On Tuesday, January 11, 2011, Thomas Renninger wrote:
> On Tuesday 11 January 2011 01:05:53 Rafael J. Wysocki wrote:
> > On Tuesday, January 11, 2011, Rafael J. Wysocki wrote:
> > > On Tuesday, January 11, 2011, Len Brown wrote:
> > > > > /**
> > > > > * cpuidle_enable_device - enables idle PM for a CPU
> > > > > * @dev: the CPU
> > > > > @@ -176,6 +215,8 @@ int cpuidle_enable_device(struct cpuidle
> > > > > ret = __cpuidle_register_device(dev);
> > > > > if (ret)
> > > > > return ret;
> > > > > + } else {
> > > > > + poll_idle_init(dev);
> > > > > }
> > > >
> > > > how about calling poll_idle_init() unconditionally here
> > > > and deleting the call to it from within __cpuidle_register_device()?
> > >
> > > Fine by me, as long as poll_idle_init() is called before the conditional. :-)
> >
> > In fact, it even doesn't need to be called before the conditional.
> >
> > So fine by me anyway.
> What exactly was broken?
> Is it only sysfs values?
Not only that, the entire state[0] was busted.
> Looks like an uninitialized "poll" state can cause cpuidle
> to not enter "poll" state when it should or enter "poll" when
> it should not.
> Hm, if cpuidle would try to call state[0]->enter,
> it might even segfault?
Yes, in theory.
> Even not that many machines might be affected because most won't
> implement runtime C-state changes, shouldn't this still be
> submitted for stable@ kernels?
I think it should.
Thanks,
Rafael
_______________________________________________
linux-pm mailing list
linux-pm@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linux-foundation.org/mailman/listinfo/linux-pm
