Windows guest crashes on Linux host if VT-x/AMD-V enabled

[f.hoefling]

Booting Windows XP Prof. (32 bit) in Virtualbox 3.0.6 with VT-x/AMD-V results in a crash of the virtual machine. The problem is solved by disabeling the VT-x/AMD-V option.

Booting and running memtest86 with the same configuration works without problems.

Host configuration: Ubuntu Linux (Jaunty, kernel 2.6.28), AMD PhenomII X4. Virtualization is enabled in the Bios.

[f.hoefling]

Log file

[f.hoefling]

Log file

[f.hoefling]

the relevant error message might be this:

fatal error in recompiler cpu: triple fault

[Sander van Leeuwen]

Right and I happened to be working on a similar bug. Fixed now.

[f.hoefling]

Log file for version 3.0.8 with VT-x/AMD-V enabled

[f.hoefling]

The bug persists in version 3.0.8.

I've updated to virtualbox-3.0_3.0.8-53138_Ubuntu_jaunty, enabled the VT-x/AMD-V option, and Windows still crashes.

[Sander van Leeuwen]

Which service pack level of Windows XP? Any unusual programs installed (virus scanners)?

[f.hoefling]

Thanks for the quick reply. I'm running Windows XP Prof. SP3. The crash occurs during the boot procedure, presumably when Windows switches to graphics mode for the first time. (When the Windows logo appears.) Before, I can select the hardware configuration (text mode). I'm booting from a raw disk (AHCI) - not sure, whether this makes a difference.

My graphics card is nVidia GeForce 8500 GT. The VM has 48 MB graphics memory and 3D acceleration enabled.

For comparison, I have attached the log-file without the VT-x/AMD-V option.

[f.hoefling]

Log file for version 3.0.8 with VT-x/AMD-V disabled

[Sander van Leeuwen]

Did you install SATA drivers in the VM?

[f.hoefling]

The disks are running in SATA mode, and I've installed SATA drivers in the native boot configuration of Windows.

In the VM, I can't find that SATA drivers are installed. But there is an unconfigured PCI device: 0x2829 AHCI Controller 0x8086 Intel Corporation

[f.hoefling]

I've updated the Virtualbox GuestAdditions, but nothing has changed.

I've installed the driver for the Intel AHCI controller, no change - it still crashes with virtualization enabled.

[Carles]

Confirm the situation in Virtualbox 3.0.8 and earlier. If enabled VT-x/AMD-V on a disk with two installations, windows xp professional sp3 and Linux, when Windows starts we get the following error: fatal error in recompiler cpu: triple fault

If not enabled the VT-x/AMD-V, windows xp working properly. This error occurs both with a virtual IDE and SATA disk. I think the system can not access the NTFS partition of wind, and therefore, load the NT loader.

Excuse my poor English

Thanks

[Carles]

Log "fatal error in recompiler cpu"

[Carles]

Full log in .virtualbox

[kaatmanduu]

I was experiencing the same issue, XP would crash with VT-x/AMD-V enabled, and boot without it. I installed the Intel SATA drivers from ​http://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&DwnldID=17882&lang=eng, and I can boot with VT-x/AMD-V enabled. I now have a multi-processor enabled guest, very nice!

[f.hoefling]

Replying to kaatmanduu:

I was experiencing the same issue, XP would crash with VT-x/AMD-V enabled, and boot without it. I installed the Intel SATA drivers from ​http://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&DwnldID=17882&lang=eng, and I can boot with VT-x/AMD-V enabled. I now have a multi-processor enabled guest, very nice!

I just wanted to confirm that the Intel driver (version 8.9.0.1023) doesn't solve the problem for me - but big surprise! Now it works for me too, even with 2 processors enabled :-) But I don't know why.

I'd installed the Intel driver already two weeks ago, and it kept crashing. (See my previous post.) The only thing I've changed since is that Windows boots directly with its standard boot loader, not via Grub. (I'm using raw disk access with "virtual" MBR.)

[Carles]

Replying to kaatmanduu:

From the Windows MBR bootloader starts correctly with VT-X enabled or not enabled. With GRUB starts well without VT-X enabled, with VT-X enabled does not start. Linux works always with or without VT-X from GRUB.

Greetings

[kaatmanduu]

Replying to carlesgs:

From the Windows MBR bootloader starts correctly with VT-X enabled or not enabled. With GRUB starts well without VT-X enabled, with VT-X enabled does not start.

I'd changed to use the native Windows boot loader a few days before I installed the drivers. It never occurred to me that this might have an affect on a crash that late in the boot process. I booted using grub, and XP indeed crashes again when AMD-V is enabled.

[Carles]

I confirm that this error occurs in the new version 3.0.10

[Carles]

This error occurs in grub, grub2 and lilo, confirmed and tested

[Carles]

Even if the computer does not have a processor with hardware virtualization, grub can not boot windows xp with enabled VT-x extension. If this extension is not enabled, it starts correctly.

Curious ...

[Markus Schaber]

I can also reproduce that problem here in 3.10 - it also happened with the 3.0.6 OSE edition coming with debian. I can assist in debugging (provide logfiles, doing diagnostics etc.) if wanted.

As non-virtualized booting is unconsiderably slow, I'll try to find out infos about the direct boot (virtual mbr) mentioned here.

[Carles]

The error also occurs in version 3.1.0 beta

[Carles]

Solution "temporary" to the malfunctioning of grub and VT-X

Situation: 1 disk virtualbox ide or sata, with 2 operating systems installed: Windows and Linux. Grub, Grub 2 and Lilo not function properly if you enable VT-X on the virtual machine, Windows does not start with VT-x enabled.

