Re: [PATCH] Fix: disable sys_membarrier when nohz_full is enabled

[Paul E. McKenney]

On Tue, Nov 08, 2016 at 11:15:51AM +0000, Mathieu Desnoyers wrote:
> ----- On Nov 7, 2016, at 3:03 PM, Paul E. McKenney paulmck@xxxxxxxxxxxxxxxxxx wrote:
> 
> > On Mon, Nov 07, 2016 at 06:10:14PM +0000, Mathieu Desnoyers wrote:
> >> 
> >> ----- On Nov 7, 2016, at 1:06 PM, Paul E. McKenney paulmck@xxxxxxxxxxxxxxxxxx
> >> wrote:
> >> 
> >> > On Mon, Nov 07, 2016 at 05:08:59PM +0000, Mathieu Desnoyers wrote:
> >> >> ----- On Nov 3, 2016, at 1:49 PM, Paul E. McKenney paulmck@xxxxxxxxxxxxxxxxxx
> >> >> wrote:
> >> >> 
> >> >> > On Thu, Nov 03, 2016 at 10:29:28AM -0600, Mathieu Desnoyers wrote:
> >> >> >> Userspace applications should be allowed to expect the membarrier system
> >> >> >> call with MEMBARRIER_CMD_SHARED command to issue memory barriers on
> >> >> >> nohz_full CPUs, but synchronize_sched() does not take those into
> >> >> >> account.
> >> >> >> 
> >> >> >> Given that we do not want unrelated processes to be able to affect
> >> >> >> real-time sensitive nohz_full CPUs, simply return ENOSYS when membarrier
> >> >> >> is invoked on a kernel with enabled nohz_full CPUs.
> >> >> >> 
> >> >> >> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@xxxxxxxxxxxx>
> >> >> >> CC: "Paul E. McKenney" <paulmck@xxxxxxxxxxxxxxxxxx>
> >> >> > 
> >> >> > Acked-by: "Paul E. McKenney" <paulmck@xxxxxxxxxxxxxxxxxx>
> >> >> 
> >> >> Hi Paul,
> >> >> 
> >> >> Do you plan to pick it up through your tree, or I should sent
> >> >> it directly to Linus ?
> >> > 
> >> > Your choice.  I believe that the original went some other way, but I
> >> > would be fine carrying this one.
> >> 
> >> Not sure what you mean by "the original" ? And which other way ?
> >> I have not been notified about this.
> > 
> > If I remember correctly, you sent the original sys_membarrier()
> > patch through akpm or similar.
> 
> Ah right, the original implementation, yes.
> 
> > 
> >> If you can carry this patch it would be very much appreciated,
> > 
> > Will do!
> 
> Especially since the regression is somewhat related to RCU
> behavior wrt nohz_full, getting it through your tree seems
> relevant.

I have it queued for 4.11, thank you!

							Thanx, Paul