Hey everyone, I realize that people asking about PCI passthrough has been thoroughly beaten to death many a time, however I couldn''t find the answer to these questions anywhere (and I''d really like to know :P) so please just bare with me. I know the GART built into AMD chipsets can act as a "mini IOMMU" and do DMA translation but not protection/isolation. However, last I checked PCI passthrough only worked on Xen 3.2 (unstable) if a VT-d supporting chipset was detected. So what I wanted to know: 1) If PCI passthrough can be done using GART, then is this being worked on? Also can PCI hotplug work as with VT-d? 2) Are there any technological limitations which prevent the GART from being used with PCI passthrough or is this just not coded? 2) If it is not coded, is it a matter of someone not wanting to do this or is it a matter of such a patch never being merged into mainline because of potential stability issues due to lack of isolation? I would like to know since AMD''s real IOMMU won''t be in chipsets until 2010 and you can''t get features like SLI on Intel''s chipsets. If I can use the GART on a Phenom say, then I can get a nVidia SLI board and have the benefits of the more advanced features already present in the Phenom like RVI. What do you guys think? I really appreciate your work - keep it up! Jeff _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Jeff Williams
2008-May-15 23:29 UTC
[Xen-devel] Re: PCI Passthrough using GART on AMD CPUs?
Also, I apologize if I should have posted this in one of the other groups like xen.user. Jeff _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Han, Weidong
2008-May-16 01:13 UTC
RE: [Xen-devel] PCI Passthrough using GART on AMD CPUs?
> Also can PCI hotplug work as with VT-d?Yes. VT-d hotplug has been supported. docs/misc/vtd.txt tells you how to use it. Randy (Weidong) Jeff Williams wrote:> Hey everyone, > > I realize that people asking about PCI passthrough has been > thoroughly beaten to death many a time, however I couldn''t find the > answer to these questions anywhere (and I''d really like to know :P) > so please just bare with me. > > I know the GART built into AMD chipsets can act as a "mini IOMMU" and > do DMA translation but not protection/isolation. However, last I > checked PCI passthrough only worked on Xen 3.2 (unstable) if a VT-d > supporting chipset was detected. > > So what I wanted to know: > > 1) If PCI passthrough can be done using GART, then is this being > worked on? Also can PCI hotplug work as with VT-d? > 2) Are there any technological limitations which prevent the GART > from being used with PCI passthrough or is this just not coded? > 2) If it is not coded, is it a matter of someone not wanting to do > this or is it a matter of such a patch never being merged into > mainline because of potential stability issues due to lack of > isolation? > > I would like to know since AMD''s real IOMMU won''t be in chipsets > until 2010 and you can''t get features like SLI on Intel''s chipsets. > If I can use the GART on a Phenom say, then I can get a nVidia SLI > board and have the benefits of the more advanced features already > present in the Phenom like RVI. > > What do you guys think? I really appreciate your work - keep it up! > > Jeff > > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Jeff Williams
2008-May-16 01:28 UTC
[Xen-devel] Re: PCI Passthrough using GART on AMD CPUs?
Han, Weidong <weidong.han <at> intel.com> writes:> > > Also can PCI hotplug work as with VT-d? > > Yes. VT-d hotplug has been supported. docs/misc/vtd.txt tells you how to > use it. > > Randy (Weidong) >Actually I''ve read this already but thanks. I was wondering if PCI hotplug could be make to work with the GART mini-IOMMU on the AMD platform instead of using VT-d. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Li, Xin B
2008-May-16 02:08 UTC
RE: [Xen-devel] Re: PCI Passthrough using GART on AMD CPUs?
we don''t have that kind of environment, but the code is designed to support both platform. -Xin>-----Original Message----- >From: xen-devel-bounces@lists.xensource.com >[mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of >Jeff Williams >Sent: 2008年5月16日 9:28 >To: xen-devel@lists.xensource.com >Subject: [Xen-devel] Re: PCI Passthrough using GART on AMD CPUs? > >Han, Weidong <weidong.han <at> intel.com> writes: > >> >> > Also can PCI hotplug work as with VT-d? >> >> Yes. VT-d hotplug has been supported. docs/misc/vtd.txt >tells you how to >> use it. >> >> Randy (Weidong) >> > >Actually I''ve read this already but thanks. I was wondering if >PCI hotplug >could be make to work with the GART mini-IOMMU on the AMD >platform instead of >using VT-d. > > >_______________________________________________ >Xen-devel mailing list >Xen-devel@lists.xensource.com >http://lists.xensource.com/xen-devel >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Yosuke Iwamatsu
2008-May-16 04:05 UTC
Re: [Xen-devel] Re: PCI Passthrough using GART on AMD CPUs?
