> -----Original Message-----
> From: Geoff Streeter [mailto:geoff@xxxxxxxxxx]
> Sent: 25 September 2006 15:04
> To: Petersson, Mats; Derek; xen ml
> Subject: RE: [Xen-users] Which Permutations of 64/32 work?
> At 2006-09-25 11:30 +0200, Petersson, Mats wrote:
> > > -----Original Message-----
> > > From: xen-users-bounces@xxxxxxxxxxxxxxxxxxx
> > > [mailto:xen-users-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of Derek
> > > Sent: 24 September 2006 19:02
> > > To: xen ml
> > > Subject: [Xen-users] Which Permutations of 64/32 work?
> > >
> > > Hi,
> > >
> > > I'm sure this must be documented somewhere, but I can't find it.
> > >
> > > Is there a document somewhere that shows which
> permutations of 64/32
> > > bits are known to work or known not to work with XEN
> 3.0.2 on 64-bit
> > > hardware?
> > >
> > > I.e. which of the following eight permutations work:
> > >
> > > - 64-bit dom0 OS with 64 bit HVM domU OS (?)
> > > - 64-bit dom0 OS with 32 bit HVM domU OS (?)
> > > - 64-bit dom0 OS with 64 bit paravirt domU OS (?)
> > > - 64-bit dom0 OS with 32 bit paravirt domU OS (?)
> > > - 32-bit dom0 OS with 64 bit HVM domU OS (?)
> > > - 32-bit dom0 OS with 32 bit HVM domU OS (WORKS)
> > > - 32-bit dom0 OS with 64 bit paravirt domU OS (?)
> > > - 32-bit dom0 OS with 32 bit paravirt domU OS (WORKS)
> >You've also missed the "intermediate" 32-bit PAE mode [supports > 4GB
> >RAM for the Dom0 and DomU, but at any given time 4GB only is
> >which applies as:
> >32-bit PAE Dom0 + 32-bit PAE HVM DomU.
> >32-bit PAE Dom0 + 32-bit HVM DomU.
> >32-bit PAE Dom0 + 32-bit PAE PV DomU.
> >All other combinations (32-bit PV DomU, 64-bit any DomU)
> doesn't work.
> >The generic rule is:
> >For Paravirtual, the guest has to be equal to Dom0.
> >For HVM, the guest has to be equal or less than Dom0.
> >Xen Hypervisor itself must be EQUAL to Dom0.
> >To explain how this works: Xen itself must be able to access the same
> >memory as the guest domain [Dom0 and DomU]. In the case of full (HVM)
> >virtualization, it's possible to have different page-tables (and thus
> >different "bitness") for the guest than the hypervisor, because the
> >transition (VMRUN/VMEXIT) swaps CR3 (Page-table base address) when
> >switching between DomU and Xen. In Paravirtual, the same CR3
> is used for
> >both guest and hypervisor at the same time, which in turn means that
> >they have to be in the same mode.
> >This is of course assuming there are no bugs in the Xen hypervisor or
> >tools that cause some particular combination(s) to be broken
> - but that
> >applies to all things working/not working!
> > If you're really into how 32-bit/32-bit PAE/64-bit
> page-tables work,
> >you'll be aware that not ONLY CR3 is affecting the operation of the
> >page-tables, and as you may expect, the remaining registers that do
> >affect this are ALSO swapped during the transition from/to guest.
> > >
> > > If this is not documented somewhere, maybe folks on this
> list wouldn't
> > > mind replying with any knowledge they may have. For
> example, if you
> > > know one works because you've tried it yourself, or you know one
> > > doesn't because it's excluded by some architectural
> design decision,
> > > or whatever.
> >I'm not sure if it's documented "properly" somewhere, but
> certainly it
> >has been discussed previously.
> My experience with Suse 10.1 64 bit Dom0 (I think) and
> Windows 64 bit HVM
> is that it doesn't work with Xen 3.0.2.
That is a slightly different question (into which you'd also have to
factor of which processor type you're using AMD or Intel) - because
Win64 is not one of the originally targetted systems, so only "lately"
have there been any work at all towards making that work. Of course, if
no one is working on making a particular OS work, there's chances that
SOMETHING in the complex system of hardware emulation that is needed
"behind the scenes" in Xen will not work correctly - such as MMIO
operations that are unsupported, or some hardware features that aren't
working as expected.
64-bit HVM & PV guests on 64-bit Dom0 works - but individual guest OS's
may not due to "compatibility" problems between the guest OS and the
virtual world that the Xen environment differs vs. the "bare-metal".
Obviously, those incompatibilities should be considered bugs, and
reported in bugzilla and eventually will hopefully be fixed (certainly
something making Win64 not run should be fixed).
Also, 3.0.2 is several months old now, and I'm sure I've seen several
fixes in the unstable tree... "If it doesn't work in the current
release, try unstable" - I'm not guaranteeing that it will work, but
it's got a better chance...
> Anybody know better?
Xen-users mailing list