MacOSX 10.8 Guest crashes after EFI Boot with "Divide" error after upgrading from 4.3.6 to 4.3.8 [FIXED in SVN]

[ThetaPhi]

I know, those requests are closed quite often with "running MacOSX guests is not supported", but this is a new regression with a guest that was running before the upgrade.

See the attached screenshot of the guest window. After failing. The OSX guest is booting with EFI and is an unmodified OSX/Darwin installation, so definitely no Hackintosh! The "Divide error" occurs directly after EFI handling over to Darwin, no additional log output.

The strange thing is: This only happens on one host CPU type (Intel(R) Xeon(R) CPU X3440 @ 2.53GHz), on my other computer (Intel(R) Core(TM) i7-3610QM CPU @ 2.30GHz) the guest starts after upgrading 4.3.6 to 4.3.8 like it did before.

I also attached the VBox log of the failing guest.

This could be related to ticket #8651 - the failing CPU has no SSE4 features, but the working one has. But I did not enable this experimental feature! It is the unmodified guest VM reverted to snapshot that was running before.

After downgrading virtualbox to 4.3.6, it is booting again.

[ThetaPhi]

Failure in guest window (via RDP).

[ThetaPhi]

log file, showing no errors, but gives more details on the failing guest VM

[ThetaPhi]

Sorry, the failing CPU *has* SSE4 features, but those are not eanbled in the guest. See log file.

[Frank Mehnert]

For SSE 4.1/4.2 see here. You have to enable it manually for each VM.

[bird]

I've reproduced it here, I don't think it's SSE 4.x related, I believe it's some issue related to CPUID and/or MSR changes that went into 4.3.8. Stay tuned...

[ThetaPhi]

Thanks. If you have any news or a fix, I can do a test if you have a .deb package for Ubuntu Precise 64bits. This was a quick reply, many thanks!

Greetings also to the Oracle Java team, this is the virtual machine doing the Apache Lucene/Solr CI builds on MacOSX running a Jenkins Slave!

[bird]

Apple was reading MSR 0x35 to get core & thread counts, we weren't giving it the right values. Fix committed.

Until we publish a build containing fix, the issue can be worked around like this:

VBoxManage setextradata <vmname> VBoxInternal/CPUM/MSRs/MsrCoreThreadCount/First 0x00000035

[ThetaPhi]

Hi, the work around seems to work. I updated to latest version and applied this extradata key to the VM configuration. It starts up successfully.

[Socratis]

THANK YOU !!! I couldn't use 4.3.8 because of this. I waited a couple of days, to see if anyone else had that issue, before filing a bug report, but ThetaPhi was ahead of me. 10.5, 10.6, 10.7, 10.8 and 10.9 clients and from all of them only 10.6 was booting. Now all of them do, except 10.5, but that I believe is another bug report that I have to open.

[Wikicoda]

Where do i add the below command?

VBoxManage setextradata <vmname> VBoxInternal/CPUM/MSRs/MsrCoreThreadCount/First 0x00000035