On Sat, Nov 18, 2006 at 10:34:26PM +0300, Oleg Nesterov wrote:
> On 11/18, Paul E. McKenney wrote:
> >
> > On Sat, Nov 18, 2006 at 11:15:27AM -0500, Alan Stern wrote:
> > > > + smp_processor_id())->c[idx]++;
> > > > + smp_mb();
> > > > + preempt_enable();
> > > > + return idx;
> > > > + }
> > > > + if (mutex_trylock(&sp->mutex)) {
> > > > + preempt_enable();
> > >
> > > Move the preempt_enable() before the "if", then get rid of the
> > > preempt_enable() after the "if" block.
> >
> > No can do. The preempt_enable() must follow the increment and
> > the memory barrier, otherwise the synchronize_sched() inside
> > synchronize_srcu() can't do its job.
>
> Given that srcu_read_lock() does smp_mb() after ->c[idx]++, what
> is the purpose of synchronize_srcu() ? It seems to me it could be
> replaced by smp_mb().
>
> synchronize_srcu:
>
> sp->completed++;
>
> mb();
>
> // if the reader did any memory access _after_
> // srcu_read_lock()->mb() we must see the changes.
> while (srcu_readers_active_idx(sp, idx))
> sleep();
>
> No?
I believe that this could run afoul of the example I sent out earlier
(based on Alan's example). In my mind, the key difference between
this and Jens's suggestion is that in Jens's case, we check for -all-
the counters being zero, not just the old ones. (But I still don't
trust Jen's optimization -- I just have not yet come up with an example
showing breakage, possibly because there isn't one, but...)
Thanx, Paul
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
