Magenheimer, Dan (HP Labs Fort Collins) wrote:
>>> However, I agree with Matt that a PMT for other domains
>>> (domU) is a bad idea as it creates many problems for migration,
>>> save/restore, ballooning, and adding new domains to an already
>>> loaded system. Further, the grant table abstraction is the primary
>>> mechanism for page sharing for domU in Xen (on Xen/x86).
>>> I think if domU has any knowledge of actual machine addresses,
>>> the Xen team would consider this a bug that should be fixed.
>>>
>> Dan:
>> I think you get right reverse solution, without PMT in
>> domU is a nightmare for migration, balloning and etc.. We
>> (Matt, Kevin and me) believe PMT for domU is a must, because
>> domU don't want to know where the physical page is located,
>> so gpn will always be from 0 for example while mfn may start
>> from any address, this is what PMT does to translate from gpn
>> to mfn. Today's dom0 is assume gpn=mfn so no PMT table yet,
>> that is what we are discussing to let dom0 have PMT same with domU.
>> I guess you are probably assuming the PMT is a Xenlinux
>> stuff, actually PMT is a HV data structure even in Xen/X86
>> and Xen/IA64-VTi. HV need to use PMT to insert machine side
>> TLB for example, and sharing HV PMT to paravirtualized guest
>> should be done so that VBD and VNIF can refer to. With PMT in
>> dom0, we are just stepping toward close to Xen/X86 and reduce
>> various maintaince effort and deviation.
>
> Well clearly we are not on the same page regarding the definition
> of PMT, so it is good for you to define what you mean. I did
> assume that you were referring to a data structure that is
> accessible to both domain and hypervisor. This is what Matt and I are
> arguing against for domU. If it is not visible, there is already
> a "PMT" in Xen which translates domU physical addresses to machine
> addresses. It is just implemented as a multi-level page table
> rather than one large 1-1 mapping table. (And this is necessary
> because we don't want to require allocation of large contiguous memory
> blocks after dom0 boots.)
>
> So, I guess we are in agreement. We will look at the possibility
> of adding a PMT visible to both Xen and domain0 (and in the future
> also driver domains) and domU's will not have access to any
> machine address information via a PMT or any other data structure
> (but Xen of course will need to maintain this information).
> Correct?
>
> Thanks,
> Dan
Yes! :-)
Eddie
_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel
|
Home