lkml.org 
[lkml]   [2010]   [Jun]   [29]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC][PATCH 09/11] perf: Default PMU ops
On Tue, Jun 29, 2010 at 04:59:51PM +0200, Peter Zijlstra wrote:
> On Tue, 2010-06-29 at 16:58 +0200, Frederic Weisbecker wrote:
> > On Thu, Jun 24, 2010 at 04:28:13PM +0200, Peter Zijlstra wrote:
> > > Provide default implementations for the pmu txn methods, this allows
> > > us to remove some conditional code.
> > >
> > > Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
> > > ---
> > > kernel/perf_event.c | 48 ++++++++++++++++++++++++++++++++++++------------
> > > 1 file changed, 36 insertions(+), 12 deletions(-)
> > >
> > > Index: linux-2.6/kernel/perf_event.c
> > > ===================================================================
> > > --- linux-2.6.orig/kernel/perf_event.c
> > > +++ linux-2.6/kernel/perf_event.c
> > > @@ -656,21 +656,14 @@ group_sched_in(struct perf_event *group_
> > > {
> > > struct perf_event *event, *partial_group = NULL;
> > > struct pmu *pmu = group_event->pmu;
> > > - bool txn = false;
> > >
> > > if (group_event->state == PERF_EVENT_STATE_OFF)
> > > return 0;
> > >
> > > - /* Check if group transaction availabe */
> > > - if (pmu->start_txn)
> > > - txn = true;
> > > -
> > > - if (txn)
> > > - pmu->start_txn(pmu);
> > > + pmu->start_txn(pmu);
> > >
> > > if (event_sched_in(group_event, cpuctx, ctx)) {
> > > - if (txn)
> > > - pmu->cancel_txn(pmu);
> > > + pmu->cancel_txn(pmu);
> > > return -EAGAIN;
> > > }
> > >
> > > @@ -684,7 +677,7 @@ group_sched_in(struct perf_event *group_
> > > }
> > > }
> > >
> > > - if (!txn || !pmu->commit_txn(pmu))
> > > + if (!pmu->commit_txn(pmu))
> > > return 0;
> > >
> > > group_error:
> > > @@ -699,8 +692,7 @@ group_error:
> > > }
> > > event_sched_out(group_event, cpuctx, ctx);
> > >
> > > - if (txn)
> > > - pmu->cancel_txn(pmu);
> > > + pmu->cancel_txn(pmu);
> > >
> > > return -EAGAIN;
> > > }
> > > @@ -4755,6 +4747,26 @@ static struct list_head pmus;
> > > static DEFINE_MUTEX(pmus_lock);
> > > static struct srcu_struct pmus_srcu;
> > >
> > > +static void perf_pmu_nop(struct pmu *pmu)
> > > +{
> > > +}
> > > +
> > > +static void perf_pmu_start_txn(struct pmu *pmu)
> > > +{
> > > + perf_pmu_disable(pmu);
> > > +}
> > > +
> > > +static int perf_pmu_commit_txn(struct pmu *pmu)
> > > +{
> > > + perf_pmu_enable(pmu);
> > > + return 0;
> > > +}
> > > +
> > > +static void perf_pmu_cancel_txn(struct pmu *pmu)
> > > +{
> > > + perf_pmu_enable(pmu);
> > > +}
> >
> >
> > So why do you need perf_pmu_*able wrappers now that you brings stubs
> > if none is provided?
> >
> > Actually, one problem is that it makes calling two indirect nops
> > for software events.
> >
> > Should the txn things really map to the enable/disable ops is the
> > off-case? Probably better let pmu implementations deal with that.
> > If they didn't provide txn implementations, it means they don't need it,
> > hence it should directly map to a nop.
> >
>
> You mean, if (!pmu->start_txn && pmu->pmu_enable) { /* install defaults
> */ } ?


Not really. pmu_*able and txn are there for different purposes.
A pmu implementation may want to provide enable/disable things
but not require any txn. Or one may just not need any of those,
like software events.

It should simply map to a nop if nothing is provided.



\
 
 \ /
  Last update: 2010-06-29 17:05    [from the cache]
©2003-2014 Jasper Spaans. Advertise on this site