Ubuntu 10.04 64-bit host cannot install fresh Win7Pro 64-bit guest

[Adam Bolte]

Steps to reproduce:

1. Install Ubuntu 10.04 64-bit.
2. Install VirtualBox 3.2 from the official repository.
3. Create a new VM from scratch. Use all defaults, except select "Windows 7 (64 bit)" as the guest. I also increased my dynamic disk to 120 GB instead of 20 GB, but don't expect that would matter. Install the guest from either an ISO or a DVD image (I tried both ways).
4. Go through the installation selecting the defaults.

As soon as you get to the section "Installing features", (or thereabouts) you will always get the error "Windows could not set the offline locale information 0x80FF0000". The installation will fail.

I'm running on an LGA1366 i7 with 6Gb of RAM and hundreds of GB free space on the host.

The above procedure works in VirtualBox 3.1, so this would be a regression.

[Adam Bolte]

VBoxManage showvminfo "Windows 7 Professional (64 bit)" --details --machinereadable > ~/win7pro-64.log

[Technologov]

Duplicate of bug #6742

boltronics: disable "Nested Paging" until fix is released. (VM settings->System->Acceleration)

-Technologov

[Adam Bolte]

Strange... after installing successfully on 3.1, I upgraded again to 3.2 and launched the same VM. Nested paging was on by default... and it boots fine.

So I only need to disable Nested Paging for the installation, and then I can re-enable it?

[Adam Bolte]

Confirming that this is not a duplicate.

Upgraded to VirtualBox 3.2.2, performed the same steps as above and it failed (so 3.2.2 didn't fix the issue, as expected anyway from #6742).

Then I started re-installing Windows 7 again in the same VM, but with Nested Paging turned off. This option had no effect, and the problem remained.

The only work-around at this point is to install using 3.1. Once installed, Windows runs fine in 3.2 or 3.2.2 (AFAICT). If this is the same bug as #6742, why do the machines run normally after installation?

[Technologov]

Then please provide vbox.log (you have provided only VM details log).

-Technologov

[Frank Mehnert]

It is also possible that the guest runs at normal speed after the guest additions are active.

[Adam Bolte]

Hi Frank. Not sure exactly what you mean, as I'm not concerned at all with performance (in this bug report, at least). I just needed to install Windows 7 64-bit (which I achieved by downgrading, installing and then upgrading).

Sorry I haven't had a chance to redo the installation and generate the log. I note that 3.2.4 was just released too so I'll test with that when I can. For all I know it's already fixed.

[Mikael Andersson]

Installation of 64bit Windows 7 Home Premium

[Mikael Andersson]

I have a similar problem with 64bit Windows 7 Home Premium. However, this problem seems to be related to SATA storage. If I manually move the hard disk to IDE controller, the installation seems to proceed further.

[Adam Bolte]

Original reporter's vbox.log file.

[Frank Mehnert]

boltronics, is your VM disk image located on an ext4 partition? If so, then this is a known issue (ext4 bug). And enabling the host cache keeping it attached to the SATA controller would fix the problem as well.

[Adam Bolte]

Interesting. It could very well be. I'll check it out tomorrow, and test your work around. Thanks.

[Mikael Andersson]

Confirmed: Windows 7 Home Premium 64bit was successfully installed using SATA controller with host cache enabled.

[Adam Bolte]

Confirming here too. I was using ext4 and enabling host I/O cache on the SATA controller allowed the installation to complete. Good analysis.

Mind linking the ext4 bug report here? It wasn't clear to me which ext4 bug it was.

An incompatibility with ext4 has pretty big implications, as I imagine installing VB on Ubuntu (which uses ext4 by default) is quite common. I also wonder what other guest OSs would see problems due to the host using ext4.

I see now why my Win7 install succeeded in VB3.1 - it configured the guest with the PIIX4 IDE controller by default. In the next VB release, can we detect a known-broken ext4 version and disable the host I/O cache checkbox to ensure it cannot be unticked (or pop-up a warning or some such)? Alternatively/in addition, perhaps a broken ext4 install should default to using the old IDE controller instead?

Even if this ext4 bug was fixed today and a new version of ext4 released, would all popular distributions immediately backport patches to fix it (or provide an update to the latest version)? If not, perhaps the most user-acceptable solution would be to implement one of those [horrible, I agree] workarounds?

[Frank Mehnert]

VBox 3.2.6 contains a workaround. It will warn the user and automatically enable the host I/O cache if the disk image is located on an ext4 partition or an xfs partition. Regarding the kernel bug: There are already patches available but we still have to evaluate them. And surely it will still take some time until the bug is finally fixed in the kernel.

[Adam Bolte]

Thanks Frank. Does the job nicely.

[Frank Mehnert]

Thanks for the feedback!