Re: [PATCH] perf_events: fix event scheduling issues introduced by transactional API (take 2)

[Peter Zijlstra <peterz@infradead.org>]

On Tue, 2010-05-25 at 18:10 +0200, Stephane Eranian wrote:
> > Index: linux-2.6/kernel/perf_event.c
> > ===================================================================
> > --- linux-2.6.orig/kernel/perf_event.c
> > +++ linux-2.6/kernel/perf_event.c
> > @@ -668,15 +668,9 @@ group_sched_in(struct perf_event *group_
> >                }
> >        }
> >
> > -       if (!txn)
> > +       if (!txn || !pmu->commit_txn(pmu))
> >                return 0;
> >
> > -       ret = pmu->commit_txn(pmu);
> > -       if (!ret) {
> > -               pmu->cancel_txn(pmu);
> > -               return 0;
> > -       }
> > -
> >  group_error:
> >        /*
> >         * Groups can be scheduled in as one unit only, so undo any
> >
> Looks okay.
> 
> I believe you can also drop the txn test in group_sched_in() after group_error:,
> given you have the if !(txn) return 0.

Can't we still get in the group_error: branch with either scenario?

!txn will get there if ->enable() of a group sibling fails,
txn will get there if either ->enable() or ->commit_txn() fails.

(Although typically ->enable() would not fail for txn)