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

RE: [Xen-users] gaming in a windows domU?


  • To: "Igor Chubin" <igor@xxxxxxx>
  • From: "Petersson, Mats" <Mats.Petersson@xxxxxxx>
  • Date: Fri, 13 Apr 2007 13:06:15 +0200
  • Cc: xen-users@xxxxxxxxxxxxxxxxxxx
  • Delivery-date: Fri, 13 Apr 2007 04:05:15 -0700
  • List-id: Xen user discussion <xen-users.lists.xensource.com>
  • Thread-index: Acd9ubsnD0gpD6dDSX6ohrKcgczmswAAWHdA
  • Thread-topic: [Xen-users] gaming in a windows domU?

 

> -----Original Message-----
> From: Igor Chubin [mailto:igor@xxxxxxx] 
> Sent: 13 April 2007 11:52
> To: Petersson, Mats
> Cc: xen-users@xxxxxxxxxxxxxxxxxxx
> Subject: Re: [Xen-users] gaming in a windows domU?
> 
> On Fr, Apr 13, 2007 at 11:53:05 +0200, Petersson, Mats wrote:
> >  
> > 
> > > -----Original Message-----
> > > From: xen-users-bounces@xxxxxxxxxxxxxxxxxxx 
> > > [mailto:xen-users-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of 
> > > Igor Chubin
> > > Sent: 13 April 2007 10:29
> > > To: Andrew Yates
> > > Cc: xen-users@xxxxxxxxxxxxxxxxxxx
> > > Subject: Re: [Xen-users] gaming in a windows domU?
> > > 
> > > On Do, Apr 12, 2007 at 05:34:29 -0400, Andrew Yates wrote:
> > > > I've been considering buying a second video and sound 
> card, and then
> > > > using PCI passthrough with a Windows domU for gaming.
> > > > Has anyone tried this? I'm curious as to whether it's fast 
> > > enough for
> > > > recent games to be playable. The video card and CPU can 
> > > handle gaming
> > > > fine, but I'm not sure how much Xen would effect that. 
> So has anyone
> > > > done this? How well did it work?
> > > > 
> > > 
> > > Do HVM domain support PCI passthrough?
> > 
> > No. But that's the EASY part to solve. The hard part is 
> that the HVM OS
> > isn't aware that what it thinks is a physical address is in fact not
> > really the address that the memory controller is accessing when the
> > software accesses a virtual address. 
> > 
> 
> Mats, thank you for the answer.
> 
> As far as I know the problem will be eliminated
> when as soon as IOMMU will be available.
> Does IOMMU really solve the problem?

Yes, it solves this and also the security problem of having a device
owned by guest X read/write memory of guest Y (which is another
potential problem). 

> 
> Question related to previous one:
> HVM Windows Domains have poor IO-performance due to 
> necessity of usage of qemu device model (and correspondent drivers).
> Will the problem be eliminated (or alleviated at least)
> with appearance of the IOMMU?

Yes. IOMMU will also allow better performance. However, para-virtual
drivers can also solve this (at least improve it dramatically). PV are
drivers that are added to the HVM OS to allow 
> 
> 
> And question #3:
> How is Intel virtualization technology [1] related with IOMMU?
> 
> [1] http://www.intel.com/technology/itj/2006/v10i3/2-io/1-abstract.htm
> 
Intel's VT-d is the same concept as AMD's IOMMU concept. I haven't
looked at the details of it, so I can't say if it's EXACTLY the same, or
just a similar solution. [There's almost certainly SOME differences,
because the two systems are developed by different groups of people, who
have a similar problem to solve, which usually means a similar solution
with some differences]. 

--
Mats
> 
> > If we simplify things[1] to make it easy to explain: Dom0 
> uses 0..256MB,
> > HVM DomU uses 256..512MB, but thinks it got memory from 
> 0..256MB. DomU
> > tells the hardware that there is some data at 0x100000 (1MB), but we
> > should really add 256MB to that address. So the hardware will use
> > something that lives at 1MB in the Dom0 instead. Most 
> likely, this will
> > just lead to complete garbage being sent to the hardware, 
> which isn't so
> > bad. If it happens to be something that actually works with 
> the hardware
> > it may send your list of passwords to a random address on 
> the internet
> > or some such!
> > 
> > [1] In reality, each 4KB of guest memory could come from 
> any 4K section
> > of the machines memory, so it's not as simple as just 
> adding an offset -
> > but that's just a slight complication on top of the fact that DomU
> > doesn't actually know that it's memory isn't where it thinks it is. 
> > 
> > --
> > Mats
> > > 
> > > > _______________________________________________
> > > > Xen-users mailing list
> > > > Xen-users@xxxxxxxxxxxxxxxxxxx
> > > > http://lists.xensource.com/xen-users
> > > 
> > > -- 
> > > WBR, i.m.chubin
> > > 
> > > 
> > > _______________________________________________
> > > Xen-users mailing list
> > > Xen-users@xxxxxxxxxxxxxxxxxxx
> > > http://lists.xensource.com/xen-users
> > > 
> > > 
> > > 
> > 
> > 
> > 
> > _______________________________________________
> > Xen-users mailing list
> > Xen-users@xxxxxxxxxxxxxxxxxxx
> > http://lists.xensource.com/xen-users
> 
> -- 
> WBR, i.m.chubin
> 
> 
> 
> 



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


 


Rackspace

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