Hi, I am trying to make the vt-d and pci pass-through work and I have some problems. I just try to give a NIC to an HVM guest. When I create the vm the hypervisor goes to an infinite loop, it displays in loop the message enclosed (console_on_xm_create). I use xen x86_64, debian etch dom0 with linux-2.6.18.8-xen 32. The hvm guest is a debian etch 64. Cheers, -- Jean Guyader _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Jean Guyader wrote:> Hi, > > I am trying to make the vt-d and pci pass-through work and I have some > problems. I just try to give a NIC to an HVM guest. > > When I create the vm the hypervisor goes to an infinite loop, > it displays in loop the message enclosed (console_on_xm_create). > > I use xen x86_64, debian etch dom0 with linux-2.6.18.8-xen 32. > The hvm guest is a debian etch 64. > >I have the same result with xen-3.2.0. -- Jean Guyader _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Does this happen during xen/dom0 boot time or during HVM guest boot? Are you using and desktop or server platform? Are you using the latest BIOS? Some older BIOS does not initialize VT-d HW properly. Allen>-----Original Message----- >From: xen-devel-bounces@lists.xensource.com >[mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of >Jean Guyader >Sent: Friday, March 28, 2008 4:16 AM >To: xen-devel@lists.xensource.com >Subject: Re: [Xen-devel] pci pass-through, NIC card > >Jean Guyader wrote: >> Hi, >> >> I am trying to make the vt-d and pci pass-through work and I >have some >> problems. I just try to give a NIC to an HVM guest. >> >> When I create the vm the hypervisor goes to an infinite loop, >> it displays in loop the message enclosed (console_on_xm_create). >> >> I use xen x86_64, debian etch dom0 with linux-2.6.18.8-xen 32. >> The hvm guest is a debian etch 64. >> >> > >I have the same result with xen-3.2.0. > >-- >Jean Guyader > >_______________________________________________ >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
Kay, Allen M wrote:> Does this happen during xen/dom0 boot time or during HVM guest boot? > > Are you using and desktop or server platform? Are you using the latest > BIOS? Some older BIOS does not initialize VT-d HW properly. > > Allen >It''s a server platform, I could indeed check if it is the latest bios. This is happen during the HVM guest boot, all initialisation in dom0 are ok. The device I tried to pass-through was a pci NIC. I have tried to pass-through a PCIe NIC and it works better, I mean the hvm guest boot. The NIC card is well detected by the hvm guest but, there is still a problem. When I tried to get an ip address from dhcp I received a storm of "iommu_page_fault" (enclosed the log). The pci device I try to pass-through is at the address 8:0:0. I have disable the pci msi in the guest. Thanks, -- Jean Guyader _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
I would assume you see "Intel VT-d has been enabled" in you full log file, correct? Given that dom0 devices are working OK - this means VT-d translation is working correctly for dom0. The current code maps entire guest memory in VT-d by sharing with p2m table. The vt-d page fault you are seeing should only happen if the guest is using some DMA address that is not coverred by the P2m page table - which is shouldn''t have happened. I have seen this type of fault if the BIOS is not programming VT-d HW correctly. Who''s the vendor of this Stoakley system? We can try to duplicate the problem here if we can get hold of the same system. Allen>-----Original Message----- >From: Jean Guyader [mailto:jean.guyader@eu.citrix.com] >Sent: Friday, March 28, 2008 10:50 AM >To: Kay, Allen M >Cc: xen-devel@lists.xensource.com >Subject: Re: [Xen-devel] pci pass-through, NIC card > >Kay, Allen M wrote: >> Does this happen during xen/dom0 boot time or during HVM guest boot? >> >> Are you using and desktop or server platform? Are you using >the latest >> BIOS? Some older BIOS does not initialize VT-d HW properly. >> >> Allen >> > >It''s a server platform, I could