Re: [ANNOUNCE] 3.0-rt4

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Thu, Jul 28, 2011 at 12:34:22PM -0700, Paul E. McKenney wrote:
> On Thu, Jul 28, 2011 at 09:05:14PM +0200, Thomas Gleixner wrote:
> > On Thu, 28 Jul 2011, Paul E. McKenney wrote:
> > 
> > > On Thu, Jul 28, 2011 at 07:43:43PM +0200, Thomas Gleixner wrote:
> > > > 
> > > > On Thu, 28 Jul 2011, Paul E. McKenney wrote:
> > > > 
> > > > > On Wed, Jul 27, 2011 at 11:37:09PM +0200, Thomas Gleixner wrote:
> > > > > > Dear RT Folks,
> > > > > > 
> > > > > > I'm pleased to announce the 3.0-rt4 release.
> > > > > 
> > > > > 3.0-rt5 boots single-CPU under KVM on a 32-bit laptop.  Also with "-smp
> > > > > 2".  It even survived a minute or so of rcutorture in both cases.  ;-)
> > > > > 
> > > > > Some weirdness with interactivity -- if the system is idle aside from
> > > > > rcutorture, the rcutorture.stat_interval console messages don't come
> > > > > out unless I hit "enter" on the console.  But if I run "top", they do
> > > > > come out regularly.  Might well be KVM's fault, except that running
> > > > > rcutorture on non-rt kernels doesn't exhibit this behavior.
> > > > 
> > > > Nah, that's my lazy "avoid printing from atomic context" hack :)
> > > 
> > > Well, then, that does seem to be working...  ;-)
> > > 
> > > Except that these printk()s are from process level -- rcu_torture_stats()
> > > in kernel/rcutorture.c.
> > > 
> > > > > I guess I should try RCU_BOOST=y, then...
> > > > 
> > > > Good luck :)
> > > 
> > > Compiled without error and has been running rcutorture smp=2 for
> > > about 30 minutes now.
> > 
> > Hmm, perhaps it was some weird interaction with NOHZ which is a
> > separate issue.
> 
> OK, I guess I have no alternative but to try enabling NOHZ...

And the combination of RCU_BOOST and NO_HZ runs fine as well.  30 minutes
of rcutorture and no complaints.

I only enabled CONFIG_PREEMPT_RT_FULL.  Are there other related config
variables I should be setting?

							Thanx, Paul