WIndows 7 64 bits guests crash on 32 bits Linux host with VirtualBox v5.0

[dinosaur0]

I just upgraded to VirtualBox v5.0 today (I previously tried v5.0RC1 and v5.0RC2 and did not run into any issue), from VirtualBox v4.3.28.

After upgrade (including VBox additions upgrade) I got *systematic* BSOD when attempting to run a Windows 7 64 bits guest from my Linux 32 bits host (the 64 bits guest is running fine when launched from a 64 bits Linux host). I thirst thought it was because of the old VBox additions (v4.3.28) in the 64 bits guest, so I upgraded these while running the guest on the 64 bits Linux host, but even after the upgrade to additions v5.0 in the guest, the latter kept crashing on boot when launched from the 32 bits host.

I'm attaching 3 screen shots of the BSODs: there doesn't seem to be any common crash point (the VM simply crashes at a seemingly random time during the boot process).

For now, I reverted to VirtualBox v4.3.28 which runs all my guests systems (32 or 64 bits, Windows or Linux) just fine on 32 bits Linux hosts...

[dinosaur0]

BSOD n°1

[dinosaur0]

BSOD n°2

[dinosaur0]

BSOD n°3

[andi77]

maybe this is the same problem (64bit guest with 32bit host): ​https://forums.virtualbox.org/viewtopic.php?f=7&t=68859

[Neurones67]

Having the same problem...

I could run the VM by disabling nested paging in the Acceleration tab in the VM configuration.

Hope that helps :)

[Frank Mehnert]

Could you add VBox.log files?

[Artem S. Tashkinov]

Replying to frank:

Could you add VBox.log files?

I have added a log file to ticket #14263.

[dinosaur0]

@Frank

New screen shots of BSODs (Crash1.png to Crash3.png) added, together with their respective VBox.log files (Crash1.log to Crash3.log).

@Neurones67

Disabling nested paging does not entirely solve the issue for me: while the VM doesn't crash any more, it stays stuck at the "Please wait..." login screen.

@Birdie

Ticket #14263 is totally unrelated... Not the same bug at all, obviously, since the symptoms are entirely different... I therefore don't see why you are commenting about 14263's log in this ticket...

For now, I'm using 4.3.30, which works just fine. v5.0.0 is totally unusable as far as I am concerned...

[Frank Mehnert]

dinosaur0, please could you try again using the latest 5.0 test build from here? There is a known problem with 64-bit guests on 32-bit hosts if AVX is passed to the guest, therefore the link to #14263. The latest builds on the test builds page have a workaround for tihs (AVX is explicitly disabled in that case).

Does that build work for you?

[dinosaur0]

Replying to frank:

dinosaur0, please could you try again using the latest 5.0 test build from here? There is a known problem with 64-bit guests on 32-bit hosts if AVX is passed to the guest, therefore the link to #14263. The latest builds on the test builds page have a workaround for tihs (AVX is explicitly disabled in that case).

Does that build work for you?

Yep, it looks like it works ! :-)

Strange however, that AVX instructions get called at all at boot time... Aren't drivers supposed to work even on an old i486 without even MMX ?...

[Artem S. Tashkinov]

Replying to dinosaur0:

Yep, it looks like it works ! :-)

Strange however, that AVX instructions get called at all at boot time... Aren't drivers supposed to work even on an old i486 without even MMX ?...

I suspect Windows uses AVX instructions to speed up, for instance, hashing algorithms which are used to verify drivers signatures.

[StarboardBow]

Changelog for VBox 5.0.2 includes following bullet-item:

VMM: This release has AVX2 passthrough disabled on every host and AVX passthrough disabled for 64-bit VMs on 32-bit hosts. This will be properly fixed in a future 5.0.x maintenance release (see e.g. bug #14262).

Changelogs for subsequent VBox 5.0.4 and 5.0.6 do not mention fix to bug #14262 -- assume bug fix is still pending, correct?

Is there any planned release and timeline available for issuing bug fix?

[anotherbugmaster]

Fix it, please

​John the Ripper requires this feature. Not so convenient to crack your o5logon without it. :D

[Frank Mehnert]

We are still looking at a simple testcase to reproduce this problem.

[Willard Dawson]

It's easy enough to comment out -DAVX2 and -mavx... from ​http://www.openwall.com/john/ Makefile once configured before compiling, but it is annoying to have to do so. +1 from me on this request, I hope progress can be made.

[Frank Mehnert]

No immediate fix planned, too many other things to do. If you insist on using AVX with a 64-bit guest, use a 64-bit host operating system. 64-bit guests on 32-bit VirtualBox hosts are supposed to work but are less tested.