indeed check if it is the latest bios. > >This is happen during the HVM guest boot, all initialisation >in dom0 are >ok. The device I tried to pass-through was a pci NIC. > >I have tried to pass-through a PCIe NIC and it works better, I >mean the >hvm guest boot. The NIC card is well detected by the hvm guest but, >there is still a problem. When I tried to get an ip address >from dhcp I >received a storm of "iommu_page_fault" (enclosed the log). > >The pci device I try to pass-through is at the address 8:0:0. > >I have disable the pci msi in the guest. > >Thanks, > >-- >Jean Guyader >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Kay, Allen M wrote:> I would assume you see "Intel VT-d has been enabled" in you full log > file, correct? >Yes, there is this line.> Given that dom0 devices are working OK - this means VT-d translation is > working correctly for dom0. > > The current code maps entire guest memory in VT-d by sharing with p2m > table. The vt-d page fault you are seeing should only happen if the > guest is using some DMA address that is not coverred by the P2m page > table - which is shouldn''t have happened. > > I have seen this type of fault if the BIOS is not programming VT-d HW > correctly. > > Who''s the vendor of this Stoakley system? We can try to duplicate the > problem here if we can get hold of the same system.Here, the reference of the machine : Phoenix TrustedCore(tm) Server Copyright 1985-2007 Phoenix Technologies Ltd. All Rights Reserved Supermicro X7DWE+/X7DWN+ BIOS Rev 1.0 CPU = 2 Processor(s) Detected, Cores per Processor = 4 Intel(R) Xeon(R) CPU X5376 @ 2.80GHz Intel(R) Xeon(R) CPU X5376 @ 2.80GHz DRAM Type : DDR2-800, FSB at 1600MHz 8192M System RAM Passed 8192 KB L2 Cache System BIOS shadowed Video BIOS shadowed Fixed Disk 0: Maxtor 6Y160M0 ATAPI CD-ROM: MATSHITADVD-ROM SR-8178 -- Jean Guyader _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Kay, Allen M wrote:> I would assume you see "Intel VT-d has been enabled" in you full log > file, correct? > > Given that dom0 devices are working OK - this means VT-d translation is > working correctly for dom0. > > The current code maps entire guest memory in VT-d by sharing with p2m > table. The vt-d page fault you are seeing should only happen if the > guest is using some DMA address that is not coverred by the P2m page > table - which is shouldn''t have happened. > > I have seen this type of fault if the BIOS is not programming VT-d HW > correctly. > > Who''s the vendor of this Stoakley system? We can try to duplicate the > problem here if we can get hold of the same system. > > Allen > >> -----Original Message----- >> From: Jean Guyader [mailto:jean.guyader@eu.citrix.com] >> Sent: Friday, March 28, 2008 10:50 AM >> To: Kay, Allen M >> Cc: xen-devel@lists.xensource.com >> Subject: Re: [Xen-devel] pci pass-through, NIC card >> >> Kay, Allen M wrote: >>> Does this happen during xen/dom0 boot time or during HVM guest boot? >>> >>> Are you using and desktop or server platform? Are you using >> the latest >>> BIOS? Some older BIOS does not initialize VT-d HW properly. >>> >>> Allen >>> >> It''s a server platform, I could indeed check if it is the latest bios. >> >> This is happen during the HVM guest boot, all initialisation >> in dom0 are >> ok. The device I tried to pass-through was a pci NIC. >> >> I have tried to pass-through a PCIe NIC and it works better, I >> mean the >> hvm guest boot. The NIC card is well detected by the hvm guest but, >> there is still a problem. When I tried to get an ip address >>from dhcp I >> received a storm of "iommu_page_fault" (enclosed the log). >> >> The pci device I try to pass-through is at the address 8:0:0. >> >> I have disable the pci msi in the guest. >> >> Thanks, >>May be it fails because, the NIC I tried to pass-through is not a PCie e1000 card. The MIC card is a 82575EB and it works fine in dom0 with the igb driver (http://sourceforge.net/project/showfiles.php?group_id=42302). -- Jean Guyader _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel