Re: [RFC] Re-introduce VCPUOP_register_vcpu_info back for PVHVM guests. (v1)

[George Dunlap <george.dunlap@eu.citrix.com>]

On 17/04/13 17:23, Jan Beulich wrote:
>>>> On 17.04.13 at 17:31, George Dunlap <George.Dunlap@eu.citrix.com> wrote:
>> On Tue, Apr 16, 2013 at 10:09 PM, Konrad Rzeszutek Wilk
>> <konrad.wilk@oracle.com> wrote:
>>> Hey Jan,
>>>
>>> While I've been digging around CPU hotplug for PVHVM I noticed an oddity
>>> when running with different versions of hypervisor. I found out that the
>>> big change you did in Xen 4.2 of splitting the PV and HVM arch structure
>>> introduce the regression wherein VCPUOP_register_vcpu_info will not
>>> work for PVHVM guests.
>>>
>>> Interestingly enough if I introduce this back to the hypervisor the
>>> CPU hotplug path gets even more broken for PVHVM :-).. But that is a Linux
>>> problem.
>> Except that we really don't want to be trashing people's existing,
>> working versions of Linux, especially if they're distro-provided
>> kernels.  I haven't tested a distro kernel yet, but for my own
>> locally-built 2.6.37 kernel, this c/s breaks it running in PVHVM mode.
> Hmm - that's minimally a reason to not backport it to 4.2, but
> perhaps even a reason to revert it from staging.

Hold off on reverting it -- I thought I had clearly shown it to be the 
c/s that breaks things, but now it's not so clear... really frustrating 
dealing w/ the build system right now...

  -George