[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] Xen 4.3 development update

>>> On 21.06.13 at 10:08, "Zhang, Yang Z" <yang.z.zhang@xxxxxxxxx> wrote:
> Jan Beulich wrote on 2013-06-20:
>>>>> On 17.06.13 at 13:13, "Jan Beulich" <JBeulich@xxxxxxxx> wrote:
>>>>>> On 17.06.13 at 12:58, George Dunlap <George.Dunlap@xxxxxxxxxxxxx>
>> wrote:
>>>> * Win2k8 fails install on many IvyBridge EP systems
>>>>> Not reproducible by Intel
>>>>> Reproducible on 4.2
>>> Reproducible only on 4.2 with APIC virtualization commits backported.
>>> So this is a regression due to new functionality (which to date cannot
>>> be worked around).
>> Update: This is a bad interaction between Viridian mode and APIC-V.
>> Disabling either makes things work again. Yang narrowed this down
>> to the EOI handling, and created a non-postable patch that deals
>> with the issue (non-postable because it does things in ways only
>> suitable for temporary testing). I do think, though, that this isn't the
>> way to go anyway - instead we likely will want to suppress Windows
>> using the APIC related MSRs when APIC register virtualization is in
>> use. Yang seems to agree with this, but this approach wasn't tested
> Just do a testing and it is working to not set CPUID3A_MSR_APIC_ACCESS and 

But after some more thought I think we ought to still use your
original patch (suitably converted to proper shape), and keep the
leaf 3 bit set (i.e. only turn off the respective leaf 4 bit). This
_should_ make Windows not use the MSRs, but would still cope
with it nevertheless doing so for whatever reason.

>> yet (partly because for it to be complete we'd also want to
>> populate Viridian CPUID leaf 6, but the APIC related bit there is
>> insufficiently specified, so we first need to find out from MS when
>> exactly this bit is supposed to get set).
> I think CPUID leaf 6 just indicate which hardware feature is used by 
> hypervisor. It should have nothing to do with this issue. We can consider it 
> separately. 

Sure, this would be a separate patch, but still belonging here as
potentially influencing Windows' decision on how to set up certain
aspects of its operation, including APIC handling.

I'm in the process of putting all these together, in case you haven't


Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.