Solution: Modify boot.ini in windows to start linux. I used Wingrub (​http://sourceforge.net/projects/grub4dos/files/) to create the grub file to use boot.ini . Wingrub has a graphical interface, and options are virtually identical to the grub menu.lst file in linux, indicate: kernel, partition and initrd. Finally, modify boot.ini to start linux, example:

[boot loader] timeout=-1 default=multi(0)disk(0)rdisk(0)partition(1)\WINDOWS [operating systems] multi(0)disk(0)rdisk(0)partition(1)\WINDOWS="Microsoft Windows" /noexecute=optin /fastdetect C:\GRLDR="my linux"

Note: The binary file created with wingrub must be located in c: , and is called by the entry GRLDR in boot.ini .

Excuse my English

[Carles]

Sorry, boot.ini has to be well

[boot loader] timeout=-1 default=multi(0)disk(0)rdisk(0)partition(1)\WINDOWS [operating systems] multi(0)disk(0)rdisk(0)partition(1)\WINDOWS="Microsoft Windows" /noexecute=optin /fastdetect
C:\GRLDR="my linux" <---new line

[Markus Schaber]

It still happens with 3.1.0-55467_Debian_lenny

[Frank Mehnert]

Well, if the VM crashes (unexpected process termination), then please try to generate a ​core dump and contact me via frank _dot_ mehnert _at_ sun _dot_ com, then I can provide you a server for uploading the file.

[Frank Mehnert]

schabi, thanks for the core dump. Actually this was a guru mediation and therefore no real process crash (segfault or something like that). Therefore the core dump is not helpful in that case, sorry. Just another (unsolved) fatal error in recompiler cpu: triple fault problem ...

[Markus Schaber]

Anything I can do do help debugging?

[Sander van Leeuwen]

#6017 has been marked as a duplicate.

[Sander van Leeuwen]

Still unable to reproduce. Windows XP SP3 with SMP enabled boots fine here. The instruction that causes the problem in the attached logs is executed properly here.

Access to a disk image that always causes the guru meditation might be helpful.

[Markus Schaber]

In my case, it is not a disk image, but a native partition which is booted. I don't know whether I'm in the position to provide you with that native partition, as it contains some copyrighted and some private contents.

I tried to create a dual-boot scenario (booting the same windows installation both natively and in VirtualBox), and while I know that this is not officially supported, a guru meditation is not what I expect :-)

[Markus Schaber]

Maybe one relevant fact is that my Host is a 64 bit lenny version from download.virtualbox.org on a debian testing system on a Intel(R) Core(TM)2 Duo CPU, while the guest is 32 bit.

There is no download version for debian testing, and the one for unstable is broken.

If I can do anything help you tracing down the problem, please tell me. A packed dump of my windows partition is about 25 gig, but as it might be that the bootloader is also part of the problem, and I don't know whether I'm allowed to give you the windows partition (it contains copyright protected software), I don't know whether that will suffice. But I also could run some debug version of virtualbox, or create some trace files, etc...

[Sander van Leeuwen]

schabi: 25 GB is rather large and then there's the matter of privacy and copyright, so that's not really an option.

Could you also attach a log file of a session that crashed?

[Frank Mehnert]

Replying to schabi:

There is no download version for debian testing, and the one for unstable is broken.

I'm sorry? There are packages for Etch and Lenny, both should work. For Debian/testing and Debian/unstable use the corresponding Lenny package of course.

[Markus Schaber]

Hi,

Replying to sandervl73:

schabi: 25 GB is rather large and then there's the matter of privacy and copyright, so that's not really an option.

Could you also attach a log file of a session that crashed?

I was in contact with Frank, and uploaded all the log files he wanted. His Conclusion was that it was no crash, but a Guru Meditation - which reminds me of my good, old Amiga times long ago, but seemingly does not help.

[Markus Schaber]

Replying to frank:

Replying to schabi:

There is no download version for debian testing, and the one for unstable is broken.

I'm sorry? There are packages for Etch and Lenny, both should work. For Debian/testing and Debian/unstable use the corresponding Lenny package of course.

Sorry, I meant that the download for unstable does not work (at least not with debian testing from today): sarge (aka unstable) as mentioned on ​http://www.alldomusa.eu.org/wiki/Linux_Downloads gives the following error: W: Failed to fetch ​http://download.virtualbox.org/virtualbox/debian/dists/sarge/Release Unable to find expected entry non-free/binary-amd64/Packages in Meta-index file (malformed Release file?)

E: Some index files failed to download, they have been ignored, or old ones used instead.

The lenny one is the one I use: 3.1.2-56127_Debian_lenny. Debian testing is squeeze which is not available as a repository. And I thought that running a lenny version on squeeze might be the reason for the bug.

[Frank Mehnert]

schabi, Debian/unstable is sid not sarge. And no, running lenny on squeeze or sid should be fine.

[Frank Mehnert]

Oh, and sarge is Debian 3.1 so outdated.

[Markus Schaber]

Yes, you're right, sorry. But you might consider removing the sarge link from your download page then.

[Frank Mehnert]

Could you check if VBox 3.1.6 changes anything?

[Markus Schaber]

I just tried, and it worked - I booted into the login screen, and after login the activation message came, no Guru mediation any more.

I cannot guarantee that there were no other relevant changes (GRUB updates or windows updates) in the meantime, but the problem is fixed for me.

Thanks!

[Technologov]

should be CLOSED as FIXED

-Technologov