[Win7 w/o SP1] VMM: openSUSE 13.2 boot failed on Windows

[frank]

Could be very well related to the new AVX instruction set. AVX is not passed to the guest if nested paging is not enabled for this VM. To collect more information for reproducing: You said that re-installing openSUSE 13.2 crashes the kernel during boot. So you installed the guest from DVD then rebooted and the guest kernel crashed quite immediately during boot, is that correct? Also, are you able to boot the very same VM image with VBox 5.0.0 BETA2 or older?

[Technologov]

It is fully 100% reproducible on my host.
Either booting ready openSUSE 13.2 VM (installed with BETA2), or during installation from DVD it fails on first stage Yast, before X11 Yast, when loading hard disk driver or file system driver (better fs), causes guest kernel crash.
Degrade to BETA2 fixes it, -or- disabling Intel EPT fixes it also. (RC1/RC2)

I have not tried:
-upgrading CPU firmware on Windows 7 (host)
-disabling better fs driver in guest

[loukingjr]

OpenSUSE 13.2 still boots without a problem for me on a Windows 8.1 host - VB 5.0 RC2.

[Technologov]

Same problem on this host with Fedora 22 x64 guest.
Disabling Intel EPT solves this.

[michaln]

For the record, no problem with openSUSE 13.2 on my Core i7-2600 and Win7 x64 host. Guest sees and uses AVX.

[Technologov]

Then there's only one logical explanation: I have an earlier CPU revision / firmware. Which might have some dumb bugs with AVX and VT-x combined, that I only experience on Windows host for whatever reason.
I can't reproduce this on my dad's Core i3 2100 (Sandy Bridge) either, on Win 7 x64. Meh.

[loukingjr]

FWIW, I have no problem with OpenSUSE 13.2 on a Windows 8.1 host.

[michaln]

Then there's only one logical explanation: I have an earlier CPU revision / firmware. Which might have some dumb bugs with AVX and VT-x combined, that I only experience on Windows host for whatever reason.

How about a different explanation... the VBox.log you posted suggests that you're running Windows 7 without SP1 on the affected system. Is that really true?

[Technologov]

Yes, true.
My security is based on newest browser (Chrome) + Firewall + careful thinking of what I download. Without OS updates. In this scenario, even Windows 2000 is secure. (with modern browser)
BTW: if so, then Windows Vista might be affected also. (but very few machines in the entire world, that have AVX, and running Vista)

Anyway, how can Windows 7 affect AVX ? AVX was invented *after* Windows 7. Does Win 7 kernel forgets to clear some CPU registers on VM exit / process switch / context switch ?

UPDATE: changed bug to INVALID.

[michaln]

Now you just need to check that all your working Win7 hosts do in fact have SP1... but this is almost certainly the problem. AVX support was officially introduced in SP1, so it shouldn't work (shouldn't be exposed to a VM), but who knows what really happens.

Anyway, thanks for confirming that this is not an important problem.

[Technologov]

Thanks-a-lot !
It works now ! Both openSUSE 13.2 and Fedora 22 x64 boot now !
Hard for me to imagine, that a slightly older host kernel can break VT-x in such a subtle way.
CPU is doing well, VT-x, Intel EPT, Unrestricted VMX, AVX, and even KVM paravirt works !

[bird]

Got a fix for this in SVN.

Problem was that W7 w/o SP1 enabled XSAVE but not for the AVX state. So, we were giving the guest mixed signals; AVX and AVX-2 instruction available in CPUID(1), but CPUID(0xd) indicated that there is not way of enabling them.

Cheers,
bird.

[Technologov]

Thanks !
Will a similar problem re-appear again with Win 7 (SP0 or SP1) and AVX-512 (ZMM) in your opinion ?
(It is likely, that next year I will buy some Intel Cannonlake PC and still stay on Windows 7, as will many Enterprises.)