On 2011-04-25 03:48, Ben Hutchings wrote:
On Thu, 2011-04-21 at 22:53 +0200, Robin Axelsson wrote:On 2011-04-21 21:22, Ben Hutchings wrote:On Thu, Apr 21, 2011 at 09:05:14PM +0200, Robin Axelsson wrote:On 2011-04-21 20:03, Debian Bug Tracking System wrote:n Thu, Apr 21, 2011 at 05:41:29PM +0200, Robin Axelsson wrote:Package: Debian-testing-amd64-DC-1.iso (latest version converted to boot USB) Severity: Critical Platform: x86_64 Linux (also .386/i686) > Description of problem: Since I updated the firmware of my BIOS to a revision that supports AMD IOMMU and turned that feature on, pretty much anything that runs Linux fails to boot All I see about half a second after I pass the GRUB bootloader is a blinking cursor in the upper left corner. The mouse gets disabled (i.e. powered off in spite of being plugged in) and the keyboard doesn't even respond to caps-lock, num-lock and scroll-lock toggles (numlock light is constantly on). > I assume that it is needless to say that this is anon-issue whenthe IOMMU feature is disabled.[...] So there is a bug in the BIOS, and you have a workaround for it. I don't see a bug in Debian. Ben.I don't understand your statement. IOMMU is a feature that allows for passthrough of I/O devices to virtual machines. I need this feature on my computer but Debian doesn't work with that feature.Like many new hardware features, IOMMUs must be correctly described by the BIOS before an operating system can use them.If this was due to a bug in the BIOS why the does it not crash Citrix Xenserver 5.6, Windows 7 x64 or even FreeDOS/MS-DOS?Because they don't try to use the IOMMU. Ben.Ok, I have failed to activate the IOMMU feature in the XenServer (by adding iommu=1 after xen.gz in the /boot/extlinux.conf and running extlinux /boot).[...] I'm not sure what you mean by that. If you *do* use that option, does XenServer boot and are you able to use PCI pass-through? Ben.
I was very surprised to find out that XenServer booted at all since it is linux based and everything else with linux on it had failed so far. Moreover, since XenServer is with the horse's mouth of Xen I assumed that it is the most technologically advanced Xen implementation. So to me it was implied that it would use IOMMU by default whenever such feature would be available in the hardware.
When I read the documentation more thoroughly I found that the version of Xen is merely 3.4.2 and not the latest 4.1. I also found it that it runs with an as lean a configuration as possible.
So when I added that option in the bootloader config, the system simply failed to boot with the error message "dom0 kernel not specified. Check bootloader configuration". This error message doesn't make any sense to me since the only change that I did was adding the iommu=1 option. I tried to add this option on several different locations on the arguments line yielding this very same result. Removing this option makes the system bootable again.
Sander Eikelenboom who is active on the xen-devel mailing list have reported (on the Xen VTd HowTo page) IOMMU to work with this particular motherboard. He was using a beta 1.75 revision of the BIOS firmware. I'm currently using revision 1.9 so I'm considering to downgrade to the older beta BIOS to see if that helps.
Robin.