Re: WARN... Device 'cpu1' does not have a release() function, it is broken and must be fixed. when doing 'xl vcpu-set <guest_id> 1'

[Greg KH <gregkh@suse.de>]

On Fri, Jan 27, 2012 at 01:22:46AM +0100, Kay Sievers wrote:
> On Fri, Jan 27, 2012 at 01:06, Greg KH <gregkh@suse.de> wrote:
> > On Mon, Jan 23, 2012 at 01:06:01PM -0500, Konrad Rzeszutek Wilk wrote:
> 
> >> Is anybody else hitting this with ACPI CPU hot-unplug? Or do I have
> >> the privilige of being the first? Oh, I hadn't done a full bisection
> >> but v3.2 does not have this.
> >
> > Kay, this is a mess.
> >
> > This cpu system device is is interconnected with the different arches
> > and their cpu-specific structures.  Some arches have a static array,
> > some allocate a huge structure (struct arch_cpu * NUM_CPUS), and others
> > try to do the right thing with DECLARE_PER_CPU() but don't quite get it
> > right, making that a static array per cpu.
> >
> > To unwind all of this, is much beyond 3.3 material, as I'm sure I'll get
> > it wrong, and have a bunch of non-x86-64 build problems along the way.
> >
> > Any objection to me just doing the "hack" of the empty release function
> > at the moment to get rid of this warning, and then clean it all up
> > properly for 3.4?
> 
> No problem at all.
> 
> It would be nice if we get all that to the usual model some day, but I
> can totally see that CPU devices try to deal with statically allocated
> per-cpu memory. It seems fine, as long as they know what they are
> doing.
> 
> Just silencing the driver-core warning here sounds fine to me.

Ok, I'll do that tomorrow and send a patch out and then start working on
making all of this dynamic, as it really should be.

thanks,

greg k-h