Messages in this thread | | | Date | Thu, 25 Aug 2011 01:34:24 +0200 | From | Frederic Weisbecker <> | Subject | Re: 3.0-git15 Atomic scheduling in pidmap_init |
| |
On Wed, Aug 24, 2011 at 04:12:55PM -0700, Paul E. McKenney wrote: > Indeed, this was a bogus patch. The version I posted on the -rcu git > tree a few days ago has the correct "rdp->qs_pending = 0". I thought > that I had chased down all the bogus copies, but obviously not. :-(
Ah ok.
> > Should it perhaps set it to 1 only if we have rnp->gpnum > rnp->completed ? > > I would rather keep it simple. If rnp->gpnum > rnp->completed, then > the newly onlined CPU will notice and adjust appropriately soon enough.
Ah, just to ensure I understood well is that because rcu_pending() (called from tick) would detect rnp->gpnum != rdp->gpnum and thus trigger the softirq that notes the new gpnum and toggles qs_pending?
> Thanx, Paul > > > > rdp->passed_quiesc_completed = rnp->completed - 1; > > > } > > > raw_spin_unlock(&rnp->lock); /* irqs already disabled. */
| |