Re: [RFC] perf_events: support for uncore a.k.a. nest units

[Lin Ming <ming.m.lin@intel.com>]

On Wed, 2010-04-21 at 22:22 +0800, Peter Zijlstra wrote:
> On Wed, 2010-04-21 at 22:12 +0000, Lin Ming wrote:
> > +       ret = pmu->commit_txn(pmu);
> > +       if (!ret) {
> > +               pmu->stop_txn(pmu);
> > +               return 0;
> > +       }
> >  
> >  group_error:
> > +       pmu->stop_txn(pmu); 
> 
> If you let commit_txn() also clear the state you can save some logic and
> a method.

I add this ->stop_txn(pmu) because the rollback code also need to clear
the state.

> 
> But yes, this looks good. If you don't remove the weak interface just
> yet, you can do a patch per architecture that uses this (at least
> powerpc and sparc do), and remove the weak thing at the end once all
> users are gone.
> 

OK, I'll do that for powerpc and sparc.

And need to check if there are transaction methods in group_sched_in for
other arch, as below.

group_sched_in(...)
{
	if (pmu->start_txn)
		pmu->start_txn(pmu);

	if (pmu->commit_txn)
		pmu->commit_txn(pmu)
}

Thanks,
Lin Ming