On Fri, Nov 25, 2016 at 12:34:26AM -0800, Paul E. McKenney wrote:
> On Tue, Nov 22, 2016 at 05:12:20PM +0900, Byungchul Park wrote:
> > On Wed, Nov 09, 2016 at 03:32:15PM +0900, Byungchul Park wrote:
> > > Currently rcu code forces CPU into scheduler when jiffies >=
> > > rcu_state.gp_start + jiffies_till_sched_qs, via resched_cpu().
> > > 
> > > It would be better to force CPU into scheduler when jiffies >=
> > > rcu_state.jiffies_resched, too.
> > 
> > Hello,
> > 
> > I think these two patches are necessary to call resched_cpu() even in
> > case of jiffies >= rcu_state.jiffies_resched, too. Am I wrong?
> > 
> > It would be appriciated if you let me know if I was wrong.
> 
> My current thought is that both the "if" statement and the call to
> resched_cpu() should be removed, but I am still testing and working
> through the timing.
> 
> Either way, I do very much appreciate your having called my attention
> to this code!

And the patch shown below seems to do the trick.  Thoughts?

                                                        Thanx, Paul

------------------------------------------------------------------------

commit d2db185bfee894c573faebed93461e9938bdbb61
Author: Paul E. McKenney <paul...@linux.vnet.ibm.com>
Date:   Fri Nov 25 00:07:23 2016 -0800

    rcu: Remove short-term CPU kicking
    
    Commit 4914950aaa12d ("rcu: Stop treating in-kernel CPU-bound workloads
    as errors") added a (relatively) short-timeout call to resched_cpu().
    This was inspired by as issue that was fixed by b7e7ade34e61 ("sched/core:
    Fix remote wakeups").  But given that this issue was fixed, it is time
    for the current commit to remove this call to resched_cpu().
    
    Reported-by: Byungchul Park <byungchul.p...@lge.com>
    Signed-off-by: Paul E. McKenney <paul...@linux.vnet.ibm.com>

diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index 5f4f80d07852..91a68e4e6671 100644
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -1345,12 +1345,6 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp,
                rdp->rsp->jiffies_resched += 5; /* Re-enable beating. */
        }
 
-       /* And if it has been a really long time, kick the CPU as well. */
-       if (ULONG_CMP_GE(jiffies,
-                        rdp->rsp->gp_start + 2 * jiffies_till_sched_qs) ||
-           ULONG_CMP_GE(jiffies, rdp->rsp->gp_start + jiffies_till_sched_qs))
-               resched_cpu(rdp->cpu);  /* Force CPU into scheduler. */
-
        return 0;
 }
 

Reply via email to