com>
> > Cc: Oleg Nesterov <oleg@redhat.com>
> > Cc: Peter Zijlstra <peterz@infradead.org>
> > Cc: Thomas Gleixner <tglx@linutronix.de>
> > Cc: linux-kernel@vger.kernel.org
> > Signed-off-by: Ingo Molnar <mingo@kernel.org>
> > ---
> >  include/linux/sched.h | 4 ++++
> >  kernel/sched/core.c   | 2 ++
> >  2 files changed, 6 insertions(+)
> > 
> > diff --git a/include/linux/sched.h b/include/linux/sched.h
> > index f00ee8e..b99fcd1 100644
> > --- a/include/linux/sched.h
> > +++ b/include/linux/sched.h
> > @@ -3209,7 +3209,11 @@ static inline int signal_pending_state(long state, struct task_struct *p)
> >   * cond_resched_lock() will drop the spinlock before scheduling,
> >   * cond_resched_softirq() will enable bhs before scheduling.
> >   */
> > +#ifndef CONFIG_PREEMPT
> >  extern int _cond_resched(void);
> > +#else
> > +static inline int _cond_resched(void) { return 0; }
> > +#endif
> >  
> >  #define cond_resched() ({			\
> >  	___might_sleep(__FILE__, __LINE__, 0);	\
> > diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> > index b2ec53c..d7babcc 100644
> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> > @@ -4883,6 +4883,7 @@ SYSCALL_DEFINE0(sched_yield)
> >  	return 0;
> >  }
> >  
> > +#ifndef CONFIG_PREEMPT
> >  int __sched _cond_resched(void)
> >  {
> >  	if (should_resched(0)) {
> > @@ -4892,6 +4893,7 @@ int __sched _cond_resched(void)
> >  	return 0;
> >  }
> >  EXPORT_SYMBOL(_cond_resched);
> > +#endif
> >  
> >  /*
> >   * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
> > 
> 

Re: [tip:sched/core] sched/core: Avoid _cond_resched() for PREEMPT=y

[Peter Zijlstra]

On Thu, Sep 22, 2016 at 02:25:01PM -0400, Mikulas Patocka wrote:
> I've found a document that says that cond_resched() is needed on
> preemptible kernels to mark RCU quiescent states:
> 
> https://lwn.net/Articles/603252/
> 
> Paul, is cond_resched() still needed on current RCU implementation? Or not?

No, we ripped that out. See 4a81e8328d37 ("rcu: Reduce overhead of
cond_resched() checks for RCU")

Re: [tip:sched/core] sched/core: Avoid _cond_resched() for PREEMPT=y

[Paul E. McKenney]

On Thu, Sep 22, 2016 at 08:42:28PM +0200, Peter Zijlstra wrote:
> On Thu, Sep 22, 2016 at 02:25:01PM -0400, Mikulas Patocka wrote:
> > I've found a document that says that cond_resched() is needed on
> > preemptible kernels to mark RCU quiescent states:
> > 
> > https://lwn.net/Articles/603252/
> > 
> > Paul, is cond_resched() still needed on current RCU implementation? Or not?
> 
> No, we ripped that out. See 4a81e8328d37 ("rcu: Reduce overhead of
> cond_resched() checks for RCU")

I would instead say that we changed the algorithm to reduce the overhead
in the common case, but yes, this commit did change things quite a bit.

							Thanx, Paul

Re: [tip:sched/core] sched/core: Avoid _cond_resched() for PREEMPT=y

[Peter Zijlstra]

On Thu, Sep 22, 2016 at 01:28:17PM -0700, Paul E. McKenney wrote:
> On Thu, Sep 22, 2016 at 08:42:28PM +0200, Peter Zijlstra wrote:
> > On Thu, Sep 22, 2016 at 02:25:01PM -0400, Mikulas Patocka wrote:
> > > I've found a document that says that cond_resched() is needed on
> > > preemptible kernels to mark RCU quiescent states:
> > > 
> > > https://lwn.net/Articles/603252/
> > > 
> > > Paul, is cond_resched() still needed on current RCU implementation? Or not?
> > 
> > No, we ripped that out. See 4a81e8328d37 ("rcu: Reduce overhead of
> > cond_resched() checks for RCU")
> 
> I would instead say that we changed the algorithm to reduce the overhead
> in the common case, but yes, this commit did change things quite a bit.

Sure, but critical point (for this discussion) is that cond_resched() no
longer implies anything RCU, it really only is a voluntary reschedule
point.

Re: [tip:sched/core] sched/core: Avoid _cond_resched() for PREEMPT=y

[Paul E. McKenney]

On Fri, Sep 23, 2016 at 10:44:42AM +0200, Peter Zijlstra wrote:
> On Thu, Sep 22, 2016 at 01:28:17PM -0700, Paul E. McKenney wrote:
> > On Thu, Sep 22, 2016 at 08:42:28PM +0200, Peter Zijlstra wrote:
> > > On Thu, Sep 22, 2016 at 02:25:01PM -0400, Mikulas Patocka wrote:
> > > > I've found a document that says that cond_resched() is needed on
> > > > preemptible kernels to mark RCU quiescent states:
> > > > 
> > > > https://lwn.net/Articles/603252/
> > > > 
> > > > Paul, is cond_resched() still needed on current RCU implementation? Or not?
> > > 
> > > No, we ripped that out. See 4a81e8328d37 ("rcu: Reduce overhead of
> > > cond_resched() checks for RCU")
> > 
> > I would instead say that we changed the algorithm to reduce the overhead
> > in the common case, but yes, this commit did change things quite a bit.
> 
> Sure, but critical point (for this discussion) is that cond_resched() no
> longer implies anything RCU, it really only is a voluntary reschedule
> point.

Agreed, cond_resched_rcu_qs() is what you use to do both a cond_resched()
and an RCU quiescent state.

							Thanx, Paul