In another thread (@xen-devel), On Sun, Jan 24, 2010 at 5:44 PM, Weidong Han <weidong.han@intel.com> wrote:> Pls refer to section 3.6.1 in VT-d spec.Getting to that , reading @, http://wiki.xensource.com/xenwiki/VTdHowTo I note, "Caveat on Conventional PCI Device Pass-Through * The VT-d specification states that all conventional PCI devices behind a PCIe-to-PCI bridge have to be assigned to the same domain. * PCIe devices do not have this restriction. " but, at that Wiki page, there''s *no* mention of AMD -- at all. *Are*AMDs chipsets, and AMD-V, ''supported'' by Xen ? if so, is there a similar doc, listing compatible/support AMD platforms? or, at least, some clarification as to whether everything that''s documented for VT-d (parameters, exceptions, etc) holds true for AMD as well? thanks. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Mon, Jan 25, 2010 at 09:16:24AM -0800, 0bo0 wrote:> In another thread (@xen-devel), > > On Sun, Jan 24, 2010 at 5:44 PM, Weidong Han <weidong.han@intel.com> wrote: > > Pls refer to section 3.6.1 in VT-d spec. > > Getting to that , reading @, > > http://wiki.xensource.com/xenwiki/VTdHowTo > > I note, > > "Caveat on Conventional PCI Device Pass-Through > > * The VT-d specification states that all conventional PCI devices > behind a PCIe-to-PCI bridge have to be assigned to the same domain. > * PCIe devices do not have this restriction. " > > but, at that Wiki page, there''s *no* mention of AMD -- at all. > > *Are*AMDs chipsets, and AMD-V, ''supported'' by Xen ? > > if so, is there a similar doc, listing compatible/support AMD > platforms? or, at least, some clarification as to whether everything > that''s documented for VT-d (parameters, exceptions, etc) holds true > for AMD as well? >I''m not sure if AMD even has IOMMU capable hardware out yet? I think recently there was a patch to enable AMD IOMMU with Xen, on "800" series chipsets. -- Pasi _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Mon, Jan 25, 2010 at 10:00 AM, Pasi Kärkkäinen <pasik@iki.fi> wrote:> I''m not sure if AMD even has IOMMU capable hardware out yet? > > I think recently there was a patch to enable AMD IOMMU with Xen, > on "800" series chipsets.i''m drowning in acronyms and lack of documentation ... let''s try this: does Xen support whatever AMD cpus/chipsets DO have, that i''m currently using to successfully do pci-passthrough (which that wiki page seems to be about) to PV''d DomUs? I keep being told "The VT-d specification states that all conventional PCI devices behind a PCIe-to-PCI bridge have to be assigned to the same domain." and to "refer to the VT-d" spec, but I''m not ON Intel, don''t HAVE Vt-d, etc etc. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
hi On Mon, Jan 25, 2010 at 10:59 AM, chris <tknchris@gmail.com> wrote:> vt-d is for passthrough to hvm domains, pv should always work afaik with no > special cpu features...if that''s true, then i''m even more unclear as to why I''m being referred to the VT-d spec to explain away the issues i''m seeing/having ... in any case, for my box reading @ "HVM refactors Xen''s full virtualization support into a common layer and defines an interface to architecture specific components. This enables a single Xen hypervisor to dynamically detect and support AMD''s SVM or Intel''s VT technology while providing a single uniform management and user interface. " and cat /proc/cpuinfo | grep -i svm flags : ... svm ... _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Mon, Jan 25, 2010 at 11:22:09AM -0800, 0bo0 wrote:> hi > > On Mon, Jan 25, 2010 at 10:59 AM, chris <tknchris@gmail.com> wrote: > > vt-d is for passthrough to hvm domains, pv should always work afaik with no > > special cpu features... > > if that''s true, then i''m even more unclear as to why I''m being > referred to the VT-d spec to explain away the issues i''m seeing/having > ... > > in any case, for my box > > reading @ > > "HVM refactors Xen''s full virtualization support into a common layer > and defines an interface to architecture specific components. This > enables a single Xen hypervisor to dynamically detect and support > AMD''s SVM or Intel''s VT technology while providing a single uniform > management and user interface. " > > and > > cat /proc/cpuinfo | grep -i svm > flags : ... svm ... >Xen requires Intel VT or AMD-V to run HVM guests. These features are found from Intel and AMD CPUs. IOMMU I/O virtualization (Intel VT-D, or AMD IOMMU) is different and additional technology, allowing secure PCI