Re: [RFC][PATCH 2/9] perf: core, remove hw_perf_event_init

[Frederic Weisbecker <fweisbec@gmail.com>]

On Mon, May 10, 2010 at 02:54:14PM +0200, Peter Zijlstra wrote:
> On Mon, 2010-05-10 at 14:27 +0200, Frederic Weisbecker wrote:
> > On Mon, May 10, 2010 at 12:19:29PM +0200, Peter Zijlstra wrote:
> > > On Mon, 2010-05-10 at 18:17 +0800, Lin Ming wrote:
> > > > On Mon, 2010-05-10 at 17:40 +0800, Peter Zijlstra wrote:
> > > > > On Mon, 2010-05-10 at 17:26 +0800, Lin Ming wrote:
> > > > > > +static struct pmu *perf_event_lookup_pmu(struct perf_event *event)
> > > > > > +{
> > > > > > +       struct pmu *pmu;
> > > > > > +       int pmu_id = event->attr.pmu_id;
> > > > > > +
> > > > > > +       list_for_each_entry(pmu, &pmus, entry) {
> > > > > > +               if (pmu->id == pmu_id)
> > > > > > +                       return pmu;
> > > > > > +       }
> > > > > > +
> > > > > > +       return NULL;
> > > > > > +}
> > > > > 
> > > > > > +void perf_event_register_pmu(struct pmu *pmu)
> > > > > > +{
> > > > > > +       pmu->id = pmu_id_curr++;
> > > > > > +       list_add_tail(&pmu->entry, &pmus);
> > > > > > +} 
> > > > > 
> > > > > That will be wanting some sort of synchronization
> > > > 
> > > > Will add a mutex to protect the list of pmus.
> > > 
> > > I'm thinking RCU might be better suited, a mutex for lookup doesn't
> > > sound ideal.
> > 
> > 
> > Is it really needed? I expect this function to be called on boot
> > only.
> > 
> > In fact I would even suggest to tag it as __init.
> 
> Loadable modules as well as PCI-Hotplug need supporting.


Which module do you have in mind that could register a pmu?
And I don't understand the problem with pci-hotplug.

Thanks.