Re: [PATCH 2/2] rcu: Keep gpnum and completed fields synchronized

[Frederic Weisbecker <fweisbec@gmail.com>]

On Fri, Dec 10, 2010 at 04:48:43PM -0800, Paul E. McKenney wrote:
> On Sat, Dec 11, 2010 at 01:00:39AM +0100, Frederic Weisbecker wrote:
> > On Fri, Dec 10, 2010 at 03:02:00PM -0800, Paul E. McKenney wrote:
> > > On Fri, Dec 10, 2010 at 10:11:11PM +0100, Frederic Weisbecker wrote:
> > > > When a CPU that was in an extended quiescent state wakes
> > > > up and catches up with grace periods that remote CPUs
> > > > completed on its behalf, we update the completed field
> > > > but not the gpnum that keeps a stale value of a backward
> > > > grace period ID.
> > > > 
> > > > Later, note_new_gpnum() will interpret the shift between
> > > > the local CPU and the node grace period ID as some new grace
> > > > period to handle and will then start to hunt quiescent state.
> > > > 
> > > > But if every grace periods have already been completed, this
> > > > interpretation becomes broken. And we'll be stuck in clusters
> > > > of spurious softirqs because rcu_report_qs_rdp() will make
> > > > this broken state run into infinite loop.
> > > > 
> > > > The solution, as suggested by Lai Jiangshan, is to ensure that
> > > > the gpnum and completed fields are well synchronized when we catch
> > > > up with completed grace periods on their behalf by other cpus.
> > > > This way we won't start noting spurious new grace periods.
> > > 
> > > Also good, queued!
> > > 
> > > One issue -- this approach is vulnerable to overflow.  I therefore
> > > followed up with a patch that changes the condition to
> > > 
> > > 	if (ULONG_CMP_LT(rdp->gpnum, rdp->completed))
> > > 
> > > And I clearly need to make RCU defend itself against the scenario where
> > > a CPU stays in dyntick-idle mode long enough for the grace-period number
> > > to wrap halfway around its range of possible values.  Not a problem at
> > > the moment, and never will be for 64-bit systems, but...
> > > 
> > > I will fix that up.
> > 
> > Oh you're right of course. I did not think about possible overflows.
> > 
> > Now looking at ULONG_CMP_LT() definition, if it wraps more than halfways
> > we are screwed anyway. I suspect it won't ever happen, but it can. Perhaps
> > we need some watchguard code in note_new_gpnum() to fixup that corner case.
> 
> We still have to guard against a full wrap, though there are lots of other
> things that break if you stay in dyntick-idle mode that long.  My plan is
> to have a counter in the rcu_state structure that cycles through the CPUs.
> Check the current CPU at the start of each grace period.  If a given
> CPU is more than one-quarter of the way behind, force it to wake up
> and catch up.  This gets easier to do given my in-progress changes to
> convert from softirq to kthread, so I will combine it with those changes.
> 
> So I will cover this.

Fine!

Thanks.

> 
> 							Thanx, Paul
> 
> > > > Suggested-by: Lai Jiangshan <laijs@cn.fujitsu.com>
> > > > Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
> > > > Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> > > > Cc: Ingo Molnar <mingo@elte.hu>
> > > > Cc: Thomas Gleixner <tglx@linutronix.de>
> > > > Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
> > > > Cc: Steven Rostedt <rostedt@goodmis.org
> > > > ---
> > > >  kernel/rcutree.c |    9 +++++++++
> > > >  1 files changed, 9 insertions(+), 0 deletions(-)
> > > > 
> > > > diff --git a/kernel/rcutree.c b/kernel/rcutree.c
> > > > index 8c4ed60..2e16da3 100644
> > > > --- a/kernel/rcutree.c
> > > > +++ b/kernel/rcutree.c
> > > > @@ -683,6 +683,15 @@ __rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_dat
> > > >  		rdp->completed = rnp->completed;
> > > > 
> > > >  		/*
> > > > +		 * If we were in an extended quiescent state, we may have
> > > > +		 * missed some grace periods that others CPUs took care on
> > > > +		 * our behalf. Catch up with this state to avoid noting
> > > > +		 * spurious new grace periods.
> > > > +		 */
> > > > +		if (rdp->completed > rdp->gpnum)
> > > > +			rdp->gpnum = rdp->completed;
> > > > +
> > > > +		/*
> > > >  		 * If another CPU handled our extended quiescent states and
> > > >  		 * we have no more grace period to complete yet, then stop
> > > >  		 * chasing quiescent states.
> > > > -- 
> > > > 1.7.3.2
> > > > 
> > --
> > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> > the body of a message to majordomo@vger.kernel.org
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> > Please read the FAQ at  http://www.tux.org/lkml/