Re: [PATCH] schedule: use unlikely()

[Ingo Molnar <mingo@kernel.org>]

* Mikulas Patocka <mpatocka@redhat.com> wrote:

> 
> 
> On Fri, 24 Nov 2017, Ingo Molnar wrote:
> 
> > * Mikulas Patocka <mpatocka@redhat.com> wrote:
> > 
> > > A small patch for schedule(), so that the code goes straght in the common
> > > case.
> > > 
> > > Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
> > > 
> > > ---
> > >  include/linux/blkdev.h |    2 +-
> > >  kernel/sched/core.c    |    2 +-
> > >  2 files changed, 2 insertions(+), 2 deletions(-)
> > > 
> > > Index: linux-2.6/include/linux/blkdev.h
> > > ===================================================================
> > > --- linux-2.6.orig/include/linux/blkdev.h
> > > +++ linux-2.6/include/linux/blkdev.h
> > > @@ -1308,7 +1308,7 @@ static inline bool blk_needs_flush_plug(
> > >  {
> > >  	struct blk_plug *plug = tsk->plug;
> > >  
> > > -	return plug &&
> > > +	return unlikely(plug != NULL) &&
> > >  		(!list_empty(&plug->list) ||
> > >  		 !list_empty(&plug->mq_list) ||
> > >  		 !list_empty(&plug->cb_list));
> > 
> > That's an unrelated change.
> 
> It is related, because this function gets inlined into schedule().

That needs to be in the changelog and the patch needs to be Cc:-ed to the 
affected maintainer as well.

> > > Index: linux-2.6/kernel/sched/core.c
> > > ===================================================================
> > > --- linux-2.6.orig/kernel/sched/core.c
> > > +++ linux-2.6/kernel/sched/core.c
> > > @@ -3405,7 +3405,7 @@ void __noreturn do_task_dead(void)
> > >  
> > >  static inline void sched_submit_work(struct task_struct *tsk)
> > >  {
> > > -	if (!tsk->state || tsk_is_pi_blocked(tsk))
> > > +	if (!tsk->state || unlikely(tsk_is_pi_blocked(tsk)))
> > >  		return;
> > >  	/*
> > >  	 * If we are going to sleep and we have plugged IO queued,
> > 
> > Do these changes actually change the generated assembly code?
> > 
> > Thanks,
> > 
> > 	Ingo
> 
> Yes. It saves two jumps because the compiler falsely assumes that 
> tsk_is_pi_blocked would return true. Likewise, the compiler falsely 
> assumes that tsk->plug would be true.
> 
> The static branch prediction in gcc assumes that pointers are usually not 
> NULL, but in this case tsk->pi_blocked_on and tsk->plug are usually NULL.

That too should be in the changelog.

Thanks,

	Ingo