Hi, I''ve found an old post about GART as mini-IOMMU. Please see below. As far as I read, the patch was not checked in because it was not complete. If the code comes to the main tree, there is no reason we don''t support PCI hotplug for GART. http://xen.markmail.org/message/ylcrn2qgvgzlpajd Thanks, Yosuke Li, Xin B wrote:> we don''t have that kind of environment, but the code is designed to support both platform. > -Xin > >> -----Original Message----- >> From: xen-devel-bounces@lists.xensource.com >> [mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of >> Jeff Williams >> Sent: 2008年5月16日 9:28 >> To: xen-devel@lists.xensource.com >> Subject: [Xen-devel] Re: PCI Passthrough using GART on AMD CPUs? >> >> Han, Weidong <weidong.han <at> intel.com> writes: >> >>>> Also can PCI hotplug work as with VT-d? >>> Yes. VT-d hotplug has been supported. docs/misc/vtd.txt >> tells you how to >>> use it. >>> >>> Randy (Weidong) >>> >> Actually I''ve read this already but thanks. I was wondering if >> PCI hotplug >> could be make to work with the GART mini-IOMMU on the AMD >> platform instead of >> using VT-d. >> >> >> _______________________________________________ >> Xen-devel mailing list >> Xen-devel@lists.xensource.com >> http://lists.xensource.com/xen-devel >> > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Keir Fraser
2008-May-16 07:06 UTC
Re: [Xen-devel] Re: PCI Passthrough using GART on AMD CPUs?
The functionality of that patch is to allow dom0 to use the GART as a target aperture for 32-bit PCI devices to access memory above 4GB. The GART cannot possibly implement the functionality of the new AMD or Intel IOMMUs, as it does not intercept *every* DMA access issued by guest-controlled devices. The patch you would like to see is an impossibility. -- Keir On 16/5/08 05:05, "Yosuke Iwamatsu" <y-iwamatsu@ab.jp.nec.com> wrote:> Hi, > > I''ve found an old post about GART as mini-IOMMU. Please see below. > As far as I read, the patch was not checked in because it was not > complete. If the code comes to the main tree, there is no reason > we don''t support PCI hotplug for GART. > > http://xen.markmail.org/message/ylcrn2qgvgzlpajd > > Thanks, > Yosuke > > Li, Xin B wrote: >> we don''t have that kind of environment, but the code is designed to support >> both platform. >> -Xin >> >>> -----Original Message----- >>> From: xen-devel-bounces@lists.xensource.com >>> [mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of >>> Jeff Williams >>> Sent: 2008年5月16日 9:28 >>> To: xen-devel@lists.xensource.com >>> Subject: [Xen-devel] Re: PCI Passthrough using GART on AMD CPUs? >>> >>> Han, Weidong <weidong.han <at> intel.com> writes: >>> >>>>> Also can PCI hotplug work as with VT-d? >>>> Yes. VT-d hotplug has been supported. docs/misc/vtd.txt >>> tells you how to >>>> use it. >>>> >>>> Randy (Weidong) >>>> >>> Actually I''ve read this already but thanks. I was wondering if >>> PCI hotplug >>> could be make to work with the GART mini-IOMMU on the AMD >>> platform instead of >>> using VT-d. >>> >>> >>> _______________________________________________ >>> Xen-devel mailing list >>> Xen-devel@lists.xensource.com >>> http://lists.xensource.com/xen-devel >>> >> >> _______________________________________________ >> Xen-devel mailing list >> Xen-devel@lists.xensource.com >> http://lists.xensource.com/xen-devel > > > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Jeff Williams
2008-May-17 19:29 UTC
[Xen-devel] Re: PCI Passthrough using GART on AMD CPUs?
Keir Fraser <keir.fraser <at> eu.citrix.com> writes:> > The functionality of that patch is to allow dom0 to use the GART as a target > aperture for 32-bit PCI devices to access memory above 4GB. The GART cannot > possibly implement the functionality of the new AMD or Intel IOMMUs, as it > does not intercept *every* DMA access issued by guest-controlled devices. > The patch you would like to see is an impossibility. > > -- Keir >Thank you Keir for clarifying this. I guess I''ll just have to wait, or get an Intel VT-d board. Too bad Skulltrail is still so expensive.. Jeff _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Reasonably Related Threads
- Megaraid SAS driver failing in Xen-3.3.0 but was working in Xen-3.2.2-rc3
- FW: megaraid sas driver failing in Xen-3.4 unstable version
- RE: Xen-3.2.1 VT-d Support (NOT SURE WHETHER IT''S A BUG OR...)
- iommu: mapping reserved region failed - Q35 - VT-D Issue
- [VTD][PATCH] Change xc_assign_device()