Mac OSX 10.15.2 guest freezes on startup

[Daniel Bonner]

Mac OSX 10.15.2 as a guest freezes when you try to start it or complete an upgrade from a previous version (combo update or app store update) using VirtualBox 6.1.0 or 6.1.1. Mac OSX 10.15.1 works well as a guest with VirtualBox 6.1.0 (release build) or 6.1.1 (development build). Would it be possible to implement compatibility for Mac OSX 10.15.2 guests in VirtualBox? Many thanks, Dan

[Vietwoojagig2]

VirtualBox 6.1.2 does not solve the problem.

[aeichner]

We need at least a VBox.log from the affected VM to start looking into this issue.

[Jaanus]

I attached a number of logs and screenshots. This bug is fully reproducible for me using this script: ​https://github.com/AlexanderWillner/runMacOSinVirtualBox I attempt a clean install with the script, and end up with the result described in this bug. My related report in Github: ​https://github.com/AlexanderWillner/runMacOSinVirtualBox/issues/95

[aeichner]

We need a VBox.log of the affected VM. You attached the VBoxSVC.log which is of no use for this particular issue unfortunately. The VBox.log can be usually found in ~/VirtualBoxVMs/<VM name>/Logs.

[Jack_Smith]

This appears to be reproducible in all versions of VirtualBox on all macOS 10.15.2 guests. A workaround involves using the "boot.efi" file from an earlier version of macOS.

[Jaanus]

Attached VBox.log files for the affected VM.

[Vietwoojagig2]

Tried to update from 10.15.1 to 10.15.3 in 6.1.12

[Vietwoojagig2]

Also does not work with to update 10.15.3 from 10.15.1 in 6.1.2

[michaln]

The source of the boot hangs has been identified, although fixing it correctly isn't entirely trivial. It's triggered by a very, very old problem in VirtualBox. Apple switches to 32-bit mode (i.e. out of long mode) with junk in the high byte of RDI register and expects that when it switches back to long mode, the junk will be zeroed out because the EDI register was written in 32-bit mode. That is relying on processor behavior which is not documented, or rather behavior that is explicitly undocumented. Neither Intel nor AMD say what should happen in this case, although the behavior of actual CPUs appears to be consistent.

At any rate it will be fixed for the next maintenance release.

[Vietwoojagig2]

Was able to update to 10.15.3 with 6.1.4. It seems they solved the problem.

[aeichner]

Indeed, this was fixed for 6.1.4, thanks for confirming!