Does Xen support Intel''s IOMMU technology (VT-d)? -- Roberto Jung Drebes - Tokyo, Japan <drebes@gmail.com> _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
> Does Xen support Intel''s IOMMU technology (VT-d)?Patches have been posted on the xen-devel mailing list. They may have been merged into xen-unstable by now; if not, then I''m sure the Intel folks would be aiming for a merge before Xen 3.2 or so. 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@lists.xensource.com http://lists.xensource.com/xen-users
On 8/31/07, Mark Williamson <mark.williamson@cl.cam.ac.uk> wrote:> > Does Xen support Intel''s IOMMU technology (VT-d)? > > Patches have been posted on the xen-devel mailing list. They may have been > merged into xen-unstable by now; if not, then I''m sure the Intel folks would > be aiming for a merge before Xen 3.2 or so.And does Xen support AMD''s IOMMU? I''m trying to decide on a hardware to make some experiments with HVM and DMA transfers, but can''t find much information on which would be the best hardware onto which to try it out. I get some information from presentations and stuff but am not sure if they are still current or already outdated. Thanks for your answer, it has already helped me. -- Roberto Jung Drebes - Tokyo, Japan <drebes@gmail.com> http://dorebesu.blogspot.com/ _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Would this VT-d let us assign graphic cards to DomUs? Are we going to be able to switch between Dom0 and DomU while this card is being used? We can talk some kind of "sharing" here, when DomU has the "screen", it uses the card, if not, Dom0. Or, somehow we should see the output of DomU in Dom0''s screen, but with unmodified driver in the DomU. Possible? or not yet even with VT-d? Emre On 8/31/07, Roberto Jung Drebes <drebes.lists@gmail.com> wrote:> > On 8/31/07, Mark Williamson <mark.williamson@cl.cam.ac.uk> wrote: > > > Does Xen support Intel''s IOMMU technology (VT-d)? > > > > Patches have been posted on the xen-devel mailing list. They may have > been > > merged into xen-unstable by now; if not, then I''m sure the Intel folks > would > > be aiming for a merge before Xen 3.2 or so. > > And does Xen support AMD''s IOMMU? I''m trying to decide on a hardware > to make some experiments with HVM and DMA transfers, but can''t find > much information on which would be the best hardware onto which to try > it out. I get some information from presentations and stuff but am not > sure if they are still current or already outdated. > > Thanks for your answer, it has already helped me. > > -- > Roberto Jung Drebes - Tokyo, Japan > <drebes@gmail.com> > http://dorebesu.blogspot.com/ > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users >-- Emre Erenoglu erenoglu@gmail.com _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
> > Patches have been posted on the xen-devel mailing list. They may have > > been merged into xen-unstable by now; if not, then I''m sure the Intel > > folks would be aiming for a merge before Xen 3.2 or so. > > And does Xen support AMD''s IOMMU? I''m trying to decide on a hardware > to make some experiments with HVM and DMA transfers, but can''t find > much information on which would be the best hardware onto which to try > it out. I get some information from presentations and stuff but am not > sure if they are still current or already outdated. > > Thanks for your answer, it has already helped me.I don''t know if AMD have submitted patches yet. There was an AMD guy working on using the GART as a translation-only (not protection) IOMMU to allow passthrough to work (albeit without isolation) on existing non-IOMMU hardware. I''m not sure what the status of that is. Some folks have recently been trying to get PCI passthrough to HVM working without any form of translation hardware whatsoever; patches have been floated on the list for this. Basically the idea is to hide the areas of memory used by Xen, dom0, etc from an HVM guest by changing the E820 memory map to just say "nothing interesting here". You''re still relying on the guest''s obeying that, since it has DMA access to all memory. But one day it could be handy for passing PCI devices to Windows domains, etc... 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@lists.xensource.com http://lists.xensource.com/xen-users
> Would this VT-d let us assign graphic cards to DomUs?You already can do this in principle (I''m not sure if it works in practice; a few people have tried playing with it though) for a PV domU. With an IOMMU and some other moving parts (changes to Qemu device models, etc) it ought to be possible to do this for HVM domUs too. Note my other e-mail about being able to pass through devices on systems without an IOMMU... This could prove useful at some point in the future.> Are we going to be > able to switch between Dom0 and DomU while this card is being used? > We can talk some kind of "sharing" here, when DomU has the "screen", it > uses the card, if not, Dom0. Or, somehow we should see the output of DomU > in Dom0''s screen, but with unmodified driver in the DomU. > > Possible? or not yet even with VT-d?It''d need more coding in order to get the switching / sharing working. VT-d could be one part of a solution make it safe and practical to pass graphics cards directly to HVM guests, but in itself it doesn''t make switching on-the-fly possible. 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@lists.xensource.com http://lists.xensource.com/xen-users