passthrough to HVM guests. IOMMU (VT-d) support is required from the chipset. -- Pasi _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Mon, Jan 25, 2010 at 10:51:57AM -0800, 0bo0 wrote:> On Mon, Jan 25, 2010 at 10:00 AM, Pasi Kärkkäinen <pasik@iki.fi> wrote: > > I''m not sure if AMD even has IOMMU capable hardware out yet? > > > > I think recently there was a patch to enable AMD IOMMU with Xen, > > on "800" series chipsets. > > i''m drowning in acronyms and lack of documentation ... > > let''s try this: > > does Xen support whatever AMD cpus/chipsets DO have, that i''m > currently using to successfully do pci-passthrough (which that wiki > page seems to be about) to PV''d DomUs? >Xen PCI passthrough to PV guests has been supported for years, and it doesn''t require IOMMU or VT-d. On both Intel and AMD. Xen PCI passthrough to PV guests was actually available years before there even was a VT-d spec! The way it''s done for PV guests can be a bit unsafe though, since the guest gets DMA access and could use that maliciously..> I keep being told "The VT-d specification states that all conventional > PCI devices behind a PCIe-to-PCI bridge have to be assigned to the > same domain." and to "refer to the VT-d" spec, but I''m not ON Intel, > don''t HAVE Vt-d, etc etc. >VT-d (or IOMMU in general) is required for PCI passthrough to HVM guests. Xen requires IOMMU (VT-d on Intel) to do PCI passthrough to HVM guests. I think IOMMU can also be used to do _secure_ PCI passthrough to PV guests. (to fix the security problem I mentioned above). -- Pasi _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
hi, On Mon, Jan 25, 2010 at 12:38 PM, Pasi Kärkkäinen <pasik@iki.fi> wrote:> Xen PCI passthrough to PV guests has been supported for years, and it > doesn''t require IOMMU or VT-d. On both Intel and AMD. > > Xen PCI passthrough to PV guests was actually available years before > there even was a VT-d spec!good. esp since I''ve been doing it for ages ... ;-) so all that said (thx) ... why is it, that when I''m trying to "get around" the pci-passthrough co-assignment requirement -- and both fixes for 34x (as an unimplemented patch) and 4x (option in config file) are referenced/available -- AND,i''m (i) on AMD, and (ii) only using passthrough to PV guests, that I''m being referred to Intel VT-d documentation for the rationale/explanation that I''m *not* seeing a bug? not trying to be difficult here, just seems like the finger''s being pointed at something that''s not relevant to my situtation .... thanks. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Mon, Jan 25, 2010 at 12:54:04PM -0800, 0bo0 wrote:> hi, > > > On Mon, Jan 25, 2010 at 12:38 PM, Pasi Kärkkäinen <pasik@iki.fi> wrote: > > Xen PCI passthrough to PV guests has been supported for years, and it > > doesn''t require IOMMU or VT-d. On both Intel and AMD. > > > > Xen PCI passthrough to PV guests was actually available years before > > there even was a VT-d spec! > > good. esp since I''ve been doing it for ages ... ;-) > > so all that said (thx) ... > > why is it, that when I''m trying to "get around" the pci-passthrough > co-assignment requirement -- and both fixes for 34x (as an > unimplemented patch) and 4x (option in config file) are > referenced/available -- AND,i''m (i) on AMD, and (ii) only using > passthrough to PV guests, that I''m being referred to Intel VT-d > documentation for the rationale/explanation that I''m *not* seeing a > bug? > > not trying to be difficult here, just seems like the finger''s being > pointed at something that''s not relevant to my situtation .... >Well maybe you should point it out to be a regression compared to previous functionality.. ie. when you don''t use VT-d. -- Pasi _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Mon, Jan 25, 2010 at 1:05 PM, Pasi Kärkkäinen <pasik@iki.fi> wrote:>> not trying to be difficult here, just seems like the finger''s being >> pointed at something that''s not relevant to my situtation .... >> > > Well maybe you should point it out to be a regression compared to previous functionality.. > > ie. when you don''t use VT-d.tbh, because i''m not sure what is, isn''t a regression. or what is/isn''t to be expected. the relevant thread/comment @xen-devel, http://lists.xensource.com/archives/html/xen-devel/2010-01/msg00870.html i find totally confusing. sooo ... i brought the question "here" to get some understanding/clarity before making further statements there. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users