Re: [Xen-users] HVM and IO devices

[Mark Williamson <mark.williamson@xxxxxxxxxxxx>]

> I need to make some experiments using Xen and HVM. My question is
> regarding IO devices. I know that Xen supports unmodified guest OSs
> when running in HVM mode, but does it include unmodified device
> drivers? What kind of hardware can be transparently used by the guest
> OS?

Generally an HVM guest will be given access to emulated models of a set of 
real devices (disk, network, CD-ROM, graphics, and possibly sound and USB 
although I don't think many people use those).  These all look like real 
devices (although the host won't necessarily have those actual devices) and 
the guest will just use it's normal drivers for those devices.

HVM guests are not given direct access to real hardware devices.  It's 
technically not very easy to do this for most interesting devices; you can't 
pass a PCI device to a guest because it won't know where to issue DMA 
commands to (since it doesn't understand memory layout).

Some guys are working on getting HVM guests to run in such a way that they can 
do DMA directly, and correctly (although this doesn't address the problem of 
isolating them from taking over the machine using DMA access).

Passing through USB devices to guests through the Qemu device model is 
possible, although I've heard varying reports of how well this actually 
works.  In principle it ought to be fairly straightforward to let an HVM 
guest access serial / parallel ports, but I'm not sure if this is supported.

> My experiments will probably run with a single domU in adition to
> dom0, so providing exclusive access to a device in domU wouldn't be a
> problem.

Sounds like the HVM PCI passthrough work I mentioned above might help you 
eventually.  Patches are being posted on xen-devel by developers at Neocleus 
who are doing this stuff.  It doesn't provide isolation for that DomU to stop 
it taking over the system though.

> Also, will the introduction of IOMMU support in CPUs change anything
> about this scenario? And if so, when are CPUs expected to begin
> shipping with IOMMU support (I see announces from both Intel and AMD
> but not any concrete information).

IOMMU support solves both the translation and the protection part of this 
problem, whereas the HVM PCI passthrough work from Neocleus and some patches 
from AMD that utilise the gart only fix the problem of translation (to give 
correct behaviour) not protection (to give isolation).

I also don't know when the hardware will be available.  Probably the 
representatives from those companies won't be able to tell us anything more 
than what's on the websites due to NDAs.

Hope that helps some,
Cheers,
Mark


-- 
Dave: Just a question. What use is a unicyle with no seat?  And no pedals!
Mark: To answer a question with a question: What use is a skateboard?
Dave: Skateboards have wheels.
Mark: My wheel has a wheel!

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users