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

Re: [Xen-devel] [PATCH] Xen/MCE: adjust for future new vMCE model

Jan Beulich wrote:
>>>> On 05.07.12 at 09:02, Ian Campbell <Ian.Campbell@xxxxxxxxxx> wrote:
>> On Thu, 2012-07-05 at 07:44 +0100, Ian Campbell wrote:
>>> On Thu, 2012-07-05 at 07:39 +0100, Jan Beulich wrote:
>>>>>>> On 05.07.12 at 05:03, "Liu, Jinsong" <jinsong.liu@xxxxxxxxx>
>>>>>>> wrote: 
>>>>> Jan Beulich wrote:
>>>>>> Also I seem to recall that there was going to be a need to
>>>>>> save/restore MCi_CTL2, but there's nothing being done in that
>>>>>> direction.
>>>>> Yes, but that would be done at new vMCE.
>>>>> Current vMCE didn't enable MCG_CMCI_P, and MCi_CTL2 are not used,
>>>>> so we don't save/restore MCi_CTL2 at this version.
>>>> But _that_ is the major save/restore related issue to be
>>>> addressed, at least if backward migration (4.2 -> 4.1) is also
>>>> to be working reasonably. If such compatibility is entirely
>>>> unneeded, then I agree that it likely doesn't need any
>>>> consideration at this point.
>>> We don't generally worry about N->N-1 migration, just N->N+1.
>> If there's nothing to save in 4.2 (and therefore to restore in 4.3)
>> what is it we are giving a freeze exception to again?
>> If there's no save/restore can you even migrate a guest with vMCE
>> enabled at all?
> We're currently saving MCG_CAP, with the bank count in it
> reflecting the host bank count, and with other bits in there also
> matching the host's. That's what we'd like to simplify, _but_ as
> said we're already shipping the current interface, so backward
> compatibility will be needed anyway.
> The main issue really was with MCi_CTL, which if not saved by
> 4.2 wouldn't be restorable in 4.3. Now that we decided that they
> will get fixed values anyway, simply enforcing these fixed values
> (to not surprise the guest) would seem sufficient for 4.2. MCi_CTL2
> are different in that afaiu setting them to 0 for a guest with no
> saved values (i.e. 4.2, where their presence isn't being advertised,
> since CMCI doesn't get surfaced) would be correct.
> Jan

So Jan and Ian, do we really care migrate N -> N-1 ?
If yes, I will save/restore MCi_CTL2, but that would involve more code change.

Xen-devel mailing list



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