Here''s the link to the top of the VT-d external stuff at Intel: http://www.intel.com/technology/itj/2006/v10i3/2-io/5-platform-hardware-support.htm VT-d makes it possible to assign devices (B:D:F) to a Dom, Xen may be limited to 8 devices due to the current architecture of pciback.hide, but it''s difficult to tell because this technology is still in the really early stages of development. Yes, the tech will allow you to assign video cards (or even ports on cards) provided they can be uniquely addressed (see PCI IOV sig, iirc for the prerequisites). We will *not* be sharing. Go re-read the specs, this is device *reassignment* and *isolation* to a Dom as the desirable objective of this new silicon technology. With respect to the scenario you''re describing, I don''t think I understand it. My main interest in this technology is not from a desktop user perspective but as a big server administrator who wants to segregate encrypted network sessions and i/o controllers. </luddite> ndex On 8/31/07, Emre Erenoglu <erenoglu@gmail.com> wrote:> > Would this VT-d let us assign graphic cards to DomUs? Are we going to be > able to switch between Dom0 and DomU while this card is being used? > > We can talk some kind of "sharing" here, when DomU has the "screen", it > uses the card, if not, Dom0. Or, somehow we should see the output of DomU in > Dom0''s screen, but with unmodified driver in the DomU. > > Possible? or not yet even with VT-d? > > Emre > > > On 8/31/07, Roberto Jung Drebes <drebes.lists@gmail.com> wrote: > > > > On 8/31/07, Mark Williamson <mark.williamson@cl.cam.ac.uk > wrote: > > > > Does Xen support Intel''s IOMMU technology (VT-d)? > > > > > > Patches have been posted on the xen-devel mailing list. They may have > > been > > > merged into xen-unstable by now; if not, then I''m sure the Intel folks > > would > > > be aiming for a merge before Xen 3.2 or so. > > > > And does Xen support AMD''s IOMMU? I''m trying to decide on a hardware > > to make some experiments with HVM and DMA transfers, but can''t find > > much information on which would be the best hardware onto which to try > > it out. I get some information from presentations and stuff but am not > > sure if they are still current or already outdated. > > > > Thanks for your answer, it has already helped me. > > > > -- > > Roberto Jung Drebes - Tokyo, Japan > > <drebes@gmail.com> > > http://dorebesu.blogspot.com/ > > > > _______________________________________________ > > Xen-users mailing list > > Xen-users@lists.xensource.com > > http://lists.xensource.com/xen-users > > > > > > -- > Emre Erenoglu > erenoglu@gmail.com > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users >_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Hi, Thanks for the response, it''s very informative. One thing though, aren''t we already able to make "device assignment" to a specific DomU with pciback.hide, what will VT-d bring in addition to this? More secure assignment? Emre On 9/2/07, Ndex Server <ndex.srvr@gmail.com> wrote:> > Here''s the link to the top of the VT-d external stuff at Intel: > http://www.intel.com/technology/itj/2006/v10i3/2-io/5-platform-hardware-support.htm > > > VT-d makes it possible to assign devices (B:D:F) to a Dom, Xen may be > limited to 8 devices due to the current architecture of pciback.hide, but > it''s difficult to tell because this technology is still in the really early > stages of development. > > Yes, the tech will allow you to assign video cards (or even ports on > cards) provided they can be uniquely addressed (see PCI IOV sig, iirc for > the prerequisites). > > We will *not* be sharing. Go re-read the specs, this is device > *reassignment* and *isolation* to a Dom as the desirable objective of this > new silicon technology. > > With respect to the scenario you''re describing, I don''t think I understand > it. My main interest in this technology is not from a desktop user > perspective but as a big server administrator who wants to segregate > encrypted network sessions and i/o controllers. </luddite> > > ndex > > On 8/31/07, Emre Erenoglu <erenoglu@gmail.com> wrote: > > > > Would this VT-d let us assign graphic cards to DomUs? Are we going to be > > able to switch between Dom0 and DomU while this card is being used? > > > > We can talk some kind of "sharing" here, when DomU has the "screen", it > > uses the card, if not, Dom0. Or, somehow we should see the output of DomU in > > Dom0''s screen, but with unmodified driver in the DomU. > > > > Possible? or not yet even with VT-d? > > > > Emre > > > > > > On 8/31/07, Roberto Jung Drebes <drebes.lists@gmail.com > wrote: > > > > > > On 8/31/07, Mark Williamson < mark.williamson@cl.cam.ac.uk > wrote: > > > > > Does Xen support Intel''s IOMMU technology (VT-d)? > > > > > > > > Patches have been posted on the xen-devel mailing list. They may > > > have been > > > > merged into xen-unstable by now; if not, then I''m sure the Intel > > > folks would > > > > be aiming for a merge before Xen 3.2 or so. > > > > > > And does Xen support AMD''s IOMMU? I''m trying to decide on a hardware > > > to make some experiments with HVM and DMA transfers, but can''t find > > > much information on which would be the best hardware onto which to try > > > > > > it out. I get some information from presentations and stuff but am not > > > sure if they are still current or already outdated. > > > > > > Thanks for your answer, it has already helped me. > > > > > > -- > > > Roberto Jung Drebes - Tokyo, Japan > > > <drebes@gmail.com> > > > http://dorebesu.blogspot.com/ > > > > > > _______________________________________________ > > > Xen-users mailing list > > > Xen-users@lists.xensource.com > > > http://lists.xensource.com/xen-users > > > > > > > > > > > -- > > Emre Erenoglu > > erenoglu@gmail.com > > _______________________________________________ > > Xen-users mailing list > > Xen-users@lists.xensource.com > > http://lists.xensource.com/xen-users > > > >-- Emre Erenoglu erenoglu@gmail.com _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users