Re: Warning in during hotplug on 2.6.27-rc2-git5

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Sunday, 17 of August 2008, Rafael J. Wysocki wrote:
> On Sunday, 17 of August 2008, Andi Kleen wrote:
> > > > > I'm still seeing it on 2.6.27-rc2, even with the 
> > > > > patch here http://lkml.org/lkml/2008/7/30/171 and the
> > > > > wbinvd_halt code patch applied.  Maybe something else
> > > > > broke in some of the recent hotplug changes?
> > > > 
> > > > My guess is that MCE does somthing that is not allowed by sysfs any more.
> > > 
> > > Hm, sysfs hasn't changed any in 2.6.27-rcX that I know of.
> > 
> > mce hasn't either in this regard. My current theory is that the CPU 
> > up/down notifiers are not balanced anymore (as in duplicated up events) 
> 
> It doesn't look like this is the case.  Moreover, had that been the case, we'd
> have had many reports from people doing suspend/hibernation, but it doesn't
> happen.
> 
> I think that cpu_down() fails for some reason and that causes the subsequent
> onlining to fail.

Well, no.  If my understanding of the CPU hotplug code is correct, this is not
possible.

The next possibility is that for some 'i' mce_attributes[i] is NULL, although
there are non-NULL values for some j > i.  In that case, mce_remove_device()
would fail to remove device_mce for given CPU and the subsequent
mce_create_device() would cause the observed failure.

Thanks,
Rafael