Re: Xen 4.2 Release Plan / TODO

[Ian Campbell <Ian.Campbell@citrix.com>]

On Tue, 2012-06-26 at 23:57 +0100, Matt Wilson wrote:
> On Tue, Jun 26, 2012 at 02:09:06PM -0700, Konrad Rzeszutek Wilk wrote:
> > On Tue, Jun 26, 2012 at 01:31:51PM -0700, Matt Wilson wrote:
> > > On Tue, Jun 26, 2012 at 01:39:42AM -0700, Ian Campbell wrote:
> > > >     * [BUG] vcpu-set doesn't take effect on guest, reported by Intel:
> > > >       http://bugzilla.xen.org/bugzilla/show_bug.cgi?id=1822
> > > 
> > > This may be a problem with the guest, rather than a problem with Xen
> > > or the toolstack. I tested on a domU running a 3.2.20-based Linux
> > > kernel. The domU kernel didn't automatically enable the vCPUs after
> > > calling "xl vcpu-set" to increase the allocation.
> > > [...]
> > > 
> > > I've added this as a comment to the bugzilla.
> > 
> > That is a generic bug - its a race in the generic code where
> > the cpuX directory is created; then udev kicks off and tries to
> > echo 1 > cpuX/online (but online hasn't been created yet); the cpu-hotplug
> > code creates 'online' directory.
> > 
> > I've asked Greg KH about it, and here is his take:
> > https://lkml.org/lkml/2012/4/30/198
> > 
> > But I never got to prep a patch. If somebody gets to do it before me,
> > I owe them a beer!
> 
> My problem was that I didn't even have a udev rule to auto-online
> hotplugged CPUs. Everything works as expected after adding the rule
> suggested on the wiki: http://wiki.xen.org/wiki/Paravirt_Linux_CPU_Hotplug
> 
> I'm thinking that the bug report is just this well hashed out
> problem. For historical reference, here's the thread discussing the
> behavior change in pv-ops kernels:
>   http://lists.xen.org/archives/html/xen-devel/2010-05/msg00516.html

I've added a link to that thread to the wiki page.